[llvm] [X86] Fold BLEND(PERMUTE(X), PERMUTE(Y)) -> PERMUTE(BLEND(X, Y)) (PR #90219)

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Mon Apr 29 06:22:35 PDT 2024


https://github.com/RKSimon updated https://github.com/llvm/llvm-project/pull/90219

>From c92980c2036ccc98acd88bae11456a2caeb3572d Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev at redking.me.uk>
Date: Fri, 26 Apr 2024 15:44:59 +0100
Subject: [PATCH] [X86] Fold BLEND(PERMUTE(X),PERMUTE(Y)) ->
 PERMUTE(BLEND(X,Y))

If we don't demand the same element from both single source shuffles (permutes), then attempt to blend the sources together first and then perform a merged permute.

We don't attempt this for vXi16 blends as these are much more likely to involve byte/word vector shuffles that will result in the creation of a lot more variable-mask shuffles (PSHUFB etc.).

This fold might be worth it for VSELECT with constant masks on AVX512 targets, but I haven't investigated this yet.

The PR34592 -O0 regression is an unfortunate failure to cleanup with a later SimplifyDemandedElts pass like the -O3 does - I'm not sure how worried we should be tbh.
---
 llvm/lib/Target/X86/X86ISelLowering.cpp       |   55 +
 llvm/test/CodeGen/X86/horizontal-sum.ll       |   20 +-
 llvm/test/CodeGen/X86/oddshuffles.ll          |   76 +-
 llvm/test/CodeGen/X86/pr34592.ll              |   46 +-
 .../vector-interleaved-load-i32-stride-3.ll   | 1847 +++--
 .../vector-interleaved-load-i32-stride-5.ll   | 3672 +++++----
 .../vector-interleaved-load-i32-stride-6.ll   | 6296 ++++++++-------
 .../vector-interleaved-load-i32-stride-7.ll   | 6791 ++++++++---------
 .../vector-interleaved-store-i16-stride-8.ll  | 2568 +++----
 .../vector-interleaved-store-i32-stride-6.ll  |    5 +-
 .../vector-interleaved-store-i32-stride-7.ll  |   20 +-
 .../X86/vector-shuffle-combining-avx.ll       |    6 +-
 .../X86/vector-shuffle-combining-sse41.ll     |   11 +-
 13 files changed, 10045 insertions(+), 11368 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 3dca306a7642c9..86dbdf1b9b3f47 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -41754,6 +41754,61 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
     KnownUndef = SrcUndef.zextOrTrunc(NumElts);
     break;
   }
+  case X86ISD::BLENDI: {
+    SDValue N0 = Op.getOperand(0);
+    SDValue N1 = Op.getOperand(1);
+    if (VT.getScalarSizeInBits() < 32 || !N0.hasOneUse() || !N1.hasOneUse())
+      break;
+
+    // Attempt to fold BLEND(PERMUTE(X),PERMUTE(Y)) -> PERMUTE(BLEND(X,Y))
+    // iff we don't demand the same element index for both X and Y.
+    SDValue BC0 = peekThroughOneUseBitcasts(N0);
+    SDValue BC1 = peekThroughOneUseBitcasts(N1);
+    SmallVector<SDValue, 2> Ops, Ops0, Ops1;
+    SmallVector<int, 32> Mask, Mask0, Mask1, ScaledMask0, ScaledMask1;
+    if (!getTargetShuffleMask(Op, false, Ops, Mask) ||
+        !getTargetShuffleMask(BC0, false, Ops0, Mask0) ||
+        !getTargetShuffleMask(BC1, false, Ops1, Mask1) ||
+        !scaleShuffleElements(Mask0, NumElts, ScaledMask0) ||
+        !scaleShuffleElements(Mask1, NumElts, ScaledMask1))
+      break;
+
+    // Determine the demanded elts from both permutes, confirm that we only use
+    // a single operand and that we don't demand the same index from both.
+    APInt Demanded0, DemandedLHS0, DemandedRHS0;
+    APInt Demanded1, DemandedLHS1, DemandedRHS1;
+    if (getShuffleDemandedElts(NumElts, Mask, DemandedElts, Demanded0,
+                               Demanded1, /*AllowUndefElts=*/true) &&
+        getShuffleDemandedElts(NumElts, ScaledMask0, Demanded0, DemandedLHS0,
+                               DemandedRHS0, /*AllowUndefElts=*/true) &&
+        getShuffleDemandedElts(NumElts, ScaledMask1, Demanded1, DemandedLHS1,
+                               DemandedRHS1, /*AllowUndefElts=*/true) &&
+        DemandedRHS0.isZero() && DemandedRHS1.isZero() &&
+        !DemandedLHS0.intersects(DemandedLHS1)) {
+      // Use the permute demanded elts masks as the new blend mask.
+      uint64_t NewBlendMask = DemandedLHS1.getZExtValue();
+
+      // Create the new permute mask as a blend of the 2 original permute masks.
+      SmallVector<int, 32> NewPermuteMask(NumElts, SM_SentinelUndef);
+      for (int I = 0; I != NumElts; ++I) {
+        if (Demanded0[I])
+          NewPermuteMask[I] = ScaledMask0[I];
+        else if (Demanded1[I])
+          NewPermuteMask[I] = ScaledMask1[I];
+      }
+      assert(isUndefOrInRange(NewPermuteMask, 0, NumElts) && "Bad permute");
+
+      SDLoc DL(Op);
+      SDValue NewBlend = TLO.DAG.getNode(
+          X86ISD::BLENDI, DL, VT, TLO.DAG.getBitcast(VT, Ops0[0]),
+          TLO.DAG.getBitcast(VT, Ops1[0]),
+          TLO.DAG.getTargetConstant(NewBlendMask, DL, MVT::i8));
+      return TLO.CombineTo(Op, TLO.DAG.getVectorShuffle(VT, DL, NewBlend,
+                                                        TLO.DAG.getUNDEF(VT),
+                                                        NewPermuteMask));
+    }
+    break;
+  }
   case X86ISD::BLENDV: {
     APInt SelUndef, SelZero;
     if (SimplifyDemandedVectorElts(Op.getOperand(0), DemandedElts, SelUndef,
diff --git a/llvm/test/CodeGen/X86/horizontal-sum.ll b/llvm/test/CodeGen/X86/horizontal-sum.ll
index a026757a0264d6..5fe1e2996ee9b0 100644
--- a/llvm/test/CodeGen/X86/horizontal-sum.ll
+++ b/llvm/test/CodeGen/X86/horizontal-sum.ll
@@ -679,9 +679,8 @@ define <4 x i32> @sequential_sum_v4i32_v4i32(<4 x i32> %0, <4 x i32> %1, <4 x i3
 ; AVX1-SLOW-NEXT:    vphaddd %xmm1, %xmm0, %xmm4
 ; AVX1-SLOW-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
 ; AVX1-SLOW-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX1-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
-; AVX1-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX1-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
+; AVX1-SLOW-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX1-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
 ; AVX1-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[1,1,1,1]
 ; AVX1-SLOW-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
 ; AVX1-SLOW-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm4[0],xmm1[0]
@@ -704,9 +703,8 @@ define <4 x i32> @sequential_sum_v4i32_v4i32(<4 x i32> %0, <4 x i32> %1, <4 x i3
 ; AVX1-FAST-NEXT:    vphaddd %xmm1, %xmm0, %xmm4
 ; AVX1-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
 ; AVX1-FAST-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX1-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
-; AVX1-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX1-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5,6,7]
+; AVX1-FAST-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX1-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
 ; AVX1-FAST-NEXT:    vphaddd %xmm2, %xmm2, %xmm1
 ; AVX1-FAST-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm4[0],xmm1[0]
 ; AVX1-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm2[3,3,3,3]
@@ -727,9 +725,8 @@ define <4 x i32> @sequential_sum_v4i32_v4i32(<4 x i32> %0, <4 x i32> %1, <4 x i3
 ; AVX2-SLOW-NEXT:    vphaddd %xmm1, %xmm0, %xmm4
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
+; AVX2-SLOW-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[1,1,1,1]
 ; AVX2-SLOW-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
 ; AVX2-SLOW-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm4[0],xmm1[0]
@@ -752,9 +749,8 @@ define <4 x i32> @sequential_sum_v4i32_v4i32(<4 x i32> %0, <4 x i32> %1, <4 x i3
 ; AVX2-FAST-NEXT:    vphaddd %xmm1, %xmm0, %xmm4
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
 ; AVX2-FAST-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
+; AVX2-FAST-NEXT:    vpunpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
 ; AVX2-FAST-NEXT:    vphaddd %xmm2, %xmm2, %xmm1
 ; AVX2-FAST-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm4[0],xmm1[0]
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm2[3,3,3,3]
diff --git a/llvm/test/CodeGen/X86/oddshuffles.ll b/llvm/test/CodeGen/X86/oddshuffles.ll
index 01056a8b2c24a9..44f965d1c6e84e 100644
--- a/llvm/test/CodeGen/X86/oddshuffles.ll
+++ b/llvm/test/CodeGen/X86/oddshuffles.ll
@@ -1450,19 +1450,17 @@ define void @interleave_24i32_out(ptr %p, ptr %q1, ptr %q2, ptr %q3) nounwind {
 ; SSE42-NEXT:    pshufd {{.*#+}} xmm9 = xmm5[2,3,2,3]
 ; SSE42-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,3],xmm1[2,3]
 ; SSE42-NEXT:    insertps {{.*#+}} xmm5 = xmm5[0,1,2],xmm0[1]
-; SSE42-NEXT:    pshufd {{.*#+}} xmm10 = xmm3[2,2,2,2]
-; SSE42-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[1,0,3,3]
-; SSE42-NEXT:    pblendw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,5],xmm10[6,7]
-; SSE42-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[1,0,3,3]
-; SSE42-NEXT:    pshufd {{.*#+}} xmm10 = xmm0[2,2,2,2]
-; SSE42-NEXT:    pblendw {{.*#+}} xmm10 = xmm8[0,1,2,3,4,5],xmm10[6,7]
+; SSE42-NEXT:    pblendw {{.*#+}} xmm6 = xmm6[0,1,2,3],xmm3[4,5],xmm6[6,7]
+; SSE42-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[1,0,3,2]
+; SSE42-NEXT:    pblendw {{.*#+}} xmm8 = xmm8[0,1,2,3],xmm0[4,5],xmm8[6,7]
+; SSE42-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[1,0,3,2]
 ; SSE42-NEXT:    pblendw {{.*#+}} xmm7 = xmm7[0,1],xmm2[2,3],xmm7[4,5,6,7]
 ; SSE42-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,1],xmm3[0,3]
 ; SSE42-NEXT:    pblendw {{.*#+}} xmm9 = xmm9[0,1],xmm1[2,3],xmm9[4,5,6,7]
 ; SSE42-NEXT:    shufps {{.*#+}} xmm9 = xmm9[0,1],xmm0[0,3]
 ; SSE42-NEXT:    movups %xmm5, 16(%rsi)
 ; SSE42-NEXT:    movups %xmm4, (%rsi)
-; SSE42-NEXT:    movdqu %xmm10, 16(%rdx)
+; SSE42-NEXT:    movdqu %xmm8, 16(%rdx)
 ; SSE42-NEXT:    movdqu %xmm6, (%rdx)
 ; SSE42-NEXT:    movups %xmm9, 16(%rcx)
 ; SSE42-NEXT:    movups %xmm7, (%rcx)
@@ -1504,19 +1502,14 @@ define void @interleave_24i32_out(ptr %p, ptr %q1, ptr %q2, ptr %q3) nounwind {
 ; AVX2-SLOW-NEXT:    vmovups (%rdi), %ymm0
 ; AVX2-SLOW-NEXT:    vmovups 32(%rdi), %ymm1
 ; AVX2-SLOW-NEXT:    vmovups 64(%rdi), %ymm2
-; AVX2-SLOW-NEXT:    vbroadcastsd {{.*#+}} ymm3 = [2,5,2,5,2,5,2,5]
-; AVX2-SLOW-NEXT:    vpermps %ymm2, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
-; AVX2-SLOW-NEXT:    vmovaps {{.*#+}} ymm5 = [0,3,6,1,4,7,u,u]
+; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
+; AVX2-SLOW-NEXT:    vmovaps {{.*#+}} ymm4 = [0,3,6,1,4,7,2,5]
+; AVX2-SLOW-NEXT:    vpermps %ymm3, %ymm4, %ymm3
+; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0],ymm4[1,2],ymm2[3],ymm4[4,5],ymm2[6],ymm4[7]
+; AVX2-SLOW-NEXT:    vmovaps {{.*#+}} ymm5 = [1,4,7,2,5,0,3,6]
 ; AVX2-SLOW-NEXT:    vpermps %ymm4, %ymm5, %ymm4
-; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-SLOW-NEXT:    vbroadcastf128 {{.*#+}} ymm4 = [0,0,3,6,0,0,3,6]
-; AVX2-SLOW-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX2-SLOW-NEXT:    vpermps %ymm2, %ymm4, %ymm4
-; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
-; AVX2-SLOW-NEXT:    vmovaps {{.*#+}} ymm6 = [1,4,7,2,5,u,u,u]
-; AVX2-SLOW-NEXT:    vpermps %ymm5, %ymm6, %ymm5
-; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm4[5,6,7]
 ; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
 ; AVX2-SLOW-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
 ; AVX2-SLOW-NEXT:    vpermps %ymm0, %ymm1, %ymm0
@@ -1534,26 +1527,18 @@ define void @interleave_24i32_out(ptr %p, ptr %q1, ptr %q2, ptr %q3) nounwind {
 ; AVX2-FAST-ALL-NEXT:    vmovups (%rdi), %ymm0
 ; AVX2-FAST-ALL-NEXT:    vmovups 32(%rdi), %ymm1
 ; AVX2-FAST-ALL-NEXT:    vmovups 64(%rdi), %ymm2
-; AVX2-FAST-ALL-NEXT:    vbroadcastsd {{.*#+}} ymm3 = [2,5,2,5,2,5,2,5]
-; AVX2-FAST-ALL-NEXT:    vpermps %ymm2, %ymm3, %ymm3
-; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
-; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm5 = [0,3,6,1,4,7,u,u]
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm4 = [0,3,6,1,4,7,2,5]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm3, %ymm4, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0],ymm4[1,2],ymm2[3],ymm4[4,5],ymm2[6],ymm4[7]
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm5 = [1,4,7,2,5,0,3,6]
 ; AVX2-FAST-ALL-NEXT:    vpermps %ymm4, %ymm5, %ymm4
-; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FAST-ALL-NEXT:    vbroadcastf128 {{.*#+}} ymm4 = [0,0,3,6,0,0,3,6]
-; AVX2-FAST-ALL-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX2-FAST-ALL-NEXT:    vpermps %ymm2, %ymm4, %ymm4
-; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
-; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm6 = [1,4,7,2,5,u,u,u]
-; AVX2-FAST-ALL-NEXT:    vpermps %ymm5, %ymm6, %ymm5
-; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FAST-ALL-NEXT:    vbroadcastf128 {{.*#+}} ymm5 = [0,1,4,7,0,1,4,7]
-; AVX2-FAST-ALL-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX2-FAST-ALL-NEXT:    vpermps %ymm2, %ymm5, %ymm2
 ; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
-; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2,3],ymm2[4],ymm0[5,6],ymm2[7]
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,1,4,7]
 ; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FAST-ALL-NEXT:    vmovups %ymm3, (%rsi)
 ; AVX2-FAST-ALL-NEXT:    vmovups %ymm4, (%rdx)
 ; AVX2-FAST-ALL-NEXT:    vmovups %ymm0, (%rcx)
@@ -1565,19 +1550,14 @@ define void @interleave_24i32_out(ptr %p, ptr %q1, ptr %q2, ptr %q3) nounwind {
 ; AVX2-FAST-PERLANE-NEXT:    vmovups (%rdi), %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovups 32(%rdi), %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovups 64(%rdi), %ymm2
-; AVX2-FAST-PERLANE-NEXT:    vbroadcastsd {{.*#+}} ymm3 = [2,5,2,5,2,5,2,5]
-; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm2, %ymm3, %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
-; AVX2-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm5 = [0,3,6,1,4,7,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm4 = [0,3,6,1,4,7,2,5]
+; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm3, %ymm4, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0],ymm4[1,2],ymm2[3],ymm4[4,5],ymm2[6],ymm4[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm5 = [1,4,7,2,5,0,3,6]
 ; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm4, %ymm5, %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FAST-PERLANE-NEXT:    vbroadcastf128 {{.*#+}} ymm4 = [0,0,3,6,0,0,3,6]
-; AVX2-FAST-PERLANE-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm2, %ymm4, %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
-; AVX2-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm6 = [1,4,7,2,5,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm5, %ymm6, %ymm5
-; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm4[5,6,7]
 ; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
 ; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm0, %ymm1, %ymm0
diff --git a/llvm/test/CodeGen/X86/pr34592.ll b/llvm/test/CodeGen/X86/pr34592.ll
index 23de746ecb3539..bce060c4c6f56a 100644
--- a/llvm/test/CodeGen/X86/pr34592.ll
+++ b/llvm/test/CodeGen/X86/pr34592.ll
@@ -8,38 +8,40 @@ define <16 x i64> @pluto(<16 x i64> %arg, <16 x i64> %arg1, <16 x i64> %arg2, <1
 ; CHECK-O0-NEXT:    pushq %rbp
 ; CHECK-O0-NEXT:    movq %rsp, %rbp
 ; CHECK-O0-NEXT:    andq $-32, %rsp
-; CHECK-O0-NEXT:    subq $32, %rsp
+; CHECK-O0-NEXT:    subq $64, %rsp
 ; CHECK-O0-NEXT:    vmovaps %ymm4, %ymm10
 ; CHECK-O0-NEXT:    vmovaps %ymm3, %ymm9
+; CHECK-O0-NEXT:    vmovaps %ymm2, (%rsp) # 32-byte Spill
 ; CHECK-O0-NEXT:    vmovaps %ymm1, %ymm8
+; CHECK-O0-NEXT:    vmovaps %ymm0, %ymm3
+; CHECK-O0-NEXT:    vmovaps (%rsp), %ymm0 # 32-byte Reload
 ; CHECK-O0-NEXT:    vmovaps 240(%rbp), %ymm4
-; CHECK-O0-NEXT:    vmovaps 208(%rbp), %ymm3
-; CHECK-O0-NEXT:    vmovaps 176(%rbp), %ymm1
-; CHECK-O0-NEXT:    vmovaps 144(%rbp), %ymm1
+; CHECK-O0-NEXT:    vmovaps 208(%rbp), %ymm1
+; CHECK-O0-NEXT:    vmovaps 176(%rbp), %ymm2
+; CHECK-O0-NEXT:    vmovaps 144(%rbp), %ymm2
 ; CHECK-O0-NEXT:    vmovaps 112(%rbp), %ymm11
 ; CHECK-O0-NEXT:    vmovaps 80(%rbp), %ymm11
 ; CHECK-O0-NEXT:    vmovaps 48(%rbp), %ymm11
 ; CHECK-O0-NEXT:    vmovaps 16(%rbp), %ymm11
-; CHECK-O0-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3,4,5],ymm2[6,7]
-; CHECK-O0-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[2],ymm3[2]
-; CHECK-O0-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,1,3]
+; CHECK-O0-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3,4,5],ymm0[6,7]
+; CHECK-O0-NEXT:    vpunpcklqdq {{.*#+}} ymm2 = ymm2[0],ymm1[0],ymm2[2],ymm1[2]
+; CHECK-O0-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,1,3]
 ; CHECK-O0-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[3,1,2,1]
-; CHECK-O0-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5],ymm0[6,7]
-; CHECK-O0-NEXT:    vperm2i128 {{.*#+}} ymm1 = ymm7[2,3],ymm6[0,1]
-; CHECK-O0-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; CHECK-O0-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1],ymm2[2,3],ymm1[4,5,6,7]
+; CHECK-O0-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3,4,5],ymm0[6,7]
+; CHECK-O0-NEXT:    vperm2i128 {{.*#+}} ymm2 = ymm7[2,3],ymm6[0,1]
+; CHECK-O0-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; CHECK-O0-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2,3],ymm2[4,5,6,7]
+; CHECK-O0-NEXT:    vmovaps %xmm1, %xmm3
+; CHECK-O0-NEXT:    vmovaps %xmm7, %xmm1
+; CHECK-O0-NEXT:    vpblendd {{.*#+}} xmm3 = xmm1[0,1],xmm3[2,3]
+; CHECK-O0-NEXT:    # implicit-def: $ymm1
+; CHECK-O0-NEXT:    vmovaps %xmm3, %xmm1
+; CHECK-O0-NEXT:    vpermq {{.*#+}} ymm3 = ymm1[0,0,1,3]
+; CHECK-O0-NEXT:    vpslldq {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,zero,zero,ymm5[0,1,2,3,4,5,6,7],zero,zero,zero,zero,zero,zero,zero,zero,ymm5[16,17,18,19,20,21,22,23]
+; CHECK-O0-NEXT:    vpblendd {{.*#+}} ymm3 = ymm1[0,1],ymm3[2,3,4,5],ymm1[6,7]
 ; CHECK-O0-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm7[0],ymm5[0],ymm7[2],ymm5[2]
-; CHECK-O0-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,2,3]
-; CHECK-O0-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[1,1,1,1]
-; CHECK-O0-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm4[2,3,4,5],ymm1[6,7]
-; CHECK-O0-NEXT:    vmovaps %xmm3, %xmm4
-; CHECK-O0-NEXT:    vmovaps %xmm7, %xmm3
-; CHECK-O0-NEXT:    vpblendd {{.*#+}} xmm4 = xmm3[0,1],xmm4[2,3]
-; CHECK-O0-NEXT:    # implicit-def: $ymm3
-; CHECK-O0-NEXT:    vmovaps %xmm4, %xmm3
-; CHECK-O0-NEXT:    vpermq {{.*#+}} ymm4 = ymm3[0,0,1,3]
-; CHECK-O0-NEXT:    vpslldq {{.*#+}} ymm3 = zero,zero,zero,zero,zero,zero,zero,zero,ymm5[0,1,2,3,4,5,6,7],zero,zero,zero,zero,zero,zero,zero,zero,ymm5[16,17,18,19,20,21,22,23]
-; CHECK-O0-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3,4,5],ymm3[6,7]
+; CHECK-O0-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm4[2,3],ymm1[4,5,6,7]
+; CHECK-O0-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,1,3]
 ; CHECK-O0-NEXT:    movq %rbp, %rsp
 ; CHECK-O0-NEXT:    popq %rbp
 ; CHECK-O0-NEXT:    retq
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-3.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-3.ll
index a0ea6ddeca7dfd..afdeebc45ed0ab 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-3.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-3.ll
@@ -519,19 +519,14 @@ define void @load_i32_stride3_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-NEXT:    vmovaps (%rdi), %ymm0
 ; AVX2-NEXT:    vmovaps 32(%rdi), %ymm1
 ; AVX2-NEXT:    vmovaps 64(%rdi), %ymm2
-; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm3 = [2,5,2,5,2,5,2,5]
-; AVX2-NEXT:    vpermps %ymm2, %ymm3, %ymm3
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm5 = [0,3,6,1,4,7,u,u]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm4 = [0,3,6,1,4,7,2,5]
+; AVX2-NEXT:    vpermps %ymm3, %ymm4, %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0],ymm4[1,2],ymm2[3],ymm4[4,5],ymm2[6],ymm4[7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm5 = [1,4,7,2,5,0,3,6]
 ; AVX2-NEXT:    vpermps %ymm4, %ymm5, %ymm4
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm4 = [0,0,3,6,0,0,3,6]
-; AVX2-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermps %ymm2, %ymm4, %ymm4
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm6 = [1,4,7,2,5,u,u,u]
-; AVX2-NEXT:    vpermps %ymm5, %ymm6, %ymm5
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm4[5,6,7]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
 ; AVX2-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
 ; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
@@ -549,19 +544,14 @@ define void @load_i32_stride3_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm2
-; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm3 = [2,5,2,5,2,5,2,5]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm3, %ymm3
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm5 = [0,3,6,1,4,7,u,u]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm4 = [0,3,6,1,4,7,2,5]
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm4, %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0],ymm4[1,2],ymm2[3],ymm4[4,5],ymm2[6],ymm4[7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm5 = [1,4,7,2,5,0,3,6]
 ; AVX2-FP-NEXT:    vpermps %ymm4, %ymm5, %ymm4
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm4 = [0,0,3,6,0,0,3,6]
-; AVX2-FP-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm4, %ymm4
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm6 = [1,4,7,2,5,u,u,u]
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm6, %ymm5
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm4[5,6,7]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
 ; AVX2-FP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
@@ -579,26 +569,18 @@ define void @load_i32_stride3_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm0
 ; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm3 = [2,5,2,5,2,5,2,5]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm3, %ymm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm5 = [0,3,6,1,4,7,u,u]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm4 = [0,3,6,1,4,7,2,5]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm4, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0],ymm4[1,2],ymm2[3],ymm4[4,5],ymm2[6],ymm4[7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm5 = [1,4,7,2,5,0,3,6]
 ; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm5, %ymm4
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm4 = [0,0,3,6,0,0,3,6]
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm4, %ymm4
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm6 = [1,4,7,2,5,u,u,u]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm6, %ymm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm5 = [0,1,4,7,0,1,4,7]
-; AVX2-FCP-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm2
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2,3],ymm2[4],ymm0[5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,1,4,7]
 ; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rsi)
 ; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rdx)
 ; AVX2-FCP-NEXT:    vmovaps %ymm0, (%rcx)
@@ -899,33 +881,26 @@ define void @load_i32_stride3_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-LABEL: load_i32_stride3_vf16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovaps 128(%rdi), %ymm1
 ; AVX2-NEXT:    vmovaps 160(%rdi), %ymm0
+; AVX2-NEXT:    vmovaps 128(%rdi), %ymm1
 ; AVX2-NEXT:    vmovaps (%rdi), %ymm2
 ; AVX2-NEXT:    vmovaps 32(%rdi), %ymm3
 ; AVX2-NEXT:    vmovaps 64(%rdi), %ymm4
 ; AVX2-NEXT:    vmovaps 96(%rdi), %ymm5
-; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm6 = [2,5,2,5,2,5,2,5]
-; AVX2-NEXT:    vpermps %ymm4, %ymm6, %ymm7
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6],ymm3[7]
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm9 = [0,3,6,1,4,7,u,u]
-; AVX2-NEXT:    vpermps %ymm8, %ymm9, %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm8[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-NEXT:    vpermps %ymm0, %ymm6, %ymm6
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6],ymm3[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm4[2],ymm6[3,4],ymm4[5],ymm6[6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm7 = [0,3,6,1,4,7,2,5]
+; AVX2-NEXT:    vpermps %ymm6, %ymm7, %ymm6
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0],ymm1[1],ymm5[2,3],ymm1[4],ymm5[5,6],ymm1[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7]
+; AVX2-NEXT:    vpermps %ymm8, %ymm7, %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0],ymm8[1,2],ymm4[3],ymm8[4,5],ymm4[6],ymm8[7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm9 = [1,4,7,2,5,0,3,6]
 ; AVX2-NEXT:    vpermps %ymm8, %ymm9, %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm8 = [0,0,3,6,0,0,3,6]
-; AVX2-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermps %ymm4, %ymm8, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm11 = [1,4,7,2,5,u,u,u]
-; AVX2-NEXT:    vpermps %ymm10, %ymm11, %ymm10
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm10[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-NEXT:    vpermps %ymm0, %ymm8, %ymm8
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm5[0,1],ymm1[2],ymm5[3,4],ymm1[5],ymm5[6,7]
-; AVX2-NEXT:    vpermps %ymm10, %ymm11, %ymm10
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm10[0,1,2,3,4],ymm8[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0],ymm10[1,2],ymm0[3],ymm10[4,5],ymm0[6],ymm10[7]
+; AVX2-NEXT:    vpermps %ymm10, %ymm9, %ymm9
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
 ; AVX2-NEXT:    vmovaps {{.*#+}} ymm3 = [2,5,0,3,6,u,u,u]
 ; AVX2-NEXT:    vpermps %ymm2, %ymm3, %ymm2
@@ -937,10 +912,10 @@ define void @load_i32_stride3_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,3,4,5,4,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovaps %ymm6, 32(%rsi)
-; AVX2-NEXT:    vmovaps %ymm7, (%rsi)
-; AVX2-NEXT:    vmovaps %ymm8, 32(%rdx)
-; AVX2-NEXT:    vmovaps %ymm9, (%rdx)
+; AVX2-NEXT:    vmovaps %ymm7, 32(%rsi)
+; AVX2-NEXT:    vmovaps %ymm6, (%rsi)
+; AVX2-NEXT:    vmovaps %ymm9, 32(%rdx)
+; AVX2-NEXT:    vmovaps %ymm8, (%rdx)
 ; AVX2-NEXT:    vmovaps %ymm0, 32(%rcx)
 ; AVX2-NEXT:    vmovaps %ymm2, (%rcx)
 ; AVX2-NEXT:    vzeroupper
@@ -948,33 +923,26 @@ define void @load_i32_stride3_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-FP-LABEL: load_i32_stride3_vf16:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm0
+; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vmovaps 96(%rdi), %ymm5
-; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm6 = [2,5,2,5,2,5,2,5]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm6, %ymm7
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm9 = [0,3,6,1,4,7,u,u]
-; AVX2-FP-NEXT:    vpermps %ymm8, %ymm9, %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm8[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm0, %ymm6, %ymm6
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6],ymm3[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm4[2],ymm6[3,4],ymm4[5],ymm6[6,7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm7 = [0,3,6,1,4,7,2,5]
+; AVX2-FP-NEXT:    vpermps %ymm6, %ymm7, %ymm6
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0],ymm1[1],ymm5[2,3],ymm1[4],ymm5[5,6],ymm1[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm7, %ymm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0],ymm8[1,2],ymm4[3],ymm8[4,5],ymm4[6],ymm8[7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm9 = [1,4,7,2,5,0,3,6]
 ; AVX2-FP-NEXT:    vpermps %ymm8, %ymm9, %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm8 = [0,0,3,6,0,0,3,6]
-; AVX2-FP-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm8, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm11 = [1,4,7,2,5,u,u,u]
-; AVX2-FP-NEXT:    vpermps %ymm10, %ymm11, %ymm10
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm10[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm0, %ymm8, %ymm8
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm5[0,1],ymm1[2],ymm5[3,4],ymm1[5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm10, %ymm11, %ymm10
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm10[0,1,2,3,4],ymm8[5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0],ymm10[1,2],ymm0[3],ymm10[4,5],ymm0[6],ymm10[7]
+; AVX2-FP-NEXT:    vpermps %ymm10, %ymm9, %ymm9
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
 ; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm3 = [2,5,0,3,6,u,u,u]
 ; AVX2-FP-NEXT:    vpermps %ymm2, %ymm3, %ymm2
@@ -986,10 +954,10 @@ define void @load_i32_stride3_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,3,4,5,4,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%rsi)
-; AVX2-FP-NEXT:    vmovaps %ymm7, (%rsi)
-; AVX2-FP-NEXT:    vmovaps %ymm8, 32(%rdx)
-; AVX2-FP-NEXT:    vmovaps %ymm9, (%rdx)
+; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%rsi)
+; AVX2-FP-NEXT:    vmovaps %ymm6, (%rsi)
+; AVX2-FP-NEXT:    vmovaps %ymm9, 32(%rdx)
+; AVX2-FP-NEXT:    vmovaps %ymm8, (%rdx)
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rcx)
 ; AVX2-FP-NEXT:    vmovaps %ymm2, (%rcx)
 ; AVX2-FP-NEXT:    vzeroupper
@@ -997,47 +965,37 @@ define void @load_i32_stride3_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-FCP-LABEL: load_i32_stride3_vf16:
 ; AVX2-FCP:       # %bb.0:
-; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm0
-; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm4
 ; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm6 = [2,5,2,5,2,5,2,5]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm6, %ymm7
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm9 = [0,3,6,1,4,7,u,u]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm9, %ymm8
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm8[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm6, %ymm6
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0],ymm0[1],ymm5[2,3],ymm0[4],ymm5[5,6],ymm0[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm4[2],ymm6[3,4],ymm4[5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm7 = [0,3,6,1,4,7,2,5]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm7, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0],ymm1[1],ymm5[2,3],ymm1[4],ymm5[5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm7, %ymm7
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0],ymm8[1,2],ymm4[3],ymm8[4,5],ymm4[6],ymm8[7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm9 = [1,4,7,2,5,0,3,6]
 ; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm9, %ymm8
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm8 = [0,0,3,6,0,0,3,6]
-; AVX2-FCP-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm8, %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm11 = [1,4,7,2,5,u,u,u]
-; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm11, %ymm10
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm10[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm8, %ymm8
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm5[0,1],ymm0[2],ymm5[3,4],ymm0[5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm11, %ymm10
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm10[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm10 = [0,1,0,3,0,1,4,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm10, %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm5[0,1],ymm1[2],ymm5[3,4],ymm1[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0],ymm10[1,2],ymm0[3],ymm10[4,5],ymm0[6],ymm10[7]
+; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm9, %ymm9
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm3 = [2,5,0,3,6,u,u,u]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm4[1],ymm2[2,3],ymm4[4],ymm2[5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm3 = [2,5,0,3,6,1,4,7]
 ; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm3, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm10, %ymm1
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm5[2],ymm0[3,4],ymm5[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm5[2],ymm1[3,4],ymm5[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6],ymm0[7]
 ; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm3, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%rsi)
-; AVX2-FCP-NEXT:    vmovaps %ymm7, (%rsi)
-; AVX2-FCP-NEXT:    vmovaps %ymm8, 32(%rdx)
-; AVX2-FCP-NEXT:    vmovaps %ymm9, (%rdx)
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 32(%rsi)
+; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rsi)
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 32(%rdx)
+; AVX2-FCP-NEXT:    vmovaps %ymm8, (%rdx)
 ; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%rcx)
 ; AVX2-FCP-NEXT:    vmovaps %ymm2, (%rcx)
 ; AVX2-FCP-NEXT:    vzeroupper
@@ -1608,317 +1566,261 @@ define void @load_i32_stride3_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-LABEL: load_i32_stride3_vf32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    subq $136, %rsp
-; AVX2-NEXT:    vmovaps 224(%rdi), %ymm0
-; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 320(%rdi), %ymm4
-; AVX2-NEXT:    vmovaps 288(%rdi), %ymm7
-; AVX2-NEXT:    vmovaps 352(%rdi), %ymm5
-; AVX2-NEXT:    vmovaps 128(%rdi), %ymm10
-; AVX2-NEXT:    vmovaps (%rdi), %ymm12
-; AVX2-NEXT:    vmovaps 32(%rdi), %ymm6
-; AVX2-NEXT:    vmovaps 64(%rdi), %ymm3
-; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 96(%rdi), %ymm13
-; AVX2-NEXT:    vmovaps 160(%rdi), %ymm11
-; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm1 = [2,5,2,5,2,5,2,5]
-; AVX2-NEXT:    vpermps %ymm11, %ymm1, %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm13[0],ymm10[1],ymm13[2,3],ymm10[4],ymm13[5,6],ymm10[7]
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm14 = [0,3,6,1,4,7,u,u]
-; AVX2-NEXT:    vpermps %ymm9, %ymm14, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-NEXT:    subq $40, %rsp
+; AVX2-NEXT:    vmovaps 256(%rdi), %ymm15
+; AVX2-NEXT:    vmovaps 224(%rdi), %ymm5
+; AVX2-NEXT:    vmovaps 192(%rdi), %ymm3
+; AVX2-NEXT:    vmovaps 352(%rdi), %ymm4
+; AVX2-NEXT:    vmovaps 320(%rdi), %ymm8
+; AVX2-NEXT:    vmovaps 288(%rdi), %ymm10
+; AVX2-NEXT:    vmovaps 160(%rdi), %ymm13
+; AVX2-NEXT:    vmovaps 128(%rdi), %ymm1
+; AVX2-NEXT:    vmovaps (%rdi), %ymm6
+; AVX2-NEXT:    vmovaps 32(%rdi), %ymm9
+; AVX2-NEXT:    vmovaps 64(%rdi), %ymm7
+; AVX2-NEXT:    vmovaps 96(%rdi), %ymm0
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm13[2],ymm11[3,4],ymm13[5],ymm11[6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm12 = [0,3,6,1,4,7,2,5]
+; AVX2-NEXT:    vpermps %ymm11, %ymm12, %ymm2
+; AVX2-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm10[0],ymm8[1],ymm10[2,3],ymm8[4],ymm10[5,6],ymm8[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm4[2],ymm11[3,4],ymm4[5],ymm11[6,7]
+; AVX2-NEXT:    vpermps %ymm11, %ymm12, %ymm2
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm5, %ymm1, %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm7[0],ymm4[1],ymm7[2,3],ymm4[4],ymm7[5,6],ymm4[7]
-; AVX2-NEXT:    vpermps %ymm9, %ymm14, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm6[0],ymm9[1],ymm6[2,3],ymm9[4],ymm6[5,6],ymm9[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm7[2],ymm11[3,4],ymm7[5],ymm11[6,7]
+; AVX2-NEXT:    vpermps %ymm11, %ymm12, %ymm2
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm3, %ymm1, %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm12[0],ymm6[1],ymm12[2,3],ymm6[4],ymm12[5,6],ymm6[7]
-; AVX2-NEXT:    vmovaps %ymm12, %ymm3
-; AVX2-NEXT:    vpermps %ymm9, %ymm14, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm3[0],ymm5[1],ymm3[2,3],ymm5[4],ymm3[5,6],ymm5[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1],ymm15[2],ymm14[3,4],ymm15[5],ymm14[6,7]
+; AVX2-NEXT:    vpermps %ymm14, %ymm12, %ymm2
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 192(%rdi), %ymm12
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm12[0],ymm0[1],ymm12[2,3],ymm0[4],ymm12[5,6],ymm0[7]
-; AVX2-NEXT:    vpermps %ymm8, %ymm14, %ymm8
-; AVX2-NEXT:    vmovaps 256(%rdi), %ymm15
-; AVX2-NEXT:    vpermps %ymm15, %ymm1, %ymm1
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm13[0,1],ymm10[2],ymm13[3,4],ymm10[5],ymm13[6,7]
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm2 = [1,4,7,2,5,u,u,u]
-; AVX2-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,0,3,6,0,0,3,6]
-; AVX2-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermps %ymm11, %ymm0, %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1],ymm4[2],ymm7[3,4],ymm4[5],ymm7[6,7]
-; AVX2-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-NEXT:    vpermps %ymm5, %ymm0, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm3[0,1],ymm6[2],ymm3[3,4],ymm6[5],ymm3[6,7]
-; AVX2-NEXT:    vmovaps %ymm6, %ymm8
-; AVX2-NEXT:    vpermps %ymm9, %ymm2, %ymm9
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm6, %ymm0, %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm12[0,1],ymm1[2],ymm12[3,4],ymm1[5],ymm12[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0],ymm12[1,2],ymm13[3],ymm12[4,5],ymm13[6],ymm12[7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm2 = [1,4,7,2,5,0,3,6]
+; AVX2-NEXT:    vpermps %ymm12, %ymm2, %ymm11
+; AVX2-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm10[0,1],ymm8[2],ymm10[3,4],ymm8[5],ymm10[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm4[0],ymm11[1,2],ymm4[3],ymm11[4,5],ymm4[6],ymm11[7]
+; AVX2-NEXT:    vpermps %ymm11, %ymm2, %ymm11
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm6[0,1],ymm9[2],ymm6[3,4],ymm9[5],ymm6[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm7[0],ymm12[1,2],ymm7[3],ymm12[4,5],ymm7[6],ymm12[7]
+; AVX2-NEXT:    vpermps %ymm12, %ymm2, %ymm12
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm3[0,1],ymm5[2],ymm3[3,4],ymm5[5],ymm3[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0],ymm14[1,2],ymm15[3],ymm14[4,5],ymm15[6],ymm14[7]
 ; AVX2-NEXT:    vpermps %ymm14, %ymm2, %ymm2
-; AVX2-NEXT:    vpermps %ymm15, %ymm0, %ymm0
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm10[0,1],ymm13[2],ymm10[3,4],ymm13[5],ymm10[6,7]
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm10 = [2,5,0,3,6,u,u,u]
-; AVX2-NEXT:    vpermps %ymm2, %ymm10, %ymm2
-; AVX2-NEXT:    vshufps {{.*#+}} ymm11 = ymm11[0,1,0,3,4,5,4,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm11 = ymm11[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm11[5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm7[2],ymm4[3,4],ymm7[5],ymm4[6,7]
-; AVX2-NEXT:    vpermps %ymm4, %ymm10, %ymm4
-; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[0,1,0,3,4,5,4,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm8[0,1],ymm3[2],ymm8[3,4],ymm3[5],ymm8[6,7]
-; AVX2-NEXT:    vpermps %ymm5, %ymm10, %ymm5
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm6[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
+; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm13 = ymm13[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm13[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm10[2],ymm8[3,4],ymm10[5],ymm8[6,7]
+; AVX2-NEXT:    vpermps %ymm8, %ymm1, %ymm8
+; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm8[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm9[0,1],ymm6[2],ymm9[3,4],ymm6[5],ymm9[6,7]
+; AVX2-NEXT:    vpermps %ymm6, %ymm1, %ymm6
+; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2],ymm5[3,4],ymm3[5],ymm5[6,7]
+; AVX2-NEXT:    vpermps %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm15[0,1,0,3,4,5,4,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm1[0,1],ymm12[2],ymm1[3,4],ymm12[5],ymm1[6,7]
-; AVX2-NEXT:    vpermps %ymm5, %ymm10, %ymm5
-; AVX2-NEXT:    vshufps {{.*#+}} ymm6 = ymm15[0,1,0,3,4,5,4,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7]
-; AVX2-NEXT:    vmovups (%rsp), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 64(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, (%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 96(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 32(%rsi)
-; AVX2-NEXT:    vmovaps %ymm0, 64(%rdx)
-; AVX2-NEXT:    vmovaps %ymm9, (%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 96(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 32(%rdx)
-; AVX2-NEXT:    vmovaps %ymm5, 64(%rcx)
-; AVX2-NEXT:    vmovaps %ymm3, (%rcx)
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 64(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, (%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 96(%rsi)
+; AVX2-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 32(%rsi)
+; AVX2-NEXT:    vmovaps %ymm2, 64(%rdx)
+; AVX2-NEXT:    vmovaps %ymm12, (%rdx)
+; AVX2-NEXT:    vmovaps %ymm11, 96(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm2, 32(%rdx)
+; AVX2-NEXT:    vmovaps %ymm1, 64(%rcx)
+; AVX2-NEXT:    vmovaps %ymm6, (%rcx)
 ; AVX2-NEXT:    vmovaps %ymm4, 96(%rcx)
-; AVX2-NEXT:    vmovaps %ymm2, 32(%rcx)
-; AVX2-NEXT:    addq $136, %rsp
+; AVX2-NEXT:    vmovaps %ymm0, 32(%rcx)
+; AVX2-NEXT:    addq $40, %rsp
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: load_i32_stride3_vf32:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    subq $136, %rsp
-; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm0
-; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm4
-; AVX2-FP-NEXT:    vmovaps 288(%rdi), %ymm7
-; AVX2-FP-NEXT:    vmovaps 352(%rdi), %ymm5
-; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm10
-; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm12
-; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm6
-; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm3
-; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 96(%rdi), %ymm13
-; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm11
-; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm1 = [2,5,2,5,2,5,2,5]
-; AVX2-FP-NEXT:    vpermps %ymm11, %ymm1, %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm13[0],ymm10[1],ymm13[2,3],ymm10[4],ymm13[5,6],ymm10[7]
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm14 = [0,3,6,1,4,7,u,u]
-; AVX2-FP-NEXT:    vpermps %ymm9, %ymm14, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    subq $40, %rsp
+; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm15
+; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm5
+; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovaps 352(%rdi), %ymm4
+; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm8
+; AVX2-FP-NEXT:    vmovaps 288(%rdi), %ymm10
+; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm13
+; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm6
+; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm9
+; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm7
+; AVX2-FP-NEXT:    vmovaps 96(%rdi), %ymm0
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm13[2],ymm11[3,4],ymm13[5],ymm11[6,7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm12 = [0,3,6,1,4,7,2,5]
+; AVX2-FP-NEXT:    vpermps %ymm11, %ymm12, %ymm2
+; AVX2-FP-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm10[0],ymm8[1],ymm10[2,3],ymm8[4],ymm10[5,6],ymm8[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm4[2],ymm11[3,4],ymm4[5],ymm11[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm11, %ymm12, %ymm2
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm1, %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm7[0],ymm4[1],ymm7[2,3],ymm4[4],ymm7[5,6],ymm4[7]
-; AVX2-FP-NEXT:    vpermps %ymm9, %ymm14, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm6[0],ymm9[1],ymm6[2,3],ymm9[4],ymm6[5,6],ymm9[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm7[2],ymm11[3,4],ymm7[5],ymm11[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm11, %ymm12, %ymm2
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm1, %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm12[0],ymm6[1],ymm12[2,3],ymm6[4],ymm12[5,6],ymm6[7]
-; AVX2-FP-NEXT:    vmovaps %ymm12, %ymm3
-; AVX2-FP-NEXT:    vpermps %ymm9, %ymm14, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm3[0],ymm5[1],ymm3[2,3],ymm5[4],ymm3[5,6],ymm5[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1],ymm15[2],ymm14[3,4],ymm15[5],ymm14[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm14, %ymm12, %ymm2
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm12
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm12[0],ymm0[1],ymm12[2,3],ymm0[4],ymm12[5,6],ymm0[7]
-; AVX2-FP-NEXT:    vpermps %ymm8, %ymm14, %ymm8
-; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm15
-; AVX2-FP-NEXT:    vpermps %ymm15, %ymm1, %ymm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm13[0,1],ymm10[2],ymm13[3,4],ymm10[5],ymm13[6,7]
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm2 = [1,4,7,2,5,u,u,u]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,0,3,6,0,0,3,6]
-; AVX2-FP-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermps %ymm11, %ymm0, %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1],ymm4[2],ymm7[3,4],ymm4[5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm0, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm3[0,1],ymm6[2],ymm3[3,4],ymm6[5],ymm3[6,7]
-; AVX2-FP-NEXT:    vmovaps %ymm6, %ymm8
-; AVX2-FP-NEXT:    vpermps %ymm9, %ymm2, %ymm9
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm6, %ymm0, %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm12[0,1],ymm1[2],ymm12[3,4],ymm1[5],ymm12[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0],ymm12[1,2],ymm13[3],ymm12[4,5],ymm13[6],ymm12[7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm2 = [1,4,7,2,5,0,3,6]
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm2, %ymm11
+; AVX2-FP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm10[0,1],ymm8[2],ymm10[3,4],ymm8[5],ymm10[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm4[0],ymm11[1,2],ymm4[3],ymm11[4,5],ymm4[6],ymm11[7]
+; AVX2-FP-NEXT:    vpermps %ymm11, %ymm2, %ymm11
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm6[0,1],ymm9[2],ymm6[3,4],ymm9[5],ymm6[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm7[0],ymm12[1,2],ymm7[3],ymm12[4,5],ymm7[6],ymm12[7]
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm2, %ymm12
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm3[0,1],ymm5[2],ymm3[3,4],ymm5[5],ymm3[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0],ymm14[1,2],ymm15[3],ymm14[4,5],ymm15[6],ymm14[7]
 ; AVX2-FP-NEXT:    vpermps %ymm14, %ymm2, %ymm2
-; AVX2-FP-NEXT:    vpermps %ymm15, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm10[0,1],ymm13[2],ymm10[3,4],ymm13[5],ymm10[6,7]
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm10 = [2,5,0,3,6,u,u,u]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm10, %ymm2
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm11 = ymm11[0,1,0,3,4,5,4,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm11 = ymm11[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm11[5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm7[2],ymm4[3,4],ymm7[5],ymm4[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm10, %ymm4
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[0,1,0,3,4,5,4,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm8[0,1],ymm3[2],ymm8[3,4],ymm3[5],ymm8[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm10, %ymm5
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm6[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm13 = ymm13[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm13[5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm10[2],ymm8[3,4],ymm10[5],ymm8[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm1, %ymm8
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm8[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm9[0,1],ymm6[2],ymm9[3,4],ymm6[5],ymm9[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm6, %ymm1, %ymm6
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2],ymm5[3,4],ymm3[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm1, %ymm1
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm15[0,1,0,3,4,5,4,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm1[0,1],ymm12[2],ymm1[3,4],ymm12[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm10, %ymm5
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm6 = ymm15[0,1,0,3,4,5,4,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7]
-; AVX2-FP-NEXT:    vmovups (%rsp), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, (%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%rsi)
-; AVX2-FP-NEXT:    vmovaps %ymm0, 64(%rdx)
-; AVX2-FP-NEXT:    vmovaps %ymm9, (%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 96(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rdx)
-; AVX2-FP-NEXT:    vmovaps %ymm5, 64(%rcx)
-; AVX2-FP-NEXT:    vmovaps %ymm3, (%rcx)
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 64(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, (%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 96(%rsi)
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rsi)
+; AVX2-FP-NEXT:    vmovaps %ymm2, 64(%rdx)
+; AVX2-FP-NEXT:    vmovaps %ymm12, (%rdx)
+; AVX2-FP-NEXT:    vmovaps %ymm11, 96(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm2, 32(%rdx)
+; AVX2-FP-NEXT:    vmovaps %ymm1, 64(%rcx)
+; AVX2-FP-NEXT:    vmovaps %ymm6, (%rcx)
 ; AVX2-FP-NEXT:    vmovaps %ymm4, 96(%rcx)
-; AVX2-FP-NEXT:    vmovaps %ymm2, 32(%rcx)
-; AVX2-FP-NEXT:    addq $136, %rsp
+; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rcx)
+; AVX2-FP-NEXT:    addq $40, %rsp
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
 ; AVX2-FCP-LABEL: load_i32_stride3_vf32:
 ; AVX2-FCP:       # %bb.0:
-; AVX2-FCP-NEXT:    subq $104, %rsp
-; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm6
+; AVX2-FCP-NEXT:    subq $72, %rsp
+; AVX2-FCP-NEXT:    vmovaps 256(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovaps 352(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm15
-; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %ymm13
-; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm12
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm8 = [2,5,2,5,2,5,2,5]
-; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm8, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm13[0],ymm10[1],ymm13[2,3],ymm10[4],ymm13[5,6],ymm10[7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm14 = [0,3,6,1,4,7,u,u]
-; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm14, %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm8, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm6[0],ymm3[1],ymm6[2,3],ymm3[4],ymm6[5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm14, %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm8, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm2[0],ymm15[1],ymm2[2,3],ymm15[4],ymm2[5,6],ymm15[7]
-; AVX2-FCP-NEXT:    vmovaps %ymm2, %ymm7
-; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm14, %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0],ymm1[1],ymm11[2,3],ymm1[4],ymm11[5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm14, %ymm9
-; AVX2-FCP-NEXT:    vmovaps 256(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm8, %ymm8
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm6
+; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm7
+; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm13
+; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm14
+; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm8
+; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm9
+; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm10
+; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %ymm15
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm15[0],ymm14[1],ymm15[2,3],ymm14[4],ymm15[5,6],ymm14[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm13[2],ymm3[3,4],ymm13[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm12 = [0,3,6,1,4,7,2,5]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm12, %ymm3
+; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0],ymm6[1],ymm7[2,3],ymm6[4],ymm7[5,6],ymm6[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1],ymm4[2],ymm5[3,4],ymm4[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm12, %ymm3
+; AVX2-FCP-NEXT:    vmovups %ymm3, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm8[0],ymm9[1],ymm8[2,3],ymm9[4],ymm8[5,6],ymm9[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm12, %ymm3
+; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm0[2],ymm3[3,4],ymm0[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vmovaps %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm12, %ymm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm13[0,1],ymm10[2],ymm13[3,4],ymm10[5],ymm13[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm1 = [1,4,7,2,5,u,u,u]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm1, %ymm8
-; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,0,3,6,0,0,3,6]
-; AVX2-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm0, %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm6[0,1],ymm3[2],ymm6[3,4],ymm3[5],ymm6[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm1, %ymm8
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm0, %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm7[0,1],ymm15[2],ymm7[3,4],ymm15[5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vmovaps %ymm7, %ymm9
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm1, %ymm8
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm0, %ymm14
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm11[0,1],ymm7[2],ymm11[3,4],ymm7[5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm1, %ymm1
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm10[0,1],ymm13[2],ymm10[3,4],ymm13[5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm10 = [0,1,0,3,0,1,4,7]
-; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm10, %ymm12
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm13 = [2,5,0,3,6,u,u,u]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm13, %ymm1
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm12[5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm6[2],ymm3[3,4],ymm6[5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm10, %ymm6
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm13, %ymm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm6[5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm15[0,1],ymm9[2],ymm15[3,4],ymm9[5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm10, %ymm5
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm13, %ymm4
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm10, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0,1],ymm11[2],ymm7[3,4],ymm11[5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm13, %ymm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm5[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%rsi)
-; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, (%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%rsi)
-; AVX2-FCP-NEXT:    vmovaps %ymm0, 64(%rdx)
-; AVX2-FCP-NEXT:    vmovaps %ymm8, (%rdx)
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm15[0,1],ymm14[2],ymm15[3,4],ymm14[5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0],ymm12[1,2],ymm13[3],ymm12[4,5],ymm13[6],ymm12[7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm3 = [1,4,7,2,5,0,3,6]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm3, %ymm5
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0,1],ymm6[2],ymm7[3,4],ymm6[5],ymm7[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm4[0],ymm5[1,2],ymm4[3],ymm5[4,5],ymm4[6],ymm5[7]
+; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm3, %ymm5
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm8[0,1],ymm9[2],ymm8[3,4],ymm9[5],ymm8[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm10[0],ymm11[1,2],ymm10[3],ymm11[4,5],ymm10[6],ymm11[7]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm3, %ymm11
+; AVX2-FCP-NEXT:    vmovaps %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm2[0],ymm12[1,2],ymm2[3],ymm12[4,5],ymm2[6],ymm12[7]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm3, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm14[0,1],ymm15[2],ymm14[3,4],ymm15[5],ymm14[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0],ymm13[1],ymm12[2,3],ymm13[4],ymm12[5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm7[2],ymm6[3,4],ymm7[5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0],ymm4[1],ymm6[2,3],ymm4[4],ymm6[5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm9[0,1],ymm8[2],ymm9[3,4],ymm8[5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0],ymm10[1],ymm6[2,3],ymm10[4],ymm6[5,6],ymm10[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm2 = [2,5,0,3,6,1,4,7]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm2, %ymm7
+; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm2, %ymm4
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm2, %ymm6
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm0, 96(%rdx)
+; AVX2-FCP-NEXT:    vmovaps %ymm0, 64(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm2, (%rsi)
+; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm2, 96(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm2, 32(%rsi)
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 64(%rdx)
+; AVX2-FCP-NEXT:    vmovaps %ymm11, (%rdx)
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rdx)
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%rdx)
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 64(%rcx)
-; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rcx)
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 96(%rcx)
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 32(%rcx)
-; AVX2-FCP-NEXT:    addq $104, %rsp
+; AVX2-FCP-NEXT:    vmovaps %ymm1, 64(%rcx)
+; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rcx)
+; AVX2-FCP-NEXT:    vmovaps %ymm4, 96(%rcx)
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 32(%rcx)
+; AVX2-FCP-NEXT:    addq $72, %rsp
 ; AVX2-FCP-NEXT:    vzeroupper
 ; AVX2-FCP-NEXT:    retq
 ;
@@ -3014,245 +2916,229 @@ define void @load_i32_stride3_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-LABEL: load_i32_stride3_vf64:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    subq $1032, %rsp # imm = 0x408
+; AVX2-NEXT:    vmovaps 736(%rdi), %ymm2
+; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 704(%rdi), %ymm3
 ; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 512(%rdi), %ymm4
+; AVX2-NEXT:    vmovaps 672(%rdi), %ymm4
 ; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 480(%rdi), %ymm6
-; AVX2-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 544(%rdi), %ymm5
 ; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 320(%rdi), %ymm7
+; AVX2-NEXT:    vmovaps 512(%rdi), %ymm6
+; AVX2-NEXT:    vmovups %ymm6, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 480(%rdi), %ymm7
 ; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 288(%rdi), %ymm10
+; AVX2-NEXT:    vmovaps 352(%rdi), %ymm8
+; AVX2-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 320(%rdi), %ymm10
 ; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 352(%rdi), %ymm9
+; AVX2-NEXT:    vmovaps 288(%rdi), %ymm11
+; AVX2-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 160(%rdi), %ymm9
 ; AVX2-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 128(%rdi), %ymm13
-; AVX2-NEXT:    vmovaps 96(%rdi), %ymm14
-; AVX2-NEXT:    vmovaps 160(%rdi), %ymm1
+; AVX2-NEXT:    vmovaps 128(%rdi), %ymm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 96(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [2,5,2,5,2,5,2,5]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6],ymm0[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1],ymm9[2],ymm0[3,4],ymm9[5],ymm0[6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm0 = [0,3,6,1,4,7,2,5]
 ; AVX2-NEXT:    vpermps %ymm1, %ymm0, %ymm1
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm14[0],ymm13[1],ymm14[2,3],ymm13[4],ymm14[5,6],ymm13[7]
-; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm8 = [0,3,6,1,4,7,u,u]
-; AVX2-NEXT:    vpermps %ymm2, %ymm8, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm9, %ymm0, %ymm1
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm10[0],ymm7[1],ymm10[2,3],ymm7[4],ymm10[5,6],ymm7[7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm8, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm5, %ymm0, %ymm1
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm6[0],ymm4[1],ymm6[2,3],ymm4[4],ymm6[5,6],ymm4[7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm8, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm11[0],ymm10[1],ymm11[2,3],ymm10[4],ymm11[5,6],ymm10[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm8[2],ymm1[3,4],ymm8[5],ymm1[6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 672(%rdi), %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0],ymm6[1],ymm7[2,3],ymm6[4],ymm7[5,6],ymm6[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm5[2],ymm1[3,4],ymm5[5],ymm1[6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6],ymm3[7]
-; AVX2-NEXT:    vpermps %ymm1, %ymm8, %ymm1
-; AVX2-NEXT:    vmovaps 736(%rdi), %ymm2
-; AVX2-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps (%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 32(%rdi), %ymm3
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6],ymm3[7]
-; AVX2-NEXT:    vmovaps %ymm3, %ymm5
-; AVX2-NEXT:    vpermps %ymm1, %ymm8, %ymm1
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 64(%rdi), %ymm2
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6],ymm3[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm2[2],ymm4[3,4],ymm2[5],ymm4[6,7]
+; AVX2-NEXT:    vmovaps %ymm2, %ymm8
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm4
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 224(%rdi), %ymm7
-; AVX2-NEXT:    vmovaps 192(%rdi), %ymm15
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm15[0],ymm7[1],ymm15[2,3],ymm7[4],ymm15[5,6],ymm7[7]
-; AVX2-NEXT:    vmovups %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm1, %ymm8, %ymm1
-; AVX2-NEXT:    vmovaps 256(%rdi), %ymm2
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm6
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 416(%rdi), %ymm1
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 384(%rdi), %ymm3
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0],ymm1[1],ymm3[2,3],ymm1[4],ymm3[5,6],ymm1[7]
-; AVX2-NEXT:    vmovaps %ymm3, %ymm6
-; AVX2-NEXT:    vpermps %ymm1, %ymm8, %ymm1
-; AVX2-NEXT:    vmovaps 448(%rdi), %ymm2
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm12
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-NEXT:    vpermps %ymm4, %ymm0, %ymm4
+; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 256(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 608(%rdi), %ymm9
-; AVX2-NEXT:    vmovaps 576(%rdi), %ymm11
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm11[0],ymm9[1],ymm11[2,3],ymm9[4],ymm11[5,6],ymm9[7]
-; AVX2-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm1, %ymm8, %ymm1
-; AVX2-NEXT:    vmovaps 640(%rdi), %ymm2
+; AVX2-NEXT:    vmovaps 224(%rdi), %ymm2
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-NEXT:    vmovaps 192(%rdi), %ymm3
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6],ymm2[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm1[2],ymm6[3,4],ymm1[5],ymm6[6,7]
+; AVX2-NEXT:    vpermps %ymm6, %ymm0, %ymm6
+; AVX2-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 448(%rdi), %ymm13
+; AVX2-NEXT:    vmovaps 416(%rdi), %ymm12
+; AVX2-NEXT:    vmovaps 384(%rdi), %ymm14
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm14[0],ymm12[1],ymm14[2,3],ymm12[4],ymm14[5,6],ymm12[7]
+; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1],ymm13[2],ymm10[3,4],ymm13[5],ymm10[6,7]
+; AVX2-NEXT:    vpermps %ymm10, %ymm0, %ymm10
+; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 640(%rdi), %ymm6
+; AVX2-NEXT:    vmovaps 608(%rdi), %ymm5
+; AVX2-NEXT:    vmovaps 576(%rdi), %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm7[0],ymm5[1],ymm7[2,3],ymm5[4],ymm7[5,6],ymm5[7]
+; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm6[2],ymm15[3,4],ymm6[5],ymm15[6,7]
+; AVX2-NEXT:    vpermps %ymm15, %ymm0, %ymm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1],ymm4[2],ymm9[3,4],ymm4[5],ymm9[6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0],ymm0[1,2],ymm11[3],ymm0[4,5],ymm11[6],ymm0[7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm15 = [1,4,7,2,5,0,3,6]
+; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
+; AVX2-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
+; AVX2-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
+; AVX2-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1],ymm13[2],ymm14[3,4],ymm13[5],ymm14[6,7]
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm1 = [1,4,7,2,5,u,u,u]
-; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm2
-; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,0,3,6,0,0,3,6]
-; AVX2-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm10, %ymm0, %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm12[2],mem[3,4],ymm12[5],mem[6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm13, %ymm0, %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm2[2],mem[3,4],ymm2[5],mem[6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm2[2],mem[3,4],ymm2[5],mem[6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-NEXT:    vpermps (%rsp), %ymm0, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps %ymm5, %ymm4
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0],ymm0[1,2],ymm8[3],ymm0[4,5],ymm8[6],ymm0[7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm5[2],ymm3[3,4],ymm5[5],ymm3[6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm5, %ymm0, %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm15[0,1],ymm7[2],ymm15[3,4],ymm7[5],ymm15[6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm15, %ymm0, %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0,1],ymm3[2],ymm8[3,4],ymm3[5],ymm8[6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0],ymm0[1,2],ymm10[3],ymm0[4,5],ymm10[6],ymm0[7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1],ymm12[2],ymm14[3,4],ymm12[5],ymm14[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm13[0],ymm0[1,2],ymm13[3],ymm0[4,5],ymm13[6],ymm0[7]
+; AVX2-NEXT:    vmovaps %ymm13, %ymm14
+; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm13
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm7[0,1],ymm5[2],ymm7[3,4],ymm5[5],ymm7[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0],ymm0[1,2],ymm6[3],ymm0[4,5],ymm6[6],ymm0[7]
 ; AVX2-NEXT:    vmovaps %ymm6, %ymm7
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm7[0,1],ymm6[2],ymm7[3,4],ymm6[5],ymm7[6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm8, %ymm0, %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm11[0,1],ymm9[2],ymm11[3,4],ymm9[5],ymm11[6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm9, %ymm0, %ymm0
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm15, %ymm15
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm4[0,1],ymm9[2],ymm4[3,4],ymm9[5],ymm4[6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm12 = [2,5,0,3,6,u,u,u]
+; AVX2-NEXT:    vpermps %ymm0, %ymm12, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm9 = ymm11[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm9 = ymm9[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm0[0,1,2,3,4],ymm9[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm12, %ymm0
+; AVX2-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,1],ymm1[2],mem[3,4],ymm1[5],mem[6,7]
-; AVX2-NEXT:    vmovaps {{.*#+}} ymm14 = [2,5,0,3,6,u,u,u]
-; AVX2-NEXT:    vpermps %ymm1, %ymm14, %ymm1
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm10[0,1,0,3,4,5,4,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1],ymm3[2],ymm4[3,4],ymm3[5],ymm4[6,7]
-; AVX2-NEXT:    vpermps %ymm1, %ymm14, %ymm1
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm5[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm12, %ymm1
+; AVX2-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = mem[0,1,0,3,4,5,4,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm12[0,1],mem[2],ymm12[3,4],mem[5],ymm12[6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm14, %ymm2
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm13[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm8[2],ymm3[3,4],ymm8[5],ymm3[6,7]
+; AVX2-NEXT:    vpermps %ymm2, %ymm12, %ymm2
+; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm10[0,1,0,3,4,5,4,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm2[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm2[2],mem[3,4],ymm2[5],mem[6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm14, %ymm2
-; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm15[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = ymm3[0,1],mem[2],ymm3[3,4],mem[5],ymm3[6,7]
+; AVX2-NEXT:    vpermps %ymm3, %ymm12, %ymm3
+; AVX2-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = mem[0,1,0,3,4,5,4,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = ymm4[0,1],mem[2],ymm4[3,4],mem[5],ymm4[6,7]
-; AVX2-NEXT:    vpermps %ymm4, %ymm14, %ymm4
-; AVX2-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = mem[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = mem[0,1],ymm4[2],mem[3,4],ymm4[5],mem[6,7]
+; AVX2-NEXT:    vpermps %ymm4, %ymm12, %ymm4
+; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm14[0,1,0,3,4,5,4,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1],ymm7[2],ymm6[3,4],ymm7[5],ymm6[6,7]
-; AVX2-NEXT:    vpermps %ymm5, %ymm14, %ymm5
-; AVX2-NEXT:    vshufps {{.*#+}} ymm6 = ymm8[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = ymm5[0,1],mem[2],ymm5[3,4],mem[5],ymm5[6,7]
+; AVX2-NEXT:    vpermps %ymm5, %ymm12, %ymm5
+; AVX2-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm6 = mem[0,1,0,3,4,5,4,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm6 = ymm6[0,1],mem[2],ymm6[3,4],mem[5],ymm6[6,7]
-; AVX2-NEXT:    vpermps %ymm6, %ymm14, %ymm6
-; AVX2-NEXT:    vpermilps $196, (%rsp), %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = mem[0,1,0,3,4,5,4,7]
+; AVX2-NEXT:    vpermps %ymm6, %ymm12, %ymm6
+; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[0,1,0,3,4,5,4,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5,6,7]
-; AVX2-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = mem[0,1],ymm11[2],mem[3,4],ymm11[5],mem[6,7]
-; AVX2-NEXT:    vpermps %ymm7, %ymm14, %ymm7
-; AVX2-NEXT:    vshufps {{.*#+}} ymm8 = ymm9[0,1,0,3,4,5,4,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 192(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 128(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 64(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, (%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 224(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 160(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 96(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 32(%rsi)
-; AVX2-NEXT:    vmovaps %ymm0, 192(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 128(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 64(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, (%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 224(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 160(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 96(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 32(%rdx)
-; AVX2-NEXT:    vmovaps %ymm7, 192(%rcx)
-; AVX2-NEXT:    vmovaps %ymm6, 224(%rcx)
-; AVX2-NEXT:    vmovaps %ymm5, 128(%rcx)
-; AVX2-NEXT:    vmovaps %ymm4, 160(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 192(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 128(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 64(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, (%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 224(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 160(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 96(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 32(%rsi)
+; AVX2-NEXT:    vmovaps %ymm15, 192(%rdx)
+; AVX2-NEXT:    vmovaps %ymm13, 128(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 64(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, (%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 224(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 160(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 96(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 32(%rdx)
+; AVX2-NEXT:    vmovaps %ymm6, 192(%rcx)
+; AVX2-NEXT:    vmovaps %ymm5, 224(%rcx)
+; AVX2-NEXT:    vmovaps %ymm4, 128(%rcx)
+; AVX2-NEXT:    vmovaps %ymm3, 160(%rcx)
 ; AVX2-NEXT:    vmovaps %ymm2, 64(%rcx)
-; AVX2-NEXT:    vmovaps %ymm3, 96(%rcx)
-; AVX2-NEXT:    vmovaps %ymm1, (%rcx)
-; AVX2-NEXT:    vmovaps %ymm10, 32(%rcx)
+; AVX2-NEXT:    vmovaps %ymm1, 96(%rcx)
+; AVX2-NEXT:    vmovaps %ymm0, (%rcx)
+; AVX2-NEXT:    vmovaps %ymm9, 32(%rcx)
 ; AVX2-NEXT:    addq $1032, %rsp # imm = 0x408
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
@@ -3260,245 +3146,229 @@ define void @load_i32_stride3_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-LABEL: load_i32_stride3_vf64:
 ; AVX2-FP:       # %bb.0:
 ; AVX2-FP-NEXT:    subq $1032, %rsp # imm = 0x408
+; AVX2-FP-NEXT:    vmovaps 736(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 704(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 512(%rdi), %ymm4
+; AVX2-FP-NEXT:    vmovaps 672(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 480(%rdi), %ymm6
-; AVX2-FP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 544(%rdi), %ymm5
 ; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm7
+; AVX2-FP-NEXT:    vmovaps 512(%rdi), %ymm6
+; AVX2-FP-NEXT:    vmovups %ymm6, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 480(%rdi), %ymm7
 ; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 288(%rdi), %ymm10
+; AVX2-FP-NEXT:    vmovaps 352(%rdi), %ymm8
+; AVX2-FP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm10
 ; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 352(%rdi), %ymm9
+; AVX2-FP-NEXT:    vmovaps 288(%rdi), %ymm11
+; AVX2-FP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm9
 ; AVX2-FP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm13
-; AVX2-FP-NEXT:    vmovaps 96(%rdi), %ymm14
-; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 96(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [2,5,2,5,2,5,2,5]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6],ymm0[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1],ymm9[2],ymm0[3,4],ymm9[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm0 = [0,3,6,1,4,7,2,5]
 ; AVX2-FP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm14[0],ymm13[1],ymm14[2,3],ymm13[4],ymm14[5,6],ymm13[7]
-; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm8 = [0,3,6,1,4,7,u,u]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm8, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm9, %ymm0, %ymm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm10[0],ymm7[1],ymm10[2,3],ymm7[4],ymm10[5,6],ymm7[7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm8, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm0, %ymm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm6[0],ymm4[1],ymm6[2,3],ymm4[4],ymm6[5,6],ymm4[7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm8, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm11[0],ymm10[1],ymm11[2,3],ymm10[4],ymm11[5,6],ymm10[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm8[2],ymm1[3,4],ymm8[5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 672(%rdi), %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0],ymm6[1],ymm7[2,3],ymm6[4],ymm7[5,6],ymm6[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm5[2],ymm1[3,4],ymm5[5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6],ymm3[7]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm8, %ymm1
-; AVX2-FP-NEXT:    vmovaps 736(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm3
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovaps %ymm3, %ymm5
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm8, %ymm1
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6],ymm3[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm2[2],ymm4[3,4],ymm2[5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovaps %ymm2, %ymm8
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm4
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm7
-; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm15
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm15[0],ymm7[1],ymm15[2,3],ymm7[4],ymm15[5,6],ymm7[7]
-; AVX2-FP-NEXT:    vmovups %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm8, %ymm1
-; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm6
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 416(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 384(%rdi), %ymm3
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0],ymm1[1],ymm3[2,3],ymm1[4],ymm3[5,6],ymm1[7]
-; AVX2-FP-NEXT:    vmovaps %ymm3, %ymm6
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm8, %ymm1
-; AVX2-FP-NEXT:    vmovaps 448(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm12
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm0, %ymm4
+; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 608(%rdi), %ymm9
-; AVX2-FP-NEXT:    vmovaps 576(%rdi), %ymm11
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm11[0],ymm9[1],ymm11[2,3],ymm9[4],ymm11[5,6],ymm9[7]
-; AVX2-FP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm8, %ymm1
-; AVX2-FP-NEXT:    vmovaps 640(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6],ymm2[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm1[2],ymm6[3,4],ymm1[5],ymm6[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm6, %ymm0, %ymm6
+; AVX2-FP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 448(%rdi), %ymm13
+; AVX2-FP-NEXT:    vmovaps 416(%rdi), %ymm12
+; AVX2-FP-NEXT:    vmovaps 384(%rdi), %ymm14
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm14[0],ymm12[1],ymm14[2,3],ymm12[4],ymm14[5,6],ymm12[7]
+; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1],ymm13[2],ymm10[3,4],ymm13[5],ymm10[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm10, %ymm0, %ymm10
+; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 640(%rdi), %ymm6
+; AVX2-FP-NEXT:    vmovaps 608(%rdi), %ymm5
+; AVX2-FP-NEXT:    vmovaps 576(%rdi), %ymm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm7[0],ymm5[1],ymm7[2,3],ymm5[4],ymm7[5,6],ymm5[7]
+; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm6[2],ymm15[3,4],ymm6[5],ymm15[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm15, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1],ymm4[2],ymm9[3,4],ymm4[5],ymm9[6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0],ymm0[1,2],ymm11[3],ymm0[4,5],ymm11[6],ymm0[7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm15 = [1,4,7,2,5,0,3,6]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
+; AVX2-FP-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
+; AVX2-FP-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
+; AVX2-FP-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1],ymm13[2],ymm14[3,4],ymm13[5],ymm14[6,7]
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm1 = [1,4,7,2,5,u,u,u]
-; AVX2-FP-NEXT:    vpermps %ymm0, %ymm1, %ymm2
-; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,0,3,6,0,0,3,6]
-; AVX2-FP-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm10, %ymm0, %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm12[2],mem[3,4],ymm12[5],mem[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm13, %ymm0, %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm2[2],mem[3,4],ymm2[5],mem[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm2[2],mem[3,4],ymm2[5],mem[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FP-NEXT:    vpermps (%rsp), %ymm0, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps %ymm5, %ymm4
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0],ymm0[1,2],ymm8[3],ymm0[4,5],ymm8[6],ymm0[7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm5[2],ymm3[3,4],ymm5[5],ymm3[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm0, %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm15[0,1],ymm7[2],ymm15[3,4],ymm7[5],ymm15[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm15, %ymm0, %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0,1],ymm3[2],ymm8[3,4],ymm3[5],ymm8[6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0],ymm0[1,2],ymm10[3],ymm0[4,5],ymm10[6],ymm0[7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1],ymm12[2],ymm14[3,4],ymm12[5],ymm14[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm13[0],ymm0[1,2],ymm13[3],ymm0[4,5],ymm13[6],ymm0[7]
+; AVX2-FP-NEXT:    vmovaps %ymm13, %ymm14
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm13
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm7[0,1],ymm5[2],ymm7[3,4],ymm5[5],ymm7[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0],ymm0[1,2],ymm6[3],ymm0[4,5],ymm6[6],ymm0[7]
 ; AVX2-FP-NEXT:    vmovaps %ymm6, %ymm7
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm7[0,1],ymm6[2],ymm7[3,4],ymm6[5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm8, %ymm0, %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm11[0,1],ymm9[2],ymm11[3,4],ymm9[5],ymm11[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm1, %ymm1
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm9, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm15, %ymm15
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm4[0,1],ymm9[2],ymm4[3,4],ymm9[5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm12 = [2,5,0,3,6,u,u,u]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm12, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm9 = ymm11[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm9 = ymm9[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm0[0,1,2,3,4],ymm9[5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm12, %ymm0
+; AVX2-FP-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,1],ymm1[2],mem[3,4],ymm1[5],mem[6,7]
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm14 = [2,5,0,3,6,u,u,u]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm14, %ymm1
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm10[0,1,0,3,4,5,4,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1],ymm3[2],ymm4[3,4],ymm3[5],ymm4[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm14, %ymm1
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm5[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm12, %ymm1
+; AVX2-FP-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = mem[0,1,0,3,4,5,4,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm12[0,1],mem[2],ymm12[3,4],mem[5],ymm12[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm14, %ymm2
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm13[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm8[2],ymm3[3,4],ymm8[5],ymm3[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm2, %ymm12, %ymm2
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm10[0,1,0,3,4,5,4,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm2[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm2[2],mem[3,4],ymm2[5],mem[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm14, %ymm2
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm15[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = ymm3[0,1],mem[2],ymm3[3,4],mem[5],ymm3[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm12, %ymm3
+; AVX2-FP-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = mem[0,1,0,3,4,5,4,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = ymm4[0,1],mem[2],ymm4[3,4],mem[5],ymm4[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm14, %ymm4
-; AVX2-FP-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = mem[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = mem[0,1],ymm4[2],mem[3,4],ymm4[5],mem[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm12, %ymm4
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm14[0,1,0,3,4,5,4,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1],ymm7[2],ymm6[3,4],ymm7[5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm14, %ymm5
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm6 = ymm8[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = ymm5[0,1],mem[2],ymm5[3,4],mem[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm12, %ymm5
+; AVX2-FP-NEXT:    vpermilps $196, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm6 = mem[0,1,0,3,4,5,4,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm6 = ymm6[0,1],mem[2],ymm6[3,4],mem[5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm6, %ymm14, %ymm6
-; AVX2-FP-NEXT:    vpermilps $196, (%rsp), %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = mem[0,1,0,3,4,5,4,7]
+; AVX2-FP-NEXT:    vpermps %ymm6, %ymm12, %ymm6
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[0,1,0,3,4,5,4,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5,6,7]
-; AVX2-FP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = mem[0,1],ymm11[2],mem[3,4],ymm11[5],mem[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm7, %ymm14, %ymm7
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm8 = ymm9[0,1,0,3,4,5,4,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 192(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 128(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 64(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, (%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 224(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 160(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 96(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 32(%rsi)
-; AVX2-FP-NEXT:    vmovaps %ymm0, 192(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 128(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 64(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, (%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 224(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 160(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 96(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rdx)
-; AVX2-FP-NEXT:    vmovaps %ymm7, 192(%rcx)
-; AVX2-FP-NEXT:    vmovaps %ymm6, 224(%rcx)
-; AVX2-FP-NEXT:    vmovaps %ymm5, 128(%rcx)
-; AVX2-FP-NEXT:    vmovaps %ymm4, 160(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 192(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 128(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 64(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, (%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 224(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 160(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 96(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%rsi)
+; AVX2-FP-NEXT:    vmovaps %ymm15, 192(%rdx)
+; AVX2-FP-NEXT:    vmovaps %ymm13, 128(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 64(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, (%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 224(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 160(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 96(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%rdx)
+; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%rcx)
+; AVX2-FP-NEXT:    vmovaps %ymm5, 224(%rcx)
+; AVX2-FP-NEXT:    vmovaps %ymm4, 128(%rcx)
+; AVX2-FP-NEXT:    vmovaps %ymm3, 160(%rcx)
 ; AVX2-FP-NEXT:    vmovaps %ymm2, 64(%rcx)
-; AVX2-FP-NEXT:    vmovaps %ymm3, 96(%rcx)
-; AVX2-FP-NEXT:    vmovaps %ymm1, (%rcx)
-; AVX2-FP-NEXT:    vmovaps %ymm10, 32(%rcx)
+; AVX2-FP-NEXT:    vmovaps %ymm1, 96(%rcx)
+; AVX2-FP-NEXT:    vmovaps %ymm0, (%rcx)
+; AVX2-FP-NEXT:    vmovaps %ymm9, 32(%rcx)
 ; AVX2-FP-NEXT:    addq $1032, %rsp # imm = 0x408
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
@@ -3506,233 +3376,214 @@ define void @load_i32_stride3_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-LABEL: load_i32_stride3_vf64:
 ; AVX2-FCP:       # %bb.0:
 ; AVX2-FCP-NEXT:    subq $1032, %rsp # imm = 0x408
+; AVX2-FCP-NEXT:    vmovaps 736(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 704(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 512(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vmovaps 672(%rdi), %ymm4
 ; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 480(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vmovaps 544(%rdi), %ymm5
 ; AVX2-FCP-NEXT:    vmovups %ymm5, (%rsp) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 544(%rdi), %ymm7
+; AVX2-FCP-NEXT:    vmovaps 512(%rdi), %ymm6
+; AVX2-FCP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 480(%rdi), %ymm7
 ; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vmovaps 352(%rdi), %ymm8
 ; AVX2-FCP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 352(%rdi), %ymm10
+; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm9
+; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm10
 ; AVX2-FCP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %ymm14
-; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [2,5,2,5,2,5,2,5]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm14[0],ymm11[1],ymm14[2,3],ymm11[4],ymm14[5,6],ymm11[7]
-; AVX2-FCP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm11
 ; AVX2-FCP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm6 = [0,3,6,1,4,7,u,u]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm6, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %ymm13
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm13[0],ymm1[1],ymm13[2,3],ymm1[4],ymm13[5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovaps %ymm1, %ymm14
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1],ymm11[2],ymm0[3,4],ymm11[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm0 = [0,3,6,1,4,7,2,5]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0],ymm8[1],ymm9[2,3],ymm8[4],ymm9[5,6],ymm8[7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm6, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm10[0],ymm9[1],ymm10[2,3],ymm9[4],ymm10[5,6],ymm9[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm8[2],ymm1[3,4],ymm8[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm5[0],ymm4[1],ymm5[2,3],ymm4[4],ymm5[5,6],ymm4[7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm6, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0],ymm6[1],ymm7[2,3],ymm6[4],ymm7[5,6],ymm6[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm5[2],ymm1[3,4],ymm5[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 672(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
-; AVX2-FCP-NEXT:    vmovaps 736(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0],ymm5[1],ymm4[2,3],ymm5[4],ymm4[5,6],ymm5[7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
-; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovaps %ymm2, %ymm5
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm15
-; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm3[2],ymm1[3,4],ymm3[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm15[1],ymm1[2,3],ymm15[4],ymm1[5,6],ymm15[7]
-; AVX2-FCP-NEXT:    vmovups %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
 ; AVX2-FCP-NEXT:    vmovaps 256(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm7
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm7[6,7]
+; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 416(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vmovaps %ymm3, %ymm4
+; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 384(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0],ymm1[1],ymm7[2,3],ymm1[4],ymm7[5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
-; AVX2-FCP-NEXT:    vmovaps 448(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm0, %ymm12
-; AVX2-FCP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-FCP-NEXT:    vmovaps 448(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 608(%rdi), %ymm13
-; AVX2-FCP-NEXT:    vmovaps 576(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm9[0],ymm13[1],ymm9[2,3],ymm13[4],ymm9[5,6],ymm13[7]
-; AVX2-FCP-NEXT:    vmovups %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
-; AVX2-FCP-NEXT:    vmovaps 640(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovaps 416(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vmovaps 384(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm1[2],ymm3[3,4],ymm1[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 640(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vmovaps 608(%rdi), %ymm7
+; AVX2-FCP-NEXT:    vmovaps 576(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm11[0],ymm7[1],ymm11[2,3],ymm7[4],ymm11[5,6],ymm7[7]
+; AVX2-FCP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm8[2],ymm15[3,4],ymm8[5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm0, %ymm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1],ymm11[2],ymm14[3,4],ymm11[5],ymm14[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm1 = [1,4,7,2,5,u,u,u]
-; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm2
-; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,0,3,6,0,0,3,6]
-; AVX2-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm0, %ymm14
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps %ymm14, %ymm10
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm13[0,1],ymm14[2],ymm13[3,4],ymm14[5],ymm13[6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm12[2],mem[3,4],ymm12[5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm0, %ymm14
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $219, (%rsp), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm2[2],mem[3,4],ymm2[5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm2[2],mem[3,4],ymm2[5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm0, %ymm14
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm15[2],mem[3,4],ymm15[5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm0, %ymm14
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm7[0,1],ymm15[2],ymm7[3,4],ymm15[5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm0, %ymm14
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0,1],ymm13[2],ymm9[3,4],ymm13[5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vmovaps %ymm9, %ymm13
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm1
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm12[0],ymm0[1,2],ymm12[3],ymm0[4,5],ymm12[6],ymm0[7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm15 = [1,4,7,2,5,0,3,6]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps %ymm9, %ymm14
+; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm9[2],mem[3,4],ymm9[5],mem[6,7]
+; AVX2-FCP-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm14 = [0,1,0,3,0,1,4,7]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm14, %ymm0
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm2[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm5[0,1],ymm4[2],ymm5[3,4],ymm4[5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm14, %ymm2
-; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm0 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm0 = ymm12[0,1],mem[2],ymm12[3,4],mem[5],ymm12[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm14, %ymm2
-; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm0 = ymm0[0,1],mem[2],ymm0[3,4],mem[5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm14, %ymm2
-; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm0[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
+; AVX2-FCP-NEXT:    vblendps $73, (%rsp), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $36, (%rsp), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm0 = ymm0[0,1],mem[2],ymm0[3,4],mem[5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm15[0,1],ymm7[2],ymm15[3,4],ymm7[5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm7 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm7[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm7 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm7 = ymm6[0,1],mem[2],ymm6[3,4],mem[5],ymm6[6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm8 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm1, %ymm7
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm14, %ymm6
-; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm8 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm8 = mem[0,1],ymm13[2],mem[3,4],ymm13[5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm1, %ymm1
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm6[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 192(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 128(%rsi)
+; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4],ymm0[5],mem[6,7]
+; AVX2-FCP-NEXT:    vblendps $73, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0],ymm0[1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm5[0,1],ymm1[2],ymm5[3,4],ymm1[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0],ymm0[1,2],ymm2[3],ymm0[4,5],ymm2[6],ymm0[7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 64(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 224(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 160(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 96(%rsi)
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm4[0],ymm0[1,2],ymm4[3],ymm0[4,5],ymm4[6],ymm0[7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 192(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 128(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 64(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 224(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 160(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 96(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 32(%rdx)
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 192(%rcx)
-; AVX2-FCP-NEXT:    vmovaps %ymm7, 224(%rcx)
-; AVX2-FCP-NEXT:    vmovaps %ymm0, 128(%rcx)
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 160(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1],ymm5[2],ymm6[3,4],ymm5[5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0],ymm0[1,2],ymm9[3],ymm0[4,5],ymm9[6],ymm0[7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1],ymm7[2],ymm11[3,4],ymm7[5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0],ymm0[1,2],ymm8[3],ymm0[4,5],ymm8[6],ymm0[7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm15, %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0,1],ymm13[2],ymm10[3,4],ymm13[5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm12[1],ymm0[2,3],ymm12[4],ymm0[5,6],ymm12[7]
+; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm8 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm8 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm8[0],ymm2[1],ymm8[2,3],ymm2[4],ymm8[5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm8 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm8 = ymm14[0,1],mem[2],ymm14[3,4],mem[5],ymm14[6,7]
+; AVX2-FCP-NEXT:    vblendps $146, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm8 = ymm8[0],mem[1],ymm8[2,3],mem[4],ymm8[5,6],mem[7]
+; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm3[2],mem[3,4],ymm3[5],mem[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm4[1],ymm2[2,3],ymm4[4],ymm2[5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vblendps $146, (%rsp), %ymm7, %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = ymm7[0],mem[1],ymm7[2,3],mem[4],ymm7[5,6],mem[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1],ymm6[2],ymm5[3,4],ymm6[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm9[1],ymm1[2,3],ymm9[4],ymm1[5,6],ymm9[7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm3[0,1],mem[2],ymm3[3,4],mem[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vblendps $146, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm6[0],mem[1],ymm6[2,3],mem[4],ymm6[5,6],mem[7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1],ymm3[2],mem[3,4],ymm3[5],mem[6,7]
+; AVX2-FCP-NEXT:    vblendps $146, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3],mem[4],ymm3[5,6],mem[7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm5 = [2,5,0,3,6,1,4,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm5, %ymm4
+; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm5, %ymm8
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm2
+; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm5, %ymm7
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm5, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm5, %ymm6
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm5, %ymm3
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 192(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, (%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 224(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 160(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%rsi)
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 192(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, (%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 224(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 160(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%rdx)
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 192(%rcx)
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 224(%rcx)
+; AVX2-FCP-NEXT:    vmovaps %ymm1, 128(%rcx)
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 160(%rcx)
 ; AVX2-FCP-NEXT:    vmovaps %ymm2, 64(%rcx)
-; AVX2-FCP-NEXT:    vmovaps %ymm11, 96(%rcx)
-; AVX2-FCP-NEXT:    vmovaps %ymm5, (%rcx)
-; AVX2-FCP-NEXT:    vmovaps %ymm10, 32(%rcx)
+; AVX2-FCP-NEXT:    vmovaps %ymm8, 96(%rcx)
+; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rcx)
+; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%rcx)
 ; AVX2-FCP-NEXT:    addq $1032, %rsp # imm = 0x408
 ; AVX2-FCP-NEXT:    vzeroupper
 ; AVX2-FCP-NEXT:    retq
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-5.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-5.ll
index f01aa90e3efc5b..dd94dffa859329 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-5.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-5.ll
@@ -851,14 +851,11 @@ define void @load_i32_stride5_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0,1,2,3],ymm5[4,5,6,7]
 ; AVX2-NEXT:    vinserti128 $1, 128(%rdi), %ymm6, %ymm6
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm6[3],ymm5[4,5,6],ymm6[7]
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm6 = [1,6,3,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm1[0,1],ymm2[2,3],ymm1[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm7, %ymm6, %ymm6
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm1[0,1],ymm2[2,3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]
-; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [5,2,7,0,5,2,7,0]
-; AVX2-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermd %ymm7, %ymm8, %ymm7
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm7[0],ymm6[1],ymm7[2],ymm6[3],ymm7[4,5],ymm6[6],ymm7[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm7 = [1,6,3,0,5,2,7,0]
+; AVX2-NEXT:    vpermd %ymm6, %ymm7, %ymm6
 ; AVX2-NEXT:    vpbroadcastd 144(%rdi), %ymm7
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm7[7]
 ; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [2,7,4,0]
@@ -874,10 +871,10 @@ define void @load_i32_stride5_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm8 = [1,6,1,6,1,6,1,6]
 ; AVX2-NEXT:    vpermd %ymm0, %ymm8, %ymm8
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm3[0,1,2,3],ymm4[4,5],ymm3[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[3,0,2,2,7,4,6,6]
 ; AVX2-NEXT:    vpalignr {{.*#+}} ymm10 = ymm1[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm1[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0,1,2],ymm9[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0,1],ymm9[2],ymm10[3],ymm9[4],ymm10[5,6],ymm9[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm10 = [0,1,6,2,7,4,0,0]
+; AVX2-NEXT:    vpermd %ymm9, %ymm10, %ymm9
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1,2,3,4,5],ymm8[6,7]
 ; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm3 = ymm3[0,1],ymm4[0,1]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5],ymm3[6,7]
@@ -911,14 +908,11 @@ define void @load_i32_stride5_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0,1,2,3],ymm5[4,5,6,7]
 ; AVX2-FP-NEXT:    vinserti128 $1, 128(%rdi), %ymm6, %ymm6
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm6[3],ymm5[4,5,6],ymm6[7]
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm6 = [1,6,3,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm1[0,1],ymm2[2,3],ymm1[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm7, %ymm6, %ymm6
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm1[0,1],ymm2[2,3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [5,2,7,0,5,2,7,0]
-; AVX2-FP-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermd %ymm7, %ymm8, %ymm7
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm7[0],ymm6[1],ymm7[2],ymm6[3],ymm7[4,5],ymm6[6],ymm7[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm7 = [1,6,3,0,5,2,7,0]
+; AVX2-FP-NEXT:    vpermd %ymm6, %ymm7, %ymm6
 ; AVX2-FP-NEXT:    vpbroadcastd 144(%rdi), %ymm7
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm7[7]
 ; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [2,7,4,0]
@@ -934,10 +928,10 @@ define void @load_i32_stride5_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm8 = [1,6,1,6,1,6,1,6]
 ; AVX2-FP-NEXT:    vpermd %ymm0, %ymm8, %ymm8
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm3[0,1,2,3],ymm4[4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[3,0,2,2,7,4,6,6]
 ; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm10 = ymm1[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm1[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0,1,2],ymm9[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0,1],ymm9[2],ymm10[3],ymm9[4],ymm10[5,6],ymm9[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm10 = [0,1,6,2,7,4,0,0]
+; AVX2-FP-NEXT:    vpermd %ymm9, %ymm10, %ymm9
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1,2,3,4,5],ymm8[6,7]
 ; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm3 = ymm3[0,1],ymm4[0,1]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5],ymm3[6,7]
@@ -971,14 +965,11 @@ define void @load_i32_stride5_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0,1,2,3],ymm5[4,5,6,7]
 ; AVX2-FCP-NEXT:    vinserti128 $1, 128(%rdi), %ymm6, %ymm6
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm6[3],ymm5[4,5,6],ymm6[7]
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm6 = [1,6,3,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm1[0,1],ymm2[2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm7, %ymm6, %ymm6
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm1[0,1],ymm2[2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [5,2,7,0,5,2,7,0]
-; AVX2-FCP-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermd %ymm7, %ymm8, %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm7[0],ymm6[1],ymm7[2],ymm6[3],ymm7[4,5],ymm6[6],ymm7[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm7 = [1,6,3,0,5,2,7,0]
+; AVX2-FCP-NEXT:    vpermd %ymm6, %ymm7, %ymm6
 ; AVX2-FCP-NEXT:    vpbroadcastd 144(%rdi), %ymm7
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm7[7]
 ; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [2,7,4,0]
@@ -994,10 +985,10 @@ define void @load_i32_stride5_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm8 = [1,6,1,6,1,6,1,6]
 ; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm8, %ymm8
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm3[0,1,2,3],ymm4[4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[3,0,2,2,7,4,6,6]
 ; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm10 = ymm1[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm1[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0,1,2],ymm9[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0,1],ymm9[2],ymm10[3],ymm9[4],ymm10[5,6],ymm9[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm10 = [0,1,6,2,7,4,0,0]
+; AVX2-FCP-NEXT:    vpermd %ymm9, %ymm10, %ymm9
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1,2,3,4,5],ymm8[6,7]
 ; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm3 = ymm3[0,1],ymm4[0,1]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5],ymm3[6,7]
@@ -1672,349 +1663,355 @@ define void @load_i32_stride5_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-LABEL: load_i32_stride5_vf16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    pushq %rax
-; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm0
+; AVX2-NEXT:    subq $72, %rsp
+; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm1
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm4
 ; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm6
 ; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm8
 ; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm9
-; AVX2-NEXT:    vmovdqa 224(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqa 224(%rdi), %ymm0
 ; AVX2-NEXT:    vmovdqa 256(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 192(%rdi), %ymm3
 ; AVX2-NEXT:    vmovdqa 160(%rdi), %ymm5
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [0,5,2,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm5[0,1],ymm3[2,3],ymm5[4,5],ymm3[6,7]
-; AVX2-NEXT:    vpermd %ymm10, %ymm7, %ymm10
-; AVX2-NEXT:    vinserti128 $1, 288(%rdi), %ymm10, %ymm11
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm10 = [0,5,2,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm5[0,1],ymm3[2,3],ymm5[4,5],ymm3[6,7]
+; AVX2-NEXT:    vpermd %ymm7, %ymm10, %ymm7
+; AVX2-NEXT:    vinserti128 $1, 288(%rdi), %ymm7, %ymm11
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm2[0,1,0,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3],ymm1[4],ymm12[5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3],ymm12[4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2],ymm11[3],ymm10[4,5,6],ymm11[7]
-; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm9[0,1,0,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3],ymm8[4],ymm10[5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm4[0,1],ymm6[2,3],ymm4[4,5],ymm6[6,7]
-; AVX2-NEXT:    vpermd %ymm11, %ymm7, %ymm7
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm7[0,1,2,3],ymm10[4,5,6,7]
-; AVX2-NEXT:    vinserti128 $1, 128(%rdi), %ymm7, %ymm7
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1,2],ymm7[3],ymm10[4,5,6],ymm7[7]
-; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [1,6,3,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm4[0,1],ymm6[2,3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm11, %ymm7, %ymm11
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm8[0,1],ymm9[2,3],ymm8[4,5],ymm9[6,7]
-; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [5,2,7,0,5,2,7,0]
-; AVX2-NEXT:    # ymm13 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermd %ymm12, %ymm13, %ymm12
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 144(%rdi), %ymm12
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5,6],ymm12[7]
-; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm5[0,1],ymm3[2,3],ymm5[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm12, %ymm7, %ymm7
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
-; AVX2-NEXT:    vpermd %ymm12, %ymm13, %ymm12
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 304(%rdi), %ymm12
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3],ymm0[4],ymm12[5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm12[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2],ymm11[3],ymm7[4,5,6],ymm11[7]
 ; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [2,7,4,0]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm11 = ymm9[0,1,0,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3],ymm8[4],ymm11[5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm4[0,1],ymm6[2,3],ymm4[4,5],ymm6[6,7]
+; AVX2-NEXT:    vpermd %ymm12, %ymm10, %ymm10
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm10[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-NEXT:    vinserti128 $1, 128(%rdi), %ymm10, %ymm10
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm11[0,1,2],ymm10[3],ymm11[4,5,6],ymm10[7]
+; AVX2-NEXT:    vmovdqu %ymm7, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm5[0,1],ymm3[2,3],ymm5[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-NEXT:    vmovdqa %ymm0, %ymm7
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm12[0],ymm11[1],ymm12[2],ymm11[3],ymm12[4,5],ymm11[6],ymm12[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm12 = [1,6,3,0,5,2,7,0]
+; AVX2-NEXT:    vpermd %ymm11, %ymm12, %ymm11
+; AVX2-NEXT:    vpbroadcastd 304(%rdi), %ymm13
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm11[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm4[0,1],ymm6[2,3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm14 = ymm8[0,1],ymm9[2,3],ymm8[4,5],ymm9[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm14[0],ymm13[1],ymm14[2],ymm13[3],ymm14[4,5],ymm13[6],ymm14[7]
+; AVX2-NEXT:    vpermd %ymm13, %ymm12, %ymm12
+; AVX2-NEXT:    vpbroadcastd 144(%rdi), %ymm13
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm14 = [2,7,4,0]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm4[0,1,2,3],ymm6[4,5],ymm4[6,7]
-; AVX2-NEXT:    vpermd %ymm13, %ymm7, %ymm13
-; AVX2-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm14
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm14 = ymm8[0,1,2,3,4,5,6],ymm14[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[2,3,0,1,6,7,4,5]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm14 = [0,5,0,5,0,5,0,5]
-; AVX2-NEXT:    vpermd %ymm0, %ymm14, %ymm15
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vpermd %ymm13, %ymm14, %ymm13
+; AVX2-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm15
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm8[0,1,2,3,4,5,6],ymm15[7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,3,0,1,6,7,4,5]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [0,5,0,5,0,5,0,5]
+; AVX2-NEXT:    vmovdqa %ymm1, %ymm12
+; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm15
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm5[0,1,2,3],ymm3[4,5],ymm5[6,7]
-; AVX2-NEXT:    vpermd %ymm15, %ymm7, %ymm7
+; AVX2-NEXT:    vpermd %ymm15, %ymm14, %ymm14
 ; AVX2-NEXT:    vinserti128 $1, 256(%rdi), %ymm0, %ymm15
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm1[0,1,2,3,4,5,6],ymm15[7]
+; AVX2-NEXT:    vmovdqa %ymm7, %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm7[0,1,2,3,4,5,6],ymm15[7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,3,0,1,6,7,4,5]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 288(%rdi), %ymm15
-; AVX2-NEXT:    vpermd %ymm15, %ymm14, %ymm14
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm14 = ymm7[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3],ymm9[4,5],ymm8[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[3,0,2,2,7,4,6,6]
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm10 = ymm4[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1,2],ymm7[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm10 = [1,6,1,6,1,6,1,6]
-; AVX2-NEXT:    vpermd %ymm0, %ymm10, %ymm11
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0,1,2,3],ymm2[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[3,0,2,2,7,4,6,6]
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm12 = ymm5[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm5[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm12[0,1,2],ymm11[3,4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm15, %ymm10, %ymm10
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm8 = ymm8[0,1],ymm9[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm9[5],ymm8[6,7]
+; AVX2-NEXT:    vpermd %ymm15, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm9[4,5],ymm8[6,7]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm7 = ymm4[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm0[2],ymm7[3],ymm0[4],ymm7[5,6],ymm0[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm7 = [0,1,6,2,7,4,0,0]
+; AVX2-NEXT:    vpermd %ymm0, %ymm7, %ymm2
+; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [1,6,1,6,1,6,1,6]
+; AVX2-NEXT:    vpermd %ymm12, %ymm0, %ymm10
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm2[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm1[0,1,2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm11 = ymm5[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm5[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3],ymm10[4],ymm11[5,6],ymm10[7]
+; AVX2-NEXT:    vpermd %ymm10, %ymm7, %ymm7
+; AVX2-NEXT:    vpermd %ymm15, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm7 = ymm8[0,1],ymm9[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm9[5],ymm7[6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm4[4,5],ymm6[6,7]
 ; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm6 = [4,1,6,0]
 ; AVX2-NEXT:    vpermd %ymm4, %ymm6, %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm8[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm8 = [2,7,2,7,2,7,2,7]
-; AVX2-NEXT:    vpermd %ymm0, %ymm8, %ymm0
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm7 = [2,7,2,7,2,7,2,7]
+; AVX2-NEXT:    vpermd %ymm12, %ymm7, %ymm8
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm8[6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm5[4,5],ymm3[6,7]
 ; AVX2-NEXT:    vpermd %ymm3, %ymm6, %ymm3
-; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm15, %ymm8, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm2, 32(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm2, (%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm2, 32(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm2, (%rdx)
+; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm1[0,1],ymm2[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4],ymm2[5],ymm5[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-NEXT:    vpermd %ymm15, %ymm7, %ymm3
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 32(%rsi)
+; AVX2-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, (%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 32(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm1, (%rdx)
 ; AVX2-NEXT:    vmovdqa %ymm14, 32(%rcx)
-; AVX2-NEXT:    vmovdqa %ymm13, (%rcx)
-; AVX2-NEXT:    vmovdqa %ymm10, 32(%r8)
-; AVX2-NEXT:    vmovdqa %ymm7, (%r8)
-; AVX2-NEXT:    vmovdqa %ymm1, 32(%r9)
-; AVX2-NEXT:    vmovdqa %ymm0, (%r9)
-; AVX2-NEXT:    popq %rax
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm1, (%rcx)
+; AVX2-NEXT:    vmovdqa %ymm0, 32(%r8)
+; AVX2-NEXT:    vmovdqa %ymm13, (%r8)
+; AVX2-NEXT:    vmovdqa %ymm2, 32(%r9)
+; AVX2-NEXT:    vmovdqa %ymm4, (%r9)
+; AVX2-NEXT:    addq $72, %rsp
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: load_i32_stride5_vf16:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    pushq %rax
-; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm0
+; AVX2-FP-NEXT:    subq $72, %rsp
+; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm4
 ; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm6
 ; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm8
 ; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm9
-; AVX2-FP-NEXT:    vmovdqa 224(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqa 224(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 192(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vmovdqa 160(%rdi), %ymm5
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [0,5,2,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm5[0,1],ymm3[2,3],ymm5[4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm10, %ymm7, %ymm10
-; AVX2-FP-NEXT:    vinserti128 $1, 288(%rdi), %ymm10, %ymm11
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm10 = [0,5,2,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm5[0,1],ymm3[2,3],ymm5[4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm7, %ymm10, %ymm7
+; AVX2-FP-NEXT:    vinserti128 $1, 288(%rdi), %ymm7, %ymm11
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm2[0,1,0,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3],ymm1[4],ymm12[5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3],ymm12[4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2],ymm11[3],ymm10[4,5,6],ymm11[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm9[0,1,0,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3],ymm8[4],ymm10[5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm4[0,1],ymm6[2,3],ymm4[4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm11, %ymm7, %ymm7
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm7[0,1,2,3],ymm10[4,5,6,7]
-; AVX2-FP-NEXT:    vinserti128 $1, 128(%rdi), %ymm7, %ymm7
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1,2],ymm7[3],ymm10[4,5,6],ymm7[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [1,6,3,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm4[0,1],ymm6[2,3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm11, %ymm7, %ymm11
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm8[0,1],ymm9[2,3],ymm8[4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [5,2,7,0,5,2,7,0]
-; AVX2-FP-NEXT:    # ymm13 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermd %ymm12, %ymm13, %ymm12
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 144(%rdi), %ymm12
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5,6],ymm12[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm5[0,1],ymm3[2,3],ymm5[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm12, %ymm7, %ymm7
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm12, %ymm13, %ymm12
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 304(%rdi), %ymm12
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3],ymm0[4],ymm12[5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm12[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2],ymm11[3],ymm7[4,5,6],ymm11[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [2,7,4,0]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm11 = ymm9[0,1,0,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3],ymm8[4],ymm11[5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm4[0,1],ymm6[2,3],ymm4[4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm12, %ymm10, %ymm10
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm10[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-FP-NEXT:    vinserti128 $1, 128(%rdi), %ymm10, %ymm10
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm11[0,1,2],ymm10[3],ymm11[4,5,6],ymm10[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm7, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm5[0,1],ymm3[2,3],ymm5[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm0, %ymm7
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm12[0],ymm11[1],ymm12[2],ymm11[3],ymm12[4,5],ymm11[6],ymm12[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm12 = [1,6,3,0,5,2,7,0]
+; AVX2-FP-NEXT:    vpermd %ymm11, %ymm12, %ymm11
+; AVX2-FP-NEXT:    vpbroadcastd 304(%rdi), %ymm13
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm11[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm4[0,1],ymm6[2,3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm8[0,1],ymm9[2,3],ymm8[4,5],ymm9[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm14[0],ymm13[1],ymm14[2],ymm13[3],ymm14[4,5],ymm13[6],ymm14[7]
+; AVX2-FP-NEXT:    vpermd %ymm13, %ymm12, %ymm12
+; AVX2-FP-NEXT:    vpbroadcastd 144(%rdi), %ymm13
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm14 = [2,7,4,0]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm4[0,1,2,3],ymm6[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm13, %ymm7, %ymm13
-; AVX2-FP-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm14
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm8[0,1,2,3,4,5,6],ymm14[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[2,3,0,1,6,7,4,5]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm14 = [0,5,0,5,0,5,0,5]
-; AVX2-FP-NEXT:    vpermd %ymm0, %ymm14, %ymm15
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm13, %ymm14, %ymm13
+; AVX2-FP-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm15
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm8[0,1,2,3,4,5,6],ymm15[7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,3,0,1,6,7,4,5]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [0,5,0,5,0,5,0,5]
+; AVX2-FP-NEXT:    vmovdqa %ymm1, %ymm12
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm15
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm5[0,1,2,3],ymm3[4,5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm15, %ymm7, %ymm7
+; AVX2-FP-NEXT:    vpermd %ymm15, %ymm14, %ymm14
 ; AVX2-FP-NEXT:    vinserti128 $1, 256(%rdi), %ymm0, %ymm15
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm1[0,1,2,3,4,5,6],ymm15[7]
+; AVX2-FP-NEXT:    vmovdqa %ymm7, %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm7[0,1,2,3,4,5,6],ymm15[7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,3,0,1,6,7,4,5]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %ymm15
-; AVX2-FP-NEXT:    vpermd %ymm15, %ymm14, %ymm14
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm7[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3],ymm9[4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[3,0,2,2,7,4,6,6]
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm10 = ymm4[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1,2],ymm7[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm10 = [1,6,1,6,1,6,1,6]
-; AVX2-FP-NEXT:    vpermd %ymm0, %ymm10, %ymm11
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0,1,2,3],ymm2[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[3,0,2,2,7,4,6,6]
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm12 = ymm5[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm5[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm12[0,1,2],ymm11[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm15, %ymm10, %ymm10
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm8 = ymm8[0,1],ymm9[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm9[5],ymm8[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm15, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm9[4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm7 = ymm4[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm0[2],ymm7[3],ymm0[4],ymm7[5,6],ymm0[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm7 = [0,1,6,2,7,4,0,0]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm7, %ymm2
+; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [1,6,1,6,1,6,1,6]
+; AVX2-FP-NEXT:    vpermd %ymm12, %ymm0, %ymm10
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm2[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm1[0,1,2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm11 = ymm5[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm5[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3],ymm10[4],ymm11[5,6],ymm10[7]
+; AVX2-FP-NEXT:    vpermd %ymm10, %ymm7, %ymm7
+; AVX2-FP-NEXT:    vpermd %ymm15, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm7 = ymm8[0,1],ymm9[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm9[5],ymm7[6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm4[4,5],ymm6[6,7]
 ; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm6 = [4,1,6,0]
 ; AVX2-FP-NEXT:    vpermd %ymm4, %ymm6, %ymm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm8[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm8 = [2,7,2,7,2,7,2,7]
-; AVX2-FP-NEXT:    vpermd %ymm0, %ymm8, %ymm0
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm7 = [2,7,2,7,2,7,2,7]
+; AVX2-FP-NEXT:    vpermd %ymm12, %ymm7, %ymm8
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm8[6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm5[4,5],ymm3[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm3, %ymm6, %ymm3
-; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm15, %ymm8, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm2, 32(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm2, (%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm2, 32(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm2, (%rdx)
+; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm1[0,1],ymm2[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4],ymm2[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermd %ymm15, %ymm7, %ymm3
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rsi)
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, (%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm1, (%rdx)
 ; AVX2-FP-NEXT:    vmovdqa %ymm14, 32(%rcx)
-; AVX2-FP-NEXT:    vmovdqa %ymm13, (%rcx)
-; AVX2-FP-NEXT:    vmovdqa %ymm10, 32(%r8)
-; AVX2-FP-NEXT:    vmovdqa %ymm7, (%r8)
-; AVX2-FP-NEXT:    vmovdqa %ymm1, 32(%r9)
-; AVX2-FP-NEXT:    vmovdqa %ymm0, (%r9)
-; AVX2-FP-NEXT:    popq %rax
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm1, (%rcx)
+; AVX2-FP-NEXT:    vmovdqa %ymm0, 32(%r8)
+; AVX2-FP-NEXT:    vmovdqa %ymm13, (%r8)
+; AVX2-FP-NEXT:    vmovdqa %ymm2, 32(%r9)
+; AVX2-FP-NEXT:    vmovdqa %ymm4, (%r9)
+; AVX2-FP-NEXT:    addq $72, %rsp
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
 ; AVX2-FCP-LABEL: load_i32_stride5_vf16:
 ; AVX2-FCP:       # %bb.0:
-; AVX2-FCP-NEXT:    pushq %rax
-; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm0
+; AVX2-FCP-NEXT:    subq $72, %rsp
+; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovdqa (%rdi), %ymm4
 ; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %ymm6
 ; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm8
 ; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vmovdqa 224(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 224(%rdi), %ymm0
 ; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 192(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqa 160(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [0,5,2,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm5[0,1],ymm3[2,3],ymm5[4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm10, %ymm7, %ymm10
-; AVX2-FCP-NEXT:    vinserti128 $1, 288(%rdi), %ymm10, %ymm11
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm10 = [0,5,2,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm5[0,1],ymm3[2,3],ymm5[4,5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm7, %ymm10, %ymm7
+; AVX2-FCP-NEXT:    vinserti128 $1, 288(%rdi), %ymm7, %ymm11
 ; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm12 = ymm2[0,1,0,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3],ymm1[4],ymm12[5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3],ymm12[4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2],ymm11[3],ymm10[4,5,6],ymm11[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm10 = ymm9[0,1,0,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3],ymm8[4],ymm10[5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm4[0,1],ymm6[2,3],ymm4[4,5],ymm6[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm11, %ymm7, %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm7[0,1,2,3],ymm10[4,5,6,7]
-; AVX2-FCP-NEXT:    vinserti128 $1, 128(%rdi), %ymm7, %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1,2],ymm7[3],ymm10[4,5,6],ymm7[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [1,6,3,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm4[0,1],ymm6[2,3],ymm4[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm11, %ymm7, %ymm11
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm8[0,1],ymm9[2,3],ymm8[4,5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [5,2,7,0,5,2,7,0]
-; AVX2-FCP-NEXT:    # ymm13 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermd %ymm12, %ymm13, %ymm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 144(%rdi), %ymm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5,6],ymm12[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm5[0,1],ymm3[2,3],ymm5[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm12, %ymm7, %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm12, %ymm13, %ymm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 304(%rdi), %ymm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3],ymm0[4],ymm12[5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm12[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2],ymm11[3],ymm7[4,5,6],ymm11[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [2,7,4,0]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm11 = ymm9[0,1,0,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3],ymm8[4],ymm11[5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm4[0,1],ymm6[2,3],ymm4[4,5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm12, %ymm10, %ymm10
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm10[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-FCP-NEXT:    vinserti128 $1, 128(%rdi), %ymm10, %ymm10
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm11[0,1,2],ymm10[3],ymm11[4,5,6],ymm10[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm7, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm5[0,1],ymm3[2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm0, %ymm7
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm12[0],ymm11[1],ymm12[2],ymm11[3],ymm12[4,5],ymm11[6],ymm12[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm12 = [1,6,3,0,5,2,7,0]
+; AVX2-FCP-NEXT:    vpermd %ymm11, %ymm12, %ymm11
+; AVX2-FCP-NEXT:    vpbroadcastd 304(%rdi), %ymm13
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm11[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm4[0,1],ymm6[2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm8[0,1],ymm9[2,3],ymm8[4,5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm14[0],ymm13[1],ymm14[2],ymm13[3],ymm14[4,5],ymm13[6],ymm14[7]
+; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm12, %ymm12
+; AVX2-FCP-NEXT:    vpbroadcastd 144(%rdi), %ymm13
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm14 = [2,7,4,0]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm4[0,1,2,3],ymm6[4,5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm7, %ymm13
-; AVX2-FCP-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm14
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm8[0,1,2,3,4,5,6],ymm14[7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[2,3,0,1,6,7,4,5]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm14 = [0,5,0,5,0,5,0,5]
-; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm14, %ymm15
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm14, %ymm13
+; AVX2-FCP-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm15
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm8[0,1,2,3,4,5,6],ymm15[7]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,3,0,1,6,7,4,5]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [0,5,0,5,0,5,0,5]
+; AVX2-FCP-NEXT:    vmovdqa %ymm1, %ymm12
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm15
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm5[0,1,2,3],ymm3[4,5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm7, %ymm7
+; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm14, %ymm14
 ; AVX2-FCP-NEXT:    vinserti128 $1, 256(%rdi), %ymm0, %ymm15
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm1[0,1,2,3,4,5,6],ymm15[7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm7, %ymm1
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm7[0,1,2,3,4,5,6],ymm15[7]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,3,0,1,6,7,4,5]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %ymm15
-; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm14, %ymm14
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm7[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3],ymm9[4,5],ymm8[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[3,0,2,2,7,4,6,6]
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm10 = ymm4[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1,2],ymm7[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm10 = [1,6,1,6,1,6,1,6]
-; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm10, %ymm11
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0,1,2,3],ymm2[4,5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[3,0,2,2,7,4,6,6]
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm12 = ymm5[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm5[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm12[0,1,2],ymm11[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm10, %ymm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm8 = ymm8[0,1],ymm9[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm9[5],ymm8[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm9[4,5],ymm8[6,7]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm7 = ymm4[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm0[2],ymm7[3],ymm0[4],ymm7[5,6],ymm0[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm7 = [0,1,6,2,7,4,0,0]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm7, %ymm2
+; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [1,6,1,6,1,6,1,6]
+; AVX2-FCP-NEXT:    vpermd %ymm12, %ymm0, %ymm10
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm2[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm1[0,1,2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm11 = ymm5[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm5[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3],ymm10[4],ymm11[5,6],ymm10[7]
+; AVX2-FCP-NEXT:    vpermd %ymm10, %ymm7, %ymm7
+; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm7 = ymm8[0,1],ymm9[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm9[5],ymm7[6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm4[4,5],ymm6[6,7]
 ; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm6 = [4,1,6,0]
 ; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm6, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm8[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm8 = [2,7,2,7,2,7,2,7]
-; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm8, %ymm0
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm7 = [2,7,2,7,2,7,2,7]
+; AVX2-FCP-NEXT:    vpermd %ymm12, %ymm7, %ymm8
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm8[6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm5[4,5],ymm3[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm6, %ymm3
-; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm8, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 32(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, (%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 32(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, (%rdx)
+; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm1[0,1],ymm2[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4],ymm2[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm7, %ymm3
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rsi)
+; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm1, (%rdx)
 ; AVX2-FCP-NEXT:    vmovdqa %ymm14, 32(%rcx)
-; AVX2-FCP-NEXT:    vmovdqa %ymm13, (%rcx)
-; AVX2-FCP-NEXT:    vmovdqa %ymm10, 32(%r8)
-; AVX2-FCP-NEXT:    vmovdqa %ymm7, (%r8)
-; AVX2-FCP-NEXT:    vmovdqa %ymm1, 32(%r9)
-; AVX2-FCP-NEXT:    vmovdqa %ymm0, (%r9)
-; AVX2-FCP-NEXT:    popq %rax
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm1, (%rcx)
+; AVX2-FCP-NEXT:    vmovdqa %ymm0, 32(%r8)
+; AVX2-FCP-NEXT:    vmovdqa %ymm13, (%r8)
+; AVX2-FCP-NEXT:    vmovdqa %ymm2, 32(%r9)
+; AVX2-FCP-NEXT:    vmovdqa %ymm4, (%r9)
+; AVX2-FCP-NEXT:    addq $72, %rsp
 ; AVX2-FCP-NEXT:    vzeroupper
 ; AVX2-FCP-NEXT:    retq
 ;
@@ -3354,33 +3351,31 @@ define void @load_i32_stride5_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-LABEL: load_i32_stride5_vf32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    subq $1000, %rsp # imm = 0x3E8
-; AVX2-NEXT:    vmovdqa 384(%rdi), %ymm14
+; AVX2-NEXT:    subq $968, %rsp # imm = 0x3C8
+; AVX2-NEXT:    vmovdqa 384(%rdi), %ymm4
 ; AVX2-NEXT:    vmovdqa 416(%rdi), %ymm5
-; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 352(%rdi), %ymm6
-; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 320(%rdi), %ymm7
-; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 544(%rdi), %ymm8
+; AVX2-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 576(%rdi), %ymm9
 ; AVX2-NEXT:    vmovdqa 512(%rdi), %ymm10
-; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu %ymm10, (%rsp) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 480(%rdi), %ymm15
 ; AVX2-NEXT:    vmovdqa 224(%rdi), %ymm13
-; AVX2-NEXT:    vmovdqa 256(%rdi), %ymm11
+; AVX2-NEXT:    vmovdqa 256(%rdi), %ymm14
 ; AVX2-NEXT:    vmovdqa 192(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 160(%rdi), %ymm1
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,5,2,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
+; AVX2-NEXT:    vmovdqa %ymm2, %ymm11
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vinserti128 $1, 288(%rdi), %ymm1, %ymm2
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm11[0,1,0,3]
-; AVX2-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm14[0,1,0,3]
+; AVX2-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm13[4],ymm3[5,6,7]
-; AVX2-NEXT:    vmovdqa %ymm13, %ymm12
+; AVX2-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3],ymm1[4,5,6],ymm2[7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -3388,105 +3383,99 @@ define void @load_i32_stride5_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm9[0,1,0,3]
-; AVX2-NEXT:    vmovdqa %ymm9, %ymm13
 ; AVX2-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm8[4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovdqa %ymm8, %ymm10
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-NEXT:    vinserti128 $1, 608(%rdi), %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm6[2,3],ymm7[4,5],ymm6[6,7]
+; AVX2-NEXT:    vmovdqa %ymm7, %ymm10
+; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm5[0,1,0,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm14[4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4],ymm2[5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-NEXT:    vinserti128 $1, 448(%rdi), %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm7
-; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm1[2,3],ymm7[4,5],ymm1[6,7]
+; AVX2-NEXT:    vmovdqa (%rdi), %ymm2
+; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm3
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm3[2,3],ymm2[4,5],ymm3[6,7]
+; AVX2-NEXT:    vmovdqa %ymm3, %ymm8
+; AVX2-NEXT:    vmovdqa %ymm2, %ymm7
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm6
-; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm9
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm9[0,1,0,3]
-; AVX2-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm6[4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm12
+; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm12[4],ymm1[5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vinserti128 $1, 128(%rdi), %ymm0, %ymm0
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [1,6,3,0]
-; AVX2-NEXT:    vmovdqu (%rsp), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm8[2,3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm11[2,3],ymm12[4,5],ymm11[6,7]
-; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [5,2,7,0,5,2,7,0]
-; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm11[2,3],ymm3[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1],ymm14[2,3],ymm13[4,5],ymm14[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [1,6,3,0,5,2,7,0]
+; AVX2-NEXT:    vpermd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpbroadcastd 304(%rdi), %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm15[0,1],ymm5[2,3],ymm15[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0,1],ymm13[2,3],ymm10[4,5],ymm13[6,7]
-; AVX2-NEXT:    vmovdqa %ymm10, %ymm11
-; AVX2-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu (%rsp), %ymm14 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm14[2,3],ymm15[4,5,6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1],ymm9[2,3],ymm15[4,5],ymm9[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2],ymm0[3],ymm2[4,5],ymm0[6],ymm2[7]
+; AVX2-NEXT:    vpermd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpbroadcastd 624(%rdi), %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm6[2,3],ymm10[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm5[2,3],ymm4[4,5],ymm5[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2],ymm0[3],ymm2[4,5],ymm0[6],ymm2[7]
+; AVX2-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpbroadcastd 464(%rdi), %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm10[2,3],ymm7[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm9[2,3],ymm6[4,5],ymm9[6,7]
-; AVX2-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 144(%rdi), %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,1],ymm15[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = ymm14[0,1],mem[2,3],ymm14[4,5],mem[6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm3, %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 464(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm8[2,3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vmovdqa %ymm12, %ymm9
+; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = ymm12[0,1],mem[2,3],ymm12[4,5],mem[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2],ymm0[3],ymm2[4,5],ymm0[6],ymm2[7]
+; AVX2-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpbroadcastd 144(%rdi), %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm9 = [2,7,4,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm8[4,5],ymm4[6,7]
-; AVX2-NEXT:    vmovdqa %ymm8, %ymm14
-; AVX2-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [2,7,4,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm11[4,5],ymm3[6,7]
+; AVX2-NEXT:    vmovdqa %ymm3, %ymm12
+; AVX2-NEXT:    vmovdqa %ymm11, %ymm13
+; AVX2-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-NEXT:    vinserti128 $1, 256(%rdi), %ymm0, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-NEXT:    vmovdqa %ymm12, %ymm8
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqa 288(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqa 288(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [0,5,0,5,0,5,0,5]
-; AVX2-NEXT:    vpermd %ymm3, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa %ymm3, %ymm12
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm5[4,5],ymm3[6,7]
-; AVX2-NEXT:    vmovdqa %ymm5, %ymm13
-; AVX2-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm14[4,5],ymm4[6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-NEXT:    vinserti128 $1, 576(%rdi), %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa %ymm11, %ymm4
-; AVX2-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm11[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqa %ymm15, %ymm14
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 608(%rdi), %ymm2
@@ -3494,112 +3483,110 @@ define void @load_i32_stride5_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3],ymm10[4,5],ymm7[6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3],ymm8[4,5],ymm7[6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa %ymm6, %ymm7
-; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm6
-; AVX2-NEXT:    vpermd %ymm6, %ymm0, %ymm10
-; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm7
+; AVX2-NEXT:    vpermd %ymm7, %ymm0, %ymm10
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm10[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm15[4,5],ymm2[6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm2[4,5],ymm3[6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm5, %ymm8
 ; AVX2-NEXT:    vinserti128 $1, 416(%rdi), %ymm0, %ymm5
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm9[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm1[0,1,2,3,4,5,6],ymm5[7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[2,3,0,1,6,7,4,5]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm1[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqa 448(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm8[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-NEXT:    vmovdqa 448(%rdi), %ymm8
+; AVX2-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermd %ymm8, %ymm0, %ymm0
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm1[4,5],ymm8[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm5 = ymm11[12,13,14,15],ymm14[0,1,2,3,4,5,6,7,8,9,10,11],ymm11[28,29,30,31],ymm14[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [1,6,1,6,1,6,1,6]
-; AVX2-NEXT:    vpermd %ymm12, %ymm5, %ymm10
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm11[4,5],ymm6[6,7]
+; AVX2-NEXT:    vmovdqa %ymm12, %ymm6
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm5 = ymm12[12,13,14,15],ymm13[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm13[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1],ymm0[2],ymm5[3],ymm0[4],ymm5[5,6],ymm0[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm5 = [0,1,6,2,7,4,0,0]
+; AVX2-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm10 = [1,6,1,6,1,6,1,6]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm12[4,5],ymm4[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm10 = ymm3[12,13,14,15],ymm13[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm13[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpermd %ymm3, %ymm5, %ymm10
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-NEXT:    vpermd %ymm12, %ymm10, %ymm15
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1,2,3],ymm13[4,5],ymm7[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm10 = ymm14[12,13,14,15],ymm15[0,1,2,3,4,5,6,7,8,9,10,11],ymm14[28,29,30,31],ymm15[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm6, %ymm5, %ymm10
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm14[0,1,2,3],ymm13[4,5],ymm14[6,7]
+; AVX2-NEXT:    vmovdqu (%rsp), %ymm8 # 32-byte Reload
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm4[12,13,14,15],ymm8[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm8[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3],ymm0[4],ymm15[5,6],ymm0[7]
+; AVX2-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm0 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm0 = ymm9[0,1,2,3],mem[4,5],ymm9[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm15 = mem[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3],ymm0[4],ymm15[5,6],ymm0[7]
+; AVX2-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vpermd %ymm7, %ymm10, %ymm15
+; AVX2-NEXT:    vmovdqa %ymm7, %ymm14
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm10 = ymm2[12,13,14,15],ymm7[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm7[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vpermd %ymm9, %ymm5, %ymm5
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm7[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3],ymm0[4],ymm15[5,6],ymm0[7]
+; AVX2-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-NEXT:    vpermd %ymm15, %ymm10, %ymm5
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd $207, (%rsp), %ymm11, %ymm0 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm11[4,5],mem[6,7]
-; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm8[0,1],ymm1[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm1[5],ymm5[6,7]
+; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm6[4,5],mem[6,7]
+; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = mem[0,1],ymm11[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm11[5],ymm5[6,7]
 ; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm10 = [4,1,6,0]
 ; AVX2-NEXT:    vpermd %ymm0, %ymm10, %ymm0
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm5[3,4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [2,7,2,7,2,7,2,7]
-; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    vpermd %ymm12, %ymm5, %ymm4
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
-; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm11 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm11 = mem[0,1],ymm12[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm12[5],ymm11[6,7]
+; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = ymm8[0,1,2,3],mem[4,5],ymm8[6,7]
+; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm11 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm11 = mem[0,1],ymm13[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm13[5],ymm11[6,7]
 ; AVX2-NEXT:    vpermd %ymm4, %ymm10, %ymm4
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm11[3,4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm3, %ymm5, %ymm3
+; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm3 # 32-byte Folded Reload
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,1],ymm13[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm13[5],ymm4[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm15[0,1,2,3],ymm14[4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm4 = ymm9[0,1],ymm1[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm1[5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm6 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm6 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
 ; AVX2-NEXT:    vpermd %ymm6, %ymm10, %ymm6
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    vpermd %ymm14, %ymm5, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = ymm7[0,1,2,3],mem[4,5],ymm7[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
 ; AVX2-NEXT:    vpermd %ymm4, %ymm10, %ymm4
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
 ; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm6 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm6 = mem[0,1],ymm7[0,1]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5],ymm6[6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm6[3,4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm9, %ymm5, %ymm1
+; AVX2-NEXT:    vpermd %ymm15, %ymm5, %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm4, 64(%rsi)
@@ -3637,39 +3624,37 @@ define void @load_i32_stride5_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqa %ymm2, (%r9)
 ; AVX2-NEXT:    vmovdqa %ymm3, 96(%r9)
 ; AVX2-NEXT:    vmovdqa %ymm0, 32(%r9)
-; AVX2-NEXT:    addq $1000, %rsp # imm = 0x3E8
+; AVX2-NEXT:    addq $968, %rsp # imm = 0x3C8
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: load_i32_stride5_vf32:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    subq $1000, %rsp # imm = 0x3E8
-; AVX2-FP-NEXT:    vmovdqa 384(%rdi), %ymm14
+; AVX2-FP-NEXT:    subq $968, %rsp # imm = 0x3C8
+; AVX2-FP-NEXT:    vmovdqa 384(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vmovdqa 416(%rdi), %ymm5
-; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 352(%rdi), %ymm6
-; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 320(%rdi), %ymm7
-; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 544(%rdi), %ymm8
+; AVX2-FP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 576(%rdi), %ymm9
 ; AVX2-FP-NEXT:    vmovdqa 512(%rdi), %ymm10
-; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu %ymm10, (%rsp) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 480(%rdi), %ymm15
 ; AVX2-FP-NEXT:    vmovdqa 224(%rdi), %ymm13
-; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %ymm11
+; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %ymm14
 ; AVX2-FP-NEXT:    vmovdqa 192(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 160(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,5,2,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm11
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vinserti128 $1, 288(%rdi), %ymm1, %ymm2
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm11[0,1,0,3]
-; AVX2-FP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm14[0,1,0,3]
+; AVX2-FP-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm13[4],ymm3[5,6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm13, %ymm12
+; AVX2-FP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3],ymm1[4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -3677,105 +3662,99 @@ define void @load_i32_stride5_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm9[0,1,0,3]
-; AVX2-FP-NEXT:    vmovdqa %ymm9, %ymm13
 ; AVX2-FP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm8[4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm8, %ymm10
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FP-NEXT:    vinserti128 $1, 608(%rdi), %ymm1, %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm6[2,3],ymm7[4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm7, %ymm10
+; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm5[0,1,0,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm14[4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4],ymm2[5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FP-NEXT:    vinserti128 $1, 448(%rdi), %ymm1, %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm7
-; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm1[2,3],ymm7[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm3
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm3[2,3],ymm2[4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm8
+; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm7
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm6
-; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm9
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm9[0,1,0,3]
-; AVX2-FP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm6[4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm12
+; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm12[4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vinserti128 $1, 128(%rdi), %ymm0, %ymm0
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [1,6,3,0]
-; AVX2-FP-NEXT:    vmovdqu (%rsp), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm8[2,3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm11[2,3],ymm12[4,5],ymm11[6,7]
-; AVX2-FP-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [5,2,7,0,5,2,7,0]
-; AVX2-FP-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm11[2,3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1],ymm14[2,3],ymm13[4,5],ymm14[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [1,6,3,0,5,2,7,0]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm1, %ymm0
 ; AVX2-FP-NEXT:    vpbroadcastd 304(%rdi), %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm15[0,1],ymm5[2,3],ymm15[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0,1],ymm13[2,3],ymm10[4,5],ymm13[6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm10, %ymm11
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu (%rsp), %ymm14 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm14[2,3],ymm15[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1],ymm9[2,3],ymm15[4,5],ymm9[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2],ymm0[3],ymm2[4,5],ymm0[6],ymm2[7]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm1, %ymm0
 ; AVX2-FP-NEXT:    vpbroadcastd 624(%rdi), %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm6[2,3],ymm10[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm5[2,3],ymm4[4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2],ymm0[3],ymm2[4,5],ymm0[6],ymm2[7]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FP-NEXT:    vpbroadcastd 464(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm10[2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm9[2,3],ymm6[4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 144(%rdi), %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,1],ymm15[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = ymm14[0,1],mem[2,3],ymm14[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm3, %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 464(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm8[2,3],ymm7[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm12, %ymm9
+; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = ymm12[0,1],mem[2,3],ymm12[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2],ymm0[3],ymm2[4,5],ymm0[6],ymm2[7]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FP-NEXT:    vpbroadcastd 144(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm9 = [2,7,4,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm8[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm8, %ymm14
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [2,7,4,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm11[4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm12
+; AVX2-FP-NEXT:    vmovdqa %ymm11, %ymm13
+; AVX2-FP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-FP-NEXT:    vinserti128 $1, 256(%rdi), %ymm0, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vmovdqa %ymm12, %ymm8
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [0,5,0,5,0,5,0,5]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm12
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm5[4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm5, %ymm13
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm14[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-FP-NEXT:    vinserti128 $1, 576(%rdi), %ymm0, %ymm2
-; AVX2-FP-NEXT:    vmovdqa %ymm11, %ymm4
-; AVX2-FP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm11[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqa %ymm15, %ymm14
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 608(%rdi), %ymm2
@@ -3783,112 +3762,110 @@ define void @load_i32_stride5_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3],ymm10[4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3],ymm8[4,5],ymm7[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-FP-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm2
-; AVX2-FP-NEXT:    vmovdqa %ymm6, %ymm7
-; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm6
-; AVX2-FP-NEXT:    vpermd %ymm6, %ymm0, %ymm10
-; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm7
+; AVX2-FP-NEXT:    vpermd %ymm7, %ymm0, %ymm10
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm10[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm15[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm2[4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm5, %ymm8
 ; AVX2-FP-NEXT:    vinserti128 $1, 416(%rdi), %ymm0, %ymm5
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm9[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm1[0,1,2,3,4,5,6],ymm5[7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[2,3,0,1,6,7,4,5]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm1[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 448(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm8[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqa 448(%rdi), %ymm8
+; AVX2-FP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermd %ymm8, %ymm0, %ymm0
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm1[4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm5 = ymm11[12,13,14,15],ymm14[0,1,2,3,4,5,6,7,8,9,10,11],ymm11[28,29,30,31],ymm14[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [1,6,1,6,1,6,1,6]
-; AVX2-FP-NEXT:    vpermd %ymm12, %ymm5, %ymm10
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm11[4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm12, %ymm6
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm5 = ymm12[12,13,14,15],ymm13[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm13[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1],ymm0[2],ymm5[3],ymm0[4],ymm5[5,6],ymm0[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm5 = [0,1,6,2,7,4,0,0]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm10 = [1,6,1,6,1,6,1,6]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm12[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm10 = ymm3[12,13,14,15],ymm13[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm13[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm5, %ymm10
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm12, %ymm10, %ymm15
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1,2,3],ymm13[4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm10 = ymm14[12,13,14,15],ymm15[0,1,2,3,4,5,6,7,8,9,10,11],ymm14[28,29,30,31],ymm15[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm6, %ymm5, %ymm10
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm14[0,1,2,3],ymm13[4,5],ymm14[6,7]
+; AVX2-FP-NEXT:    vmovdqu (%rsp), %ymm8 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm4[12,13,14,15],ymm8[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm8[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3],ymm0[4],ymm15[5,6],ymm0[7]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm0 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm0 = ymm9[0,1,2,3],mem[4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm15 = mem[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3],ymm0[4],ymm15[5,6],ymm0[7]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-FP-NEXT:    vpermd %ymm7, %ymm10, %ymm15
+; AVX2-FP-NEXT:    vmovdqa %ymm7, %ymm14
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm10 = ymm2[12,13,14,15],ymm7[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm7[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermd %ymm9, %ymm5, %ymm5
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm7[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3],ymm0[4],ymm15[5,6],ymm0[7]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FP-NEXT:    vpermd %ymm15, %ymm10, %ymm5
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd $207, (%rsp), %ymm11, %ymm0 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm11[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm8[0,1],ymm1[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm1[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm6[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = mem[0,1],ymm11[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm11[5],ymm5[6,7]
 ; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm10 = [4,1,6,0]
 ; AVX2-FP-NEXT:    vpermd %ymm0, %ymm10, %ymm0
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm5[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [2,7,2,7,2,7,2,7]
-; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpermd %ymm12, %ymm5, %ymm4
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm11 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm11 = mem[0,1],ymm12[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm12[5],ymm11[6,7]
+; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = ymm8[0,1,2,3],mem[4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm11 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm11 = mem[0,1],ymm13[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm13[5],ymm11[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm4, %ymm10, %ymm4
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm11[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm5, %ymm3
+; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm3 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,1],ymm13[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm13[5],ymm4[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm15[0,1,2,3],ymm14[4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm4 = ymm9[0,1],ymm1[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm1[5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm6 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm6 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm6, %ymm10, %ymm6
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpermd %ymm14, %ymm5, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = ymm7[0,1,2,3],mem[4,5],ymm7[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm4, %ymm10, %ymm4
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
 ; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm6 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm6 = mem[0,1],ymm7[0,1]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5],ymm6[6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm6[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm9, %ymm5, %ymm1
+; AVX2-FP-NEXT:    vpermd %ymm15, %ymm5, %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm4, 64(%rsi)
@@ -3926,39 +3903,37 @@ define void @load_i32_stride5_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqa %ymm2, (%r9)
 ; AVX2-FP-NEXT:    vmovdqa %ymm3, 96(%r9)
 ; AVX2-FP-NEXT:    vmovdqa %ymm0, 32(%r9)
-; AVX2-FP-NEXT:    addq $1000, %rsp # imm = 0x3E8
+; AVX2-FP-NEXT:    addq $968, %rsp # imm = 0x3C8
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
 ; AVX2-FCP-LABEL: load_i32_stride5_vf32:
 ; AVX2-FCP:       # %bb.0:
-; AVX2-FCP-NEXT:    subq $1000, %rsp # imm = 0x3E8
-; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %ymm14
+; AVX2-FCP-NEXT:    subq $968, %rsp # imm = 0x3C8
+; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %ymm4
 ; AVX2-FCP-NEXT:    vmovdqa 416(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %ymm6
-; AVX2-FCP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 544(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 576(%rdi), %ymm9
 ; AVX2-FCP-NEXT:    vmovdqa 512(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu %ymm10, (%rsp) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 480(%rdi), %ymm15
 ; AVX2-FCP-NEXT:    vmovdqa 224(%rdi), %ymm13
-; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %ymm14
 ; AVX2-FCP-NEXT:    vmovdqa 192(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 160(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,5,2,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm2, %ymm11
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vinserti128 $1, 288(%rdi), %ymm1, %ymm2
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm11[0,1,0,3]
-; AVX2-FCP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm14[0,1,0,3]
+; AVX2-FCP-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm13[4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm13, %ymm12
+; AVX2-FCP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3],ymm1[4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -3966,105 +3941,99 @@ define void @load_i32_stride5_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm9[0,1,0,3]
-; AVX2-FCP-NEXT:    vmovdqa %ymm9, %ymm13
 ; AVX2-FCP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm8[4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm8, %ymm10
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FCP-NEXT:    vinserti128 $1, 608(%rdi), %ymm1, %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm6[2,3],ymm7[4,5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm7, %ymm10
+; AVX2-FCP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm5[0,1,0,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm14[4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FCP-NEXT:    vinserti128 $1, 448(%rdi), %ymm1, %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa (%rdi), %ymm7
-; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm1[2,3],ymm7[4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vmovdqa (%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm3[2,3],ymm2[4,5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm8
+; AVX2-FCP-NEXT:    vmovdqa %ymm2, %ymm7
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm6
-; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm1 = ymm9[0,1,0,3]
-; AVX2-FCP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm6[4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm12
+; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm12[4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vinserti128 $1, 128(%rdi), %ymm0, %ymm0
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [1,6,3,0]
-; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm8[2,3],ymm4[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm11[2,3],ymm12[4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [5,2,7,0,5,2,7,0]
-; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm11[2,3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1],ymm14[2,3],ymm13[4,5],ymm14[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [1,6,3,0,5,2,7,0]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm1, %ymm0
 ; AVX2-FCP-NEXT:    vpbroadcastd 304(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm15[0,1],ymm5[2,3],ymm15[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0,1],ymm13[2,3],ymm10[4,5],ymm13[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm10, %ymm11
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm14 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm14[2,3],ymm15[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1],ymm9[2,3],ymm15[4,5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2],ymm0[3],ymm2[4,5],ymm0[6],ymm2[7]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm1, %ymm0
 ; AVX2-FCP-NEXT:    vpbroadcastd 624(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm6[2,3],ymm10[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm5[2,3],ymm4[4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2],ymm0[3],ymm2[4,5],ymm0[6],ymm2[7]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vpbroadcastd 464(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm10[2,3],ymm7[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm9[2,3],ymm6[4,5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 144(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1],ymm15[2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = ymm14[0,1],mem[2,3],ymm14[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm3, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 464(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm8[2,3],ymm7[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm12, %ymm9
+; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm12[0,1],mem[2,3],ymm12[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2],ymm0[3],ymm2[4,5],ymm0[6],ymm2[7]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vpbroadcastd 144(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm9 = [2,7,4,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm8[4,5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm8, %ymm14
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [2,7,4,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm11[4,5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm12
+; AVX2-FCP-NEXT:    vmovdqa %ymm11, %ymm13
+; AVX2-FCP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vinserti128 $1, 256(%rdi), %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm12, %ymm8
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [0,5,0,5,0,5,0,5]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm12
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm5[4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm5, %ymm13
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm14[4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vinserti128 $1, 576(%rdi), %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa %ymm11, %ymm4
-; AVX2-FCP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm11[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm15, %ymm14
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 608(%rdi), %ymm2
@@ -4072,112 +4041,110 @@ define void @load_i32_stride5_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3],ymm10[4,5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3],ymm8[4,5],ymm7[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa %ymm6, %ymm7
-; AVX2-FCP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm6
-; AVX2-FCP-NEXT:    vpermd %ymm6, %ymm0, %ymm10
-; AVX2-FCP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm7
+; AVX2-FCP-NEXT:    vpermd %ymm7, %ymm0, %ymm10
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm10[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm15[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm9, %ymm1
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm2[4,5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm5, %ymm8
 ; AVX2-FCP-NEXT:    vinserti128 $1, 416(%rdi), %ymm0, %ymm5
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm9[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm1[0,1,2,3,4,5,6],ymm5[7]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[2,3,0,1,6,7,4,5]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm1[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 448(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm8[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 448(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermd %ymm8, %ymm0, %ymm0
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm1[4,5],ymm8[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm5 = ymm11[12,13,14,15],ymm14[0,1,2,3,4,5,6,7,8,9,10,11],ymm11[28,29,30,31],ymm14[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [1,6,1,6,1,6,1,6]
-; AVX2-FCP-NEXT:    vpermd %ymm12, %ymm5, %ymm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm11[4,5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm12, %ymm6
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm5 = ymm12[12,13,14,15],ymm13[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm13[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1],ymm0[2],ymm5[3],ymm0[4],ymm5[5,6],ymm0[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm5 = [0,1,6,2,7,4,0,0]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm10 = [1,6,1,6,1,6,1,6]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm12[4,5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm10 = ymm3[12,13,14,15],ymm13[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm13[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm5, %ymm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm12, %ymm10, %ymm15
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1,2,3],ymm13[4,5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm10 = ymm14[12,13,14,15],ymm15[0,1,2,3,4,5,6,7,8,9,10,11],ymm14[28,29,30,31],ymm15[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm6, %ymm5, %ymm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm14[0,1,2,3],ymm13[4,5],ymm14[6,7]
+; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm8 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm4[12,13,14,15],ymm8[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm8[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3],ymm0[4],ymm15[5,6],ymm0[7]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm0 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm0 = ymm9[0,1,2,3],mem[4,5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm15 = mem[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3],ymm0[4],ymm15[5,6],ymm0[7]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vpermd %ymm7, %ymm10, %ymm15
+; AVX2-FCP-NEXT:    vmovdqa %ymm7, %ymm14
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm10 = ymm2[12,13,14,15],ymm7[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm7[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermd %ymm9, %ymm5, %ymm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm7[4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3],ymm0[4],ymm15[5,6],ymm0[7]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm10, %ymm5
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd $207, (%rsp), %ymm11, %ymm0 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm0 = mem[0,1,2,3],ymm11[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm8[0,1],ymm1[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm1[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1,2,3],ymm6[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = mem[0,1],ymm11[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm11[5],ymm5[6,7]
 ; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm10 = [4,1,6,0]
 ; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm10, %ymm0
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm5[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [2,7,2,7,2,7,2,7]
-; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vpermd %ymm12, %ymm5, %ymm4
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = mem[0,1],ymm12[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm12[5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = ymm8[0,1,2,3],mem[4,5],ymm8[6,7]
+; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm11 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm11 = mem[0,1],ymm13[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm13[5],ymm11[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm10, %ymm4
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm11[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm5, %ymm3
+; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm3 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,1],ymm13[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm13[5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm15[0,1,2,3],ymm14[4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm4 = ymm9[0,1],ymm1[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm1[5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm6, %ymm10, %ymm6
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vpermd %ymm14, %ymm5, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = ymm7[0,1,2,3],mem[4,5],ymm7[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm10, %ymm4
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm6 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm6 = mem[0,1],ymm7[0,1]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5],ymm6[6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm6[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm9, %ymm5, %ymm1
+; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm4, 64(%rsi)
@@ -4215,7 +4182,7 @@ define void @load_i32_stride5_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vmovdqa %ymm2, (%r9)
 ; AVX2-FCP-NEXT:    vmovdqa %ymm3, 96(%r9)
 ; AVX2-FCP-NEXT:    vmovdqa %ymm0, 32(%r9)
-; AVX2-FCP-NEXT:    addq $1000, %rsp # imm = 0x3E8
+; AVX2-FCP-NEXT:    addq $968, %rsp # imm = 0x3C8
 ; AVX2-FCP-NEXT:    vzeroupper
 ; AVX2-FCP-NEXT:    retq
 ;
@@ -6777,7 +6744,7 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 896(%rdi), %ymm5
 ; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 832(%rdi), %ymm6
+; AVX2-NEXT:    vmovdqa 832(%rdi), %ymm15
 ; AVX2-NEXT:    vmovdqa 800(%rdi), %ymm7
 ; AVX2-NEXT:    vmovdqa 544(%rdi), %ymm8
 ; AVX2-NEXT:    vmovdqa 576(%rdi), %ymm9
@@ -6787,7 +6754,7 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqa 480(%rdi), %ymm11
 ; AVX2-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 224(%rdi), %ymm12
-; AVX2-NEXT:    vmovdqu %ymm12, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 256(%rdi), %ymm13
 ; AVX2-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 192(%rdi), %ymm2
@@ -6807,16 +6774,14 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm9[0,1,0,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm8[4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovdqa %ymm8, %ymm13
-; AVX2-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa %ymm8, %ymm9
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-NEXT:    vinserti128 $1, 608(%rdi), %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm6[2,3],ymm7[4,5],ymm6[6,7]
-; AVX2-NEXT:    vmovdqa %ymm7, %ymm12
-; AVX2-NEXT:    vmovdqa %ymm6, %ymm14
-; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm15[2,3],ymm7[4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovdqa %ymm7, %ymm10
+; AVX2-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm5[0,1,0,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4],ymm2[5,6,7]
@@ -6824,11 +6789,11 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vinserti128 $1, 928(%rdi), %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 1152(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqa 1152(%rdi), %ymm14
+; AVX2-NEXT:    vmovdqa 1120(%rdi), %ymm1
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 1120(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm14[2,3],ymm1[4,5],ymm14[6,7]
+; AVX2-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vmovdqa 1184(%rdi), %ymm3
 ; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -6840,11 +6805,11 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vinserti128 $1, 1248(%rdi), %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 352(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 320(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqa 352(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-NEXT:    vmovdqa 320(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vmovdqa 384(%rdi), %ymm3
 ; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -6872,137 +6837,127 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vinserti128 $1, 768(%rdi), %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 992(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 960(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqa 992(%rdi), %ymm1
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
+; AVX2-NEXT:    vmovdqa 960(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-NEXT:    vmovdqa 1024(%rdi), %ymm3
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 1024(%rdi), %ymm13
 ; AVX2-NEXT:    vmovdqa 1056(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4],ymm2[5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm13[4],ymm2[5,6,7]
+; AVX2-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-NEXT:    vinserti128 $1, 1088(%rdi), %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm7
-; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm10
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm10[2,3],ymm7[4,5],ymm10[6,7]
-; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa (%rdi), %ymm4
+; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm1[2,3],ymm4[4,5],ymm1[6,7]
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm4
-; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm15
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm15[0,1,0,3]
-; AVX2-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm11
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm11[0,1,0,3]
+; AVX2-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4],ymm1[5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vinserti128 $1, 128(%rdi), %ymm0, %ymm0
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [1,6,3,0]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm6[0,1],ymm5[2,3],ymm6[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqu (%rsp), %ymm11 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm11[0,1],mem[2,3],ymm11[4,5],mem[6,7]
-; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [5,2,7,0,5,2,7,0]
-; AVX2-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 304(%rdi), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm8[2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm13[0,1],mem[2,3],ymm13[4,5],mem[6,7]
-; AVX2-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 624(%rdi), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm14[2,3],ymm12[4,5,6,7]
-; AVX2-NEXT:    vmovdqa %ymm12, %ymm14
-; AVX2-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm12[0,1],mem[2,3],ymm12[4,5],mem[6,7]
-; AVX2-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 944(%rdi), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm2[0,1],mem[2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm13[0,1],mem[2,3],ymm13[4,5],mem[6,7]
-; AVX2-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 1264(%rdi), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1],ymm10[2,3],ymm7[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1],ymm15[2,3],ymm4[4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 144(%rdi), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm5[2,3],ymm12[4,5,6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm6[0,1],mem[2,3],ymm6[4,5],mem[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm0 = [1,6,3,0,5,2,7,0]
+; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-NEXT:    vpbroadcastd 304(%rdi), %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm7[0,1],mem[2,3],ymm7[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1],ymm7[2,3],ymm8[4,5,6,7]
+; AVX2-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = ymm9[0,1],mem[2,3],ymm9[4,5],mem[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-NEXT:    vpbroadcastd 624(%rdi), %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm10[0,1],ymm15[2,3],ymm10[4,5,6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
-; AVX2-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 464(%rdi), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-NEXT:    vpbroadcastd 944(%rdi), %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1],ymm14[2,3],mem[4,5,6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm10[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm3[0,1],mem[2,3],ymm3[4,5],mem[6,7]
-; AVX2-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 784(%rdi), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = ymm10[0,1],mem[2,3],ymm10[4,5],mem[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-NEXT:    vpbroadcastd 1264(%rdi), %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1],ymm14[2,3],mem[4,5,6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm2 = ymm2[0,1],mem[2,3],ymm2[4,5],mem[6,7]
-; AVX2-NEXT:    vpermd %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 1104(%rdi), %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-NEXT:    vpbroadcastd 464(%rdi), %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-NEXT:    vpbroadcastd 784(%rdi), %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu (%rsp), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = ymm13[0,1],mem[2,3],ymm13[4,5],mem[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-NEXT:    vpbroadcastd 1104(%rdi), %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm3[2,3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm11[2,3],ymm4[4,5],ymm11[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpbroadcastd 144(%rdi), %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,7,4,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm5[4,5],ymm6[6,7]
-; AVX2-NEXT:    vpermd %ymm0, %ymm3, %ymm0
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm13 = [2,7,4,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1,2,3],ymm5[4,5],ymm12[6,7]
+; AVX2-NEXT:    vpermd %ymm0, %ymm13, %ymm0
 ; AVX2-NEXT:    vinserti128 $1, 256(%rdi), %ymm0, %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm11[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm6[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,0,1,6,7,4,5]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 288(%rdi), %ymm2
@@ -7011,11 +6966,10 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm9[0,1,2,3],ymm8[4,5],ymm9[6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1,2,3],ymm7[4,5],ymm8[6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-NEXT:    vinserti128 $1, 576(%rdi), %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 608(%rdi), %ymm2
@@ -7023,11 +6977,12 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm14[0,1,2,3],ymm11[4,5],ymm14[6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm12[0,1,2,3],mem[4,5],ymm12[6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-NEXT:    vinserti128 $1, 896(%rdi), %ymm0, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 928(%rdi), %ymm2
@@ -7035,312 +6990,305 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm12[0,1,2,3],ymm6[4,5],ymm12[6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm6[0,1,2,3],ymm11[4,5],ymm6[6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-NEXT:    vinserti128 $1, 1216(%rdi), %ymm0, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-NEXT:    vmovdqa %ymm13, %ymm9
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqa 1248(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vmovdqa 1248(%rdi), %ymm9
+; AVX2-NEXT:    vpermd %ymm9, %ymm0, %ymm2
+; AVX2-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm4
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm7[0,1,2,3],mem[4,5],ymm7[6,7]
-; AVX2-NEXT:    vpermd %ymm2, %ymm3, %ymm2
+; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1,2,3],ymm14[4,5],mem[6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-NEXT:    vinserti128 $1, 416(%rdi), %ymm0, %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm15[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-NEXT:    vpblendd $127, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = mem[0,1,2,3,4,5,6],ymm4[7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,3,0,1,6,7,4,5]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm2[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqa 448(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm15
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,1,2,3],ymm10[4,5],mem[6,7]
-; AVX2-NEXT:    vmovdqa %ymm10, %ymm14
-; AVX2-NEXT:    vpermd %ymm4, %ymm3, %ymm4
-; AVX2-NEXT:    vinserti128 $1, 736(%rdi), %ymm0, %ymm15
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm2[0,1,2,3,4,5,6],ymm15[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,3,0,1,6,7,4,5]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm4[0,1,2],ymm15[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqa 768(%rdi), %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm4[3,4,5,6,7]
+; AVX2-NEXT:    vmovdqa 448(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm5
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm5[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm13
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm15[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm10[0,1,2,3],mem[4,5],ymm10[6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm13, %ymm1
+; AVX2-NEXT:    vinserti128 $1, 736(%rdi), %ymm0, %ymm5
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm4[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[2,3,0,1,6,7,4,5]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-NEXT:    vmovdqa 768(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm15
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovdqu (%rsp), %ymm7 # 32-byte Reload
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm7[0,1,2,3],ymm8[4,5],ymm7[6,7]
-; AVX2-NEXT:    vpermd %ymm13, %ymm3, %ymm3
-; AVX2-NEXT:    vinserti128 $1, 1056(%rdi), %ymm0, %ymm13
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm1[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[2,3,0,1,6,7,4,5]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm3[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3],ymm8[4,5],ymm7[6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm13, %ymm1
+; AVX2-NEXT:    vinserti128 $1, 1056(%rdi), %ymm0, %ymm3
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,3,0,1,6,7,4,5]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm3[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 1088(%rdi), %ymm3
 ; AVX2-NEXT:    vpermd %ymm3, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa %ymm3, %ymm4
+; AVX2-NEXT:    vmovdqa %ymm3, %ymm14
 ; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm13[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $207, (%rsp), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5],mem[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm13 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm13 = mem[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3],ymm0[4],ymm1[5,6],ymm0[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [0,1,6,2,7,4,0,0]
+; AVX2-NEXT:    vpermd %ymm0, %ymm1, %ymm15
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [1,6,1,6,1,6,1,6]
-; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm13 # 32-byte Folded Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm15[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
 ; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm13 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm13 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm15 = mem[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm13 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm13 = ymm3[0,1,2,3],mem[4,5],ymm3[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm15 = mem[12,13,14,15],ymm11[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm11[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm15 = mem[12,13,14,15],ymm5[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm5[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm13 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm13 = ymm9[0,1,2,3],mem[4,5],ymm9[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm12[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm13 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm13 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm12[12,13,14,15],ymm5[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm5[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm13 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm13 = ymm1[0,1,2,3],mem[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm7[12,13,14,15],ymm8[0,1,2,3,4,5,6,7,8,9,10,11],ymm7[28,29,30,31],ymm8[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm4, %ymm0, %ymm15
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm13 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm13 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm6[12,13,14,15],ymm11[0,1,2,3,4,5,6,7,8,9,10,11],ymm6[28,29,30,31],ymm11[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-NEXT:    vpermd %ymm13, %ymm1, %ymm13
+; AVX2-NEXT:    vpermd %ymm9, %ymm0, %ymm15
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm13 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm13 = ymm2[0,1,2,3],mem[4,5],ymm2[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm7[12,13,14,15],ymm8[0,1,2,3,4,5,6,7,8,9,10,11],ymm7[28,29,30,31],ymm8[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-NEXT:    vpermd %ymm13, %ymm1, %ymm13
+; AVX2-NEXT:    vpermd %ymm14, %ymm0, %ymm15
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm12[12,13,14,15],ymm10[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm10[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm4[0,1,2,3],ymm12[4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm10[12,13,14,15],ymm11[0,1,2,3,4,5,6,7,8,9,10,11],ymm10[28,29,30,31],ymm11[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm5[0,1,2,3],ymm14[4,5],ymm5[6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm9[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm10[0,1,2,3],ymm4[4,5],ymm10[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
+; AVX2-NEXT:    vpermd %ymm10, %ymm0, %ymm15
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm8[0,1,2,3],ymm7[4,5],ymm8[6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm9[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpermd %ymm3, %ymm0, %ymm15
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm7[0,1,2,3],ymm6[4,5],ymm7[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm5[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm5[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-NEXT:    vpermd %ymm11, %ymm0, %ymm0
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm7[0,1],ymm6[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5],ymm0[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm1[0,1,2,3],ymm5[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm15 = ymm6[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm6[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-NEXT:    vpermd %ymm13, %ymm1, %ymm1
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-NEXT:    vpermd %ymm15, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm1[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm8[0,1],ymm7[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm7[5],ymm0[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm6[4,5],ymm2[6,7]
 ; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm8 = [4,1,6,0]
-; AVX2-NEXT:    vpermd %ymm5, %ymm8, %ymm5
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vpermd %ymm1, %ymm8, %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm7 = [2,7,2,7,2,7,2,7]
-; AVX2-NEXT:    vpermd %ymm11, %ymm7, %ymm1
+; AVX2-NEXT:    vpermd %ymm15, %ymm7, %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4,5],ymm1[6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-NEXT:    vperm2i128 $2, (%rsp), %ymm11, %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = mem[0,1],ymm11[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm11[5],ymm5[6,7]
+; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = mem[0,1],ymm3[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5],ymm2[6,7]
 ; AVX2-NEXT:    vpermd %ymm1, %ymm8, %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0,1,2,3,4,5],ymm5[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm9[4,5],ymm2[6,7]
-; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm10[0,1],ymm4[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm4[5],ymm5[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm9[4,5],ymm4[6,7]
+; AVX2-NEXT:    vperm2i128 {{.*#+}} ymm2 = ymm5[0,1],ymm14[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5],ymm2[6,7]
 ; AVX2-NEXT:    vpermd %ymm1, %ymm8, %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm3, %ymm7, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-NEXT:    vpermd %ymm10, %ymm7, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5],mem[6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = mem[0,1],ymm9[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm9[5],ymm5[6,7]
+; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = mem[0,1],ymm9[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm9[5],ymm4[6,7]
 ; AVX2-NEXT:    vpermd %ymm2, %ymm8, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm2[0,1,2,3,4,5],ymm5[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm14[0,1,2,3],ymm12[4,5],ymm14[6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = mem[0,1],ymm10[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm10[5],ymm9[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3,4,5,6,7]
+; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm2[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = ymm11[0,1,2,3],mem[4,5],ymm11[6,7]
+; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = mem[0,1],ymm12[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm12[5],ymm4[6,7]
 ; AVX2-NEXT:    vpermd %ymm2, %ymm8, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm9[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3,4,5,6,7]
 ; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm4 # 32-byte Folded Reload
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = ymm3[0,1,2,3],mem[4,5],ymm3[6,7]
+; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = mem[0,1],ymm10[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm10[5],ymm9[6,7]
+; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = mem[0,1],ymm10[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm10[5],ymm5[6,7]
 ; AVX2-NEXT:    vpermd %ymm4, %ymm8, %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm10 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm10 = mem[0,1],ymm12[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm12[5],ymm10[6,7]
-; AVX2-NEXT:    vpermd %ymm9, %ymm8, %ymm9
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2],ymm10[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-NEXT:    vmovdqu (%rsp), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm10 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm10 = mem[0,1],ymm11[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm11[5],ymm10[6,7]
+; AVX2-NEXT:    vpermd %ymm5, %ymm8, %ymm5
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm10[3,4,5,6,7]
 ; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm9[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = ymm6[0,1,2,3],mem[4,5],ymm6[6,7]
-; AVX2-NEXT:    vpermd %ymm9, %ymm8, %ymm8
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm5[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
+; AVX2-NEXT:    vpermd %ymm5, %ymm8, %ymm5
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = mem[0,1],ymm10[0,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm10[5],ymm9[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2],ymm9[3,4,5,6,7]
+; AVX2-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm8 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm8 = mem[0,1],ymm10[0,1]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm10[5],ymm8[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm8[3,4,5,6,7]
 ; AVX2-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 192(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 128(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 64(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, (%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 224(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 160(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 96(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm8, 32(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 192(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 128(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 64(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, (%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 224(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 160(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 96(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 32(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 192(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 128(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 64(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, (%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 224(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 160(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 96(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 32(%rcx)
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm7[6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 192(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 128(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 64(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, (%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 224(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 160(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 96(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 32(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 192(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 128(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 64(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, (%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 224(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 160(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 96(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 32(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 192(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 128(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 64(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, (%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 224(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 160(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 96(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 32(%rcx)
 ; AVX2-NEXT:    vmovdqa %ymm13, (%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 64(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 128(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 192(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 224(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 160(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 96(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 32(%r8)
-; AVX2-NEXT:    vmovdqa %ymm7, 224(%r9)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 64(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 128(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 192(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 224(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 160(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 96(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm7, 32(%r8)
+; AVX2-NEXT:    vmovdqa %ymm5, 224(%r9)
 ; AVX2-NEXT:    vmovdqa %ymm3, 192(%r9)
 ; AVX2-NEXT:    vmovdqa %ymm4, 160(%r9)
 ; AVX2-NEXT:    vmovdqa %ymm2, 128(%r9)
-; AVX2-NEXT:    vmovdqa %ymm5, 96(%r9)
+; AVX2-NEXT:    vmovdqa %ymm9, 96(%r9)
 ; AVX2-NEXT:    vmovdqa %ymm1, 64(%r9)
-; AVX2-NEXT:    vmovdqa %ymm11, 32(%r9)
+; AVX2-NEXT:    vmovdqa %ymm6, 32(%r9)
 ; AVX2-NEXT:    vmovdqa %ymm0, (%r9)
 ; AVX2-NEXT:    addq $2152, %rsp # imm = 0x868
 ; AVX2-NEXT:    vzeroupper
@@ -7353,7 +7301,7 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 896(%rdi), %ymm5
 ; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 832(%rdi), %ymm6
+; AVX2-FP-NEXT:    vmovdqa 832(%rdi), %ymm15
 ; AVX2-FP-NEXT:    vmovdqa 800(%rdi), %ymm7
 ; AVX2-FP-NEXT:    vmovdqa 544(%rdi), %ymm8
 ; AVX2-FP-NEXT:    vmovdqa 576(%rdi), %ymm9
@@ -7363,7 +7311,7 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqa 480(%rdi), %ymm11
 ; AVX2-FP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 224(%rdi), %ymm12
-; AVX2-FP-NEXT:    vmovdqu %ymm12, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %ymm13
 ; AVX2-FP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 192(%rdi), %ymm2
@@ -7383,16 +7331,14 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm9[0,1,0,3]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm8[4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm8, %ymm13
-; AVX2-FP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa %ymm8, %ymm9
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FP-NEXT:    vinserti128 $1, 608(%rdi), %ymm1, %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm6[2,3],ymm7[4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm7, %ymm12
-; AVX2-FP-NEXT:    vmovdqa %ymm6, %ymm14
-; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm15[2,3],ymm7[4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm7, %ymm10
+; AVX2-FP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm5[0,1,0,3]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4],ymm2[5,6,7]
@@ -7400,11 +7346,11 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vinserti128 $1, 928(%rdi), %ymm1, %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 1152(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqa 1152(%rdi), %ymm14
+; AVX2-FP-NEXT:    vmovdqa 1120(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 1120(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm14[2,3],ymm1[4,5],ymm14[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vmovdqa 1184(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -7416,11 +7362,11 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vinserti128 $1, 1248(%rdi), %ymm1, %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 352(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 320(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqa 352(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vmovdqa 320(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vmovdqa 384(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -7448,137 +7394,127 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vinserti128 $1, 768(%rdi), %ymm1, %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 992(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 960(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqa 992(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vmovdqa 960(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-FP-NEXT:    vmovdqa 1024(%rdi), %ymm3
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 1024(%rdi), %ymm13
 ; AVX2-FP-NEXT:    vmovdqa 1056(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm13[4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FP-NEXT:    vinserti128 $1, 1088(%rdi), %ymm1, %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm7
-; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm10
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm10[2,3],ymm7[4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm4
+; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm1[2,3],ymm4[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm4
-; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm15
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm15[0,1,0,3]
-; AVX2-FP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm11
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm11[0,1,0,3]
+; AVX2-FP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vinserti128 $1, 128(%rdi), %ymm0, %ymm0
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [1,6,3,0]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm6[0,1],ymm5[2,3],ymm6[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vmovdqu (%rsp), %ymm11 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm11[0,1],mem[2,3],ymm11[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [5,2,7,0,5,2,7,0]
-; AVX2-FP-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 304(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm8[2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm13[0,1],mem[2,3],ymm13[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 624(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm14[2,3],ymm12[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm12, %ymm14
-; AVX2-FP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm12[0,1],mem[2,3],ymm12[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 944(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm2[0,1],mem[2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm13[0,1],mem[2,3],ymm13[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 1264(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1],ymm10[2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1],ymm15[2,3],ymm4[4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 144(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm5[2,3],ymm12[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm6[0,1],mem[2,3],ymm6[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm0 = [1,6,3,0,5,2,7,0]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vpbroadcastd 304(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm7[0,1],mem[2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1],ymm7[2,3],ymm8[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = ymm9[0,1],mem[2,3],ymm9[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vpbroadcastd 624(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm10[0,1],ymm15[2,3],ymm10[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 464(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vpbroadcastd 944(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1],ymm14[2,3],mem[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm10[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm3[0,1],mem[2,3],ymm3[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 784(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = ymm10[0,1],mem[2,3],ymm10[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vpbroadcastd 1264(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1],ymm14[2,3],mem[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm2 = ymm2[0,1],mem[2,3],ymm2[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm1, %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 1104(%rdi), %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vpbroadcastd 464(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vpbroadcastd 784(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu (%rsp), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = ymm13[0,1],mem[2,3],ymm13[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vpbroadcastd 1104(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm3[2,3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm11[2,3],ymm4[4,5],ymm11[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpbroadcastd 144(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,7,4,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm5[4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm0, %ymm3, %ymm0
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm13 = [2,7,4,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1,2,3],ymm5[4,5],ymm12[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm13, %ymm0
 ; AVX2-FP-NEXT:    vinserti128 $1, 256(%rdi), %ymm0, %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm11[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm6[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,0,1,6,7,4,5]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %ymm2
@@ -7587,11 +7523,10 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm9[0,1,2,3],ymm8[4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1,2,3],ymm7[4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-FP-NEXT:    vinserti128 $1, 576(%rdi), %ymm0, %ymm2
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 608(%rdi), %ymm2
@@ -7599,11 +7534,12 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm14[0,1,2,3],ymm11[4,5],ymm14[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm12[0,1,2,3],mem[4,5],ymm12[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-FP-NEXT:    vinserti128 $1, 896(%rdi), %ymm0, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 928(%rdi), %ymm2
@@ -7611,312 +7547,305 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm12[0,1,2,3],ymm6[4,5],ymm12[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm6[0,1,2,3],ymm11[4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-FP-NEXT:    vinserti128 $1, 1216(%rdi), %ymm0, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vmovdqa %ymm13, %ymm9
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 1248(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vmovdqa 1248(%rdi), %ymm9
+; AVX2-FP-NEXT:    vpermd %ymm9, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-FP-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm4
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm7[0,1,2,3],mem[4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm3, %ymm2
+; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1,2,3],ymm14[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-FP-NEXT:    vinserti128 $1, 416(%rdi), %ymm0, %ymm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm15[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FP-NEXT:    vpblendd $127, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = mem[0,1,2,3,4,5,6],ymm4[7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,3,0,1,6,7,4,5]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm2[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 448(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm15
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,1,2,3],ymm10[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm10, %ymm14
-; AVX2-FP-NEXT:    vpermd %ymm4, %ymm3, %ymm4
-; AVX2-FP-NEXT:    vinserti128 $1, 736(%rdi), %ymm0, %ymm15
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm2[0,1,2,3,4,5,6],ymm15[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,3,0,1,6,7,4,5]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm4[0,1,2],ymm15[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 768(%rdi), %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm4[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqa 448(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm5
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm5[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm13
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm15[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm10[0,1,2,3],mem[4,5],ymm10[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
+; AVX2-FP-NEXT:    vinserti128 $1, 736(%rdi), %ymm0, %ymm5
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm4[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[2,3,0,1,6,7,4,5]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqa 768(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm15
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovdqu (%rsp), %ymm7 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm7[0,1,2,3],ymm8[4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm13, %ymm3, %ymm3
-; AVX2-FP-NEXT:    vinserti128 $1, 1056(%rdi), %ymm0, %ymm13
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm1[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[2,3,0,1,6,7,4,5]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm3[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3],ymm8[4,5],ymm7[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
+; AVX2-FP-NEXT:    vinserti128 $1, 1056(%rdi), %ymm0, %ymm3
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,3,0,1,6,7,4,5]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm3[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 1088(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vpermd %ymm3, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm4
+; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm14
 ; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm13[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $207, (%rsp), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm13 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm13 = mem[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3],ymm0[4],ymm1[5,6],ymm0[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [0,1,6,2,7,4,0,0]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm1, %ymm15
 ; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [1,6,1,6,1,6,1,6]
-; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm13 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm15[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
 ; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm13 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm13 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm15 = mem[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm13 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm13 = ymm3[0,1,2,3],mem[4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-FP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm15 = mem[12,13,14,15],ymm11[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm11[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm15 = mem[12,13,14,15],ymm5[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm5[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm13 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm13 = ymm9[0,1,2,3],mem[4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm12[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm13 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm13 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm12[12,13,14,15],ymm5[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm5[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm13 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm13 = ymm1[0,1,2,3],mem[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm7[12,13,14,15],ymm8[0,1,2,3,4,5,6,7,8,9,10,11],ymm7[28,29,30,31],ymm8[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm4, %ymm0, %ymm15
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm13 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm13 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm6[12,13,14,15],ymm11[0,1,2,3,4,5,6,7,8,9,10,11],ymm6[28,29,30,31],ymm11[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
+; AVX2-FP-NEXT:    vpermd %ymm9, %ymm0, %ymm15
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm13 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm13 = ymm2[0,1,2,3],mem[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm7[12,13,14,15],ymm8[0,1,2,3,4,5,6,7,8,9,10,11],ymm7[28,29,30,31],ymm8[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
+; AVX2-FP-NEXT:    vpermd %ymm14, %ymm0, %ymm15
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm12[12,13,14,15],ymm10[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm10[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm4[0,1,2,3],ymm12[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm10[12,13,14,15],ymm11[0,1,2,3,4,5,6,7,8,9,10,11],ymm10[28,29,30,31],ymm11[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm5[0,1,2,3],ymm14[4,5],ymm5[6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm9[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm10[0,1,2,3],ymm4[4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
+; AVX2-FP-NEXT:    vpermd %ymm10, %ymm0, %ymm15
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm8[0,1,2,3],ymm7[4,5],ymm8[6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm9[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm0, %ymm15
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm7[0,1,2,3],ymm6[4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm5[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm5[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermd %ymm11, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm7[0,1],ymm6[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm1[0,1,2,3],ymm5[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm6[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm6[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FP-NEXT:    vpermd %ymm13, %ymm1, %ymm1
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FP-NEXT:    vpermd %ymm15, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm1[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm8[0,1],ymm7[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm7[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm6[4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm8 = [4,1,6,0]
-; AVX2-FP-NEXT:    vpermd %ymm5, %ymm8, %ymm5
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm8, %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm7 = [2,7,2,7,2,7,2,7]
-; AVX2-FP-NEXT:    vpermd %ymm11, %ymm7, %ymm1
+; AVX2-FP-NEXT:    vpermd %ymm15, %ymm7, %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FP-NEXT:    vperm2i128 $2, (%rsp), %ymm11, %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = mem[0,1],ymm11[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm11[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm3[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm8, %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0,1,2,3,4,5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm9[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm10[0,1],ymm4[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm4[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm9[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vperm2i128 {{.*#+}} ymm2 = ymm5[0,1],ymm14[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm8, %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm7, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermd %ymm10, %ymm7, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5],mem[6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = mem[0,1],ymm9[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm9[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = mem[0,1],ymm9[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm9[5],ymm4[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm2, %ymm8, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm2[0,1,2,3,4,5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm14[0,1,2,3],ymm12[4,5],ymm14[6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = mem[0,1],ymm10[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm10[5],ymm9[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm2[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = ymm11[0,1,2,3],mem[4,5],ymm11[6,7]
+; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = mem[0,1],ymm12[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm12[5],ymm4[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm2, %ymm8, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm9[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm4 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = ymm3[0,1,2,3],mem[4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = mem[0,1],ymm10[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm10[5],ymm9[6,7]
+; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = mem[0,1],ymm10[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm10[5],ymm5[6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm4, %ymm8, %ymm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm10 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm10 = mem[0,1],ymm12[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm12[5],ymm10[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm9, %ymm8, %ymm9
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2],ymm10[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vmovdqu (%rsp), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm10 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm10 = mem[0,1],ymm11[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm11[5],ymm10[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm5, %ymm8, %ymm5
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm10[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm9[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = ymm6[0,1,2,3],mem[4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpermd %ymm9, %ymm8, %ymm8
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm5[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpermd %ymm5, %ymm8, %ymm5
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = mem[0,1],ymm10[0,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm10[5],ymm9[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2],ymm9[3,4,5,6,7]
+; AVX2-FP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm8 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm8 = mem[0,1],ymm10[0,1]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm10[5],ymm8[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm8[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 128(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 64(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, (%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 224(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 160(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 96(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm8, 32(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 128(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, (%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 224(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 160(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 128(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, (%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 224(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 160(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%rcx)
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm7[6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 192(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 128(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 64(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, (%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 224(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 160(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 96(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 192(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 128(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 64(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, (%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 224(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 160(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 96(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 192(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 128(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 64(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, (%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 224(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 160(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 96(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%rcx)
 ; AVX2-FP-NEXT:    vmovdqa %ymm13, (%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 128(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 224(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 160(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%r8)
-; AVX2-FP-NEXT:    vmovdqa %ymm7, 224(%r9)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 64(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 128(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 192(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 224(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 160(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 96(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%r8)
+; AVX2-FP-NEXT:    vmovdqa %ymm5, 224(%r9)
 ; AVX2-FP-NEXT:    vmovdqa %ymm3, 192(%r9)
 ; AVX2-FP-NEXT:    vmovdqa %ymm4, 160(%r9)
 ; AVX2-FP-NEXT:    vmovdqa %ymm2, 128(%r9)
-; AVX2-FP-NEXT:    vmovdqa %ymm5, 96(%r9)
+; AVX2-FP-NEXT:    vmovdqa %ymm9, 96(%r9)
 ; AVX2-FP-NEXT:    vmovdqa %ymm1, 64(%r9)
-; AVX2-FP-NEXT:    vmovdqa %ymm11, 32(%r9)
+; AVX2-FP-NEXT:    vmovdqa %ymm6, 32(%r9)
 ; AVX2-FP-NEXT:    vmovdqa %ymm0, (%r9)
 ; AVX2-FP-NEXT:    addq $2152, %rsp # imm = 0x868
 ; AVX2-FP-NEXT:    vzeroupper
@@ -7929,7 +7858,7 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 896(%rdi), %ymm5
 ; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 832(%rdi), %ymm6
+; AVX2-FCP-NEXT:    vmovdqa 832(%rdi), %ymm15
 ; AVX2-FCP-NEXT:    vmovdqa 800(%rdi), %ymm7
 ; AVX2-FCP-NEXT:    vmovdqa 544(%rdi), %ymm8
 ; AVX2-FCP-NEXT:    vmovdqa 576(%rdi), %ymm9
@@ -7939,7 +7868,7 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vmovdqa 480(%rdi), %ymm11
 ; AVX2-FCP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 224(%rdi), %ymm12
-; AVX2-FCP-NEXT:    vmovdqu %ymm12, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %ymm13
 ; AVX2-FCP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 192(%rdi), %ymm2
@@ -7959,16 +7888,14 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm9[0,1,0,3]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm8[4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm8, %ymm13
-; AVX2-FCP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa %ymm8, %ymm9
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FCP-NEXT:    vinserti128 $1, 608(%rdi), %ymm1, %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm6[2,3],ymm7[4,5],ymm6[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm7, %ymm12
-; AVX2-FCP-NEXT:    vmovdqa %ymm6, %ymm14
-; AVX2-FCP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm15[2,3],ymm7[4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm7, %ymm10
+; AVX2-FCP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm5[0,1,0,3]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4],ymm2[5,6,7]
@@ -7976,11 +7903,11 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vinserti128 $1, 928(%rdi), %ymm1, %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 1152(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 1152(%rdi), %ymm14
+; AVX2-FCP-NEXT:    vmovdqa 1120(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 1120(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm14[2,3],ymm1[4,5],ymm14[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vmovdqa 1184(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -7992,11 +7919,11 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vinserti128 $1, 1248(%rdi), %ymm1, %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -8024,137 +7951,127 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vinserti128 $1, 768(%rdi), %ymm1, %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 992(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 960(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 992(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovdqa 960(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vmovdqa 1024(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa 1024(%rdi), %ymm13
 ; AVX2-FCP-NEXT:    vmovdqa 1056(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm13[4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FCP-NEXT:    vinserti128 $1, 1088(%rdi), %ymm1, %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa (%rdi), %ymm7
-; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm10[2,3],ymm7[4,5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa (%rdi), %ymm4
+; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm1[2,3],ymm4[4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %ymm15
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm1 = ymm15[0,1,0,3]
-; AVX2-FCP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm1 = ymm11[0,1,0,3]
+; AVX2-FCP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vinserti128 $1, 128(%rdi), %ymm0, %ymm0
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [1,6,3,0]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm6[0,1],ymm5[2,3],ymm6[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm11 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm11[0,1],mem[2,3],ymm11[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [5,2,7,0,5,2,7,0]
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 304(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm8[2,3],ymm9[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm13[0,1],mem[2,3],ymm13[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 624(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm14[2,3],ymm12[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm12, %ymm14
-; AVX2-FCP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm12[0,1],mem[2,3],ymm12[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 944(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = ymm2[0,1],mem[2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm13[0,1],mem[2,3],ymm13[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 1264(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1],ymm10[2,3],ymm7[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1],ymm15[2,3],ymm4[4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 144(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm5[2,3],ymm12[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm6[0,1],mem[2,3],ymm6[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm0 = [1,6,3,0,5,2,7,0]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vpbroadcastd 304(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = ymm7[0,1],mem[2,3],ymm7[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1],ymm7[2,3],ymm8[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm9[0,1],mem[2,3],ymm9[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vpbroadcastd 624(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm10[0,1],ymm15[2,3],ymm10[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 464(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vpbroadcastd 944(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = mem[0,1],ymm14[2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm10[2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm3[0,1],mem[2,3],ymm3[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 784(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm10[0,1],mem[2,3],ymm10[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vpbroadcastd 1264(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = mem[0,1],ymm14[2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm2 = ymm2[0,1],mem[2,3],ymm2[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm1, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 1104(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vpbroadcastd 464(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vpbroadcastd 784(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm13[0,1],mem[2,3],ymm13[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vpbroadcastd 1104(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm3[2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm11[2,3],ymm4[4,5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpbroadcastd 144(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [2,7,4,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm5[4,5],ymm6[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm3, %ymm0
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm13 = [2,7,4,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1,2,3],ymm5[4,5],ymm12[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm13, %ymm0
 ; AVX2-FCP-NEXT:    vinserti128 $1, 256(%rdi), %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm11[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm6[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,0,1,6,7,4,5]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %ymm2
@@ -8163,11 +8080,10 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm9[0,1,2,3],ymm8[4,5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1,2,3],ymm7[4,5],ymm8[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-FCP-NEXT:    vinserti128 $1, 576(%rdi), %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 608(%rdi), %ymm2
@@ -8175,11 +8091,12 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm14[0,1,2,3],ymm11[4,5],ymm14[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm12[0,1,2,3],mem[4,5],ymm12[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-FCP-NEXT:    vinserti128 $1, 896(%rdi), %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 928(%rdi), %ymm2
@@ -8187,312 +8104,305 @@ define void @load_i32_stride5_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm12[0,1,2,3],ymm6[4,5],ymm12[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm6[0,1,2,3],ymm11[4,5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-FCP-NEXT:    vinserti128 $1, 1216(%rdi), %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm13, %ymm9
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 1248(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vmovdqa 1248(%rdi), %ymm9
+; AVX2-FCP-NEXT:    vpermd %ymm9, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-FCP-NEXT:    vinserti128 $1, 96(%rdi), %ymm0, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = ymm7[0,1,2,3],mem[4,5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm3, %ymm2
+; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm14[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
 ; AVX2-FCP-NEXT:    vinserti128 $1, 416(%rdi), %ymm0, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm15[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vpblendd $127, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = mem[0,1,2,3,4,5,6],ymm4[7]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,3,0,1,6,7,4,5]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm2[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 448(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm15
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,1,2,3],ymm10[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm10, %ymm14
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm3, %ymm4
-; AVX2-FCP-NEXT:    vinserti128 $1, 736(%rdi), %ymm0, %ymm15
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm2[0,1,2,3,4,5,6],ymm15[7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,3,0,1,6,7,4,5]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm4[0,1,2],ymm15[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 768(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm4[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 448(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm5[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm13
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm15[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm10[0,1,2,3],mem[4,5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vinserti128 $1, 736(%rdi), %ymm0, %ymm5
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm4[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[2,3,0,1,6,7,4,5]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 768(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm15
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm7 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm7[0,1,2,3],ymm8[4,5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm3, %ymm3
-; AVX2-FCP-NEXT:    vinserti128 $1, 1056(%rdi), %ymm0, %ymm13
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm1[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[2,3,0,1,6,7,4,5]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm3[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3],ymm8[4,5],ymm7[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vinserti128 $1, 1056(%rdi), %ymm0, %ymm3
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,3,0,1,6,7,4,5]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm3[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 1088(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm4
+; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm14
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm13[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $207, (%rsp), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,0,2,2,7,4,6,6]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm13 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm13 = mem[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = mem[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3],ymm0[4],ymm1[5,6],ymm0[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [0,1,6,2,7,4,0,0]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm1, %ymm15
 ; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [1,6,1,6,1,6,1,6]
-; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm13 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm15[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm13 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm13 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm15 = mem[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm13 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm13 = ymm3[0,1,2,3],mem[4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-FCP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm15 = mem[12,13,14,15],ymm11[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm11[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm15 = mem[12,13,14,15],ymm5[0,1,2,3,4,5,6,7,8,9,10,11],mem[28,29,30,31],ymm5[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm13 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm13 = ymm9[0,1,2,3],mem[4,5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm12[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm13 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm13 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm12[12,13,14,15],ymm5[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm5[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm13 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm13 = ymm1[0,1,2,3],mem[4,5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm7[12,13,14,15],ymm8[0,1,2,3,4,5,6,7,8,9,10,11],ymm7[28,29,30,31],ymm8[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm0, %ymm15
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm13 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm13 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm6[12,13,14,15],ymm11[0,1,2,3,4,5,6,7,8,9,10,11],ymm6[28,29,30,31],ymm11[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
+; AVX2-FCP-NEXT:    vpermd %ymm9, %ymm0, %ymm15
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm13 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm13 = ymm2[0,1,2,3],mem[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm7[12,13,14,15],ymm8[0,1,2,3,4,5,6,7,8,9,10,11],ymm7[28,29,30,31],ymm8[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
+; AVX2-FCP-NEXT:    vpermd %ymm14, %ymm0, %ymm15
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm12[12,13,14,15],ymm10[0,1,2,3,4,5,6,7,8,9,10,11],ymm12[28,29,30,31],ymm10[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm4[0,1,2,3],ymm12[4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm10[12,13,14,15],ymm11[0,1,2,3,4,5,6,7,8,9,10,11],ymm10[28,29,30,31],ymm11[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm5[0,1,2,3],ymm14[4,5],ymm5[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm9[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm1, %ymm13
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm10[0,1,2,3],ymm4[4,5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
+; AVX2-FCP-NEXT:    vpermd %ymm10, %ymm0, %ymm15
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm8[0,1,2,3],ymm7[4,5],ymm8[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm9[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm15
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm7[0,1,2,3],ymm6[4,5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,0,2,2,7,4,6,6]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm5[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm5[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,3,2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2],ymm13[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermd %ymm11, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm7[0,1],ymm6[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm1[0,1,2,3],ymm5[4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm15 = ymm6[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm6[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3],ymm13[4],ymm15[5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm1, %ymm1
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm1[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm8[0,1],ymm7[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm7[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm6[4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm8 = [4,1,6,0]
-; AVX2-FCP-NEXT:    vpermd %ymm5, %ymm8, %ymm5
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm8, %ymm1
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm7 = [2,7,2,7,2,7,2,7]
-; AVX2-FCP-NEXT:    vpermd %ymm11, %ymm7, %ymm1
+; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm7, %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4,5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FCP-NEXT:    vperm2i128 $2, (%rsp), %ymm11, %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = mem[0,1],ymm11[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm11[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm3[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm8, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0,1,2,3,4,5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm9[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm5 = ymm10[0,1],ymm4[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm4[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm9[4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vperm2i128 {{.*#+}} ymm2 = ymm5[0,1],ymm14[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm14[5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm8, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm7, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm10, %ymm7, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5],mem[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = mem[0,1],ymm9[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm9[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = mem[0,1],ymm9[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm9[5],ymm4[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm8, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm2[0,1,2,3,4,5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm14[0,1,2,3],ymm12[4,5],ymm14[6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,1],ymm10[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm10[5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm2[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm11[0,1,2,3],mem[4,5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = mem[0,1],ymm12[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm12[5],ymm4[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm8, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm9[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm4 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = ymm3[0,1,2,3],mem[4,5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,1],ymm10[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm10[5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = mem[0,1],ymm10[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm10[5],ymm5[6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm8, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm10 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm10 = mem[0,1],ymm12[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm12[5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm9, %ymm8, %ymm9
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2],ymm10[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $207, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = mem[0,1,2,3],ymm3[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm10 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm10 = mem[0,1],ymm11[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm11[5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm5, %ymm8, %ymm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm10[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm9[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = ymm6[0,1,2,3],mem[4,5],ymm6[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm9, %ymm8, %ymm8
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm5[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $48, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = ymm5[0,1,2,3],mem[4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm5, %ymm8, %ymm5
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,1],ymm10[0,1]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm10[5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2],ymm9[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vperm2i128 $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm8 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm8 = mem[0,1],ymm10[0,1]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm10[5],ymm8[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm8[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 192(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 128(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm8, 64(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm8, 224(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm8, 160(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm8, 96(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm8, 32(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 192(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 128(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 64(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 224(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 160(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 96(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 192(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 128(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 64(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 224(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 160(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 96(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%rcx)
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm7[6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 192(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 128(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 64(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, (%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 224(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 160(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 96(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 32(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 192(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 128(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 64(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, (%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 224(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 160(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 96(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 32(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 192(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 128(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 64(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, (%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 224(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 160(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 96(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 32(%rcx)
 ; AVX2-FCP-NEXT:    vmovdqa %ymm13, (%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 64(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 128(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 192(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 224(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 160(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 96(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%r8)
-; AVX2-FCP-NEXT:    vmovdqa %ymm7, 224(%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 64(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 128(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 192(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 224(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 160(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 96(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 32(%r8)
+; AVX2-FCP-NEXT:    vmovdqa %ymm5, 224(%r9)
 ; AVX2-FCP-NEXT:    vmovdqa %ymm3, 192(%r9)
 ; AVX2-FCP-NEXT:    vmovdqa %ymm4, 160(%r9)
 ; AVX2-FCP-NEXT:    vmovdqa %ymm2, 128(%r9)
-; AVX2-FCP-NEXT:    vmovdqa %ymm5, 96(%r9)
+; AVX2-FCP-NEXT:    vmovdqa %ymm9, 96(%r9)
 ; AVX2-FCP-NEXT:    vmovdqa %ymm1, 64(%r9)
-; AVX2-FCP-NEXT:    vmovdqa %ymm11, 32(%r9)
+; AVX2-FCP-NEXT:    vmovdqa %ymm6, 32(%r9)
 ; AVX2-FCP-NEXT:    vmovdqa %ymm0, (%r9)
 ; AVX2-FCP-NEXT:    addq $2152, %rsp # imm = 0x868
 ; AVX2-FCP-NEXT:    vzeroupper
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-6.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-6.ll
index aae4d9fa15e246..4cc129151ac800 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-6.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-6.ll
@@ -524,35 +524,32 @@ define void @load_i32_stride6_vf4(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [1,7,5,0]
 ; AVX2-NEXT:    vpermd %ymm3, %ymm5, %ymm3
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1,2],xmm4[3]
-; AVX2-NEXT:    vmovdqa (%rdi), %xmm5
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm6 = xmm5[2,3,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm2[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0],ymm7[1,2,3],ymm6[4],ymm7[5,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,3,2,3]
-; AVX2-NEXT:    vmovdqa 80(%rdi), %xmm7
-; AVX2-NEXT:    vpbroadcastd %xmm7, %xmm8
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm6 = xmm6[0,1,2],xmm8[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[3,3,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm8 = ymm2[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm8[1,2,3],ymm5[4],ymm8[5,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,3,2,3]
-; AVX2-NEXT:    vpbroadcastd 84(%rdi), %xmm8
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1,2],xmm8[3]
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm8 = xmm4[0,1],xmm7[2,3]
+; AVX2-NEXT:    vmovdqa 80(%rdi), %xmm5
+; AVX2-NEXT:    vpbroadcastd %xmm5, %xmm6
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm2[0,1],ymm1[2],ymm2[3,4,5,6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,3,2,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm6 = xmm7[0,1,2],xmm6[3]
+; AVX2-NEXT:    vpbroadcastd 84(%rdi), %xmm7
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,3,2,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm7 = xmm8[0,1,2],xmm7[3]
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm8 = xmm4[0,1],xmm5[2,3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[0,1,0,2]
 ; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm9 = [4,2,0,0]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vpermd %ymm1, %ymm9, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm8[2,3]
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm7[3]
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [5,3,0,0]
-; AVX2-NEXT:    vpermd %ymm1, %ymm7, %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm5[3]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [5,3,0,0]
+; AVX2-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3]
 ; AVX2-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX2-NEXT:    vmovdqa %xmm3, (%rdx)
 ; AVX2-NEXT:    vmovdqa %xmm6, (%rcx)
-; AVX2-NEXT:    vmovdqa %xmm5, (%r8)
+; AVX2-NEXT:    vmovdqa %xmm7, (%r8)
 ; AVX2-NEXT:    vmovdqa %xmm2, (%r9)
 ; AVX2-NEXT:    vmovdqa %xmm1, (%rax)
 ; AVX2-NEXT:    vzeroupper
@@ -572,35 +569,32 @@ define void @load_i32_stride6_vf4(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [1,7,5,0]
 ; AVX2-FP-NEXT:    vpermd %ymm3, %ymm5, %ymm3
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1,2],xmm4[3]
-; AVX2-FP-NEXT:    vmovdqa (%rdi), %xmm5
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm6 = xmm5[2,3,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm2[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0],ymm7[1,2,3],ymm6[4],ymm7[5,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,3,2,3]
-; AVX2-FP-NEXT:    vmovdqa 80(%rdi), %xmm7
-; AVX2-FP-NEXT:    vpbroadcastd %xmm7, %xmm8
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm6 = xmm6[0,1,2],xmm8[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[3,3,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm8 = ymm2[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm8[1,2,3],ymm5[4],ymm8[5,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,3,2,3]
-; AVX2-FP-NEXT:    vpbroadcastd 84(%rdi), %xmm8
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1,2],xmm8[3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm8 = xmm4[0,1],xmm7[2,3]
+; AVX2-FP-NEXT:    vmovdqa 80(%rdi), %xmm5
+; AVX2-FP-NEXT:    vpbroadcastd %xmm5, %xmm6
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm2[0,1],ymm1[2],ymm2[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,3,2,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm6 = xmm7[0,1,2],xmm6[3]
+; AVX2-FP-NEXT:    vpbroadcastd 84(%rdi), %xmm7
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,3,2,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm7 = xmm8[0,1,2],xmm7[3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm8 = xmm4[0,1],xmm5[2,3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[0,1,0,2]
 ; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm9 = [4,2,0,0]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm9, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm8[2,3]
 ; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm7[3]
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [5,3,0,0]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm7, %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm5[3]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [5,3,0,0]
+; AVX2-FP-NEXT:    vpermd %ymm1, %ymm5, %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3]
 ; AVX2-FP-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX2-FP-NEXT:    vmovdqa %xmm3, (%rdx)
 ; AVX2-FP-NEXT:    vmovdqa %xmm6, (%rcx)
-; AVX2-FP-NEXT:    vmovdqa %xmm5, (%r8)
+; AVX2-FP-NEXT:    vmovdqa %xmm7, (%r8)
 ; AVX2-FP-NEXT:    vmovdqa %xmm2, (%r9)
 ; AVX2-FP-NEXT:    vmovdqa %xmm1, (%rax)
 ; AVX2-FP-NEXT:    vzeroupper
@@ -620,35 +614,32 @@ define void @load_i32_stride6_vf4(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [1,7,5,0]
 ; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm5, %ymm3
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1,2],xmm4[3]
-; AVX2-FCP-NEXT:    vmovdqa (%rdi), %xmm5
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm6 = xmm5[2,3,2,3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm2[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0],ymm7[1,2,3],ymm6[4],ymm7[5,6,7]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,3,2,3]
-; AVX2-FCP-NEXT:    vmovdqa 80(%rdi), %xmm7
-; AVX2-FCP-NEXT:    vpbroadcastd %xmm7, %xmm8
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm6 = xmm6[0,1,2],xmm8[3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[3,3,3,3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm8 = ymm2[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm8[1,2,3],ymm5[4],ymm8[5,6,7]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,3,2,3]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [2,0,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm2[0,1],ymm1[2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm6, %ymm5, %ymm5
+; AVX2-FCP-NEXT:    vmovdqa 80(%rdi), %xmm6
+; AVX2-FCP-NEXT:    vpbroadcastd %xmm6, %xmm7
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1,2],xmm7[3]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [3,1,7,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm8, %ymm7, %ymm7
 ; AVX2-FCP-NEXT:    vpbroadcastd 84(%rdi), %xmm8
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1,2],xmm8[3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm8 = xmm4[0,1],xmm7[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm7 = xmm7[0,1,2],xmm8[3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm8 = xmm4[0,1],xmm6[2,3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[0,1,0,2]
 ; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm9 = [4,2,0,0]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm9, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm8[2,3]
 ; AVX2-FCP-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm7[3]
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [5,3,0,0]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm7, %ymm1
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm6[3]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm6 = [5,3,0,0]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm6, %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3]
 ; AVX2-FCP-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX2-FCP-NEXT:    vmovdqa %xmm3, (%rdx)
-; AVX2-FCP-NEXT:    vmovdqa %xmm6, (%rcx)
-; AVX2-FCP-NEXT:    vmovdqa %xmm5, (%r8)
+; AVX2-FCP-NEXT:    vmovdqa %xmm5, (%rcx)
+; AVX2-FCP-NEXT:    vmovdqa %xmm7, (%r8)
 ; AVX2-FCP-NEXT:    vmovdqa %xmm2, (%r9)
 ; AVX2-FCP-NEXT:    vmovdqa %xmm1, (%rax)
 ; AVX2-FCP-NEXT:    vzeroupper
@@ -1103,28 +1094,23 @@ define void @load_i32_stride6_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[2,0,2,3,6,4,6,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,2,0,3]
-; AVX2-NEXT:    vmovaps (%rdi), %xmm11
-; AVX2-NEXT:    vshufps {{.*#+}} xmm12 = xmm11[2,3,2,3]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm4[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0],ymm13[1,2,3],ymm12[4],ymm13[5,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm12[0,1,2],ymm10[3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm12 = ymm1[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm0[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0,1,2],ymm12[3],ymm13[4,5,6],ymm12[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm12[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm12 = ymm5[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0],ymm6[1],ymm12[2,3,4],ymm6[5],ymm12[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,2,0,3]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm11 = xmm11[3,3,3,3]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm4[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0],ymm13[1,2,3],ymm11[4],ymm13[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm4[0,1],ymm2[2],ymm4[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm11 = ymm11[2,0,2,3,6,4,6,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm11 = ymm11[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm12 = ymm1[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm0[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0,1,2],ymm12[3],ymm13[4,5,6],ymm12[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2],ymm10[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3],ymm1[4],ymm0[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm11 = ymm11[0,0,2,0,4,4,6,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm11 = ymm11[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm11[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm11 = ymm5[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0],ymm6[1],ymm11[2,3,4],ymm6[5],ymm11[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm11 = ymm11[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm12 = ymm12[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm12[0,1,2],ymm11[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1,2,3,4],ymm1[5],ymm0[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm12 = ymm12[0,1,3,1,4,5,7,5]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm12[5,6,7]
 ; AVX2-NEXT:    vmovaps 80(%rdi), %xmm12
@@ -1186,28 +1172,23 @@ define void @load_i32_stride6_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[2,0,2,3,6,4,6,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,2,0,3]
-; AVX2-FP-NEXT:    vmovaps (%rdi), %xmm11
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm12 = xmm11[2,3,2,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm4[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0],ymm13[1,2,3],ymm12[4],ymm13[5,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm12[0,1,2],ymm10[3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm12 = ymm1[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm0[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0,1,2],ymm12[3],ymm13[4,5,6],ymm12[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm12[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm12 = ymm5[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0],ymm6[1],ymm12[2,3,4],ymm6[5],ymm12[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,2,0,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm11 = xmm11[3,3,3,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm4[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0],ymm13[1,2,3],ymm11[4],ymm13[5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm4[0,1],ymm2[2],ymm4[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm11 = ymm11[2,0,2,3,6,4,6,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm11 = ymm11[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm12 = ymm1[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm0[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0,1,2],ymm12[3],ymm13[4,5,6],ymm12[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2],ymm10[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3],ymm1[4],ymm0[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm11 = ymm11[0,0,2,0,4,4,6,4]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm11 = ymm11[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm11[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm11 = ymm5[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0],ymm6[1],ymm11[2,3,4],ymm6[5],ymm11[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm11 = ymm11[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm12 = ymm12[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm12[0,1,2],ymm11[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1,2,3,4],ymm1[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm12 = ymm12[0,1,3,1,4,5,7,5]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm12[5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 80(%rdi), %xmm12
@@ -1269,30 +1250,26 @@ define void @load_i32_stride6_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]
 ; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm11 = [2,4,2,4,2,4,2,4]
 ; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm11, %ymm10
-; AVX2-FCP-NEXT:    vmovaps (%rdi), %xmm11
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm12 = xmm11[2,3,2,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm13 = ymm4[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0],ymm13[1,2,3],ymm12[4],ymm13[5,6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm12[0,1,2],ymm10[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm12 = ymm1[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm13 = ymm0[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0,1,2],ymm12[3],ymm13[4,5,6],ymm12[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm12[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm12 = ymm5[3,3,3,3,7,7,7,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm11 = [2,0,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm4[0,1],ymm2[2],ymm4[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm11, %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2],ymm10[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3],ymm1[4],ymm0[5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm12 = [0,0,6,4,0,0,6,4]
+; AVX2-FCP-NEXT:    # ymm12 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm12, %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm11[5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3,4],ymm1[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm12 = [0,1,7,5,0,1,7,5]
+; AVX2-FCP-NEXT:    # ymm12 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm12, %ymm11
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm12 = mem[3,3,3,3]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0],ymm6[1],ymm12[2,3,4],ymm6[5],ymm12[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,2,0,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm11 = xmm11[3,3,3,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm13 = ymm4[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0],ymm13[1,2,3],ymm11[4],ymm13[5,6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm11 = ymm11[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm12 = ymm1[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm13 = ymm0[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0,1,2],ymm12[3],ymm13[4,5,6],ymm12[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm12 = ymm12[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm12[5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm13 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0],ymm13[1,2,3,4],ymm12[5],ymm13[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm13 = [3,1,7,5,0,u,u,u]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm13, %ymm12
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm12[0,1,2,3,4],ymm11[5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 80(%rdi), %xmm12
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1,2,3],ymm5[4,5,6,7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm5[0,1],ymm12[2,3],ymm5[4,5,6,7]
@@ -2171,557 +2148,528 @@ define void @load_i32_stride6_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-LABEL: load_i32_stride6_vf16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    subq $360, %rsp # imm = 0x168
-; AVX2-NEXT:    vmovaps 288(%rdi), %ymm8
-; AVX2-NEXT:    vmovaps 224(%rdi), %ymm11
-; AVX2-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 192(%rdi), %ymm1
+; AVX2-NEXT:    subq $328, %rsp # imm = 0x148
+; AVX2-NEXT:    vmovaps 288(%rdi), %ymm13
+; AVX2-NEXT:    vmovaps 224(%rdi), %ymm2
+; AVX2-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 192(%rdi), %ymm3
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 128(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 128(%rdi), %ymm0
-; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 160(%rdi), %ymm2
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 160(%rdi), %ymm4
+; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 96(%rdi), %ymm15
-; AVX2-NEXT:    vmovaps (%rdi), %ymm3
-; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 32(%rdi), %ymm13
-; AVX2-NEXT:    vmovaps 64(%rdi), %ymm6
-; AVX2-NEXT:    vmovaps {{.*#+}} xmm14 = [0,6,4,u]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm3[0,1,2,3],ymm13[4,5],ymm3[6,7]
-; AVX2-NEXT:    vpermps %ymm4, %ymm14, %ymm3
-; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm7 = ymm6[0,1],ymm15[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm7[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm9[0,2,2,2,4,6,6,6]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2],ymm7[3,4,5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm2[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [4,2,4,2,4,2,4,2]
-; AVX2-NEXT:    vpermps %ymm10, %ymm0, %ymm12
-; AVX2-NEXT:    vmovaps %ymm0, %ymm5
+; AVX2-NEXT:    vmovaps (%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-NEXT:    vmovaps 32(%rdi), %ymm10
+; AVX2-NEXT:    vmovaps 64(%rdi), %ymm6
+; AVX2-NEXT:    vmovaps {{.*#+}} xmm12 = [0,6,4,u]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm0[0,1,2,3],ymm10[4,5],ymm0[6,7]
+; AVX2-NEXT:    vpermps %ymm8, %ymm12, %ymm7
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm9 = ymm6[0,1],ymm15[0,1]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm11 = ymm9[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm7[0,1,2],ymm11[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm1 = [4,2,4,2,4,2,4,2]
+; AVX2-NEXT:    vpermps %ymm4, %ymm1, %ymm14
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1,2,3,4,5],ymm14[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0,1,2,3],ymm11[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpermps %ymm3, %ymm14, %ymm12
-; AVX2-NEXT:    vmovaps 256(%rdi), %ymm7
-; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm14 = ymm7[0,1],ymm8[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm14 = ymm1[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm3[0,1,2,3],ymm2[4,5],ymm3[6,7]
+; AVX2-NEXT:    vpermps %ymm5, %ymm12, %ymm12
+; AVX2-NEXT:    vmovaps 256(%rdi), %ymm11
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm14 = ymm11[0,1],ymm13[0,1]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm14[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm14 = ymm3[0,2,2,2,4,6,6,6]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm12[0,1,2],ymm14[3,4,5,6,7]
 ; AVX2-NEXT:    vmovaps 320(%rdi), %ymm12
 ; AVX2-NEXT:    vmovaps 352(%rdi), %ymm14
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1,2,3],ymm12[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm0, %ymm5, %ymm11
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm7[6,7]
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps {{.*#+}} xmm2 = [1,7,5,u]
-; AVX2-NEXT:    vpermps %ymm4, %ymm2, %ymm4
-; AVX2-NEXT:    vshufps {{.*#+}} ymm9 = ymm9[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm5 = [5,3,5,3,5,3,5,3]
-; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm10, %ymm5, %ymm10
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-NEXT:    vpermps %ymm8, %ymm2, %ymm7
+; AVX2-NEXT:    vshufps {{.*#+}} ymm8 = ymm9[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2],ymm8[3,4,5,6,7]
+; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm9 = [5,3,5,3,5,3,5,3]
+; AVX2-NEXT:    vpermps %ymm4, %ymm9, %ymm4
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm7[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vpermps %ymm5, %ymm2, %ymm2
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[1,3,2,3,5,7,6,7]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vpermps %ymm0, %ymm9, %ymm0
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps (%rdi), %xmm5
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm5[2,3,2,3]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1],ymm15[2,3],ymm6[4,5],ymm15[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1],ymm15[2,3],ymm6[4,5],ymm15[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm10[0,1],ymm5[2],ymm10[3,4,5,6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm4[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm8[4],ymm2[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 192(%rdi), %xmm0
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm9[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm7[0,1],ymm8[2,3],ymm7[4,5],ymm8[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1],ymm13[2,3],ymm11[4,5],ymm13[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1],ymm4[2],ymm3[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm14[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm12[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm5[3,3,3,3]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm13[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm15[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm6[1],ymm2[2,3,4],ymm6[5],ymm2[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1,2,3],ymm14[4],ymm12[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm15[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm6[1],ymm0[2,3,4],ymm6[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm10[0,1,2],ymm5[3],ymm10[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm10[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm4[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vmovaps %ymm4, %ymm5
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm7[1],ymm1[2,3,4],ymm7[5],ymm1[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm14[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm12[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm8[5],ymm2[6,7]
+; AVX2-NEXT:    vmovaps %ymm2, %ymm7
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm15[4,5,6,7]
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm13[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovaps 80(%rdi), %xmm3
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1],ymm3[2,3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm11[1],ymm1[2,3,4],ymm11[5],ymm1[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm4[3],ymm3[4,5,6,7]
+; AVX2-NEXT:    vmovaps %ymm3, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm12[0,1,2,3,4],ymm14[5],ymm12[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,1,3,1,4,5,7,5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm6[0,1,2,3],ymm15[4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm10[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-NEXT:    vmovaps 80(%rdi), %xmm4
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm4[2,3],ymm2[4,5,6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm6
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm10[0,1],ymm5[2,3],ymm10[4,5,6,7]
-; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm11 = [0,2,0,6,0,2,0,6]
-; AVX2-NEXT:    # ymm11 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermps %ymm6, %ymm11, %ymm13
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm13[5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = ymm9[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovaps 272(%rdi), %xmm13
-; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm5[0,1],ymm13[2,3],ymm5[4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm15 = ymm15[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm7, %ymm0, %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm15[2,3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-NEXT:    vpermps %ymm3, %ymm15, %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1],ymm1[2,3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm8[0,1],ymm7[2,3],ymm8[4,5,6,7]
+; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm7 = [0,2,0,6,0,2,0,6]
+; AVX2-NEXT:    # ymm7 = mem[0,1,0,1]
+; AVX2-NEXT:    vpermps %ymm5, %ymm7, %ymm10
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm10[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm11[0,1,2,3],ymm13[4,5,6,7]
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm10 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm10 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovaps 272(%rdi), %xmm11
+; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm6[0,1],ymm11[2,3],ymm6[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vpermps %ymm10, %ymm15, %ymm15
+; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm15[0,1],ymm13[2,3,4,5,6,7]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm14[0,1],ymm12[2,3],ymm14[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm12, %ymm11, %ymm11
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm11[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm4[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm12, %ymm7, %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm13[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm3, %ymm9, %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm3 = [0,3,1,7,0,3,1,7]
+; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX2-NEXT:    vpermps %ymm5, %ymm3, %ymm4
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm6[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm11[3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm10, %ymm9, %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm12, %ymm3, %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm0, 32(%rsi)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm2, %ymm4, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [0,3,1,7,0,3,1,7]
-; AVX2-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermps %ymm6, %ymm2, %ymm3
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm5[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2],ymm13[3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm7, %ymm4, %ymm5
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm12, %ymm2, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, 32(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, (%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, 32(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, (%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, 32(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, (%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, 32(%r8)
-; AVX2-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, (%r8)
-; AVX2-NEXT:    vmovaps %ymm8, 32(%r9)
+; AVX2-NEXT:    vmovaps %ymm4, (%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm0, 32(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm0, (%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm0, 32(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm0, (%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm0, 32(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm0, (%r8)
+; AVX2-NEXT:    vmovaps %ymm7, 32(%r9)
 ; AVX2-NEXT:    vmovaps %ymm1, (%r9)
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-NEXT:    vmovaps %ymm2, 32(%rax)
-; AVX2-NEXT:    vmovaps %ymm0, (%rax)
-; AVX2-NEXT:    addq $360, %rsp # imm = 0x168
+; AVX2-NEXT:    vmovaps %ymm3, 32(%rax)
+; AVX2-NEXT:    vmovaps %ymm2, (%rax)
+; AVX2-NEXT:    addq $328, %rsp # imm = 0x148
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: load_i32_stride6_vf16:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    subq $360, %rsp # imm = 0x168
-; AVX2-FP-NEXT:    vmovaps 288(%rdi), %ymm8
-; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm11
-; AVX2-FP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm1
+; AVX2-FP-NEXT:    subq $328, %rsp # imm = 0x148
+; AVX2-FP-NEXT:    vmovaps 288(%rdi), %ymm13
+; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm0
-; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm4
+; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 96(%rdi), %ymm15
-; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm3
-; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm13
-; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm6
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm14 = [0,6,4,u]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm3[0,1,2,3],ymm13[4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm14, %ymm3
-; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm7 = ymm6[0,1],ymm15[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm7[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm9[0,2,2,2,4,6,6,6]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2],ymm7[3,4,5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm2[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [4,2,4,2,4,2,4,2]
-; AVX2-FP-NEXT:    vpermps %ymm10, %ymm0, %ymm12
-; AVX2-FP-NEXT:    vmovaps %ymm0, %ymm5
+; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm10
+; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm6
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm12 = [0,6,4,u]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm0[0,1,2,3],ymm10[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm12, %ymm7
+; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm9 = ymm6[0,1],ymm15[0,1]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm11 = ymm9[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm7[0,1,2],ymm11[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm1 = [4,2,4,2,4,2,4,2]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm1, %ymm14
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1,2,3,4,5],ymm14[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0,1,2,3],ymm11[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm14, %ymm12
-; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm7
-; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm14 = ymm7[0,1],ymm8[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm14 = ymm1[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm3[0,1,2,3],ymm2[4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm12, %ymm12
+; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm11
+; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm14 = ymm11[0,1],ymm13[0,1]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm14[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm14 = ymm3[0,2,2,2,4,6,6,6]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm12[0,1,2],ymm14[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm12
 ; AVX2-FP-NEXT:    vmovaps 352(%rdi), %ymm14
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1,2,3],ymm12[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm0, %ymm5, %ymm11
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm1, %ymm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm7[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm2 = [1,7,5,u]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm2, %ymm4
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm9 = ymm9[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm5 = [5,3,5,3,5,3,5,3]
-; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm10, %ymm5, %ymm10
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm2, %ymm7
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm8 = ymm9[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2],ymm8[3,4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm9 = [5,3,5,3,5,3,5,3]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm9, %ymm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm7[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm2, %ymm2
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm2, %ymm2
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[1,3,2,3,5,7,6,7]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm9, %ymm0
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps (%rdi), %xmm5
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm5[2,3,2,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1],ymm15[2,3],ymm6[4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1],ymm15[2,3],ymm6[4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm10[0,1],ymm5[2],ymm10[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm4[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm8[4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 192(%rdi), %xmm0
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm9[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm7[0,1],ymm8[2,3],ymm7[4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1],ymm13[2,3],ymm11[4,5],ymm13[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1],ymm4[2],ymm3[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm14[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm12[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm5[3,3,3,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm13[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm15[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm6[1],ymm2[2,3,4],ymm6[5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1,2,3],ymm14[4],ymm12[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm15[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm6[1],ymm0[2,3,4],ymm6[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm10[0,1,2],ymm5[3],ymm10[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm10[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm4[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vmovaps %ymm4, %ymm5
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm7[1],ymm1[2,3,4],ymm7[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm14[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm12[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm8[5],ymm2[6,7]
+; AVX2-FP-NEXT:    vmovaps %ymm2, %ymm7
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm15[4,5,6,7]
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm13[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 80(%rdi), %xmm3
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1],ymm3[2,3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm11[1],ymm1[2,3,4],ymm11[5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm4[3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps %ymm3, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm12[0,1,2,3,4],ymm14[5],ymm12[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,1,3,1,4,5,7,5]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm6[0,1,2,3],ymm15[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm10[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 80(%rdi), %xmm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm4[2,3],ymm2[4,5,6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm6
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm10[0,1],ymm5[2,3],ymm10[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm11 = [0,2,0,6,0,2,0,6]
-; AVX2-FP-NEXT:    # ymm11 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermps %ymm6, %ymm11, %ymm13
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm13[5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = ymm9[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 272(%rdi), %xmm13
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm5[0,1],ymm13[2,3],ymm5[4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm15 = ymm15[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm7, %ymm0, %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm15[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm15, %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1],ymm1[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm8[0,1],ymm7[2,3],ymm8[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm7 = [0,2,0,6,0,2,0,6]
+; AVX2-FP-NEXT:    # ymm7 = mem[0,1,0,1]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm7, %ymm10
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm10[5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm11[0,1,2,3],ymm13[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm10 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm10 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 272(%rdi), %xmm11
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm6[0,1],ymm11[2,3],ymm6[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[0,1,0,2,4,5,4,6]
+; AVX2-FP-NEXT:    vpermps %ymm10, %ymm15, %ymm15
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm15[0,1],ymm13[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm14[0,1],ymm12[2,3],ymm14[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm12, %ymm11, %ymm11
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm11[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm4[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm7, %ymm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm13[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm9, %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm3 = [0,3,1,7,0,3,1,7]
+; AVX2-FP-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm3, %ymm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm6[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm11[3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm10, %ymm9, %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm3, %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rsi)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm4, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [0,3,1,7,0,3,1,7]
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermps %ymm6, %ymm2, %ymm3
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm5[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2],ymm13[3],ymm3[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm7, %ymm4, %ymm5
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm12, %ymm2, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, (%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, (%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, (%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%r8)
-; AVX2-FP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, (%r8)
-; AVX2-FP-NEXT:    vmovaps %ymm8, 32(%r9)
+; AVX2-FP-NEXT:    vmovaps %ymm4, (%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm0, (%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm0, (%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm0, (%r8)
+; AVX2-FP-NEXT:    vmovaps %ymm7, 32(%r9)
 ; AVX2-FP-NEXT:    vmovaps %ymm1, (%r9)
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FP-NEXT:    vmovaps %ymm2, 32(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm0, (%rax)
-; AVX2-FP-NEXT:    addq $360, %rsp # imm = 0x168
+; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm2, (%rax)
+; AVX2-FP-NEXT:    addq $328, %rsp # imm = 0x148
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
 ; AVX2-FCP-LABEL: load_i32_stride6_vf16:
 ; AVX2-FCP:       # %bb.0:
-; AVX2-FCP-NEXT:    subq $328, %rsp # imm = 0x148
-; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm14
-; AVX2-FCP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    subq $392, %rsp # imm = 0x188
+; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm10
+; AVX2-FCP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %ymm15
 ; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm13
-; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm6
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm0 = [0,6,4,u]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm3[0,1,2,3],ymm13[4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm0, %ymm7
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm8 = ymm6[0,1],ymm15[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm9 = ymm8[0,2,2,2,4,6,6,6]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm7[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm1 = [4,2,4,2,4,2,4,2]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm1, %ymm12
-; AVX2-FCP-NEXT:    vmovaps %ymm1, %ymm5
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm9[0,1,2,3,4,5],ymm12[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm11[0,1,2,3],ymm14[4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm9
-; AVX2-FCP-NEXT:    vmovaps 256(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm12 = ymm7[0,1],ymm10[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm12 = ymm1[0,2,2,2,4,6,6,6]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2],ymm12[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm13
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm12 = [0,6,4,u]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm3[0,1,2,3],ymm4[4,5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vmovaps %ymm3, %ymm5
+; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm12, %ymm7
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm9 = ymm13[0,1],ymm15[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm11 = ymm9[0,2,2,2,4,6,6,6]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm7[0,1,2],ymm11[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [4,2,4,2,4,2,4,2]
+; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm0, %ymm14
+; AVX2-FCP-NEXT:    vmovaps %ymm0, %ymm6
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1,2,3,4,5],ymm14[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0,1,2,3],ymm10[4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm12, %ymm12
+; AVX2-FCP-NEXT:    vmovaps 256(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm14 = ymm11[0,1],ymm10[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm14 = ymm1[0,2,2,2,4,6,6,6]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm12[0,1,2],ymm14[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm12
 ; AVX2-FCP-NEXT:    vmovaps 352(%rdi), %ymm14
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1,2,3],ymm12[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm11 = [1,7,5,u]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm11, %ymm4
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm8 = ymm8[1,3,2,3,5,7,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm8[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm9 = [5,3,5,3,5,3,5,3]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm9, %ymm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm11, %ymm2
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm6, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm2 = [1,7,5,u]
+; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm2, %ymm6
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm8 = ymm9[1,3,2,3,5,7,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2],ymm8[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm7 = [5,3,5,3,5,3,5,3]
+; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm7, %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm2
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,3,2,3,5,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm9, %ymm0
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm7, %ymm0
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps (%rdi), %xmm0
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm13[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm6[0,1],ymm15[2,3],ymm6[4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm3 = [2,0,6,4,2,0,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm3, %ymm2
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm8[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm5[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1],ymm10[2,3],ymm7[4,5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm3, %ymm1
-; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm3 = xmm2[2,3,2,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm11[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0],ymm4[1,2,3],ymm3[4],ymm4[5,6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm14[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm12[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3],ymm4[4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm15[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm6[1],ymm1[2,3,4],ymm6[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm13[0,1],ymm15[2,3],ymm13[4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm1 = [2,0,6,4,2,0,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm2 = [2,0,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm7[0,1],ymm5[2],ymm7[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps %ymm5, %ymm9
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm6[4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm5 = [0,0,6,4,0,0,6,4]
+; AVX2-FCP-NEXT:    # ymm5 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm5, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm8 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1],ymm8[2],ymm3[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm11[0,1],ymm10[2,3],ymm11[4,5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm5[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3],ymm3[4,5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1,2,3],ymm14[4],ymm12[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm0 = xmm2[3,3,3,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm11[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vmovaps %ymm11, %ymm8
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm7[1],ymm1[2,3,4],ymm7[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm14[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm12[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm13[1],ymm0[2,3,4],ymm13[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1,2],ymm9[3],ymm7[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4],ymm0[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3,4],ymm6[5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vmovaps %ymm4, %ymm10
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [0,1,7,5,0,1,7,5]
+; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm5 = [3,1,7,5,0,u,u,u]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm15[4,5,6,7]
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = ymm13[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 80(%rdi), %xmm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1],ymm3[2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm1 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm11[1],ymm1[2,3,4],ymm11[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm3[0,1,2],ymm8[3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps %ymm3, %ymm0
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm4[1,2,3,4],ymm1[5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm5, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm12[0,1,2,3,4],ymm14[5],ymm12[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm13[0,1,2,3],ymm15[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm7[0,1,2,3],ymm9[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps 80(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm4[2,3],ymm2[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm6
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = mem[0,1],ymm5[2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm11 = [0,2,0,6,0,2,0,6]
-; AVX2-FCP-NEXT:    # ymm11 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm11, %ymm13
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm13[5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0,1,2,3],ymm10[4,5,6,7]
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm7 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm7 = ymm8[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm7, %ymm5
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1],ymm1[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1],ymm10[2,3],ymm6[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm6 = [0,2,0,6,0,2,0,6]
+; AVX2-FCP-NEXT:    # ymm6 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm6, %ymm10
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm10[5,6,7]
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm10 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm10 = ymm11[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3],ymm8[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 272(%rdi), %xmm13
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm5[0,1],ymm13[2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm10[0,1],ymm13[2,3],ymm10[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm15 = ymm15[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm0, %ymm10
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1],ymm15[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm7, %ymm7
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1],ymm15[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm14[0,1],ymm12[2,3],ymm14[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm11, %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4],ymm11[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm4[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3],ymm0[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm9, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [0,3,1,7,0,3,1,7]
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm2, %ymm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm5[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2],ymm13[3],ymm3[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm9, %ymm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm2, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%r8)
-; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, (%r8)
-; AVX2-FCP-NEXT:    vmovaps %ymm10, 32(%r9)
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm6, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm0, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm3 = [0,3,1,7,0,3,1,7]
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm3, %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm10[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm13[3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm0, %ymm5
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm3, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm0, (%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm0, (%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm0, (%r8)
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%r9)
 ; AVX2-FCP-NEXT:    vmovaps %ymm1, (%r9)
 ; AVX2-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 32(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm0, (%rax)
-; AVX2-FCP-NEXT:    addq $328, %rsp # imm = 0x148
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm2, (%rax)
+; AVX2-FCP-NEXT:    addq $392, %rsp # imm = 0x188
 ; AVX2-FCP-NEXT:    vzeroupper
 ; AVX2-FCP-NEXT:    retq
 ;
@@ -4501,12 +4449,12 @@ define void @load_i32_stride6_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-LABEL: load_i32_stride6_vf32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    subq $1256, %rsp # imm = 0x4E8
-; AVX2-NEXT:    vmovaps 480(%rdi), %ymm9
-; AVX2-NEXT:    vmovaps 448(%rdi), %ymm10
-; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 416(%rdi), %ymm7
+; AVX2-NEXT:    subq $1288, %rsp # imm = 0x508
+; AVX2-NEXT:    vmovaps 480(%rdi), %ymm11
+; AVX2-NEXT:    vmovaps 448(%rdi), %ymm7
 ; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 416(%rdi), %ymm10
+; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 128(%rdi), %ymm2
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 160(%rdi), %ymm3
@@ -4519,35 +4467,35 @@ define void @load_i32_stride6_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 64(%rdi), %ymm5
 ; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps {{.*#+}} xmm8 = [0,6,4,u]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-NEXT:    vpermps %ymm13, %ymm8, %ymm0
+; AVX2-NEXT:    vmovaps {{.*#+}} xmm9 = [0,6,4,u]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-NEXT:    vpermps %ymm8, %ymm9, %ymm0
 ; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm5[0,1],ymm4[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm1[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,2,2,2,4,6,6,6]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm3[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4,2,4,2,4,2,4,2]
 ; AVX2-NEXT:    vpermps %ymm6, %ymm2, %ymm1
-; AVX2-NEXT:    vmovaps %ymm2, %ymm11
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm10[0,1],ymm9[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm7[0,1],ymm11[0,1]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1,2,3,4,5],ymm11[6,7]
 ; AVX2-NEXT:    vmovaps 384(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1,2,3],ymm7[4,5],ymm0[6,7]
-; AVX2-NEXT:    vpermps %ymm4, %ymm8, %ymm0
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1,2,3],ymm10[4,5],ymm0[6,7]
+; AVX2-NEXT:    vpermps %ymm4, %ymm9, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[0,2,2,2,4,6,6,6]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovaps 512(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 544(%rdi), %ymm2
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm11, %ymm1
-; AVX2-NEXT:    vmovaps %ymm11, %ymm12
+; AVX2-NEXT:    vmovaps 544(%rdi), %ymm3
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm3, %ymm2, %ymm1
+; AVX2-NEXT:    vmovaps %ymm2, %ymm15
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 288(%rdi), %ymm1
@@ -4555,374 +4503,353 @@ define void @load_i32_stride6_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovaps 256(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm1[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-NEXT:    vmovaps 224(%rdi), %ymm1
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 192(%rdi), %ymm0
+; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vmovaps 224(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-NEXT:    vpermps %ymm1, %ymm8, %ymm14
-; AVX2-NEXT:    vshufps {{.*#+}} ymm9 = ymm10[0,2,2,2,4,6,6,6]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm14[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-NEXT:    vmovaps 320(%rdi), %ymm0
+; AVX2-NEXT:    vmovaps 192(%rdi), %ymm1
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpermps %ymm2, %ymm9, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm11 = ymm13[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2],ymm11[3,4,5,6,7]
+; AVX2-NEXT:    vmovaps 320(%rdi), %ymm1
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 352(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 352(%rdi), %ymm7
-; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm7[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm14, %ymm11, %ymm7
-; AVX2-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm9[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 608(%rdi), %ymm0
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm7, %ymm15, %ymm14
+; AVX2-NEXT:    vmovaps %ymm15, %ymm1
+; AVX2-NEXT:    vmovups %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5],ymm14[6,7]
+; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 608(%rdi), %ymm10
+; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 576(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 576(%rdi), %ymm7
-; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm0[4,5],ymm7[6,7]
-; AVX2-NEXT:    vpermps %ymm7, %ymm8, %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm0[0,1,2,3],ymm10[4,5],ymm0[6,7]
+; AVX2-NEXT:    vpermps %ymm14, %ymm9, %ymm9
 ; AVX2-NEXT:    vmovaps 672(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 640(%rdi), %ymm9
-; AVX2-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm9 = ymm9[0,1],ymm0[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm11 = ymm9[0,2,2,2,4,6,6,6]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2],ymm11[3,4,5,6,7]
-; AVX2-NEXT:    vmovaps 704(%rdi), %ymm11
-; AVX2-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 640(%rdi), %ymm10
+; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm12 = ymm10[0,1],ymm0[0,1]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm15 = ymm12[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-NEXT:    vmovaps 704(%rdi), %ymm10
+; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 736(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3],ymm11[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm11, %ymm12, %ymm15
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps {{.*#+}} xmm8 = [1,7,5,u]
-; AVX2-NEXT:    vpermps %ymm13, %ymm8, %ymm13
+; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm0[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm15, %ymm1, %ymm10
+; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps {{.*#+}} xmm9 = [1,7,5,u]
+; AVX2-NEXT:    vpermps %ymm8, %ymm9, %ymm8
+; AVX2-NEXT:    vpermilps $237, (%rsp), %ymm10 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm10 = mem[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2],ymm10[3,4,5,6,7]
+; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm10 = [5,3,5,3,5,3,5,3]
+; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermps %ymm6, %ymm10, %ymm6
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-NEXT:    vmovups %ymm6, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vpermps %ymm4, %ymm9, %ymm4
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm13[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm12 = [5,3,5,3,5,3,5,3]
-; AVX2-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm6, %ymm12, %ymm6
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm4, %ymm8, %ymm4
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm12, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm1, %ymm8, %ymm0
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm14, %ymm12, %ymm1
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm3, %ymm10, %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermps %ymm2, %ymm9, %ymm1
+; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm13[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm7, %ymm10, %ymm0
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm7, %ymm8, %ymm0
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vpermps %ymm14, %ymm9, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm12[1,3,2,3,5,7,6,7]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm11, %ymm12, %ymm1
+; AVX2-NEXT:    vpermps %ymm15, %ymm10, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps (%rdi), %xmm3
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm3[2,3,2,3]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm7[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1],ymm11[2,3],ymm12[4,5],ymm11[6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1],ymm10[2,3],ymm11[4,5],ymm10[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1],ymm4[2],ymm14[3,4,5,6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm8[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm5[4],ymm3[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 384(%rdi), %xmm1
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[2,3,2,3]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm2[1,2,3],ymm0[4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5],ymm2[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 576(%rdi), %xmm0
-; AVX2-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm2[1,2,3],ymm0[4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm2[0,1],mem[2,3],ymm2[4,5],mem[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm5[0,1,2],ymm2[3],ymm5[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1],ymm0[2,3],mem[4,5],ymm0[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 192(%rdi), %xmm0
-; AVX2-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[2,3,2,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1],ymm0[2,3],mem[4,5],ymm0[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm13[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm5[1,2,3],ymm2[4],ymm5[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm15[0,1],ymm14[2,3],ymm15[4,5],ymm14[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm10[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm6 = ymm4[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1,2],ymm5[3],ymm6[4,5,6],ymm5[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm2 = xmm3[3,3,3,3]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm7[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm3[1,2,3],ymm2[4],ymm3[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm11[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0],ymm12[1],ymm3[2,3,4],ymm12[5],ymm3[6,7]
-; AVX2-NEXT:    vmovaps %ymm12, %ymm11
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm9[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm8[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2],ymm3[3],ymm5[4,5,6],ymm3[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,3,3,3]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm12[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm6[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm5[1],ymm2[2,3,4],ymm5[5],ymm2[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1],ymm13[2],ymm12[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $239, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4],mem[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm15[0,1],ymm6[2,3],ymm15[4,5],ymm6[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm9[0,1,0,1,4,5,4,5]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm9[0,1],ymm8[2],ymm9[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm7[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX2-NEXT:    vmovaps %ymm13, %ymm3
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm14[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm15[1],ymm1[2,3,4],ymm15[5],ymm1[6,7]
-; AVX2-NEXT:    vmovaps %ymm15, %ymm14
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm4[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm7[4],ymm2[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermilps $255, (%rsp), %xmm0 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm8[1],ymm1[2,3,4],ymm8[5],ymm1[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm10[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm11[1],ymm0[2,3,4],ymm11[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2],ymm4[3],ymm14[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3,4],ymm5[5],ymm3[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm10[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm3[1],ymm0[2,3,4],ymm3[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2],ymm5[3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm11[0,1,2,3,4],ymm14[5],ymm11[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm6[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm15[1],ymm0[2,3,4],ymm15[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm9[0,1,2],ymm8[3],ymm9[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm7[5],ymm2[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm9[1],ymm0[2,3,4],ymm9[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1,2],ymm13[3],ymm12[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-NEXT:    vblendps $223, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1,2,3,4],ymm13[5],mem[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm10[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm12[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm5[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 464(%rdi), %xmm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm2, %ymm12, %ymm1
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vpermps %ymm2, %ymm7, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm9[0,1],ymm7[2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm9 = [0,2,0,6,0,2,0,6]
-; AVX2-NEXT:    # ymm9 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermps %ymm13, %ymm9, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm14[0,1],ymm11[2,3],ymm14[4,5,6,7]
+; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm8 = [0,2,0,6,0,2,0,6]
+; AVX2-NEXT:    # ymm8 = mem[0,1,0,1]
+; AVX2-NEXT:    vpermps %ymm5, %ymm8, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = ymm11[0,1,2,3],mem[4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm11 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm11 = ymm0[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovaps 80(%rdi), %xmm6
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm7[0,1],ymm6[2,3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm12 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm12 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovaps 80(%rdi), %xmm14
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm4[0,1],ymm14[2,3],ymm4[4,5,6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm11, %ymm12, %ymm1
-; AVX2-NEXT:    vmovaps %ymm12, %ymm15
+; AVX2-NEXT:    vpermps %ymm12, %ymm7, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm4, %ymm9, %ymm1
+; AVX2-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm2, %ymm8, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm10 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm10 = ymm14[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = ymm3[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm15[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm6 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm6 = mem[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-NEXT:    vmovaps 272(%rdi), %xmm3
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0,1],ymm3[2,3],ymm10[4,5,6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm5, %ymm12, %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm6, %ymm7, %ymm11
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm0[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm9, %ymm12
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1,2,3,4],ymm12[5,6,7]
+; AVX2-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm0[0,1],mem[2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm8, %ymm15
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1,2,3,4],ymm15[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm12 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm12 = ymm8[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm11 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm11 = ymm9[0,1,2,3],mem[4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm15 = mem[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-NEXT:    vmovaps 656(%rdi), %xmm0
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm12[0,1],ymm0[2,3],ymm12[4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm1, %ymm15, %ymm15
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm15 = mem[0,1],ymm15[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm15, %ymm9, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm14[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1,2],ymm6[3],ymm7[4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm11, %ymm14, %ymm7
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1],ymm6[2,3,4,5,6,7]
-; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm7 = [0,3,1,7,0,3,1,7]
-; AVX2-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermps %ymm4, %ymm7, %ymm4
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm6 = mem[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm6 = ymm6[0,1,2],mem[3],ymm6[4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm11 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm11[0,1],ymm6[2,3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm13, %ymm7, %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm8 = ymm10[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm8[0,1,2],ymm3[3],ymm8[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm5, %ymm14, %ymm5
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm7, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm12[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm11[0,1],ymm0[2,3],ymm11[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm9 = ymm9[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vpermps %ymm15, %ymm7, %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1],ymm9[2,3,4,5,6,7]
+; AVX2-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm9 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm9 = ymm13[0,1],mem[2,3],ymm13[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm9, %ymm8, %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm8[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm14[3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-NEXT:    vpermps %ymm12, %ymm13, %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm8[0,1],ymm4[2,3,4,5,6,7]
+; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm12 = [0,3,1,7,0,3,1,7]
+; AVX2-NEXT:    # ymm12 = mem[0,1,0,1]
+; AVX2-NEXT:    vpermps %ymm2, %ymm12, %ymm2
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = mem[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = ymm2[0,1,2],mem[3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm5, %ymm12, %ymm4
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm10[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm6, %ymm13, %ymm4
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm12, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm11[1,1,1,1,5,5,5,5]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm1, %ymm14, %ymm1
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm15, %ymm7, %ymm1
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 96(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 32(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 64(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, (%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 96(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 32(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 64(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, (%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 32(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 96(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 64(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, (%rcx)
-; AVX2-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 96(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 32(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 64(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, (%r8)
-; AVX2-NEXT:    vmovaps %ymm9, 96(%r9)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 32(%r9)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, (%r9)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm1, 64(%r9)
+; AVX2-NEXT:    vpermps %ymm15, %ymm13, %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm9, %ymm12, %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 96(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 32(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 64(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, (%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 96(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 32(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 64(%rdx)
+; AVX2-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, (%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 32(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 96(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 64(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, (%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 96(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 32(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 64(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, (%r8)
+; AVX2-NEXT:    vmovaps %ymm7, 96(%r9)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 32(%r9)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, (%r9)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm3, 64(%r9)
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-NEXT:    vmovaps %ymm0, 96(%rax)
-; AVX2-NEXT:    vmovaps %ymm2, 32(%rax)
-; AVX2-NEXT:    vmovaps %ymm6, 64(%rax)
-; AVX2-NEXT:    vmovaps %ymm4, (%rax)
-; AVX2-NEXT:    addq $1256, %rsp # imm = 0x4E8
+; AVX2-NEXT:    vmovaps %ymm1, 32(%rax)
+; AVX2-NEXT:    vmovaps %ymm2, 64(%rax)
+; AVX2-NEXT:    vmovaps %ymm8, (%rax)
+; AVX2-NEXT:    addq $1288, %rsp # imm = 0x508
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: load_i32_stride6_vf32:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    subq $1256, %rsp # imm = 0x4E8
-; AVX2-FP-NEXT:    vmovaps 480(%rdi), %ymm9
-; AVX2-FP-NEXT:    vmovaps 448(%rdi), %ymm10
-; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 416(%rdi), %ymm7
+; AVX2-FP-NEXT:    subq $1288, %rsp # imm = 0x508
+; AVX2-FP-NEXT:    vmovaps 480(%rdi), %ymm11
+; AVX2-FP-NEXT:    vmovaps 448(%rdi), %ymm7
 ; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 416(%rdi), %ymm10
+; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm3
@@ -4935,35 +4862,35 @@ define void @load_i32_stride6_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm5
 ; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm8 = [0,6,4,u]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm13, %ymm8, %ymm0
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm9 = [0,6,4,u]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm9, %ymm0
 ; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm5[0,1],ymm4[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm1[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,2,2,2,4,6,6,6]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm3[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4,2,4,2,4,2,4,2]
 ; AVX2-FP-NEXT:    vpermps %ymm6, %ymm2, %ymm1
-; AVX2-FP-NEXT:    vmovaps %ymm2, %ymm11
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm10[0,1],ymm9[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm7[0,1],ymm11[0,1]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1,2,3,4,5],ymm11[6,7]
 ; AVX2-FP-NEXT:    vmovaps 384(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1,2,3],ymm7[4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm8, %ymm0
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1,2,3],ymm10[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm9, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[0,2,2,2,4,6,6,6]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 512(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 544(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm11, %ymm1
-; AVX2-FP-NEXT:    vmovaps %ymm11, %ymm12
+; AVX2-FP-NEXT:    vmovaps 544(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm2, %ymm1
+; AVX2-FP-NEXT:    vmovaps %ymm2, %ymm15
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 288(%rdi), %ymm1
@@ -4971,374 +4898,353 @@ define void @load_i32_stride6_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm1[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm0
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm8, %ymm14
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm9 = ymm10[0,2,2,2,4,6,6,6]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm14[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm0
+; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm2, %ymm9, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm11 = ymm13[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2],ymm11[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 352(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 352(%rdi), %ymm7
-; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm7[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm14, %ymm11, %ymm7
-; AVX2-FP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm9[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 608(%rdi), %ymm0
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm7, %ymm15, %ymm14
+; AVX2-FP-NEXT:    vmovaps %ymm15, %ymm1
+; AVX2-FP-NEXT:    vmovups %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5],ymm14[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 608(%rdi), %ymm10
+; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 576(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 576(%rdi), %ymm7
-; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm0[4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm7, %ymm8, %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm0[0,1,2,3],ymm10[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm14, %ymm9, %ymm9
 ; AVX2-FP-NEXT:    vmovaps 672(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 640(%rdi), %ymm9
-; AVX2-FP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm9 = ymm9[0,1],ymm0[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm11 = ymm9[0,2,2,2,4,6,6,6]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2],ymm11[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 704(%rdi), %ymm11
-; AVX2-FP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 640(%rdi), %ymm10
+; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm12 = ymm10[0,1],ymm0[0,1]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm15 = ymm12[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 704(%rdi), %ymm10
+; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 736(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3],ymm11[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm11, %ymm12, %ymm15
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm8 = [1,7,5,u]
-; AVX2-FP-NEXT:    vpermps %ymm13, %ymm8, %ymm13
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm0[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm15, %ymm1, %ymm10
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm9 = [1,7,5,u]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm9, %ymm8
+; AVX2-FP-NEXT:    vpermilps $237, (%rsp), %ymm10 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm10 = mem[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2],ymm10[3,4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm10 = [5,3,5,3,5,3,5,3]
+; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermps %ymm6, %ymm10, %ymm6
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm6, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm9, %ymm4
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm13[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm12 = [5,3,5,3,5,3,5,3]
-; AVX2-FP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm6, %ymm12, %ymm6
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm8, %ymm4
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm12, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm8, %ymm0
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm14, %ymm12, %ymm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm10, %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermps %ymm2, %ymm9, %ymm1
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm13[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm7, %ymm10, %ymm0
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm7, %ymm8, %ymm0
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm14, %ymm9, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm12[1,3,2,3,5,7,6,7]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm11, %ymm12, %ymm1
+; AVX2-FP-NEXT:    vpermps %ymm15, %ymm10, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps (%rdi), %xmm3
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm3[2,3,2,3]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm7[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1],ymm11[2,3],ymm12[4,5],ymm11[6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1],ymm10[2,3],ymm11[4,5],ymm10[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1],ymm4[2],ymm14[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm8[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm5[4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 384(%rdi), %xmm1
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[2,3,2,3]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm2[1,2,3],ymm0[4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 576(%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovaps %xmm0, (%rsp) # 16-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm2[1,2,3],ymm0[4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm2[0,1],mem[2,3],ymm2[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm5[0,1,2],ymm2[3],ymm5[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm0[2,3],mem[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 192(%rdi), %xmm0
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm2 = xmm0[2,3,2,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm0[2,3],mem[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm13[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm5[1,2,3],ymm2[4],ymm5[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm15[0,1],ymm14[2,3],ymm15[4,5],ymm14[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm10[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm6 = ymm4[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1,2],ymm5[3],ymm6[4,5,6],ymm5[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm2 = xmm3[3,3,3,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm7[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm3[1,2,3],ymm2[4],ymm3[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm11[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0],ymm12[1],ymm3[2,3,4],ymm12[5],ymm3[6,7]
-; AVX2-FP-NEXT:    vmovaps %ymm12, %ymm11
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm9[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm8[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2],ymm3[3],ymm5[4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,3,3,3]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm12[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm6[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm5[1],ymm2[2,3,4],ymm5[5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1],ymm13[2],ymm12[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $239, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4],mem[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm15[0,1],ymm6[2,3],ymm15[4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm9[0,1,0,1,4,5,4,5]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm9[0,1],ymm8[2],ymm9[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm7[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX2-FP-NEXT:    vmovaps %ymm13, %ymm3
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm14[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm15[1],ymm1[2,3,4],ymm15[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vmovaps %ymm15, %ymm14
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm4[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm7[4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermilps $255, (%rsp), %xmm0 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-FP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm8[1],ymm1[2,3,4],ymm8[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm10[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm11[1],ymm0[2,3,4],ymm11[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2],ymm4[3],ymm14[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3,4],ymm5[5],ymm3[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm10[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm3[1],ymm0[2,3,4],ymm3[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2],ymm5[3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm11[0,1,2,3,4],ymm14[5],ymm11[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm6[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm15[1],ymm0[2,3,4],ymm15[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm9[0,1,2],ymm8[3],ymm9[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm7[5],ymm2[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm9[1],ymm0[2,3,4],ymm9[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1,2],ymm13[3],ymm12[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $223, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1,2,3,4],ymm13[5],mem[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm10[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm12[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm5[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 464(%rdi), %xmm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm12, %ymm1
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vpermps %ymm2, %ymm7, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm9[0,1],ymm7[2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm9 = [0,2,0,6,0,2,0,6]
-; AVX2-FP-NEXT:    # ymm9 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermps %ymm13, %ymm9, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm14[0,1],ymm11[2,3],ymm14[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm8 = [0,2,0,6,0,2,0,6]
+; AVX2-FP-NEXT:    # ymm8 = mem[0,1,0,1]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm8, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = ymm11[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm11 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm11 = ymm0[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 80(%rdi), %xmm6
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm7[0,1],ymm6[2,3],ymm7[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm12 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm12 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 80(%rdi), %xmm14
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm4[0,1],ymm14[2,3],ymm4[4,5,6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm11, %ymm12, %ymm1
-; AVX2-FP-NEXT:    vmovaps %ymm12, %ymm15
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm7, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm9, %ymm1
+; AVX2-FP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm2, %ymm8, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm10 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm10 = ymm14[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = ymm3[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm15[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm6 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm6 = mem[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 272(%rdi), %xmm3
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0,1],ymm3[2,3],ymm10[4,5,6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm12, %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm6, %ymm7, %ymm11
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm0[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm9, %ymm12
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1,2,3,4],ymm12[5,6,7]
+; AVX2-FP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm0[0,1],mem[2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm8, %ymm15
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1,2,3,4],ymm15[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm12 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm12 = ymm8[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm11 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm11 = ymm9[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm15 = mem[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 656(%rdi), %xmm0
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm12[0,1],ymm0[2,3],ymm12[4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm15, %ymm15
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm15 = mem[0,1],ymm15[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm15, %ymm9, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm14[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1,2],ymm6[3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm11, %ymm14, %ymm7
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1],ymm6[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm7 = [0,3,1,7,0,3,1,7]
-; AVX2-FP-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm7, %ymm4
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm6 = mem[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm6 = ymm6[0,1,2],mem[3],ymm6[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm11 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm11[0,1],ymm6[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm13, %ymm7, %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm8 = ymm10[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm8[0,1,2],ymm3[3],ymm8[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm14, %ymm5
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm7, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm12[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm11[0,1],ymm0[2,3],ymm11[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm9 = ymm9[0,1,0,2,4,5,4,6]
+; AVX2-FP-NEXT:    vpermps %ymm15, %ymm7, %ymm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1],ymm9[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm9 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm9 = ymm13[0,1],mem[2,3],ymm13[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm9, %ymm8, %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm8[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm14[3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm13, %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm8[0,1],ymm4[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm12 = [0,3,1,7,0,3,1,7]
+; AVX2-FP-NEXT:    # ymm12 = mem[0,1,0,1]
+; AVX2-FP-NEXT:    vpermps %ymm2, %ymm12, %ymm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = ymm2[0,1,2],mem[3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm12, %ymm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm10[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm6, %ymm13, %ymm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm12, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm11[1,1,1,1,5,5,5,5]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm14, %ymm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm15, %ymm7, %ymm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 96(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 32(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 64(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, (%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 96(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 32(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 64(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, (%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 32(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 96(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 64(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, (%rcx)
-; AVX2-FP-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 96(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 32(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 64(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, (%r8)
-; AVX2-FP-NEXT:    vmovaps %ymm9, 96(%r9)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 32(%r9)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, (%r9)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm1, 64(%r9)
+; AVX2-FP-NEXT:    vpermps %ymm15, %ymm13, %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm9, %ymm12, %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 96(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 64(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, (%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 96(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 64(%rdx)
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, (%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 96(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 64(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, (%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 96(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 64(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, (%r8)
+; AVX2-FP-NEXT:    vmovaps %ymm7, 96(%r9)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%r9)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, (%r9)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm3, 64(%r9)
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 96(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm2, 32(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm4, (%rax)
-; AVX2-FP-NEXT:    addq $1256, %rsp # imm = 0x4E8
+; AVX2-FP-NEXT:    vmovaps %ymm1, 32(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm2, 64(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm8, (%rax)
+; AVX2-FP-NEXT:    addq $1288, %rsp # imm = 0x508
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
 ; AVX2-FCP-LABEL: load_i32_stride6_vf32:
 ; AVX2-FCP:       # %bb.0:
-; AVX2-FCP-NEXT:    subq $1256, %rsp # imm = 0x4E8
+; AVX2-FCP-NEXT:    subq $1224, %rsp # imm = 0x4C8
 ; AVX2-FCP-NEXT:    vmovaps 480(%rdi), %ymm9
 ; AVX2-FCP-NEXT:    vmovaps 448(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 416(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups %ymm10, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 416(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm3
@@ -5355,12 +5261,13 @@ define void @load_i32_stride6_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm13 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-FCP-NEXT:    vpermps %ymm13, %ymm8, %ymm0
 ; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm5[0,1],ymm4[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm1[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,2,2,2,4,6,6,6]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm1[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm7[0,2,2,2,4,6,6,6]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm3[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm12 = [4,2,4,2,4,2,4,2]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm12, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm3[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4,2,4,2,4,2,4,2]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm2, %ymm1
+; AVX2-FCP-NEXT:    vmovaps %ymm2, %ymm5
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -5368,7 +5275,7 @@ define void @load_i32_stride6_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1,2,3,4,5],ymm9[6,7]
 ; AVX2-FCP-NEXT:    vmovaps 384(%rdi), %ymm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3],ymm7[4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3],ymm11[4,5],ymm0[6,7]
 ; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm8, %ymm0
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm4[0,2,2,2,4,6,6,6]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
@@ -5377,367 +5284,331 @@ define void @load_i32_stride6_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vmovaps 544(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm12, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm1
+; AVX2-FCP-NEXT:    vmovaps %ymm5, %ymm10
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 256(%rdi), %ymm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm1[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm0
-; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm8, %ymm14
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm9 = ymm10[0,2,2,2,4,6,6,6]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm14[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm0
-; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 352(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm7[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm12, %ymm7
-; AVX2-FCP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm9[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 608(%rdi), %ymm0
-; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 576(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm0[4,5],ymm7[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm8, %ymm8
-; AVX2-FCP-NEXT:    vmovaps 672(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm5[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm5[0,1,2,3],ymm0[4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm8, %ymm12
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm14 = ymm1[0,2,2,2,4,6,6,6]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm12[0,1,2],ymm14[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 352(%rdi), %ymm9
+; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm9[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm10, %ymm15
+; AVX2-FCP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm14[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 608(%rdi), %ymm9
+; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 576(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm5[0,1,2,3],ymm9[4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm8, %ymm8
+; AVX2-FCP-NEXT:    vmovaps 672(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 640(%rdi), %ymm9
 ; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm9 = ymm9[0,1],ymm0[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm11 = ymm9[0,2,2,2,4,6,6,6]
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm15 = ymm9[0,1],ymm5[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm11 = ymm15[0,2,2,2,4,6,6,6]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2],ymm11[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 704(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 736(%rdi), %ymm0
-; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3],ymm11[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm12, %ymm15
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vmovaps 704(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 736(%rdi), %ymm9
+; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm9[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm10, %ymm9
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm9[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm8 = [1,7,5,u]
-; AVX2-FCP-NEXT:    vpermps %ymm13, %ymm8, %ymm13
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,3,2,3,5,7,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm13[0,1,2],ymm6[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm13 = [5,3,5,3,5,3,5,3]
-; AVX2-FCP-NEXT:    vmovups %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm13, %ymm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm13, %ymm8, %ymm9
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,3,2,3,5,7,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm9[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm5 = [5,3,5,3,5,3,5,3]
 ; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm5, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm8, %ymm3
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[1,3,2,3,5,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm13, %ymm2
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm2
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm8, %ymm0
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[1,3,2,3,5,7,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm8, %ymm0
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,3,2,3,5,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm8, %ymm0
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[1,3,2,3,5,7,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm8, %ymm0
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm15[1,3,2,3,5,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps (%rdi), %xmm0
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm1 = xmm0[2,3,2,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm7[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm9[0,1],ymm11[2,3],ymm9[4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm10 = [2,0,6,4,2,0,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm10, %ymm3
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm2[0,0,0,0,4,4,4,4]
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = ymm11[0,1],mem[2,3],ymm11[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm1 = [2,0,6,4,2,0,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm2 = [2,0,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm10[0,1],ymm9[2],ymm10[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm8[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3],ymm4[4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 384(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm3 = xmm1[2,3,2,3]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0],ymm4[1,2,3],ymm3[4],ymm4[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,1],ymm4[2,3],mem[4,5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm10, %ymm4
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2],ymm4[3],ymm5[4,5,6],ymm4[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 576(%rdi), %xmm3
-; AVX2-FCP-NEXT:    vmovaps %xmm3, (%rsp) # 16-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm4 = xmm3[2,3,2,3]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm6[0,1,2,3],ymm8[4],ymm6[5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm4 = [0,0,6,4,0,0,6,4]
+; AVX2-FCP-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm4, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm5 = ymm13[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0],ymm5[1,2,3],ymm4[4],ymm5[5,6,7]
+; AVX2-FCP-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = ymm13[0,1],mem[2],ymm13[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm15 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm1, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = ymm3[0,1],mem[2,3],ymm3[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm10, %ymm5
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1,2],ymm5[3],ymm6[4,5,6],ymm5[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm10, %ymm4
-; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %xmm12
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm5 = xmm12[2,3,2,3]
+; AVX2-FCP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm3[0,1,2,3],mem[4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm4, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm6 = ymm14[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0],ymm6[1,2,3],ymm5[4],ymm6[5,6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm5 = ymm10[0,0,0,0,4,4,4,4]
+; AVX2-FCP-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = ymm14[0,1],mem[2],ymm14[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm2, %ymm0
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm6 = ymm3[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1,2],ymm5[3],ymm6[4,5,6],ymm5[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm7[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm4[1,2,3],ymm0[4],ymm4[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm11[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0],ymm9[1],ymm4[2,3,4],ymm9[5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vmovaps %ymm9, %ymm11
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm2[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm5 = ymm8[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2],ymm4[3],ymm5[4,5,6],ymm4[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm3[0,1],mem[2,3],ymm3[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm1, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm3[0,1,2,3],mem[4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm4, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[3,3,3,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm5[1],ymm1[2,3,4],ymm5[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[0,1,0,1,4,5,4,5]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm7[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2],ymm1[3],ymm4[4,5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm0 = xmm12[3,3,3,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm14[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm14[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm15[1],ymm1[2,3,4],ymm15[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
+; AVX2-FCP-NEXT:    vblendps $251, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm7[2],mem[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm3[0,1],mem[2,3],ymm3[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm3[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1,2,3],ymm12[4],ymm5[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $255, (%rsp), %xmm0 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[3,3,3,3,7,7,7,7]
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm11[1],ymm0[2,3,4],ymm11[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm10[0,1,2],ymm9[3],ymm10[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4],ymm0[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm6[0,1,2,3,4],ymm8[5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,1,7,5,0,1,7,5]
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm6 = [3,1,7,5,0,u,u,u]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm1 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm15[1],ymm1[2,3,4],ymm15[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm13[0,1,2],ymm4[3],ymm13[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps %ymm13, %ymm9
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3,4],ymm1[5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm10[0,1,2,3,4],ymm8[5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm1 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2,3,4],ymm3[5],ymm1[6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm13[1],ymm1[2,3,4],ymm13[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm13[0,1,2],ymm7[3],ymm13[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3,4],ymm1[5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm5[0,1,2,3,4],ymm12[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = ymm8[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm1 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm12[1],ymm1[2,3,4],ymm12[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm14[0,1,2],mem[3],ymm14[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3,4],ymm1[5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $32, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm2[0,1,2,3,4],mem[5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm15[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0,1,2,3],ymm4[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 464(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm12, %ymm1
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm9[0,1],ymm7[2,3],ymm9[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1],ymm10[2,3],ymm8[4,5,6,7]
 ; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm9 = [0,2,0,6,0,2,0,6]
 ; AVX2-FCP-NEXT:    # ymm9 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm9, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm9, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm7 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm7 = ymm11[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm10 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm10 = ymm11[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = ymm0[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 80(%rdi), %xmm6
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm7[0,1],ymm6[2,3],ymm7[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps 80(%rdi), %xmm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0,1],ymm11[2,3],ymm10[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm12, %ymm1
-; AVX2-FCP-NEXT:    vmovaps %ymm12, %ymm15
+; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm5, %ymm1
+; AVX2-FCP-NEXT:    vmovaps %ymm5, %ymm8
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm9, %ymm1
+; AVX2-FCP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm9, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm10 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm10 = mem[0,1,2,3],ymm14[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = ymm3[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = ymm13[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 272(%rdi), %xmm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0,1],ymm3[2,3],ymm10[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm5[0,1],ymm3[2,3],ymm5[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm12, %ymm14
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm8, %ymm15
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1],ymm0[2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm9, %ymm12
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1,2,3,4],ymm12[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm12 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm12 = ymm13[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm0[0,1],mem[2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm9, %ymm14
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm13 = ymm15[0,1,2,3,4],ymm14[5,6,7]
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm14 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm14 = ymm12[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm15 = mem[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 656(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm12[0,1],ymm0[2,3],ymm12[4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm15, %ymm15
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm15 = ymm13[0,1],mem[2,3],ymm13[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm9, %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm14[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1,2],ymm6[3],ymm7[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm14, %ymm7
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1],ymm6[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm7 = [0,3,1,7,0,3,1,7]
-; AVX2-FCP-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm7, %ymm4
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = mem[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = ymm6[0,1,2],mem[3],ymm6[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm11[0,1],ymm6[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm7, %ymm8
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm8[5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm14[0,1],ymm0[2,3],ymm14[4,5,6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm12 = ymm12[0,1,0,2,4,5,4,6]
+; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm8, %ymm8
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1],ymm12[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm12 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm12 = mem[0,1],ymm12[2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm9, %ymm9
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm8[0,1,2,3,4],ymm9[5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm8 = ymm10[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm8[0,1,2],ymm3[3],ymm8[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm14, %ymm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm7, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm12[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2],ymm11[3],ymm8[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm11, %ymm7
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1],ymm8[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm8 = [0,3,1,7,0,3,1,7]
+; AVX2-FCP-NEXT:    # ymm8 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm8, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm7[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vpermilps $85, (%rsp), %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = ymm7[0,1,2],mem[3],ymm7[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm10 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm10[0,1],ymm7[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm8, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2],ymm3[3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm11, %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm8, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm14[1,1,1,1,5,5,5,5]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm14, %ymm1
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm7, %ymm1
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 96(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 32(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 64(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, (%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 96(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 32(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 64(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, (%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 32(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 96(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 64(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, (%rcx)
-; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 96(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 32(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 64(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, (%r8)
+; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm11, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm8, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 96(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 64(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 96(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 64(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 96(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 64(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 96(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 64(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, (%r8)
 ; AVX2-FCP-NEXT:    vmovaps %ymm9, 96(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 32(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, (%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 64(%r9)
+; AVX2-FCP-NEXT:    vmovaps %ymm13, 32(%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, (%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 64(%r9)
 ; AVX2-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FCP-NEXT:    vmovaps %ymm0, 96(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 32(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm1, 32(%rax)
 ; AVX2-FCP-NEXT:    vmovaps %ymm6, 64(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rax)
-; AVX2-FCP-NEXT:    addq $1256, %rsp # imm = 0x4E8
+; AVX2-FCP-NEXT:    vmovaps %ymm2, (%rax)
+; AVX2-FCP-NEXT:    addq $1224, %rsp # imm = 0x4C8
 ; AVX2-FCP-NEXT:    vzeroupper
 ; AVX2-FCP-NEXT:    retq
 ;
@@ -9159,50 +9030,47 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-LABEL: load_i32_stride6_vf64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    subq $2536, %rsp # imm = 0x9E8
-; AVX2-NEXT:    vmovaps 672(%rdi), %ymm4
-; AVX2-NEXT:    vmovaps 640(%rdi), %ymm5
-; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 608(%rdi), %ymm3
-; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 320(%rdi), %ymm2
+; AVX2-NEXT:    subq $2504, %rsp # imm = 0x9C8
+; AVX2-NEXT:    vmovaps 672(%rdi), %ymm3
+; AVX2-NEXT:    vmovaps 640(%rdi), %ymm4
+; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 608(%rdi), %ymm2
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 320(%rdi), %ymm5
+; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 352(%rdi), %ymm6
 ; AVX2-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 288(%rdi), %ymm7
 ; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 256(%rdi), %ymm9
-; AVX2-NEXT:    vmovups %ymm9, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 256(%rdi), %ymm8
+; AVX2-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 224(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 192(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps {{.*#+}} xmm8 = [0,6,4,u]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
-; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm0, %ymm8, %ymm0
-; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm9[0,1],ymm7[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vmovaps {{.*#+}} xmm9 = [0,6,4,u]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpermps %ymm14, %ymm9, %ymm0
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm8[0,1],ymm7[0,1]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm1[0,1,2,3,4,5],ymm7[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm15[0,2,2,2,4,6,6,6]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1,2,3],ymm5[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4,2,4,2,4,2,4,2]
-; AVX2-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-NEXT:    vmovaps %ymm2, %ymm6
+; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm5 = [4,2,4,2,4,2,4,2]
+; AVX2-NEXT:    vpermps %ymm1, %ymm5, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm5[0,1],ymm4[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm0[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm4[0,1],ymm3[0,1]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 576(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm3[4,5],ymm0[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm0, %ymm8, %ymm0
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vpermps %ymm0, %ymm9, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2,2,2,4,6,6,6]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovaps 704(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -9210,7 +9078,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm1, %ymm6, %ymm1
+; AVX2-NEXT:    vpermps %ymm1, %ymm5, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 1056(%rdi), %ymm1
@@ -9226,7 +9094,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm0, %ymm8, %ymm0
+; AVX2-NEXT:    vpermps %ymm0, %ymm9, %ymm0
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2,2,2,4,6,6,6]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovaps 1088(%rdi), %ymm1
@@ -9235,7 +9103,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm1, %ymm6, %ymm1
+; AVX2-NEXT:    vpermps %ymm1, %ymm5, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 1440(%rdi), %ymm1
@@ -9251,7 +9119,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm0, %ymm8, %ymm0
+; AVX2-NEXT:    vpermps %ymm0, %ymm9, %ymm0
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2,2,2,4,6,6,6]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovaps 1472(%rdi), %ymm1
@@ -9260,7 +9128,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm1, %ymm6, %ymm1
+; AVX2-NEXT:    vpermps %ymm1, %ymm5, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 96(%rdi), %ymm1
@@ -9268,21 +9136,22 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovaps 64(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm1[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps (%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 32(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-NEXT:    vpermps %ymm11, %ymm8, %ymm0
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-NEXT:    vpermps %ymm13, %ymm9, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2,2,2,4,6,6,6]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovaps 128(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 160(%rdi), %ymm2
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm10, %ymm6, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm12, %ymm5, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 480(%rdi), %ymm1
@@ -9290,22 +9159,23 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovaps 448(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm1[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovaps 416(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 384(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpermps %ymm5, %ymm8, %ymm0
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm7[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpermps %ymm10, %ymm9, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm11[0,2,2,2,4,6,6,6]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovaps 512(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 544(%rdi), %ymm2
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm4, %ymm6, %ymm1
-; AVX2-NEXT:    vmovaps %ymm6, %ymm12
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm8, %ymm5, %ymm1
+; AVX2-NEXT:    vmovaps %ymm5, %ymm7
+; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 864(%rdi), %ymm1
@@ -9313,622 +9183,580 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovaps 832(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm1[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovaps 800(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 768(%rdi), %ymm1
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm8, %ymm0
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpermps %ymm5, %ymm9, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,2,2,2,4,6,6,6]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovaps 896(%rdi), %ymm1
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 928(%rdi), %ymm6
-; AVX2-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm1, %ymm12, %ymm6
-; AVX2-NEXT:    vmovaps %ymm12, %ymm14
-; AVX2-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-NEXT:    vmovaps 928(%rdi), %ymm2
+; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm4, %ymm7, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps 1184(%rdi), %ymm0
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 1152(%rdi), %ymm6
-; AVX2-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm0[4,5],ymm6[6,7]
-; AVX2-NEXT:    vpermps %ymm6, %ymm8, %ymm0
-; AVX2-NEXT:    vmovaps 1248(%rdi), %ymm9
-; AVX2-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 1216(%rdi), %ymm8
-; AVX2-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm8 = ymm8[0,1],ymm9[0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm9 = ymm8[0,2,2,2,4,6,6,6]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-NEXT:    vmovaps 1280(%rdi), %ymm9
-; AVX2-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 1312(%rdi), %ymm12
-; AVX2-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm12[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm15, %ymm14, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-NEXT:    vmovaps 1152(%rdi), %ymm1
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpermps %ymm3, %ymm9, %ymm0
+; AVX2-NEXT:    vmovaps 1248(%rdi), %ymm2
+; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 1216(%rdi), %ymm1
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[0,2,2,2,4,6,6,6]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
+; AVX2-NEXT:    vmovaps 1280(%rdi), %ymm2
+; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 1312(%rdi), %ymm1
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm7, %ymm2
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovaps {{.*#+}} xmm0 = [1,7,5,u]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = mem[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm9[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm9 = [5,3,5,3,5,3,5,3]
-; AVX2-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm12 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm14[0,1,2,3,4,5],ymm12[6,7]
-; AVX2-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm12 # 32-byte Folded Reload
-; AVX2-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = mem[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm12 # 32-byte Folded Reload
-; AVX2-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = mem[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm12 # 32-byte Folded Reload
+; AVX2-NEXT:    vpermps %ymm14, %ymm0, %ymm2
+; AVX2-NEXT:    vshufps {{.*#+}} ymm15 = ymm15[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm2[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [5,3,5,3,5,3,5,3]
+; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm14 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1,2,3,4,5],ymm14[6,7]
+; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
+; AVX2-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm15 = mem[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
+; AVX2-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm15 = mem[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
+; AVX2-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm15 = mem[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermps %ymm13, %ymm0, %ymm13
 ; AVX2-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm14 = mem[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5],ymm14[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm13[0,1,2],ymm14[3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm12, %ymm2, %ymm12
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0,1,2,3,4,5],ymm12[6,7]
 ; AVX2-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm11, %ymm0, %ymm11
-; AVX2-NEXT:    vshufps {{.*#+}} ymm12 = ymm13[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm10, %ymm9, %ymm10
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermps %ymm10, %ymm0, %ymm10
+; AVX2-NEXT:    vshufps {{.*#+}} ymm11 = ymm11[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2],ymm11[3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm8, %ymm2, %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm10[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermps %ymm5, %ymm0, %ymm5
-; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2],ymm7[3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm4, %ymm9, %ymm4
+; AVX2-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2],ymm6[3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm4, %ymm2, %ymm4
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm1, %ymm9, %ymm1
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm6, %ymm0, %ymm0
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[1,3,2,3,5,7,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm15, %ymm9, %ymm1
+; AVX2-NEXT:    vpermps %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm9[1,3,2,3,5,7,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 192(%rdi), %xmm5
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm5[2,3,2,3]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5],mem[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 576(%rdi), %xmm0
-; AVX2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1],ymm0[2,3],mem[4,5],ymm0[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5],mem[6,7]
+; AVX2-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 960(%rdi), %xmm2
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm2[2,3,2,3]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1],ymm0[2,3],mem[4,5],ymm0[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5],mem[6,7]
+; AVX2-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2],ymm1[3],ymm4[4,5,6],ymm1[7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 1344(%rdi), %xmm1
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[2,3,2,3]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm15[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm4[1,2,3],ymm0[4],ymm4[5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,1],ymm13[2,3],mem[4,5],ymm13[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,0,0,0,4,4,4,4]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm13[2,3],ymm0[4,5],ymm13[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm15[0,1],ymm12[2],ymm15[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm6 = ymm14[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2],ymm4[3],ymm6[4,5,6],ymm4[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2,3],ymm11[4],ymm14[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps (%rdi), %xmm0
-; AVX2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm4[1,2,3],ymm0[4],ymm4[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,1],ymm4[2,3],mem[4,5],ymm4[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm6 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2],ymm4[3],ymm6[4,5,6],ymm4[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1],ymm6[2,3],mem[4,5],ymm6[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm9[0,1],mem[2],ymm9[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm10[0,1,2,3],mem[4],ymm10[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 384(%rdi), %xmm0
-; AVX2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm6 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm6[1,2,3],ymm0[4],ymm6[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm6 = ymm4[0,1],mem[2,3],ymm4[4,5],mem[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm6[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm6 = mem[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1,2],ymm6[3],ymm8[4,5,6],ymm6[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $251, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1],ymm1[2],mem[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 768(%rdi), %xmm0
-; AVX2-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm6 = xmm0[2,3,2,3]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0],ymm8[1,2,3],ymm6[4],ymm8[5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm8 = ymm8[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2],ymm8[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = mem[0,0,0,0,4,4,4,4]
-; AVX2-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm10 = mem[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm10[0,1,2],ymm8[3],ymm10[4,5,6],ymm8[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1,2,3,4],ymm8[5,6,7]
+; AVX2-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 1152(%rdi), %xmm12
-; AVX2-NEXT:    vshufps {{.*#+}} xmm8 = xmm12[2,3,2,3]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm10 = ymm6[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0],ymm10[1,2,3],ymm8[4],ymm10[5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm9[0,1],ymm0[2,3],ymm9[4,5],ymm0[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[2,0,2,3,6,4,6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2],ymm10[3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm10 = ymm4[0,0,0,0,4,4,4,4]
+; AVX2-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $4, (%rsp), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0,1],ymm2[2,3],ymm8[4,5],ymm2[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm11 = ymm7[0,0,2,3,4,4,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2],ymm10[3],ymm11[4,5,6],ymm10[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm10[5,6,7]
-; AVX2-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm5 = xmm5[3,3,3,3]
-; AVX2-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = mem[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0],ymm8[1,2,3],ymm5[4],ymm8[5,6,7]
-; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = mem[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vblendps $34, (%rsp), %ymm8, %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = ymm8[0],mem[1],ymm8[2,3,4],mem[5],ymm8[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2],ymm8[3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm8 = ymm3[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm10 = mem[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm10[0,1,2],ymm8[3],ymm10[4,5,6],ymm8[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm3 = mem[3,3,3,3]
-; AVX2-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = mem[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0],ymm5[1,2,3],ymm3[4],ymm5[5,6,7]
-; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = mem[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = mem[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = mem[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm8[0,1,2],ymm5[3],ymm8[4,5,6],ymm5[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[3,3,3,3]
-; AVX2-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = mem[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm3[1,2,3],ymm2[4],ymm3[5,6,7]
-; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = mem[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = mem[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = mem[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2],ymm3[3],ymm5[4,5,6],ymm3[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,3,3,3]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm15[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm13[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm2[0],mem[1],ymm2[2,3,4],mem[5],ymm2[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1],ymm3[2],ymm7[3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm14[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm12[3,3,3,3]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm6[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm0[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm9[1],ymm2[2,3,4],ymm9[5],ymm2[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1,2,3],ymm4[4],ymm5[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2],mem[3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm4[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm7[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $32, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2,3,4],mem[5],ymm1[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm12[1],ymm1[2,3,4],ymm12[5],ymm1[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm15[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2],mem[3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $32, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2,3,4],mem[5],ymm1[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm11[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm7[1],ymm1[2,3,4],ymm7[5],ymm1[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm9[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm13[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm15[0,1,2],ymm12[3],ymm15[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2,3,4],ymm11[5],ymm14[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm14[1],ymm1[2,3,4],ymm14[5],ymm1[6,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[0,1,0,1,4,5,4,5]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm4[0,1,3,3,4,5,7,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm6[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm9[0,1,2],mem[3],ymm9[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vblendps $32, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm10[0,1,2,3,4],mem[5],ymm10[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm6[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 80(%rdi), %xmm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm2[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm8[1],ymm0[2,3,4],ymm8[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1,2],ymm3[3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1,2,3,4],ymm4[5],ymm5[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[3,3,3,3,7,7,7,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-NEXT:    vpermps %ymm2, %ymm14, %ymm1
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm5[0,1],ymm4[2,3],ymm5[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,2,0,6,0,2,0,6]
-; AVX2-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 272(%rdi), %xmm1
-; AVX2-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm3, %ymm14, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm14[1],ymm0[2,3,4],ymm14[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-NEXT:    vblendps $8, (%rsp), %ymm15, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm15[0,1,2],mem[3],ymm15[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $223, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1,2,3,4],ymm1[5],mem[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm10[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm8[1],ymm0[2,3,4],ymm8[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm9[0,1,2],ymm11[3],ymm9[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1,2,3,4],ymm13[5],ymm12[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm3[3,3,3,3,7,7,7,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2,3,4],ymm2[5],ymm0[6,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm0[0,2,0,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm1[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1,2,3,4],ymm6[5],ymm5[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm7[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm11[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 464(%rdi), %xmm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm3, %ymm14, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm10[0,1],ymm9[2,3],ymm10[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-NEXT:    vmovaps 80(%rdi), %xmm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm7[0,1],ymm0[2,3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1],ymm5[2,3],ymm6[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm3 = [0,2,0,6,0,2,0,6]
+; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX2-NEXT:    vpermps %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 272(%rdi), %xmm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vpermps %ymm4, %ymm2, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 656(%rdi), %xmm1
+; AVX2-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm3, %ymm14, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm8[0,1,2,3],ymm10[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm12[0,1,2,3],ymm13[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 848(%rdi), %xmm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm9[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 464(%rdi), %xmm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vpermps %ymm4, %ymm2, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm13[0,1],ymm12[2,3],ymm13[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm3, %ymm14, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1],ymm15[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 656(%rdi), %xmm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vpermps %ymm4, %ymm2, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm13 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm13 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm12 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm12 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovaps 1040(%rdi), %xmm11
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm13[0,1],ymm11[2,3],ymm13[4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm12, %ymm14, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm10 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm10 = ymm2[0,1],mem[2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm10, %ymm0, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovaps 1232(%rdi), %xmm6
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm8[0,1],ymm6[2,3],ymm8[4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm7, %ymm14, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = mem[0,1],ymm2[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm5, %ymm0, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm14[0,1,2,3],mem[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps $240, (%rsp), %ymm15, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = ymm15[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 848(%rdi), %xmm0
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vpermps %ymm4, %ymm2, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = ymm1[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermps %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm15 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm14 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vmovaps 1040(%rdi), %xmm13
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm15[0,1],ymm13[2,3],ymm15[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vpermps %ymm14, %ymm2, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vmovaps 1424(%rdi), %xmm2
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1],ymm2[2,3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-NEXT:    vpermps %ymm3, %ymm14, %ymm15
-; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm1[2,3,4,5,6,7]
+; AVX2-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm12 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm12 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm12, %ymm3, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm11 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm11 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm10 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm10 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovaps 1232(%rdi), %xmm9
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1],ymm9[2,3],ymm11[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vpermps %ymm10, %ymm2, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm15[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm8 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm8 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm8, %ymm3, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm0 = mem[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm6 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm6 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vmovaps 1424(%rdi), %xmm4
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1],ymm4[2,3],ymm6[4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-NEXT:    vpermps %ymm5, %ymm2, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm3, %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm7[1,1,1,1,5,5,5,5]
 ; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm0 = ymm0[0,1,2],mem[3],ymm0[4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,3,1,7,0,3,1,7]
 ; AVX2-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = mem[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps $8, (%rsp), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = ymm14[0,1,2],mem[3],ymm14[4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = mem[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = ymm14[0,1,2],mem[3],ymm14[4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = mem[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = ymm14[0,1,2],mem[3],ymm14[4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = mem[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = ymm14[0,1,2],mem[3],ymm14[4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm13[0,1,2],ymm11[3],ymm13[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm12, %ymm9, %ymm12
-; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm12[0,1],ymm11[2,3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm10, %ymm0, %ymm10
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2,3,4],ymm10[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm8 = ymm8[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1,2],ymm6[3],ymm8[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm7, %ymm9, %ymm7
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1],ymm6[2,3,4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm5, %ymm0, %ymm5
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[1,1,1,1,5,5,5,5]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm3, %ymm9, %ymm3
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = mem[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm7[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = mem[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm7[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = mem[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm7[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = mem[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm7[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm15[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2],ymm13[3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm14, %ymm2, %ymm13
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm13[0,1],ymm7[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm12, %ymm0, %ymm12
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm7[0,1,2,3,4],ymm12[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm11[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2],ymm9[3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm10, %ymm2, %ymm9
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm9[0,1],ymm7[2,3,4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm8, %ymm0, %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm8[5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2],ymm4[3],ymm6[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm5, %ymm2, %ymm2
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1],ymm4[2,3,4,5,6,7]
 ; AVX2-NEXT:    vpermps %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm0[5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
@@ -9995,7 +9823,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovaps %ymm1, 96(%r8)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm1, 32(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm1, 224(%r9)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm1, 192(%r9)
@@ -10013,9 +9841,9 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovaps %ymm1, (%r9)
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-NEXT:    vmovaps %ymm0, 224(%rax)
-; AVX2-NEXT:    vmovaps %ymm5, 192(%rax)
-; AVX2-NEXT:    vmovaps %ymm10, 160(%rax)
-; AVX2-NEXT:    vmovaps %ymm14, 128(%rax)
+; AVX2-NEXT:    vmovaps %ymm7, 192(%rax)
+; AVX2-NEXT:    vmovaps %ymm12, 160(%rax)
+; AVX2-NEXT:    vmovaps %ymm3, 128(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 96(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
@@ -10024,56 +9852,53 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovaps %ymm0, 32(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, (%rax)
-; AVX2-NEXT:    addq $2536, %rsp # imm = 0x9E8
+; AVX2-NEXT:    addq $2504, %rsp # imm = 0x9C8
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: load_i32_stride6_vf64:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    subq $2536, %rsp # imm = 0x9E8
-; AVX2-FP-NEXT:    vmovaps 672(%rdi), %ymm4
-; AVX2-FP-NEXT:    vmovaps 640(%rdi), %ymm5
-; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 608(%rdi), %ymm3
-; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm2
+; AVX2-FP-NEXT:    subq $2504, %rsp # imm = 0x9C8
+; AVX2-FP-NEXT:    vmovaps 672(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovaps 640(%rdi), %ymm4
+; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 608(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 320(%rdi), %ymm5
+; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 352(%rdi), %ymm6
 ; AVX2-FP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 288(%rdi), %ymm7
 ; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm9
-; AVX2-FP-NEXT:    vmovups %ymm9, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 256(%rdi), %ymm8
+; AVX2-FP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 224(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 192(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm8 = [0,6,4,u]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm0, %ymm8, %ymm0
-; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm9[0,1],ymm7[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm9 = [0,6,4,u]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm14, %ymm9, %ymm0
+; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm8[0,1],ymm7[0,1]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm1[0,1,2,3,4,5],ymm7[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm15[0,2,2,2,4,6,6,6]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1,2,3],ymm5[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4,2,4,2,4,2,4,2]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FP-NEXT:    vmovaps %ymm2, %ymm6
+; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm5 = [4,2,4,2,4,2,4,2]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm5, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm5[0,1],ymm4[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm0[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm4[0,1],ymm3[0,1]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 576(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm3[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm0, %ymm8, %ymm0
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm9, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2,2,2,4,6,6,6]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 704(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -10081,7 +9906,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm5, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 1056(%rdi), %ymm1
@@ -10097,7 +9922,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm0, %ymm8, %ymm0
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm9, %ymm0
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2,2,2,4,6,6,6]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 1088(%rdi), %ymm1
@@ -10106,7 +9931,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm5, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 1440(%rdi), %ymm1
@@ -10122,7 +9947,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm0, %ymm8, %ymm0
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm9, %ymm0
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2,2,2,4,6,6,6]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 1472(%rdi), %ymm1
@@ -10131,7 +9956,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm6, %ymm1
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm5, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 96(%rdi), %ymm1
@@ -10139,21 +9964,22 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovaps 64(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm1[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps (%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 32(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm11, %ymm8, %ymm0
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm13, %ymm9, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2,2,2,4,6,6,6]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 128(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 160(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm10, %ymm6, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm5, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 480(%rdi), %ymm1
@@ -10161,22 +9987,23 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovaps 448(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm1[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovaps 416(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 384(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm8, %ymm0
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm7[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm10, %ymm9, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm11[0,2,2,2,4,6,6,6]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 512(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 544(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm6, %ymm1
-; AVX2-FP-NEXT:    vmovaps %ymm6, %ymm12
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm5, %ymm1
+; AVX2-FP-NEXT:    vmovaps %ymm5, %ymm7
+; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 864(%rdi), %ymm1
@@ -10184,622 +10011,580 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovaps 832(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm1[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovaps 800(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 768(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm8, %ymm0
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm9, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,2,2,2,4,6,6,6]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovaps 896(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 928(%rdi), %ymm6
-; AVX2-FP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm12, %ymm6
-; AVX2-FP-NEXT:    vmovaps %ymm12, %ymm14
-; AVX2-FP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vmovaps 928(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm7, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 1184(%rdi), %ymm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 1152(%rdi), %ymm6
-; AVX2-FP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm0[4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpermps %ymm6, %ymm8, %ymm0
-; AVX2-FP-NEXT:    vmovaps 1248(%rdi), %ymm9
-; AVX2-FP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 1216(%rdi), %ymm8
-; AVX2-FP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm8 = ymm8[0,1],ymm9[0,1]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm9 = ymm8[0,2,2,2,4,6,6,6]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 1280(%rdi), %ymm9
-; AVX2-FP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 1312(%rdi), %ymm12
-; AVX2-FP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm12[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm15, %ymm14, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-FP-NEXT:    vmovaps 1152(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm9, %ymm0
+; AVX2-FP-NEXT:    vmovaps 1248(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 1216(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[0,2,2,2,4,6,6,6]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 1280(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 1312(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm7, %ymm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps {{.*#+}} xmm0 = [1,7,5,u]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = mem[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm9[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm9 = [5,3,5,3,5,3,5,3]
-; AVX2-FP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm12 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm14[0,1,2,3,4,5],ymm12[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm12 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = mem[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm12 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = mem[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm12 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpermps %ymm14, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm15 = ymm15[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm2[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [5,3,5,3,5,3,5,3]
+; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm14 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1,2,3,4,5],ymm14[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm15 = mem[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm15 = mem[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm15 = mem[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermps %ymm13, %ymm0, %ymm13
 ; AVX2-FP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm14 = mem[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5],ymm14[6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm13[0,1,2],ymm14[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm2, %ymm12
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0,1,2,3,4,5],ymm12[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm11, %ymm0, %ymm11
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm12 = ymm13[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm10, %ymm9, %ymm10
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermps %ymm10, %ymm0, %ymm10
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm11 = ymm11[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2],ymm11[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm2, %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm10[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermps %ymm5, %ymm0, %ymm5
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2],ymm7[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm9, %ymm4
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2],ymm6[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm2, %ymm4
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm9, %ymm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm6, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[1,3,2,3,5,7,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm15, %ymm9, %ymm1
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm9[1,3,2,3,5,7,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 192(%rdi), %xmm5
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm5[2,3,2,3]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 576(%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm0[2,3],mem[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 960(%rdi), %xmm2
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm2[2,3,2,3]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm0[2,3],mem[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2],ymm1[3],ymm4[4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 1344(%rdi), %xmm1
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[2,3,2,3]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm15[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm4[1,2,3],ymm0[4],ymm4[5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,1],ymm13[2,3],mem[4,5],ymm13[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,0,0,0,4,4,4,4]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm13[2,3],ymm0[4,5],ymm13[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm15[0,1],ymm12[2],ymm15[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm6 = ymm14[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2],ymm4[3],ymm6[4,5,6],ymm4[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2,3],ymm11[4],ymm14[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps (%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm4[1,2,3],ymm0[4],ymm4[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,1],ymm4[2,3],mem[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm6 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2],ymm4[3],ymm6[4,5,6],ymm4[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1],ymm6[2,3],mem[4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm9[0,1],mem[2],ymm9[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm10[0,1,2,3],mem[4],ymm10[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 384(%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm6 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm6[1,2,3],ymm0[4],ymm6[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm6 = ymm4[0,1],mem[2,3],ymm4[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm6[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm6 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1,2],ymm6[3],ymm8[4,5,6],ymm6[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $251, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1],ymm1[2],mem[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 768(%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm6 = xmm0[2,3,2,3]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0],ymm8[1,2,3],ymm6[4],ymm8[5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm8 = ymm8[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2],ymm8[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm10 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm10[0,1,2],ymm8[3],ymm10[4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1,2,3,4],ymm8[5,6,7]
+; AVX2-FP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 1152(%rdi), %xmm12
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm8 = xmm12[2,3,2,3]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm10 = ymm6[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0],ymm10[1,2,3],ymm8[4],ymm10[5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm9[0,1],ymm0[2,3],ymm9[4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[2,0,2,3,6,4,6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2],ymm10[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm10 = ymm4[0,0,0,0,4,4,4,4]
+; AVX2-FP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $4, (%rsp), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2],ymm1[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm8[0,1],ymm2[2,3],ymm8[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm11 = ymm7[0,0,2,3,4,4,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2],ymm10[3],ymm11[4,5,6],ymm10[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm10[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm5 = xmm5[3,3,3,3]
-; AVX2-FP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0],ymm8[1,2,3],ymm5[4],ymm8[5,6,7]
-; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vblendps $34, (%rsp), %ymm8, %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = ymm8[0],mem[1],ymm8[2,3,4],mem[5],ymm8[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2],ymm8[3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm8 = ymm3[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm10 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm10[0,1,2],ymm8[3],ymm10[4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm8 = ymm8[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm3 = mem[3,3,3,3]
-; AVX2-FP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0],ymm5[1,2,3],ymm3[4],ymm5[5,6,7]
-; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2],ymm5[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = mem[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm8[0,1,2],ymm5[3],ymm8[4,5,6],ymm5[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[3,3,3,3]
-; AVX2-FP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm3[1,2,3],ymm2[4],ymm3[5,6,7]
-; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = mem[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2],ymm3[3],ymm5[4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,3,3,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm15[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm13[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm2[0],mem[1],ymm2[2,3,4],mem[5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1],ymm3[2],ymm7[3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm14[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm12[3,3,3,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm6[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3],ymm1[4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm0[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm9[1],ymm2[2,3,4],ymm9[5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1,2,3],ymm4[4],ymm5[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,0,2,0,4,4,6,4]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2],mem[3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm4[0,1,0,1,4,5,4,5]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm7[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $32, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2,3,4],mem[5],ymm1[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-FP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm12[1],ymm1[2,3,4],ymm12[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,1,0,1,4,5,4,5]
+; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2],mem[3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $32, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2,3,4],mem[5],ymm1[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm13[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm15[0,1,2],ymm12[3],ymm15[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2,3,4],ymm11[5],ymm14[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm6[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm9[0,1,2],mem[3],ymm9[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps $32, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm10[0,1,2,3,4],mem[5],ymm10[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm2[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm8[1],ymm0[2,3,4],ymm8[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1,2],ymm3[3],ymm7[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1,2,3,4],ymm4[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm14[1],ymm0[2,3,4],ymm14[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm15[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vblendps $8, (%rsp), %ymm15, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm15[0,1,2],mem[3],ymm15[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $223, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1,2,3,4],ymm1[5],mem[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm11[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm7[1],ymm1[2,3,4],ymm7[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,1,0,1,4,5,4,5]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm10[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm8[1],ymm0[2,3,4],ymm8[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm9[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm9[0,1,2],ymm11[3],ymm9[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1,2,3,4],ymm13[5],ymm12[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm14[1],ymm1[2,3,4],ymm14[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[0,1,0,1,4,5,4,5]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm3[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2,3,4],ymm2[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm0[0,2,0,3]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm4[0,1,3,3,4,5,7,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[3,1,3,3,7,5,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm1[0,1,2],ymm5[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm5[0,1,2,3,4],ymm6[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,3,1,4,5,7,5]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm6[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovaps 80(%rdi), %xmm0
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm7[0,1],ymm0[2,3],ymm7[4,5,6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm14, %ymm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm5[0,1],ymm4[2,3],ymm5[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,2,0,6,0,2,0,6]
-; AVX2-FP-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 272(%rdi), %xmm1
-; AVX2-FP-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm14, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm7[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm11[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 464(%rdi), %xmm1
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1],ymm5[2,3],ymm6[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm14, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm10[0,1],ymm9[2,3],ymm10[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm3 = [0,2,0,6,0,2,0,6]
+; AVX2-FP-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 272(%rdi), %xmm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm2, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 656(%rdi), %xmm1
+; AVX2-FP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm14, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm2[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm8[0,1,2,3],ymm10[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm12[0,1,2,3],ymm13[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 848(%rdi), %xmm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm9[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 464(%rdi), %xmm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm2, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm13[0,1],ymm12[2,3],ymm13[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm14, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1],ymm15[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 656(%rdi), %xmm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm2, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm13 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm13 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm12 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm12 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 1040(%rdi), %xmm11
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm13[0,1],ymm11[2,3],ymm13[4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm12, %ymm14, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm10 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm10 = ymm2[0,1],mem[2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm10, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 1232(%rdi), %xmm6
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm8[0,1],ymm6[2,3],ymm8[4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm7, %ymm14, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = mem[0,1],ymm2[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm14[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps $240, (%rsp), %ymm15, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = ymm15[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 848(%rdi), %xmm0
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm2, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 1424(%rdi), %xmm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1],ymm2[2,3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,1,0,2,4,5,4,6]
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm14, %ymm15
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm1[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm15 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm14 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 1040(%rdi), %xmm13
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm15[0,1],ymm13[2,3],ymm15[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-FP-NEXT:    vpermps %ymm14, %ymm2, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm1, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm15[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm12 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm12 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm11 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm11 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm10 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm10 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 1232(%rdi), %xmm9
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm11[0,1],ymm9[2,3],ymm11[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-FP-NEXT:    vpermps %ymm10, %ymm2, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm8 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm8 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm3, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm0 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm6 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm6 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 1424(%rdi), %xmm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1],ymm4[2,3],ymm6[4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm2, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm3, %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm7[1,1,1,1,5,5,5,5]
 ; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm0 = ymm0[0,1,2],mem[3],ymm0[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm15 = ymm15[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,3,1,7,0,3,1,7]
 ; AVX2-FP-NEXT:    # ymm0 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = mem[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps $8, (%rsp), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = ymm14[0,1,2],mem[3],ymm14[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = mem[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = ymm14[0,1,2],mem[3],ymm14[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = mem[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = ymm14[0,1,2],mem[3],ymm14[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = mem[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = ymm14[0,1,2],mem[3],ymm14[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm13[0,1,2],ymm11[3],ymm13[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm12, %ymm9, %ymm12
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm12[0,1],ymm11[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm10, %ymm0, %ymm10
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2,3,4],ymm10[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm8 = ymm8[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1,2],ymm6[3],ymm8[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm7, %ymm9, %ymm7
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm7[0,1],ymm6[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm0, %ymm5
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm6[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[1,1,1,1,5,5,5,5]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm9, %ymm3
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm7[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm7[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm7[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm7[0,1],ymm3[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm15[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2],ymm13[3],ymm7[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm14, %ymm2, %ymm13
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm13[0,1],ymm7[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm0, %ymm12
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm7[0,1,2,3,4],ymm12[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm11[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2],ymm9[3],ymm7[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm10, %ymm2, %ymm9
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm9[0,1],ymm7[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm0, %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm8[5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,1,1,1,5,5,5,5]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2],ymm4[3],ymm6[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm2, %ymm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1],ymm4[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vpermps %ymm1, %ymm0, %ymm0
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm0[5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
@@ -10866,7 +10651,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovaps %ymm1, 96(%r8)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm1, 32(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm1, 224(%r9)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm1, 192(%r9)
@@ -10884,9 +10669,9 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovaps %ymm1, (%r9)
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 224(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm5, 192(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm10, 160(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm14, 128(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm7, 192(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm12, 160(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm3, 128(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 96(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
@@ -10895,7 +10680,7 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, (%rax)
-; AVX2-FP-NEXT:    addq $2536, %rsp # imm = 0x9E8
+; AVX2-FP-NEXT:    addq $2504, %rsp # imm = 0x9C8
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
@@ -10903,608 +10688,534 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP:       # %bb.0:
 ; AVX2-FCP-NEXT:    subq $2504, %rsp # imm = 0x9C8
 ; AVX2-FCP-NEXT:    vmovaps 672(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovaps 640(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 640(%rdi), %ymm6
+; AVX2-FCP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 608(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm6
-; AVX2-FCP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 352(%rdi), %ymm7
 ; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %ymm8
 ; AVX2-FCP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 256(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vmovups %ymm9, (%rsp) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 224(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm0 = [0,6,4,u]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm2[0,1,2,3],ymm1[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm1 = [0,6,4,u]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm2[0,1,2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm1, %ymm0
 ; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm9[0,1],ymm8[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm2[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm15[0,2,2,2,4,6,6,6]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm7[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm8[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm13 = [4,2,4,2,4,2,4,2]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm13, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[0,2,2,2,4,6,6,6]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm7[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm5 = [4,2,4,2,4,2,4,2]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm5[0,1],ymm4[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm1[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm6[0,1],ymm4[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 576(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4,5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovaps 576(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm3[4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm4[0,2,2,2,4,6,6,6]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 704(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 736(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm13, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 1056(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 1024(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovaps 1024(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm2[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 992(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 992(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 960(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm3[0,2,2,2,4,6,6,6]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 1088(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 1120(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm13, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 1440(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 1408(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovaps 1408(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm2[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 1376(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 1376(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 1344(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm3[0,2,2,2,4,6,6,6]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 1472(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 1504(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm13, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm2[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm0[0,1,2,3,4,5],ymm2[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps (%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 32(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm1[0,1,2,3],ymm2[4,5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm13 = ymm0[0,1,2,3],ymm2[4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm13, %ymm1, %ymm0
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm3[0,2,2,2,4,6,6,6]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 128(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 160(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm3[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm13, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm3[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm5, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 480(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 448(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovaps 416(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 448(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm2[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm0[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovaps 416(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 384(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm2[0,1,2,3],ymm1[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm10[0,2,2,2,4,6,6,6]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm2[0,1,2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm11[0,2,2,2,4,6,6,6]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 512(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 544(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm3[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm13, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm3[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm5, %ymm2
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 864(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 832(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovaps 800(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 832(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[0,1],ymm2[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm0[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovaps 800(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 768(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm2[0,1,2,3],ymm1[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm6[0,2,2,2,4,6,6,6]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm2[0,1,2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm8[0,2,2,2,4,6,6,6]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovaps 896(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups %ymm2, (%rsp) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 928(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm3[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm13, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 1184(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm3[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm5, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 1184(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 1152(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm2[0,1,2,3],ymm1[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vmovaps 1248(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm2[0,1,2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vmovaps 1248(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 1216(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[0,1],ymm2[0,1]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[0,1],ymm4[0,1]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm1[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2,2,2,4,6,6,6]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 1280(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovaps 1280(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 1312(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 1312(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm13, %ymm7
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm7[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm5, %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm4[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm0 = [1,7,5,u]
-; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm0, %ymm7
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm14 = ymm15[1,3,2,3,5,7,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm7[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm7 = [5,3,5,3,5,3,5,3]
-; AVX2-FCP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm0, %ymm4
+; AVX2-FCP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm15 = mem[1,3,2,3,5,7,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm4[0,1,2],ymm15[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm4 = [5,3,5,3,5,3,5,3]
+; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm14 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm15[0,1,2,3,4,5],ymm14[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm15 = mem[1,3,2,3,5,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm15 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm15 = mem[1,3,2,3,5,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm15 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm15 = mem[1,3,2,3,5,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2],ymm15[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm15 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm0, %ymm12
+; AVX2-FCP-NEXT:    vpermps %ymm13, %ymm0, %ymm13
 ; AVX2-FCP-NEXT:    vpermilps $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm14 = mem[1,3,2,3,5,7,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2],ymm14[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm7, %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm12[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm0, %ymm9
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[1,3,2,3,5,7,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2],ymm10[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm7, %ymm8
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm9[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm0, %ymm5
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,3,2,3,5,7,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2],ymm6[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm7, %ymm4
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm13 = ymm13[0,1,2],ymm14[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm4, %ymm12
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm0, %ymm10
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm11 = ymm11[1,3,2,3,5,7,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2],ymm11[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm4, %ymm9
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm10[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm0, %ymm7
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm8 = ymm8[1,3,2,3,5,7,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2],ymm8[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm4, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm0, %ymm0
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[1,3,2,3,5,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm7, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 192(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[2,3,2,3]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm0[0],ymm2[1,2,3],ymm0[4],ymm2[5,6,7]
-; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm0 = [2,0,6,4,2,0,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3],ymm4[4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 576(%rdi), %xmm3
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm2 = xmm3[2,3,2,3]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm4[1,2,3],ymm2[4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} xmm5 = [2,0,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0,1],ymm11[2],ymm10[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm7[0,1],mem[2,3],ymm7[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm2 = [2,0,6,4,2,0,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm3[0,1,2,3],mem[4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm3 = [0,0,6,4,0,0,6,4]
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm3, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $251, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm8[0,1],mem[2,3],ymm8[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm2, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm6[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm9[0,1,2,3],ymm12[4],ymm9[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm3, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $251, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm13[0,1],mem[2,3],ymm13[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm2, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm6[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm15[0,1,2,3],ymm14[4],ymm15[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm3, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $251, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm0[2],mem[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm1[0,1],mem[2,3],ymm1[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm2, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm0[0,1,2],ymm6[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $239, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = mem[0,1,2,3],ymm0[4],mem[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm3, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm2, %ymm0
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = ymm4[0,1],mem[2,3],ymm4[4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm0, %ymm4
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2],ymm4[3],ymm5[4,5,6],ymm4[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 960(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[2,3,2,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm5[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm4[1,2,3],ymm2[4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vblendps $251, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = mem[0,1],ymm4[2],mem[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm5, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1,2],ymm0[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,1],ymm4[2,3],mem[4,5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm0, %ymm4
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2],ymm4[3],ymm6[4,5,6],ymm4[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 1344(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[2,3,2,3]
+; AVX2-FCP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm4[0,1,2,3],mem[4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm3, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = ymm0[0,1],mem[2],ymm0[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm6 = ymm4[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm6[1,2,3],ymm2[4],ymm6[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm15[0,1],ymm14[2,3],ymm15[4,5],ymm14[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm0, %ymm6
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm6[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm8 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm8[0,1,2],ymm6[3],ymm8[4,5,6],ymm6[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm6[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps (%rdi), %xmm2
-; AVX2-FCP-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[2,3,2,3]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm6[1,2,3],ymm2[4],ymm6[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = mem[0,1],ymm6[2,3],mem[4,5],ymm6[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm0, %ymm6
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm6[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm9[0,1,2],ymm6[3],ymm9[4,5,6],ymm6[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm6[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 384(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[2,3,2,3]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm9[1,2,3],ymm2[4],ymm9[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,1],ymm6[2,3],mem[4,5],ymm6[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm0, %ymm9
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm10 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm10[0,1,2],ymm9[3],ymm10[4,5,6],ymm9[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm9 = ymm9[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 768(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm9 = xmm2[2,3,2,3]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm10 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0],ymm10[1,2,3],ymm9[4],ymm10[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm10 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm10 = mem[0,1],ymm2[2,3],mem[4,5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm0, %ymm10
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm9 = ymm9[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2],ymm10[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $0, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm10 = mem[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vpermilps $224, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = mem[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2],ymm10[3],ymm11[4,5,6],ymm10[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0,1,2,3,4],ymm10[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm12[0,1],ymm8[2,3],ymm12[4,5],ymm8[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm0, %ymm9
-; AVX2-FCP-NEXT:    vmovaps 1152(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm10 = xmm0[2,3,2,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm11 = ymm2[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0],ymm11[1,2,3],ymm10[4],ymm11[5,6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm10[0,1,2],ymm9[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm4[0,1],mem[2,3],ymm4[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm2, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm6[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm4[0,1,2,3],mem[4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm3, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $4, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = ymm0[0,1],mem[2],ymm0[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm4[0,1],mem[2,3],ymm4[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm2, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm6[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm4[0,1,2,3],mem[4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm3, %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm6[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm10 = ymm6[0,0,0,0,4,4,4,4]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm11 = ymm7[0,0,2,3,4,4,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm11[0,1,2],ymm10[3],ymm11[4,5,6],ymm10[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4],ymm10[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,3,3,3]
-; AVX2-FCP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm9[1,2,3],ymm1[4],ymm9[5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vblendps $34, (%rsp), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm9 = ymm9[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm9[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm10 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm9 = ymm10[0,1,2],ymm9[3],ymm10[4,5,6],ymm9[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm9 = ymm9[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm9[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm1 = xmm3[3,3,3,3]
-; AVX2-FCP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1,2,3],ymm1[4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm9 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm9[0,1,2],ymm3[3],ymm9[4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm1 = mem[3,3,3,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm5[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1,2,3],ymm1[4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2],ymm3[3],ymm5[4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm1 = mem[3,3,3,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm4[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1,2,3],ymm1[4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm14[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0],ymm15[1],ymm3[2,3,4],ymm15[5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3],ymm4[4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm12[1],ymm1[2,3,4],ymm12[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
+; AVX2-FCP-NEXT:    vblendps $251, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1],ymm6[2],mem[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm7[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3],ymm3[4,5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm4[0,1,2,3],mem[4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm3, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-FCP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm15[3,3,3,3,7,7,7,7]
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm7[1],ymm0[2,3,4],ymm7[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm10[0,1,2],ymm11[3],ymm10[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4],ymm0[5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm1 = [3,1,7,5,0,u,u,u]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $223, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,2,3,4],ymm2[5],mem[6,7]
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [0,1,7,5,0,1,7,5]
+; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm8[1],ymm0[2,3,4],ymm8[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm3[1,2,3,4],ymm0[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm9[0,1,2,3,4],ymm12[5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm13[1],ymm0[2,3,4],ymm13[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $247, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,2],ymm3[3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm3[1,2,3,4],ymm0[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm15[0,1,2,3,4],ymm14[5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $247, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,2],ymm3[3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm3[1,2,3,4],ymm0[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $223, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,2,3,4],ymm3[5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vblendps $247, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,2],ymm6[3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm3[1,2,3,4],ymm0[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vblendps $32, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm4[0,1,2,3,4],mem[5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[3,3,3,3]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm13[1],ymm0[2,3,4],ymm13[5],ymm0[6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm14[1],ymm1[2,3,4],ymm14[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermilps $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,0,1,4,5,4,5]
-; AVX2-FCP-NEXT:    vpermilps $244, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm14[0,1,2],mem[3],ymm14[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm3[1,2,3,4],ymm0[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $223, (%rsp), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,2,3,4],ymm3[5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm12[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm9[3,3,3,3,7,7,7,7]
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm0 = mem[3,3,3,3]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm8[1],ymm1[2,3,4],ymm8[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm11[0,1,0,1,4,5,4,5]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm8[1],ymm0[2,3,4],ymm8[5],ymm0[6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm10[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm9[0,1,2],ymm10[3],ymm9[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm3[1,2,3,4],ymm0[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm11[0,1,2,3,4],ymm12[5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm2, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $255, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm0 = mem[3,3,3,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm7[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm4[3,3,3,3,7,7,7,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2,3,4],ymm3[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,3]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,1,0,1,4,5,4,5]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1,2,3,4],ymm15[5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm0, %ymm2, %ymm7
+; AVX2-FCP-NEXT:    vpermilps {{.*#+}} xmm2 = mem[3,3,3,3]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm5[0,1,3,3,4,5,7,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
-; AVX2-FCP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm5[1],ymm2[2,3,4],ymm5[5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0],ymm3[1,2,3,4],ymm2[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm1, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm7[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = ymm7[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm5[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 80(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps %ymm1, %ymm4
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm6[0,1],ymm5[2,3],ymm6[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm15[0,1],ymm6[2,3],ymm15[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [0,2,0,6,0,2,0,6]
-; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm15 = [0,2,0,6,0,2,0,6]
+; AVX2-FCP-NEXT:    # ymm15 = mem[0,1,0,1]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm15, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm1 = ymm0[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm0[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 272(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm15, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm8[0,1,2,3],ymm9[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm8[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm12[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm9[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 464(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm11[0,1],ymm10[2,3],ymm11[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm12[0,1],ymm11[2,3],ymm12[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm15, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
@@ -11512,157 +11223,156 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    # ymm1 = ymm0[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm0[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 656(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm15, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm14[0,1,2,3],ymm15[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm13[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm0[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm14[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovaps 848(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $12, (%rsp), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm15, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm14 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm14 = ymm0[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm12 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm12 = ymm0[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 1040(%rdi), %xmm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1],ymm11[2,3],ymm14[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm13 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm13 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps 1040(%rdi), %xmm12
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm14[0,1],ymm12[2,3],ymm14[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm13, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm10 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm10 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm2, %ymm1
+; AVX2-FCP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm11 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm11 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm15, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm10 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm10 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm9 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm9 = ymm0[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm8 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm8 = ymm0[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 1232(%rdi), %xmm6
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1],ymm6[2,3],ymm9[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps 1232(%rdi), %xmm8
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm10[0,1],ymm8[2,3],ymm10[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm13, %ymm1
+; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm5, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = mem[0,1],ymm1[2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm2, %ymm1
+; AVX2-FCP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm15, %ymm1
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = ymm0[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm0[0,1,2,3],mem[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm0[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 1424(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm4[0,1],ymm1[2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps 1424(%rdi), %xmm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1],ymm2[2,3],ymm6[4,5,6,7]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,1,0,2,4,5,4,6]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm13, %ymm13
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm13[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $243, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm13 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm13 = mem[0,1],ymm7[2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm13, %ymm2, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm5, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm1[0,1],mem[2,3],ymm1[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm15, %ymm5
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm5[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm0 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm4[1,1,1,1,5,5,5,5]
 ; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm0 = ymm0[0,1,2],mem[3],ymm0[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vbroadcastf128 {{.*#+}} ymm0 = [0,3,1,7,0,3,1,7]
 ; AVX2-FCP-NEXT:    # ymm0 = mem[0,1,0,1]
 ; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps $8, (%rsp), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = ymm2[0,1,2],mem[3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm15[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm15[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm5, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = ymm5[0,1,2],mem[3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm15[0,1],ymm5[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = ymm2[0,1,2],mem[3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm15[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm15[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = ymm5[0,1,2],mem[3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm15[0,1],ymm5[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = mem[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = ymm2[0,1,2],mem[3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm15[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm15[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = ymm5[0,1,2],mem[3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm15[0,1],ymm5[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm15 = mem[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm15 = ymm15[0,1,2],mem[3],ymm15[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1],ymm15[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm15[5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermilps $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = mem[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = ymm5[0,1,2],mem[3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm15[0,1],ymm5[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm2[0,1,2,3,4],ymm15[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm14[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm11[3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm7, %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm11[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm10, %ymm0, %ymm10
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm2[0,1,2,3,4],ymm10[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm9[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2],ymm6[3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm7, %ymm6
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm6[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm0, %ymm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm2[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm4[1,1,1,1,5,5,5,5]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm7, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm13, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm15[5,6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm14[0,1,2],ymm12[3],ymm14[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm13, %ymm4, %ymm13
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm13[0,1],ymm12[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm0, %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm12[0,1,2,3,4],ymm11[5,6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm10[0,1,2],ymm8[3],ymm10[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm9, %ymm4, %ymm9
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm9[0,1],ymm8[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm0, %ymm7
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm8[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,1,1,1,5,5,5,5]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm6[0,1,2],ymm2[3],ymm6[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm4, %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm0[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm1, 192(%rsi)
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
@@ -11745,16 +11455,16 @@ define void @load_i32_stride6_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vmovaps %ymm1, (%r9)
 ; AVX2-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FCP-NEXT:    vmovaps %ymm0, 224(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 192(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm10, 160(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm15, 128(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 192(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 160(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%rax)
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm0, 96(%rax)
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm0, 64(%rax)
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm0, 32(%rax)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm0, (%rax)
 ; AVX2-FCP-NEXT:    addq $2504, %rsp # imm = 0x9C8
 ; AVX2-FCP-NEXT:    vzeroupper
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-7.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-7.ll
index b49c35e0812980..1aac5171f99ade 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-7.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-7.ll
@@ -556,24 +556,22 @@ define void @load_i32_stride7_vf4(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX-NEXT:    vmovaps 32(%rdi), %xmm4
 ; AVX-NEXT:    vmovaps 64(%rdi), %xmm5
 ; AVX-NEXT:    vblendps {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
-; AVX-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,3,2,3]
 ; AVX-NEXT:    vmovaps 80(%rdi), %xmm6
-; AVX-NEXT:    vshufps {{.*#+}} xmm7 = xmm6[0,1,0,1]
-; AVX-NEXT:    vblendps {{.*#+}} xmm2 = xmm2[0,1,2],xmm7[3]
-; AVX-NEXT:    vblendps {{.*#+}} xmm8 = xmm4[0],xmm3[1],xmm4[2,3]
-; AVX-NEXT:    vshufps {{.*#+}} xmm8 = xmm8[1,0],mem[3,3]
-; AVX-NEXT:    vinsertps {{.*#+}} xmm8 = xmm8[0,1,2],xmm6[2]
-; AVX-NEXT:    vshufps {{.*#+}} xmm9 = xmm5[0,1,0,1]
-; AVX-NEXT:    vblendps {{.*#+}} xmm6 = xmm9[0,1,2],xmm6[3]
+; AVX-NEXT:    vblendps {{.*#+}} xmm2 = xmm2[0],xmm6[1],xmm2[2,3]
+; AVX-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,3,2,1]
+; AVX-NEXT:    vblendps {{.*#+}} xmm7 = xmm4[0],xmm3[1],xmm4[2,3]
+; AVX-NEXT:    vshufps {{.*#+}} xmm7 = xmm7[1,0],mem[3,3]
+; AVX-NEXT:    vinsertps {{.*#+}} xmm7 = xmm7[0,1,2],xmm6[2]
+; AVX-NEXT:    vshufps {{.*#+}} xmm8 = xmm5[0,1,0,1]
+; AVX-NEXT:    vblendps {{.*#+}} xmm8 = xmm8[0,1,2],xmm6[3]
 ; AVX-NEXT:    vshufps {{.*#+}} xmm9 = xmm3[2,3,2,3]
 ; AVX-NEXT:    vblendps {{.*#+}} xmm9 = xmm9[0],xmm4[1],xmm9[2,3]
-; AVX-NEXT:    vblendps {{.*#+}} xmm6 = xmm9[0,1],xmm6[2,3]
+; AVX-NEXT:    vblendps {{.*#+}} xmm8 = xmm9[0,1],xmm8[2,3]
 ; AVX-NEXT:    vmovaps 96(%rdi), %xmm9
 ; AVX-NEXT:    vblendps {{.*#+}} xmm10 = xmm9[0],xmm5[1],xmm9[2,3]
-; AVX-NEXT:    vshufps {{.*#+}} xmm10 = xmm10[0,1,1,0]
 ; AVX-NEXT:    vblendps {{.*#+}} xmm3 = xmm4[0,1,2],xmm3[3]
-; AVX-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[3,2,2,3]
-; AVX-NEXT:    vblendps {{.*#+}} xmm3 = xmm3[0,1],xmm10[2,3]
+; AVX-NEXT:    vblendps {{.*#+}} xmm3 = xmm10[0,1],xmm3[2,3]
+; AVX-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[3,2,1,0]
 ; AVX-NEXT:    vshufps {{.*#+}} xmm10 = xmm9[0,1,0,1]
 ; AVX-NEXT:    vblendps {{.*#+}} xmm10 = xmm5[0,1,2],xmm10[3]
 ; AVX-NEXT:    vshufps {{.*#+}} xmm4 = xmm4[2,3,2,3]
@@ -584,14 +582,15 @@ define void @load_i32_stride7_vf4(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[2,0,2,3,6,4,6,7]
 ; AVX-NEXT:    vextractf128 $1, %ymm10, %xmm10
 ; AVX-NEXT:    vshufps {{.*#+}} xmm5 = xmm10[0,1],xmm5[3,2]
-; AVX-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1,2],xmm9[3]
+; AVX-NEXT:    vshufps {{.*#+}} xmm6 = xmm6[0,1,0,1]
+; AVX-NEXT:    vblendps {{.*#+}} xmm6 = xmm6[0,1,2],xmm9[3]
 ; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,0],ymm1[2,0],ymm0[5,4],ymm1[6,4]
 ; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,0,2,3,6,4,6,7]
 ; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm7[2,3]
+; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm6[2,3]
 ; AVX-NEXT:    vmovaps %xmm2, (%rsi)
-; AVX-NEXT:    vmovaps %xmm8, (%rdx)
-; AVX-NEXT:    vmovaps %xmm6, (%rcx)
+; AVX-NEXT:    vmovaps %xmm7, (%rdx)
+; AVX-NEXT:    vmovaps %xmm8, (%rcx)
 ; AVX-NEXT:    vmovaps %xmm3, (%r8)
 ; AVX-NEXT:    vmovaps %xmm4, (%r9)
 ; AVX-NEXT:    vmovaps %xmm5, (%r10)
@@ -625,10 +624,9 @@ define void @load_i32_stride7_vf4(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
 ; AVX2-NEXT:    vmovaps 96(%rdi), %xmm6
 ; AVX2-NEXT:    vblendps {{.*#+}} xmm9 = xmm6[0],xmm8[1],xmm6[2,3]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm9 = xmm9[0,1,1,0]
 ; AVX2-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1,2],mem[3]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm7 = xmm7[3,2,2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1],xmm9[2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm7 = xmm9[0,1],xmm7[2,3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm7 = xmm7[3,2,1,0]
 ; AVX2-NEXT:    vbroadcastss 100(%rdi), %xmm9
 ; AVX2-NEXT:    vblendps {{.*#+}} xmm9 = xmm8[0,1,2],xmm9[3]
 ; AVX2-NEXT:    vmovsd {{.*#+}} xmm10 = [4,3,0,0]
@@ -682,10 +680,9 @@ define void @load_i32_stride7_vf4(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
 ; AVX2-FP-NEXT:    vmovaps 96(%rdi), %xmm6
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm9 = xmm6[0],xmm8[1],xmm6[2,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm9 = xmm9[0,1,1,0]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1,2],mem[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm7 = xmm7[3,2,2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1],xmm9[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm7 = xmm9[0,1],xmm7[2,3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm7 = xmm7[3,2,1,0]
 ; AVX2-FP-NEXT:    vbroadcastss 100(%rdi), %xmm9
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm9 = xmm8[0,1,2],xmm9[3]
 ; AVX2-FP-NEXT:    vmovsd {{.*#+}} xmm10 = [4,3,0,0]
@@ -740,10 +737,9 @@ define void @load_i32_stride7_vf4(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3]
 ; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %xmm6
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm9 = xmm6[0],xmm8[1],xmm6[2,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm9 = xmm9[0,1,1,0]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1,2],mem[3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm7 = xmm7[3,2,2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1],xmm9[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm7 = xmm9[0,1],xmm7[2,3]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm7 = xmm7[3,2,1,0]
 ; AVX2-FCP-NEXT:    vbroadcastss 100(%rdi), %xmm9
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm9 = xmm8[0,1,2],xmm9[3]
 ; AVX2-FCP-NEXT:    vmovsd {{.*#+}} xmm10 = [4,3,0,0]
@@ -1303,12 +1299,10 @@ define void @load_i32_stride7_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,2,0]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm7[0,1,2,3,4,5,6],ymm6[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1],ymm9[2,3],ymm10[4,5],ymm9[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,2,2,2]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm1[0],ymm0[1],ymm1[2,3,4],ymm0[5],ymm1[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm11 = ymm8[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm11[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3,4],ymm7[5,6],ymm8[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm11 = [1,0,7,6,5,0,0,0]
+; AVX2-NEXT:    vpermd %ymm7, %ymm11, %ymm7
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm7[0,1,2,3,4],ymm6[5,6,7]
 ; AVX2-NEXT:    vmovdqa 80(%rdi), %xmm7
 ; AVX2-NEXT:    vpalignr {{.*#+}} ymm11 = ymm10[8,9,10,11,12,13,14,15],ymm9[0,1,2,3,4,5,6,7],ymm10[24,25,26,27,28,29,30,31],ymm9[16,17,18,19,20,21,22,23]
@@ -1406,12 +1400,10 @@ define void @load_i32_stride7_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,2,0]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm7[0,1,2,3,4,5,6],ymm6[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1],ymm9[2,3],ymm10[4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,2,2,2]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm1[0],ymm0[1],ymm1[2,3,4],ymm0[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm11 = ymm8[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm11[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3,4],ymm7[5,6],ymm8[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm11 = [1,0,7,6,5,0,0,0]
+; AVX2-FP-NEXT:    vpermd %ymm7, %ymm11, %ymm7
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm7[0,1,2,3,4],ymm6[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 80(%rdi), %xmm7
 ; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm11 = ymm10[8,9,10,11,12,13,14,15],ymm9[0,1,2,3,4,5,6,7],ymm10[24,25,26,27,28,29,30,31],ymm9[16,17,18,19,20,21,22,23]
@@ -1509,12 +1501,10 @@ define void @load_i32_stride7_vf8(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, ptr
 ; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,2,0]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm7[0,1,2,3,4,5,6],ymm6[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1],ymm9[2,3],ymm10[4,5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm8 = [5,6,5,6,5,6,5,6]
-; AVX2-FCP-NEXT:    vpermd %ymm7, %ymm8, %ymm7
-; AVX2-FCP-NEXT:    vpmovsxbq {{.*#+}} ymm11 = [1,7,0,0]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm1[0],ymm0[1],ymm1[2,3,4],ymm0[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm8, %ymm11, %ymm11
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm11[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3,4],ymm7[5,6],ymm8[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm11 = [1,0,7,6,5,0,0,0]
+; AVX2-FCP-NEXT:    vpermd %ymm7, %ymm11, %ymm7
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm7[0,1,2,3,4],ymm6[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 80(%rdi), %xmm7
 ; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm11 = ymm10[8,9,10,11,12,13,14,15],ymm9[0,1,2,3,4,5,6,7],ymm10[24,25,26,27,28,29,30,31],ymm9[16,17,18,19,20,21,22,23]
@@ -2605,225 +2595,219 @@ define void @load_i32_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-LABEL: load_i32_stride7_vf16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    subq $264, %rsp # imm = 0x108
-; AVX2-NEXT:    vmovdqa 288(%rdi), %ymm15
-; AVX2-NEXT:    vmovdqa 384(%rdi), %ymm8
-; AVX2-NEXT:    vmovdqa 352(%rdi), %ymm6
-; AVX2-NEXT:    vmovdqa 320(%rdi), %ymm5
+; AVX2-NEXT:    vmovdqa 288(%rdi), %ymm5
+; AVX2-NEXT:    vmovdqa 384(%rdi), %ymm9
+; AVX2-NEXT:    vmovdqa 352(%rdi), %ymm7
+; AVX2-NEXT:    vmovdqa 320(%rdi), %ymm4
 ; AVX2-NEXT:    vmovdqa 256(%rdi), %ymm0
-; AVX2-NEXT:    vmovdqa 224(%rdi), %ymm10
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm9
-; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm4
-; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm12
+; AVX2-NEXT:    vmovdqa 224(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqa (%rdi), %ymm10
+; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm6
+; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm15
 ; AVX2-NEXT:    vpbroadcastq 80(%rdi), %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm12[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm15[4,5,6,7]
 ; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [0,7,6,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm9[0,1,2,3,4,5],ymm4[6],ymm9[7]
-; AVX2-NEXT:    vmovdqa %ymm4, %ymm7
-; AVX2-NEXT:    vpermd %ymm3, %ymm2, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vmovdqa 128(%rdi), %xmm3
-; AVX2-NEXT:    vmovdqa 160(%rdi), %xmm4
-; AVX2-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vpunpckhqdq {{.*#+}} xmm3 = xmm3[1],xmm4[1]
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-NEXT:    vpbroadcastd 196(%rdi), %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm10[0,1,2,3,4,5],ymm6[6],ymm10[7]
+; AVX2-NEXT:    vpermd %ymm8, %ymm2, %ymm8
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1,2],ymm1[3,4,5,6,7]
+; AVX2-NEXT:    vmovdqa 128(%rdi), %xmm8
+; AVX2-NEXT:    vmovdqa 160(%rdi), %xmm11
+; AVX2-NEXT:    vmovdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-NEXT:    vpunpckhqdq {{.*#+}} xmm8 = xmm8[1],xmm11[1]
+; AVX2-NEXT:    vinserti128 $1, %xmm8, %ymm0, %ymm8
+; AVX2-NEXT:    vpbroadcastd 196(%rdi), %ymm11
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6],ymm11[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm8[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm10[0,1,2,3,4,5],ymm0[6],ymm10[7]
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm0[6],ymm3[7]
 ; AVX2-NEXT:    vpermd %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpbroadcastq 304(%rdi), %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4,5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 352(%rdi), %xmm2
-; AVX2-NEXT:    vmovdqa 384(%rdi), %xmm3
-; AVX2-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vpunpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm3[1]
+; AVX2-NEXT:    vmovdqa 384(%rdi), %xmm8
+; AVX2-NEXT:    vmovdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-NEXT:    vpunpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm8[1]
 ; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-NEXT:    vpbroadcastd 420(%rdi), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-NEXT:    vpbroadcastd 420(%rdi), %ymm8
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm8[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
 ; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm8[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm8[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm9[12,13,14,15],ymm7[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm7[16,17,18,19,20,21,22,23,24,25,26,27]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm15[2,3],ymm5[4,5],ymm15[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm0[0],ymm10[1],ymm0[2,3,4],ymm10[5],ymm0[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm5[2,3],ymm4[4,5],ymm5[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm0[0],ymm3[1],ymm0[2,3,4],ymm3[5],ymm0[6,7]
+; AVX2-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0,1,2,3,4],ymm2[5,6],ymm8[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm12 = [1,0,7,6,5,6,5,6]
+; AVX2-NEXT:    vpermd %ymm2, %ymm12, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm3
-; AVX2-NEXT:    vmovdqa 160(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm1
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm4 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm10
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm12[0,1],ymm10[2,3],ymm12[4,5],ymm10[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm4[2,2,2,2]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm7[0],ymm9[1],ymm7[2,3,4],ymm9[5],ymm7[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm14 = ymm4[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm14[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm8
+; AVX2-NEXT:    vmovdqa 160(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm2
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm11 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm11[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm11
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm11[2,3],ymm15[4,5],ymm11[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm6[0],ymm10[1],ymm6[2,3,4],ymm10[5],ymm6[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3,4],ymm0[5,6],ymm8[7]
+; AVX2-NEXT:    vpermd %ymm0, %ymm12, %ymm0
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 80(%rdi), %xmm0
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm3 = ymm12[8,9,10,11,12,13,14,15],ymm10[0,1,2,3,4,5,6,7],ymm12[24,25,26,27,28,29,30,31],ymm10[16,17,18,19,20,21,22,23]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 8(%rdi), %xmm3
-; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm14
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0],xmm14[1],xmm3[2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm3 = ymm1[0],ymm2[0],ymm1[2],ymm2[2]
-; AVX2-NEXT:    vpbroadcastd 204(%rdi), %ymm13
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm15[8,9,10,11,12,13,14,15],ymm11[0,1,2,3,4,5,6,7],ymm15[24,25,26,27,28,29,30,31],ymm11[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpbroadcastd 8(%rdi), %xmm1
+; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm12
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm12[1],xmm1[2,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm2[0],ymm3[0],ymm2[2],ymm3[2]
+; AVX2-NEXT:    vpbroadcastd 204(%rdi), %ymm14
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm14[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 304(%rdi), %xmm0
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm3 = ymm5[8,9,10,11,12,13,14,15],ymm15[0,1,2,3,4,5,6,7],ymm5[24,25,26,27,28,29,30,31],ymm15[16,17,18,19,20,21,22,23]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 232(%rdi), %xmm3
-; AVX2-NEXT:    vmovdqa 256(%rdi), %xmm13
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0],xmm13[1],xmm3[2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm3 = ymm6[0],ymm8[0],ymm6[2],ymm8[2]
-; AVX2-NEXT:    vpbroadcastd 428(%rdi), %ymm11
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm11[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm4[8,9,10,11,12,13,14,15],ymm5[0,1,2,3,4,5,6,7],ymm4[24,25,26,27,28,29,30,31],ymm5[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpbroadcastd 232(%rdi), %xmm1
+; AVX2-NEXT:    vmovdqa 256(%rdi), %xmm14
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm14[1],xmm1[2,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm7[0],ymm9[0],ymm7[2],ymm9[2]
+; AVX2-NEXT:    vpbroadcastd 428(%rdi), %ymm13
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0],ymm10[1],ymm12[2,3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm14[0,1,2],mem[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[3,2,2,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0],ymm11[1],ymm15[2,3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm12[0,1,2],mem[3]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm1[0,2],ymm2[1,3],ymm1[4,6],ymm2[5,7]
-; AVX2-NEXT:    vbroadcastss 208(%rdi), %ymm10
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm10[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2],ymm3[1,3],ymm2[4,6],ymm3[5,7]
+; AVX2-NEXT:    vbroadcastss 208(%rdi), %ymm11
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm11[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0],ymm15[1],ymm5[2,3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm13[0,1,2],mem[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[3,2,2,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0],ymm5[1],ymm4[2,3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm14[0,1,2],mem[3]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm6[0,2],ymm8[1,3],ymm6[4,6],ymm8[5,7]
-; AVX2-NEXT:    vbroadcastss 432(%rdi), %ymm5
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm5[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm7[0,2],ymm9[1,3],ymm7[4,6],ymm9[5,7]
+; AVX2-NEXT:    vbroadcastss 432(%rdi), %ymm4
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpbroadcastd 100(%rdi), %xmm0
-; AVX2-NEXT:    vmovdqa 64(%rdi), %xmm3
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm3[0,1,2],xmm0[3]
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [4,3,0,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm7[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vmovdqa %ymm7, %ymm12
-; AVX2-NEXT:    vpermd %ymm10, %ymm5, %ymm10
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm10 = xmm10[0,1],xmm0[2,3]
+; AVX2-NEXT:    vmovdqa 64(%rdi), %xmm1
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm4 = [4,3,0,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-NEXT:    vmovdqa %ymm6, %ymm15
+; AVX2-NEXT:    vpermd %ymm5, %ymm4, %ymm5
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1],xmm0[2,3]
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [0,7,0,7,0,7,0,7]
-; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm11
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-NEXT:    vpbroadcastd 212(%rdi), %ymm13
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm10[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm11
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-NEXT:    vpbroadcastd 212(%rdi), %ymm12
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm5[0,1,2,3],ymm11[4,5,6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm14[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm10, %ymm5, %ymm5
-; AVX2-NEXT:    vpbroadcastd 324(%rdi), %xmm10
-; AVX2-NEXT:    vmovdqa 288(%rdi), %xmm11
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm10 = xmm11[0,1,2],xmm10[3]
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1],xmm10[2,3]
-; AVX2-NEXT:    vpermd %ymm6, %ymm0, %ymm10
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-NEXT:    vpbroadcastd 436(%rdi), %ymm13
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm5[0,1,2,3],ymm10[4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,3]
-; AVX2-NEXT:    vpbroadcastd 216(%rdi), %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-NEXT:    vmovdqa 96(%rdi), %xmm5
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm5[0,1,2],xmm3[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,3,2]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm4[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextracti128 $1, %ymm3, %xmm3
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-NEXT:    vmovdqa 320(%rdi), %xmm3
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm3[0,1,2],xmm11[3]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm14[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-NEXT:    vpermd %ymm5, %ymm4, %ymm4
+; AVX2-NEXT:    vpbroadcastd 324(%rdi), %xmm5
+; AVX2-NEXT:    vmovdqa 288(%rdi), %xmm13
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm5 = xmm13[0,1,2],xmm5[3]
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-NEXT:    vpermd %ymm7, %ymm0, %ymm5
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-NEXT:    vpbroadcastd 436(%rdi), %ymm11
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm11[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm4[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3,4],ymm2[5],ymm3[6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,3]
+; AVX2-NEXT:    vpbroadcastd 216(%rdi), %ymm3
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-NEXT:    vmovdqa 96(%rdi), %xmm3
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm3[0,1,2],xmm1[3]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm8[1,0,2,3,5,4,6,7]
+; AVX2-NEXT:    vextracti128 $1, %ymm4, %xmm4
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm4[0,1],xmm1[2,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vmovdqa 320(%rdi), %xmm8
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm8[0,1,2],xmm13[3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,3,2]
 ; AVX2-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm4 = mem[1,0,2,3,5,4,6,7]
 ; AVX2-NEXT:    vextracti128 $1, %ymm4, %xmm4
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm8[0],ymm6[1],ymm8[2,3,4],ymm6[5],ymm8[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm9[0],ymm7[1],ymm9[2,3,4],ymm7[5],ymm9[6,7]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[1,0,3,3,5,4,7,7]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-NEXT:    vpbroadcastd 440(%rdi), %ymm6
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-NEXT:    vpbroadcastd 440(%rdi), %ymm5
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm5[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 136(%rdi), %xmm4
 ; AVX2-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
 ; AVX2-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-NEXT:    vpermd 192(%rdi), %ymm0, %ymm6
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-NEXT:    vpbroadcastd 80(%rdi), %ymm6
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm5 = xmm6[0,1,2],xmm5[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm6 = ymm9[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0],ymm12[1],ymm6[2,3,4],ymm12[5],ymm6[6,7]
-; AVX2-NEXT:    vextracti128 $1, %ymm6, %xmm6
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 360(%rdi), %xmm5
-; AVX2-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5, %xmm5 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm5 = xmm5[0],mem[1],xmm5[2,3]
-; AVX2-NEXT:    vpermd 416(%rdi), %ymm0, %ymm0
-; AVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm5
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-NEXT:    vpbroadcastd 304(%rdi), %ymm5
+; AVX2-NEXT:    vpermd 192(%rdi), %ymm0, %ymm5
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-NEXT:    vpbroadcastd 80(%rdi), %ymm5
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm5[0,1,2],xmm3[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm7[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm14[1],ymm5[2,3,4],ymm14[5],ymm5[6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm10[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm15[1],ymm5[2,3,4],ymm15[5],ymm5[6,7]
 ; AVX2-NEXT:    vextracti128 $1, %ymm5, %xmm5
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, 32(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, (%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, 32(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, (%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, 32(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, (%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, 32(%r8)
-; AVX2-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm3, (%r8)
-; AVX2-NEXT:    vmovdqa %ymm10, 32(%r9)
-; AVX2-NEXT:    vmovdqa %ymm15, (%r9)
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vpbroadcastd 360(%rdi), %xmm4
+; AVX2-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
+; AVX2-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
+; AVX2-NEXT:    vpermd 416(%rdi), %ymm0, %ymm0
+; AVX2-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-NEXT:    vpbroadcastd 304(%rdi), %ymm4
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm8[3]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm6[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm14[1],ymm5[2,3,4],ymm14[5],ymm5[6,7]
+; AVX2-NEXT:    vextracti128 $1, %ymm5, %xmm5
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm5[0,1],xmm4[2,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm4, 32(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm4, (%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm4, 32(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm4, (%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm4, 32(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm4, (%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm4, 32(%r8)
+; AVX2-NEXT:    vmovups (%rsp), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm4, (%r8)
+; AVX2-NEXT:    vmovdqa %ymm11, 32(%r9)
+; AVX2-NEXT:    vmovdqa %ymm12, (%r9)
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-NEXT:    vmovdqa %ymm2, 32(%rax)
 ; AVX2-NEXT:    vmovdqa %ymm1, (%rax)
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-NEXT:    vmovdqa %ymm0, 32(%rax)
-; AVX2-NEXT:    vmovdqa %ymm4, (%rax)
+; AVX2-NEXT:    vmovdqa %ymm3, (%rax)
 ; AVX2-NEXT:    addq $264, %rsp # imm = 0x108
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
@@ -2831,452 +2815,439 @@ define void @load_i32_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-LABEL: load_i32_stride7_vf16:
 ; AVX2-FP:       # %bb.0:
 ; AVX2-FP-NEXT:    subq $264, %rsp # imm = 0x108
-; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %ymm15
-; AVX2-FP-NEXT:    vmovdqa 384(%rdi), %ymm8
-; AVX2-FP-NEXT:    vmovdqa 352(%rdi), %ymm6
-; AVX2-FP-NEXT:    vmovdqa 320(%rdi), %ymm5
+; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %ymm5
+; AVX2-FP-NEXT:    vmovdqa 384(%rdi), %ymm9
+; AVX2-FP-NEXT:    vmovdqa 352(%rdi), %ymm7
+; AVX2-FP-NEXT:    vmovdqa 320(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %ymm0
-; AVX2-FP-NEXT:    vmovdqa 224(%rdi), %ymm10
-; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm9
-; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm4
-; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm12
+; AVX2-FP-NEXT:    vmovdqa 224(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm10
+; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm6
+; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm15
 ; AVX2-FP-NEXT:    vpbroadcastq 80(%rdi), %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm12[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm15[4,5,6,7]
 ; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [0,7,6,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm9[0,1,2,3,4,5],ymm4[6],ymm9[7]
-; AVX2-FP-NEXT:    vmovdqa %ymm4, %ymm7
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm2, %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %xmm3
-; AVX2-FP-NEXT:    vmovdqa 160(%rdi), %xmm4
-; AVX2-FP-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vpunpckhqdq {{.*#+}} xmm3 = xmm3[1],xmm4[1]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FP-NEXT:    vpbroadcastd 196(%rdi), %ymm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm10[0,1,2,3,4,5],ymm6[6],ymm10[7]
+; AVX2-FP-NEXT:    vpermd %ymm8, %ymm2, %ymm8
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1,2],ymm1[3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %xmm8
+; AVX2-FP-NEXT:    vmovdqa 160(%rdi), %xmm11
+; AVX2-FP-NEXT:    vmovdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FP-NEXT:    vpunpckhqdq {{.*#+}} xmm8 = xmm8[1],xmm11[1]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm8, %ymm0, %ymm8
+; AVX2-FP-NEXT:    vpbroadcastd 196(%rdi), %ymm11
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6],ymm11[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm8[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm10[0,1,2,3,4,5],ymm0[6],ymm10[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm0[6],ymm3[7]
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm2, %ymm1
 ; AVX2-FP-NEXT:    vpbroadcastq 304(%rdi), %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4,5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 352(%rdi), %xmm2
-; AVX2-FP-NEXT:    vmovdqa 384(%rdi), %xmm3
-; AVX2-FP-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vpunpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm3[1]
+; AVX2-FP-NEXT:    vmovdqa 384(%rdi), %xmm8
+; AVX2-FP-NEXT:    vmovdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FP-NEXT:    vpunpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm8[1]
 ; AVX2-FP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vpbroadcastd 420(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FP-NEXT:    vpbroadcastd 420(%rdi), %ymm8
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm8[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
 ; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm8[12,13,14,15],ymm6[0,1,2,3,4,5,6,7,8,9,10,11],ymm8[28,29,30,31],ymm6[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm9[12,13,14,15],ymm7[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm7[16,17,18,19,20,21,22,23,24,25,26,27]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm15[2,3],ymm5[4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm0[0],ymm10[1],ymm0[2,3,4],ymm10[5],ymm0[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm5[2,3],ymm4[4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm0[0],ymm3[1],ymm0[2,3,4],ymm3[5],ymm0[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0,1,2,3,4],ymm2[5,6],ymm8[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm12 = [1,0,7,6,5,6,5,6]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm12, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm3
-; AVX2-FP-NEXT:    vmovdqa 160(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm1
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm4 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm10
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm12[0,1],ymm10[2,3],ymm12[4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm4[2,2,2,2]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm7[0],ymm9[1],ymm7[2,3,4],ymm9[5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm14 = ymm4[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm14[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm8
+; AVX2-FP-NEXT:    vmovdqa 160(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm11 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm11[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm11
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm11[2,3],ymm15[4,5],ymm11[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm6[0],ymm10[1],ymm6[2,3,4],ymm10[5],ymm6[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3,4],ymm0[5,6],ymm8[7]
+; AVX2-FP-NEXT:    vpermd %ymm0, %ymm12, %ymm0
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 80(%rdi), %xmm0
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm12[8,9,10,11,12,13,14,15],ymm10[0,1,2,3,4,5,6,7],ymm12[24,25,26,27,28,29,30,31],ymm10[16,17,18,19,20,21,22,23]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 8(%rdi), %xmm3
-; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %xmm14
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0],xmm14[1],xmm3[2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm3 = ymm1[0],ymm2[0],ymm1[2],ymm2[2]
-; AVX2-FP-NEXT:    vpbroadcastd 204(%rdi), %ymm13
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm15[8,9,10,11,12,13,14,15],ymm11[0,1,2,3,4,5,6,7],ymm15[24,25,26,27,28,29,30,31],ymm11[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpbroadcastd 8(%rdi), %xmm1
+; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %xmm12
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm12[1],xmm1[2,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm2[0],ymm3[0],ymm2[2],ymm3[2]
+; AVX2-FP-NEXT:    vpbroadcastd 204(%rdi), %ymm14
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm14[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 304(%rdi), %xmm0
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm5[8,9,10,11,12,13,14,15],ymm15[0,1,2,3,4,5,6,7],ymm5[24,25,26,27,28,29,30,31],ymm15[16,17,18,19,20,21,22,23]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 232(%rdi), %xmm3
-; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %xmm13
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0],xmm13[1],xmm3[2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm3 = ymm6[0],ymm8[0],ymm6[2],ymm8[2]
-; AVX2-FP-NEXT:    vpbroadcastd 428(%rdi), %ymm11
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm11[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm4[8,9,10,11,12,13,14,15],ymm5[0,1,2,3,4,5,6,7],ymm4[24,25,26,27,28,29,30,31],ymm5[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpbroadcastd 232(%rdi), %xmm1
+; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %xmm14
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm14[1],xmm1[2,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm7[0],ymm9[0],ymm7[2],ymm9[2]
+; AVX2-FP-NEXT:    vpbroadcastd 428(%rdi), %ymm13
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0],ymm10[1],ymm12[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm14[0,1,2],mem[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[3,2,2,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0],ymm11[1],ymm15[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm12[0,1,2],mem[3]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm1[0,2],ymm2[1,3],ymm1[4,6],ymm2[5,7]
-; AVX2-FP-NEXT:    vbroadcastss 208(%rdi), %ymm10
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm10[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2],ymm3[1,3],ymm2[4,6],ymm3[5,7]
+; AVX2-FP-NEXT:    vbroadcastss 208(%rdi), %ymm11
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm11[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0],ymm15[1],ymm5[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm13[0,1,2],mem[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[3,2,2,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0],ymm5[1],ymm4[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm14[0,1,2],mem[3]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm6[0,2],ymm8[1,3],ymm6[4,6],ymm8[5,7]
-; AVX2-FP-NEXT:    vbroadcastss 432(%rdi), %ymm5
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm5[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm7[0,2],ymm9[1,3],ymm7[4,6],ymm9[5,7]
+; AVX2-FP-NEXT:    vbroadcastss 432(%rdi), %ymm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpbroadcastd 100(%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %xmm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm3[0,1,2],xmm0[3]
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm5 = [4,3,0,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm7[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm7, %ymm12
-; AVX2-FP-NEXT:    vpermd %ymm10, %ymm5, %ymm10
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm10 = xmm10[0,1],xmm0[2,3]
+; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %xmm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm4 = [4,3,0,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm6, %ymm15
+; AVX2-FP-NEXT:    vpermd %ymm5, %ymm4, %ymm5
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1],xmm0[2,3]
 ; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [0,7,0,7,0,7,0,7]
-; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm11
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 212(%rdi), %ymm13
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm10[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm11
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vpbroadcastd 212(%rdi), %ymm12
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm5[0,1,2,3],ymm11[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm14[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm10, %ymm5, %ymm5
-; AVX2-FP-NEXT:    vpbroadcastd 324(%rdi), %xmm10
-; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %xmm11
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm10 = xmm11[0,1,2],xmm10[3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1],xmm10[2,3]
-; AVX2-FP-NEXT:    vpermd %ymm6, %ymm0, %ymm10
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 436(%rdi), %ymm13
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm5[0,1,2,3],ymm10[4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,3]
-; AVX2-FP-NEXT:    vpbroadcastd 216(%rdi), %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %xmm5
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm5[0,1,2],xmm3[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,3,2]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm4[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextracti128 $1, %ymm3, %xmm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 320(%rdi), %xmm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm3[0,1,2],xmm11[3]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm14[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FP-NEXT:    vpermd %ymm5, %ymm4, %ymm4
+; AVX2-FP-NEXT:    vpbroadcastd 324(%rdi), %xmm5
+; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %xmm13
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm13[0,1,2],xmm5[3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-FP-NEXT:    vpermd %ymm7, %ymm0, %ymm5
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-FP-NEXT:    vpbroadcastd 436(%rdi), %ymm11
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm11[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm4[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3,4],ymm2[5],ymm3[6,7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,3]
+; AVX2-FP-NEXT:    vpbroadcastd 216(%rdi), %ymm3
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %xmm3
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm3[0,1,2],xmm1[3]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm8[1,0,2,3,5,4,6,7]
+; AVX2-FP-NEXT:    vextracti128 $1, %ymm4, %xmm4
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm4[0,1],xmm1[2,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqa 320(%rdi), %xmm8
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm8[0,1,2],xmm13[3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,3,2]
 ; AVX2-FP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm4 = mem[1,0,2,3,5,4,6,7]
 ; AVX2-FP-NEXT:    vextracti128 $1, %ymm4, %xmm4
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm8[0],ymm6[1],ymm8[2,3,4],ymm6[5],ymm8[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm9[0],ymm7[1],ymm9[2,3,4],ymm7[5],ymm9[6,7]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[1,0,3,3,5,4,7,7]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,0,3]
-; AVX2-FP-NEXT:    vpbroadcastd 440(%rdi), %ymm6
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-FP-NEXT:    vpbroadcastd 440(%rdi), %ymm5
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm5[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 136(%rdi), %xmm4
 ; AVX2-FP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
 ; AVX2-FP-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
 ; AVX2-FP-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-FP-NEXT:    vpermd 192(%rdi), %ymm0, %ymm6
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 80(%rdi), %ymm6
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm6[0,1,2],xmm5[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm6 = ymm9[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0],ymm12[1],ymm6[2,3,4],ymm12[5],ymm6[6,7]
-; AVX2-FP-NEXT:    vextracti128 $1, %ymm6, %xmm6
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 360(%rdi), %xmm5
-; AVX2-FP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5, %xmm5 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm5 = xmm5[0],mem[1],xmm5[2,3]
-; AVX2-FP-NEXT:    vpermd 416(%rdi), %ymm0, %ymm0
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm5
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm5[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 304(%rdi), %ymm5
+; AVX2-FP-NEXT:    vpermd 192(%rdi), %ymm0, %ymm5
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpbroadcastd 80(%rdi), %ymm5
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm5[0,1,2],xmm3[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm7[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm14[1],ymm5[2,3,4],ymm14[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm10[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm15[1],ymm5[2,3,4],ymm15[5],ymm5[6,7]
 ; AVX2-FP-NEXT:    vextracti128 $1, %ymm5, %xmm5
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, (%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, (%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, (%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, 32(%r8)
-; AVX2-FP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm3, (%r8)
-; AVX2-FP-NEXT:    vmovdqa %ymm10, 32(%r9)
-; AVX2-FP-NEXT:    vmovdqa %ymm15, (%r9)
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vpbroadcastd 360(%rdi), %xmm4
+; AVX2-FP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
+; AVX2-FP-NEXT:    vpermd 416(%rdi), %ymm0, %ymm0
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpbroadcastd 304(%rdi), %ymm4
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm8[3]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm6[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm14[1],ymm5[2,3,4],ymm14[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vextracti128 $1, %ymm5, %xmm5
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm5[0,1],xmm4[2,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm4, 32(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm4, (%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm4, 32(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm4, (%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm4, 32(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm4, (%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm4, 32(%r8)
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm4, (%r8)
+; AVX2-FP-NEXT:    vmovdqa %ymm11, 32(%r9)
+; AVX2-FP-NEXT:    vmovdqa %ymm12, (%r9)
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FP-NEXT:    vmovdqa %ymm2, 32(%rax)
 ; AVX2-FP-NEXT:    vmovdqa %ymm1, (%rax)
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FP-NEXT:    vmovdqa %ymm0, 32(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm4, (%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm3, (%rax)
 ; AVX2-FP-NEXT:    addq $264, %rsp # imm = 0x108
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
 ; AVX2-FCP-LABEL: load_i32_stride7_vf16:
 ; AVX2-FCP:       # %bb.0:
-; AVX2-FCP-NEXT:    subq $296, %rsp # imm = 0x128
-; AVX2-FCP-NEXT:    vmovdqa 160(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %ymm3
+; AVX2-FCP-NEXT:    subq $264, %rsp # imm = 0x108
+; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %ymm9
+; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %ymm7
+; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %ymm4
 ; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %ymm0
-; AVX2-FCP-NEXT:    vmovdqa 224(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqa (%rdi), %ymm5
+; AVX2-FCP-NEXT:    vmovdqa 224(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovdqa (%rdi), %ymm10
 ; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %ymm6
-; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vpbroadcastq 80(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm7 = [0,7,6,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm5[0,1,2,3,4,5],ymm6[6],ymm5[7]
-; AVX2-FCP-NEXT:    vpermd %ymm8, %ymm7, %ymm8
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %ymm15
+; AVX2-FCP-NEXT:    vpbroadcastq 80(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm15[4,5,6,7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm2 = [0,7,6,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm10[0,1,2,3,4,5],ymm6[6],ymm10[7]
+; AVX2-FCP-NEXT:    vpermd %ymm8, %ymm2, %ymm8
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %xmm8
-; AVX2-FCP-NEXT:    vmovdqa 160(%rdi), %xmm12
-; AVX2-FCP-NEXT:    vmovdqa %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vpunpckhqdq {{.*#+}} xmm8 = xmm8[1],xmm12[1]
+; AVX2-FCP-NEXT:    vmovdqa 160(%rdi), %xmm11
+; AVX2-FCP-NEXT:    vmovdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FCP-NEXT:    vpunpckhqdq {{.*#+}} xmm8 = xmm8[1],xmm11[1]
 ; AVX2-FCP-NEXT:    vinserti128 $1, %xmm8, %ymm0, %ymm8
-; AVX2-FCP-NEXT:    vpbroadcastd 196(%rdi), %ymm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6],ymm12[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm8[5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5],ymm0[6],ymm4[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa %ymm0, %ymm13
-; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm7, %ymm2
-; AVX2-FCP-NEXT:    vpbroadcastq 304(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm7[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %xmm7
-; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vpunpckhqdq {{.*#+}} xmm7 = xmm7[1],xmm0[1]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm7, %ymm0, %ymm7
-; AVX2-FCP-NEXT:    vpbroadcastd 420(%rdi), %ymm8
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm7[5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 196(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6],ymm11[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm8[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm0[6],ymm3[7]
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FCP-NEXT:    vpbroadcastq 304(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %xmm2
+; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %xmm8
+; AVX2-FCP-NEXT:    vmovdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FCP-NEXT:    vpunpckhqdq {{.*#+}} xmm2 = xmm2[1],xmm8[1]
 ; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm7 = ymm9[12,13,14,15],ymm10[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm10[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm1[0,1],ymm11[2,3],ymm1[4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm12 = [5,6,5,6,5,6,5,6]
-; AVX2-FCP-NEXT:    vpermd %ymm7, %ymm12, %ymm8
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm7 = [1,0,7,7,5,4,7,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0],ymm5[1],ymm6[2,3,4],ymm5[5],ymm6[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm7, %ymm14
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm14[0,1,2],ymm8[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 420(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm9[12,13,14,15],ymm7[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm7[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm5[2,3],ymm4[4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm0[0],ymm3[1],ymm0[2,3,4],ymm3[5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0,1,2,3,4],ymm2[5,6],ymm8[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm12 = [1,0,7,6,5,6,5,6]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm12, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %ymm14
-; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %ymm15
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm8 = ymm14[12,13,14,15],ymm15[0,1,2,3,4,5,6,7,8,9,10,11],ymm14[28,29,30,31],ymm15[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %ymm8
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm8[2,3],ymm3[4,5],ymm8[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm12, %ymm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0],ymm4[1],ymm13[2,3,4],ymm4[5],ymm13[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm7, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2],ymm12[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm8
+; AVX2-FCP-NEXT:    vmovdqa 160(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm11 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm11[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm11[2,3],ymm15[4,5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm6[0],ymm10[1],ymm6[2,3,4],ymm10[5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3,4],ymm0[5,6],ymm8[7]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm12, %ymm0
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 80(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm4 = ymm1[8,9,10,11,12,13,14,15],ymm11[0,1,2,3,4,5,6,7],ymm1[24,25,26,27,28,29,30,31],ymm11[16,17,18,19,20,21,22,23]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 8(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm15[8,9,10,11,12,13,14,15],ymm11[0,1,2,3,4,5,6,7],ymm15[24,25,26,27,28,29,30,31],ymm11[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 8(%rdi), %xmm1
 ; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %xmm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0],xmm12[1],xmm4[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm4 = ymm10[0],ymm9[0],ymm10[2],ymm9[2]
-; AVX2-FCP-NEXT:    vpbroadcastd 204(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm12[1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm2[0],ymm3[0],ymm2[2],ymm3[2]
+; AVX2-FCP-NEXT:    vpbroadcastd 204(%rdi), %ymm14
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm14[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 304(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm4 = ymm3[8,9,10,11,12,13,14,15],ymm8[0,1,2,3,4,5,6,7],ymm3[24,25,26,27,28,29,30,31],ymm8[16,17,18,19,20,21,22,23]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 232(%rdi), %xmm4
-; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %xmm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0],xmm7[1],xmm4[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm4 = ymm15[0],ymm14[0],ymm15[2],ymm14[2]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm4[8,9,10,11,12,13,14,15],ymm5[0,1,2,3,4,5,6,7],ymm4[24,25,26,27,28,29,30,31],ymm5[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 232(%rdi), %xmm1
+; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %xmm14
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm14[1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm7[0],ymm9[0],ymm7[2],ymm9[2]
 ; AVX2-FCP-NEXT:    vpbroadcastd 428(%rdi), %ymm13
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm11[1],ymm1[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0],ymm11[1],ymm15[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm12[0,1,2],mem[3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,2],ymm9[1,3],ymm10[4,6],ymm9[5,7]
-; AVX2-FCP-NEXT:    vbroadcastss 208(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm2[0,2],ymm3[1,3],ymm2[4,6],ymm3[5,7]
+; AVX2-FCP-NEXT:    vbroadcastss 208(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm11[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0],ymm8[1],ymm3[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm7[0,1,2],mem[3]
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0],ymm5[1],ymm4[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm14[0,1,2],mem[3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm15[0,2],ymm14[1,3],ymm15[4,6],ymm14[5,7]
-; AVX2-FCP-NEXT:    vbroadcastss 432(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm7[0,2],ymm9[1,3],ymm7[4,6],ymm9[5,7]
+; AVX2-FCP-NEXT:    vbroadcastss 432(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm4[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpbroadcastd 100(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %xmm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm3 = [4,3,0,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm5, %ymm11
-; AVX2-FCP-NEXT:    vmovdqa %ymm6, %ymm12
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm3, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm0[2,3]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm4 = [4,3,0,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm6, %ymm15
+; AVX2-FCP-NEXT:    vpermd %ymm5, %ymm4, %ymm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1],xmm0[2,3]
 ; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [0,7,0,7,0,7,0,7]
-; AVX2-FCP-NEXT:    vpermd %ymm10, %ymm0, %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 212(%rdi), %ymm8
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm11
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 212(%rdi), %ymm12
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm5[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm3, %ymm3
-; AVX2-FCP-NEXT:    vpbroadcastd 324(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm14[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm5, %ymm4, %ymm4
+; AVX2-FCP-NEXT:    vpbroadcastd 324(%rdi), %xmm5
 ; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %xmm13
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm13[0,1,2],xmm4[3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1],xmm4[2,3]
-; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm0, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 436(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm3[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm9[0],ymm10[1],ymm9[2,3,4],ymm10[5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm4 = [1,0,3,3,1,0,7,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm4, %ymm3
-; AVX2-FCP-NEXT:    vpbroadcastd 216(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %xmm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm10[0,1,2],xmm1[3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,3,2]
-; AVX2-FCP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FCP-NEXT:    vextracti128 $1, %ymm7, %xmm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm7[0,1],xmm1[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm1[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %xmm9
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm9[0,1,2],xmm13[3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm13[0,1,2],xmm5[3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-FCP-NEXT:    vpermd %ymm7, %ymm0, %ymm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 436(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm11[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm4[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3,4],ymm2[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm3 = [1,0,3,3,1,0,7,7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm3, %ymm2
+; AVX2-FCP-NEXT:    vpbroadcastd 216(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm4[0,1,2],xmm1[3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,3,2]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,2,3,5,4,6,7]
-; AVX2-FCP-NEXT:    vextracti128 $1, %ymm2, %xmm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm14[0],ymm15[1],ymm14[2,3,4],ymm15[5],ymm14[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm4, %ymm2
-; AVX2-FCP-NEXT:    vpbroadcastd 440(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 136(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm1 = xmm1[0],mem[1],xmm1[2,3]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vpermd 192(%rdi), %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 80(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1,2],xmm10[3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm11[2,3,2,3,6,7,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0],ymm12[1],ymm4[2,3,4],ymm12[5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vextracti128 $1, %ymm4, %xmm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm4[0,1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 360(%rdi), %xmm3
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm8[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vextracti128 $1, %ymm5, %xmm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm5[0,1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %xmm8
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm8[0,1,2],xmm13[3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,3,2]
+; AVX2-FCP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vextracti128 $1, %ymm5, %xmm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm5[0,1],xmm2[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm9[0],ymm7[1],ymm9[2,3,4],ymm7[5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm5, %ymm3, %ymm3
+; AVX2-FCP-NEXT:    vpbroadcastd 440(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 136(%rdi), %xmm3
 ; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
 ; AVX2-FCP-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vpermd 416(%rdi), %ymm0, %ymm0
 ; AVX2-FCP-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 304(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1,2],xmm9[3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm5[2,3,2,3,6,7,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0],ymm6[1],ymm4[2,3,4],ymm6[5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vextracti128 $1, %ymm4, %xmm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm4[0,1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rcx)
-; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 32(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, (%r8)
-; AVX2-FCP-NEXT:    vmovdqa %ymm8, 32(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm3, (%r9)
+; AVX2-FCP-NEXT:    vpermd 192(%rdi), %ymm0, %ymm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 80(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm10[2,3,2,3,6,7,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm15[1],ymm5[2,3,4],ymm15[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vextracti128 $1, %ymm5, %xmm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm5[0,1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 360(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vpermd 416(%rdi), %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 304(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm8[3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm6[2,3,2,3,6,7,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm14[1],ymm5[2,3,4],ymm14[5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vextracti128 $1, %ymm5, %xmm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm5[0,1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm4, 32(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm4, 32(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm4, 32(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm4, (%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm4, 32(%r8)
+; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm4, (%r8)
+; AVX2-FCP-NEXT:    vmovdqa %ymm11, 32(%r9)
+; AVX2-FCP-NEXT:    vmovdqa %ymm12, (%r9)
 ; AVX2-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FCP-NEXT:    vmovdqa %ymm2, 32(%rax)
-; AVX2-FCP-NEXT:    vmovdqa %ymm7, (%rax)
+; AVX2-FCP-NEXT:    vmovdqa %ymm1, (%rax)
 ; AVX2-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FCP-NEXT:    vmovdqa %ymm0, 32(%rax)
-; AVX2-FCP-NEXT:    vmovdqa %ymm1, (%rax)
-; AVX2-FCP-NEXT:    addq $296, %rsp # imm = 0x128
+; AVX2-FCP-NEXT:    vmovdqa %ymm3, (%rax)
+; AVX2-FCP-NEXT:    addq $264, %rsp # imm = 0x108
 ; AVX2-FCP-NEXT:    vzeroupper
 ; AVX2-FCP-NEXT:    retq
 ;
@@ -5516,7 +5487,7 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-LABEL: load_i32_stride7_vf32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    subq $1224, %rsp # imm = 0x4C8
+; AVX2-NEXT:    subq $1192, %rsp # imm = 0x4A8
 ; AVX2-NEXT:    vmovdqa 320(%rdi), %ymm9
 ; AVX2-NEXT:    vmovdqa 256(%rdi), %ymm4
 ; AVX2-NEXT:    vmovdqa 224(%rdi), %ymm5
@@ -5525,13 +5496,12 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqa 448(%rdi), %ymm8
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm3
 ; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm6
-; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm11
 ; AVX2-NEXT:    vpbroadcastq 80(%rdi), %ymm0
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm11[4,5,6,7]
 ; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,7,6,0]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm6[6],ymm3[7]
-; AVX2-NEXT:    vmovdqa %ymm6, %ymm12
+; AVX2-NEXT:    vmovdqa %ymm6, %ymm13
 ; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa %ymm3, %ymm14
 ; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -5552,7 +5522,7 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vpbroadcastq 528(%rdi), %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm10[4,5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa %ymm10, %ymm12
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 576(%rdi), %xmm2
 ; AVX2-NEXT:    vmovdqa 608(%rdi), %xmm3
@@ -5569,7 +5539,7 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vpbroadcastq 304(%rdi), %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vmovdqa %ymm9, %ymm13
+; AVX2-NEXT:    vmovdqa %ymm9, %ymm10
 ; AVX2-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 352(%rdi), %xmm2
@@ -5589,9 +5559,10 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqa %ymm2, %ymm3
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 768(%rdi), %ymm11
+; AVX2-NEXT:    vmovdqa 768(%rdi), %ymm2
 ; AVX2-NEXT:    vpbroadcastq 752(%rdi), %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vmovdqa %ymm2, %ymm15
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 800(%rdi), %xmm1
 ; AVX2-NEXT:    vmovdqa 832(%rdi), %xmm2
@@ -5610,371 +5581,355 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm0[7]
 ; AVX2-NEXT:    vmovdqa 512(%rdi), %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm10[0,1],ymm2[2,3],ymm10[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm2[2,3],ymm12[4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqa %ymm2, %ymm9
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
+; AVX2-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0],ymm8[1],ymm7[2,3,4],ymm8[5],ymm7[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 384(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 352(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 288(%rdi), %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm0[5,6],ymm2[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm0 = [1,0,7,6,5,6,5,6]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1],ymm1[2,3],ymm13[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0],ymm5[1],ymm4[2,3,4],ymm5[5],ymm4[6,7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 832(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqa 800(%rdi), %ymm4
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vmovdqa %ymm4, %ymm13
-; AVX2-NEXT:    vmovdqa %ymm2, %ymm15
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqa 384(%rdi), %ymm7
+; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 352(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 736(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm11[0,1],ymm2[2,3],ymm11[4,5],ymm2[6,7]
-; AVX2-NEXT:    vmovdqa %ymm2, %ymm10
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm7[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm7[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-NEXT:    vmovdqa 288(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm6[1],ymm3[2,3,4],ymm6[5],ymm3[6,7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 160(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm3
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0,1],ymm2[2,3],ymm10[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0],ymm5[1],ymm4[2,3,4],ymm5[5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4],ymm2[5,6],ymm4[7]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqa 832(%rdi), %ymm8
+; AVX2-NEXT:    vmovdqa 800(%rdi), %ymm10
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm8[12,13,14,15],ymm10[0,1,2,3,4,5,6,7,8,9,10,11],ymm8[28,29,30,31],ymm10[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-NEXT:    vmovdqa 736(%rdi), %ymm4
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1],ymm4[2,3],ymm15[4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqa %ymm4, %ymm7
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa %ymm15, %ymm5
+; AVX2-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm3[0],ymm6[1],ymm3[2,3,4],ymm6[5],ymm3[6,7]
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4],ymm2[5,6],ymm4[7]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqa 160(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm4
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vmovdqa %ymm4, %ymm15
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa %ymm3, %ymm6
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa %ymm2, %ymm5
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm1[2,3],ymm4[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm12[0],ymm14[1],ymm12[2,3,4],ymm14[5],ymm12[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm11[0,1],ymm1[2,3],ymm11[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm13[0],ymm14[1],ymm13[2,3,4],ymm14[5],ymm13[6,7]
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm3[5,6],ymm4[7]
+; AVX2-NEXT:    vpermd %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 80(%rdi), %xmm0
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm4[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm4[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm11[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 8(%rdi), %xmm2
 ; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm3
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm2 = ymm6[0],ymm5[0],ymm6[2],ymm5[2]
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm2 = ymm15[0],ymm6[0],ymm15[2],ymm6[2]
+; AVX2-NEXT:    vmovdqa %ymm6, %ymm11
 ; AVX2-NEXT:    vpbroadcastd 204(%rdi), %ymm4
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm4[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 528(%rdi), %xmm0
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm8[8,9,10,11,12,13,14,15],ymm9[0,1,2,3,4,5,6,7],ymm8[24,25,26,27,28,29,30,31],ymm9[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm12[8,9,10,11,12,13,14,15],ymm9[0,1,2,3,4,5,6,7],ymm12[24,25,26,27,28,29,30,31],ymm9[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 456(%rdi), %xmm4
 ; AVX2-NEXT:    vmovdqa 480(%rdi), %xmm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0],xmm2[1],xmm4[2,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm4 = ymm7[0],ymm9[0],ymm7[2],ymm9[2]
-; AVX2-NEXT:    vpbroadcastd 652(%rdi), %ymm12
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm4 = ymm12[0],ymm9[0],ymm12[2],ymm9[2]
+; AVX2-NEXT:    vpbroadcastd 652(%rdi), %ymm15
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm15[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm4[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 752(%rdi), %xmm0
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm4 = ymm11[8,9,10,11,12,13,14,15],ymm10[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm10[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm4 = ymm5[8,9,10,11,12,13,14,15],ymm7[0,1,2,3,4,5,6,7],ymm5[24,25,26,27,28,29,30,31],ymm7[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vpbroadcastd 680(%rdi), %xmm12
-; AVX2-NEXT:    vmovdqa 704(%rdi), %xmm4
-; AVX2-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm12 = xmm12[0],xmm4[1],xmm12[2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm12 = ymm13[0],ymm15[0],ymm13[2],ymm15[2]
-; AVX2-NEXT:    vmovdqa %ymm13, %ymm10
-; AVX2-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpbroadcastd 876(%rdi), %ymm15
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5,6],ymm15[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm12[5,6,7]
+; AVX2-NEXT:    vpbroadcastd 680(%rdi), %xmm15
+; AVX2-NEXT:    vmovdqa 704(%rdi), %xmm7
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm7[1],xmm15[2,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm10[0],ymm8[0],ymm10[2],ymm8[2]
+; AVX2-NEXT:    vpbroadcastd 876(%rdi), %ymm13
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm13[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 304(%rdi), %xmm0
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm12 = ymm5[8,9,10,11,12,13,14,15],ymm6[0,1,2,3,4,5,6,7],ymm5[24,25,26,27,28,29,30,31],ymm6[16,17,18,19,20,21,22,23]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2],ymm0[3],ymm12[4,5,6,7]
+; AVX2-NEXT:    vmovdqu (%rsp), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm13 = ymm4[8,9,10,11,12,13,14,15],ymm5[0,1,2,3,4,5,6,7],ymm4[24,25,26,27,28,29,30,31],ymm5[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm0[3],ymm13[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 232(%rdi), %xmm15
 ; AVX2-NEXT:    vmovdqa 256(%rdi), %xmm0
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm0[1],xmm15[2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm15[0,1],ymm12[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm13[0],ymm4[0],ymm13[2],ymm4[2]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2,3,4,5,6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm6[0],ymm8[0],ymm6[2],ymm8[2]
 ; AVX2-NEXT:    vpbroadcastd 428(%rdi), %ymm14
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm14 = ymm15[0,1,2,3,4,5,6],ymm14[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4],ymm14[5,6,7]
+; AVX2-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd $253, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm1 = mem[0],ymm1[1],mem[2,3,4,5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1,2],mem[3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[3,2,2,3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,1,1,0,7,5,5,4]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm14[0,2],ymm15[1,3],ymm14[4,6],ymm15[5,7]
+; AVX2-NEXT:    vmovdqa %ymm11, %ymm13
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm10[0,2],ymm11[1,3],ymm10[4,6],ymm11[5,7]
 ; AVX2-NEXT:    vbroadcastss 208(%rdi), %ymm11
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm11[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = ymm8[0],mem[1],ymm8[2,3,4,5,6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $253, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = mem[0],ymm1[1],mem[2,3,4,5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0,1,2],mem[3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[3,2,2,3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,1,1,0,7,5,5,4]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovdqa %ymm9, %ymm12
-; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm7[0,2],ymm9[1,3],ymm7[4,6],ymm9[5,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm12[0,2],ymm9[1,3],ymm12[4,6],ymm9[5,7]
 ; AVX2-NEXT:    vbroadcastss 656(%rdi), %ymm3
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm5[0],ymm6[1],ymm5[2,3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0],ymm5[1],ymm4[2,3,4,5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],mem[3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,2,2,3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,1,1,0,7,5,5,4]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[0,2],ymm4[1,3],ymm13[4,6],ymm4[5,7]
-; AVX2-NEXT:    vmovdqa %ymm4, %ymm8
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,2],ymm8[1,3],ymm6[4,6],ymm8[5,7]
+; AVX2-NEXT:    vmovaps %ymm6, %ymm15
 ; AVX2-NEXT:    vbroadcastss 432(%rdi), %ymm2
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vblendps $2, (%rsp), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1,2],mem[3]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,2,2,3]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,2],ymm9[1,3],ymm10[4,6],ymm9[5,7]
+; AVX2-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $253, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0],ymm0[1],mem[2,3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm7[0,1,2],mem[3]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[0,2],ymm6[1,3],ymm5[4,6],ymm6[5,7]
 ; AVX2-NEXT:    vbroadcastss 880(%rdi), %ymm2
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpmovsxbd {{.*#+}} xmm1 = [4,3,0,0]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovsd {{.*#+}} xmm1 = [4,3,0,0]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-NEXT:    vpbroadcastd 548(%rdi), %xmm2
-; AVX2-NEXT:    vmovdqa 512(%rdi), %xmm6
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm6[0,1,2],xmm2[3]
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3]
-; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm11 = [0,7,0,7,0,7,0,7]
-; AVX2-NEXT:    vpermd %ymm7, %ymm11, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm12[6,7]
-; AVX2-NEXT:    vpbroadcastd 660(%rdi), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vpbroadcastd 100(%rdi), %xmm2
-; AVX2-NEXT:    vmovdqa 64(%rdi), %xmm0
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm0[0,1,2],xmm2[3]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vbroadcastss 548(%rdi), %xmm2
+; AVX2-NEXT:    vmovaps 512(%rdi), %xmm7
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm7[0,1,2],xmm2[3]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3]
+; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm11 = [0,7,0,7,0,7,0,7]
+; AVX2-NEXT:    vpermps %ymm12, %ymm11, %ymm2
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-NEXT:    vmovaps %ymm9, %ymm12
+; AVX2-NEXT:    vbroadcastss 660(%rdi), %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vbroadcastss 100(%rdi), %xmm2
+; AVX2-NEXT:    vmovaps 64(%rdi), %xmm0
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm0[0,1,2],xmm2[3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm3 = mem[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
-; AVX2-NEXT:    vmovaps %ymm14, %ymm10
-; AVX2-NEXT:    vpermd %ymm14, %ymm11, %ymm3
-; AVX2-NEXT:    vmovaps %ymm15, %ymm7
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vpbroadcastd 212(%rdi), %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    vpermps %ymm3, %ymm1, %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; AVX2-NEXT:    vpermps %ymm10, %ymm11, %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-NEXT:    vbroadcastss 212(%rdi), %ymm4
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm2, %ymm1, %ymm3
-; AVX2-NEXT:    vpbroadcastd 324(%rdi), %xmm4
-; AVX2-NEXT:    vmovdqa 288(%rdi), %xmm2
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm2[0,1,2],xmm4[3]
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1],xmm4[2,3]
-; AVX2-NEXT:    vpermd %ymm13, %ymm11, %ymm4
-; AVX2-NEXT:    vmovdqa %ymm13, %ymm14
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-NEXT:    vmovdqa %ymm8, %ymm13
-; AVX2-NEXT:    vpbroadcastd 436(%rdi), %ymm8
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    vpermps %ymm2, %ymm1, %ymm3
+; AVX2-NEXT:    vbroadcastss 324(%rdi), %xmm4
+; AVX2-NEXT:    vmovaps 288(%rdi), %xmm2
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm2[0,1,2],xmm4[3]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm3[0,1],xmm4[2,3]
+; AVX2-NEXT:    vpermps %ymm15, %ymm11, %ymm4
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-NEXT:    vmovaps %ymm8, %ymm9
+; AVX2-NEXT:    vbroadcastss 436(%rdi), %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm3 = mem[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vpermd %ymm3, %ymm1, %ymm1
-; AVX2-NEXT:    vpbroadcastd 772(%rdi), %xmm4
-; AVX2-NEXT:    vmovdqa 736(%rdi), %xmm3
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm3[0,1,2],xmm4[3]
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vpermd %ymm15, %ymm11, %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-NEXT:    vpbroadcastd 884(%rdi), %ymm8
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0],ymm10[1],ymm7[2,3,4],ymm10[5],ymm7[6,7]
+; AVX2-NEXT:    vpermps %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vbroadcastss 772(%rdi), %xmm4
+; AVX2-NEXT:    vmovaps 736(%rdi), %xmm3
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm3[0,1,2],xmm4[3]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3]
+; AVX2-NEXT:    vpermps %ymm5, %ymm11, %ymm4
+; AVX2-NEXT:    vmovaps %ymm5, %ymm14
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-NEXT:    vbroadcastss 884(%rdi), %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm13[0],ymm10[1],ymm13[2,3,4],ymm10[5],ymm13[6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,0,3,3,5,4,7,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-NEXT:    vbroadcastss 216(%rdi), %ymm4
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-NEXT:    vmovdqa 96(%rdi), %xmm10
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm10[0,1,2],xmm0[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,3,2]
-; AVX2-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    vmovaps 96(%rdi), %xmm10
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm10[0,1,2],xmm0[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,1,3,2]
+; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm4 = mem[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextracti128 $1, %ymm4, %xmm4
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm4[0,1],xmm0[2,3]
+; AVX2-NEXT:    vextractf128 $1, %ymm4, %xmm4
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm4[0,1],xmm0[2,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-NEXT:    vmovdqa 544(%rdi), %xmm4
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm4[0,1,2],xmm6[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,3,2]
-; AVX2-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    vmovaps 544(%rdi), %xmm4
+; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm4[0,1,2],xmm7[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextracti128 $1, %ymm5, %xmm5
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm5[0,1],xmm1[2,3]
-; AVX2-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm5[0,1],xmm1[2,3]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm5 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm5 = ymm12[0],mem[1],ymm12[2,3,4],mem[5],ymm12[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-NEXT:    vpbroadcastd 664(%rdi), %ymm7
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-NEXT:    vmovdqa 320(%rdi), %xmm12
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm12[0,1,2],xmm2[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,3,2]
-; AVX2-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[1,0,3,3,5,4,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
+; AVX2-NEXT:    vbroadcastss 664(%rdi), %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-NEXT:    vmovaps 320(%rdi), %xmm12
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm12[0,1,2],xmm2[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,1,3,2]
+; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextracti128 $1, %ymm5, %xmm5
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm5[0,1],xmm2[2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0],ymm14[1],ymm13[2,3,4],ymm14[5],ymm13[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-NEXT:    vpbroadcastd 440(%rdi), %ymm7
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm2[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-NEXT:    vmovdqa 768(%rdi), %xmm2
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,1,3,2]
-; AVX2-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm5[0,1],xmm2[2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm9[0],ymm15[1],ymm9[2,3,4],ymm15[5],ymm9[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[1,0,3,3,5,4,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
+; AVX2-NEXT:    vbroadcastss 440(%rdi), %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm2[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-NEXT:    vmovaps 768(%rdi), %xmm2
+; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,1,3,2]
+; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextracti128 $1, %ymm7, %xmm7
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm3 = xmm7[0,1],xmm3[2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm9[0],ymm15[1],ymm9[2,3,4],ymm15[5],ymm9[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-NEXT:    vpbroadcastd 888(%rdi), %ymm8
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm3[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm7, %xmm7
+; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm7[0,1],xmm3[2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm6[0],ymm14[1],ymm6[2,3,4],ymm14[5],ymm6[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
+; AVX2-NEXT:    vbroadcastss 888(%rdi), %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm3[0,1,2,3],ymm7[4,5,6,7]
 ; AVX2-NEXT:    vbroadcastss 584(%rdi), %xmm3
 ; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-NEXT:    vpermd 640(%rdi), %ymm11, %ymm8
+; AVX2-NEXT:    vpermps 640(%rdi), %ymm11, %ymm8
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-NEXT:    vpbroadcastd 528(%rdi), %ymm8
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm8[0,1,2],xmm4[3]
-; AVX2-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
+; AVX2-NEXT:    vbroadcastss 528(%rdi), %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm8[0,1,2],xmm4[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm8 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm8 = ymm8[0],mem[1],ymm8[2,3,4],mem[5],ymm8[6,7]
-; AVX2-NEXT:    vextracti128 $1, %ymm8, %xmm8
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm8[0,1],xmm4[2,3]
+; AVX2-NEXT:    vextractf128 $1, %ymm8, %xmm8
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm8[0,1],xmm4[2,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-NEXT:    vbroadcastss 808(%rdi), %xmm3
 ; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-NEXT:    vpermd 864(%rdi), %ymm11, %ymm4
+; AVX2-NEXT:    vpermps 864(%rdi), %ymm11, %ymm4
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-NEXT:    vpbroadcastd 752(%rdi), %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm4[0,1,2],xmm2[3]
-; AVX2-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    vbroadcastss 752(%rdi), %ymm4
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm4[0,1,2],xmm2[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm4 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm4 = ymm4[0],mem[1],ymm4[2,3,4],mem[5],ymm4[6,7]
-; AVX2-NEXT:    vextracti128 $1, %ymm4, %xmm4
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3]
+; AVX2-NEXT:    vextractf128 $1, %ymm4, %xmm4
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-NEXT:    vbroadcastss 136(%rdi), %xmm3
 ; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-NEXT:    vpermd 192(%rdi), %ymm11, %ymm4
+; AVX2-NEXT:    vpermps 192(%rdi), %ymm11, %ymm4
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-NEXT:    vpbroadcastd 80(%rdi), %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm10[3]
-; AVX2-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
+; AVX2-NEXT:    vbroadcastss 80(%rdi), %ymm4
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm4[0,1,2],xmm10[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm6 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm6 = ymm6[0],mem[1],ymm6[2,3,4],mem[5],ymm6[6,7]
-; AVX2-NEXT:    vextracti128 $1, %ymm6, %xmm6
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
+; AVX2-NEXT:    vextractf128 $1, %ymm6, %xmm6
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-NEXT:    vbroadcastss 360(%rdi), %xmm4
 ; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
-; AVX2-NEXT:    vpermd 416(%rdi), %ymm11, %ymm6
+; AVX2-NEXT:    vpermps 416(%rdi), %ymm11, %ymm6
 ; AVX2-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm4
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-NEXT:    vpbroadcastd 304(%rdi), %ymm6
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm6 = xmm6[0,1,2],xmm12[3]
-; AVX2-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-NEXT:    vextracti128 $1, %ymm9, %xmm9
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm6 = xmm9[0,1],xmm6[2,3]
+; AVX2-NEXT:    vbroadcastss 304(%rdi), %ymm6
+; AVX2-NEXT:    vblendps {{.*#+}} xmm6 = xmm6[0,1,2],xmm12[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm10 = mem[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm10 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm10 = ymm10[0],mem[1],ymm10[2,3,4],mem[5],ymm10[6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm10, %xmm10
+; AVX2-NEXT:    vblendps {{.*#+}} xmm6 = xmm10[0,1],xmm6[2,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm4[4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm6, 96(%rsi)
@@ -6002,7 +5957,7 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovaps %ymm6, (%rcx)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm6, 96(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vmovups (%rsp), %ymm6 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm6, 32(%r8)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm6, 64(%r8)
@@ -6014,25 +5969,25 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovaps %ymm6, 32(%r9)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm6, (%r9)
-; AVX2-NEXT:    vmovups (%rsp), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm6, 64(%r9)
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-NEXT:    vmovdqa %ymm7, 96(%rax)
-; AVX2-NEXT:    vmovdqa %ymm5, 32(%rax)
-; AVX2-NEXT:    vmovdqa %ymm1, 64(%rax)
+; AVX2-NEXT:    vmovaps %ymm7, 96(%rax)
+; AVX2-NEXT:    vmovaps %ymm5, 32(%rax)
+; AVX2-NEXT:    vmovaps %ymm1, 64(%rax)
 ; AVX2-NEXT:    vmovaps %ymm0, (%rax)
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-NEXT:    vmovaps %ymm4, 32(%rax)
 ; AVX2-NEXT:    vmovaps %ymm3, (%rax)
 ; AVX2-NEXT:    vmovaps %ymm2, 96(%rax)
 ; AVX2-NEXT:    vmovaps %ymm8, 64(%rax)
-; AVX2-NEXT:    addq $1224, %rsp # imm = 0x4C8
+; AVX2-NEXT:    addq $1192, %rsp # imm = 0x4A8
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: load_i32_stride7_vf32:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    subq $1224, %rsp # imm = 0x4C8
+; AVX2-FP-NEXT:    subq $1192, %rsp # imm = 0x4A8
 ; AVX2-FP-NEXT:    vmovdqa 320(%rdi), %ymm9
 ; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vmovdqa 224(%rdi), %ymm5
@@ -6041,13 +5996,12 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqa 448(%rdi), %ymm8
 ; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm3
 ; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm6
-; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm11
 ; AVX2-FP-NEXT:    vpbroadcastq 80(%rdi), %ymm0
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm11[4,5,6,7]
 ; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,7,6,0]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm6[6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqa %ymm6, %ymm12
+; AVX2-FP-NEXT:    vmovdqa %ymm6, %ymm13
 ; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm14
 ; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -6068,7 +6022,7 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vpbroadcastq 528(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm10[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa %ymm10, %ymm12
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 576(%rdi), %xmm2
 ; AVX2-FP-NEXT:    vmovdqa 608(%rdi), %xmm3
@@ -6085,7 +6039,7 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vpbroadcastq 304(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm9, %ymm13
+; AVX2-FP-NEXT:    vmovdqa %ymm9, %ymm10
 ; AVX2-FP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 352(%rdi), %xmm2
@@ -6105,9 +6059,10 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm3
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 768(%rdi), %ymm11
+; AVX2-FP-NEXT:    vmovdqa 768(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vpbroadcastq 752(%rdi), %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm15
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 800(%rdi), %xmm1
 ; AVX2-FP-NEXT:    vmovdqa 832(%rdi), %xmm2
@@ -6126,371 +6081,355 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm0[7]
 ; AVX2-FP-NEXT:    vmovdqa 512(%rdi), %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm10[0,1],ymm2[2,3],ymm10[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm2[2,3],ymm12[4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm9
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
+; AVX2-FP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0],ymm8[1],ymm7[2,3,4],ymm8[5],ymm7[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 384(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 352(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm0[5,6],ymm2[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm0 = [1,0,7,6,5,6,5,6]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1],ymm1[2,3],ymm13[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0],ymm5[1],ymm4[2,3,4],ymm5[5],ymm4[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 832(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqa 800(%rdi), %ymm4
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vmovdqa %ymm4, %ymm13
-; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm15
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqa 384(%rdi), %ymm7
+; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 352(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 736(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm11[0,1],ymm2[2,3],ymm11[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm10
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm7[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm7[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm6[1],ymm3[2,3,4],ymm6[5],ymm3[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 160(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0,1],ymm2[2,3],ymm10[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0],ymm5[1],ymm4[2,3,4],ymm5[5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4],ymm2[5,6],ymm4[7]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqa 832(%rdi), %ymm8
+; AVX2-FP-NEXT:    vmovdqa 800(%rdi), %ymm10
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm8[12,13,14,15],ymm10[0,1,2,3,4,5,6,7,8,9,10,11],ymm8[28,29,30,31],ymm10[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovdqa 736(%rdi), %ymm4
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1],ymm4[2,3],ymm15[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm4, %ymm7
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa %ymm15, %ymm5
+; AVX2-FP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm3[0],ymm6[1],ymm3[2,3,4],ymm6[5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4],ymm2[5,6],ymm4[7]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqa 160(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm4
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vmovdqa %ymm4, %ymm15
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm6
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm5
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm1[2,3],ymm4[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm12[0],ymm14[1],ymm12[2,3,4],ymm14[5],ymm12[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm11[0,1],ymm1[2,3],ymm11[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm13[0],ymm14[1],ymm13[2,3,4],ymm14[5],ymm13[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm3[5,6],ymm4[7]
+; AVX2-FP-NEXT:    vpermd %ymm3, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 80(%rdi), %xmm0
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm4[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm4[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm11[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 8(%rdi), %xmm2
 ; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %xmm3
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm2 = ymm6[0],ymm5[0],ymm6[2],ymm5[2]
+; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm2 = ymm15[0],ymm6[0],ymm15[2],ymm6[2]
+; AVX2-FP-NEXT:    vmovdqa %ymm6, %ymm11
 ; AVX2-FP-NEXT:    vpbroadcastd 204(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm4[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 528(%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm8[8,9,10,11,12,13,14,15],ymm9[0,1,2,3,4,5,6,7],ymm8[24,25,26,27,28,29,30,31],ymm9[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm12[8,9,10,11,12,13,14,15],ymm9[0,1,2,3,4,5,6,7],ymm12[24,25,26,27,28,29,30,31],ymm9[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 456(%rdi), %xmm4
 ; AVX2-FP-NEXT:    vmovdqa 480(%rdi), %xmm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0],xmm2[1],xmm4[2,3]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm4 = ymm7[0],ymm9[0],ymm7[2],ymm9[2]
-; AVX2-FP-NEXT:    vpbroadcastd 652(%rdi), %ymm12
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm4 = ymm12[0],ymm9[0],ymm12[2],ymm9[2]
+; AVX2-FP-NEXT:    vpbroadcastd 652(%rdi), %ymm15
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm15[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm4[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 752(%rdi), %xmm0
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm4 = ymm11[8,9,10,11,12,13,14,15],ymm10[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm10[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm4 = ymm5[8,9,10,11,12,13,14,15],ymm7[0,1,2,3,4,5,6,7],ymm5[24,25,26,27,28,29,30,31],ymm7[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 680(%rdi), %xmm12
-; AVX2-FP-NEXT:    vmovdqa 704(%rdi), %xmm4
-; AVX2-FP-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm12 = xmm12[0],xmm4[1],xmm12[2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm12 = ymm13[0],ymm15[0],ymm13[2],ymm15[2]
-; AVX2-FP-NEXT:    vmovdqa %ymm13, %ymm10
-; AVX2-FP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpbroadcastd 876(%rdi), %ymm15
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5,6],ymm15[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm12[5,6,7]
+; AVX2-FP-NEXT:    vpbroadcastd 680(%rdi), %xmm15
+; AVX2-FP-NEXT:    vmovdqa 704(%rdi), %xmm7
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm7[1],xmm15[2,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm10[0],ymm8[0],ymm10[2],ymm8[2]
+; AVX2-FP-NEXT:    vpbroadcastd 876(%rdi), %ymm13
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm13[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 304(%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm12 = ymm5[8,9,10,11,12,13,14,15],ymm6[0,1,2,3,4,5,6,7],ymm5[24,25,26,27,28,29,30,31],ymm6[16,17,18,19,20,21,22,23]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2],ymm0[3],ymm12[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu (%rsp), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm13 = ymm4[8,9,10,11,12,13,14,15],ymm5[0,1,2,3,4,5,6,7],ymm4[24,25,26,27,28,29,30,31],ymm5[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm0[3],ymm13[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 232(%rdi), %xmm15
 ; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %xmm0
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm0[1],xmm15[2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm15[0,1],ymm12[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm13[0],ymm4[0],ymm13[2],ymm4[2]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm6[0],ymm8[0],ymm6[2],ymm8[2]
 ; AVX2-FP-NEXT:    vpbroadcastd 428(%rdi), %ymm14
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm15[0,1,2,3,4,5,6],ymm14[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3,4],ymm14[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4],ymm14[5,6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd $253, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm1 = mem[0],ymm1[1],mem[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1,2],mem[3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[3,2,2,3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,1,1,0,7,5,5,4]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm14[0,2],ymm15[1,3],ymm14[4,6],ymm15[5,7]
+; AVX2-FP-NEXT:    vmovdqa %ymm11, %ymm13
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm10[0,2],ymm11[1,3],ymm10[4,6],ymm11[5,7]
 ; AVX2-FP-NEXT:    vbroadcastss 208(%rdi), %ymm11
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm11[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = ymm8[0],mem[1],ymm8[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $253, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = mem[0],ymm1[1],mem[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0,1,2],mem[3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[3,2,2,3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,1,1,0,7,5,5,4]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm9, %ymm12
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm7[0,2],ymm9[1,3],ymm7[4,6],ymm9[5,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm12[0,2],ymm9[1,3],ymm12[4,6],ymm9[5,7]
 ; AVX2-FP-NEXT:    vbroadcastss 656(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm5[0],ymm6[1],ymm5[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0],ymm5[1],ymm4[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],mem[3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,2,2,3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,1,1,0,7,5,5,4]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm13[0,2],ymm4[1,3],ymm13[4,6],ymm4[5,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm4, %ymm8
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,2],ymm8[1,3],ymm6[4,6],ymm8[5,7]
+; AVX2-FP-NEXT:    vmovaps %ymm6, %ymm15
 ; AVX2-FP-NEXT:    vbroadcastss 432(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $253, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0],ymm0[1],mem[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm7[0,1,2],mem[3]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[0,2],ymm6[1,3],ymm5[4,6],ymm6[5,7]
+; AVX2-FP-NEXT:    vbroadcastss 880(%rdi), %ymm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovsd {{.*#+}} xmm1 = [4,3,0,0]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $2, (%rsp), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1,2],mem[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,2,2,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,2],ymm9[1,3],ymm10[4,6],ymm9[5,7]
-; AVX2-FP-NEXT:    vbroadcastss 880(%rdi), %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} xmm1 = [4,3,0,0]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FP-NEXT:    vpbroadcastd 548(%rdi), %xmm2
-; AVX2-FP-NEXT:    vmovdqa 512(%rdi), %xmm6
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm6[0,1,2],xmm2[3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3]
-; AVX2-FP-NEXT:    vpbroadcastq {{.*#+}} ymm11 = [0,7,0,7,0,7,0,7]
-; AVX2-FP-NEXT:    vpermd %ymm7, %ymm11, %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm12[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 660(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vpbroadcastd 100(%rdi), %xmm2
-; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %xmm0
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm0[0,1,2],xmm2[3]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FP-NEXT:    vbroadcastss 548(%rdi), %xmm2
+; AVX2-FP-NEXT:    vmovaps 512(%rdi), %xmm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm7[0,1,2],xmm2[3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3]
+; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm11 = [0,7,0,7,0,7,0,7]
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm11, %ymm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-FP-NEXT:    vmovaps %ymm9, %ymm12
+; AVX2-FP-NEXT:    vbroadcastss 660(%rdi), %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vbroadcastss 100(%rdi), %xmm2
+; AVX2-FP-NEXT:    vmovaps 64(%rdi), %xmm0
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm0[0,1,2],xmm2[3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm3 = mem[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
-; AVX2-FP-NEXT:    vmovaps %ymm14, %ymm10
-; AVX2-FP-NEXT:    vpermd %ymm14, %ymm11, %ymm3
-; AVX2-FP-NEXT:    vmovaps %ymm15, %ymm7
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 212(%rdi), %ymm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm1, %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; AVX2-FP-NEXT:    vpermps %ymm10, %ymm11, %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 212(%rdi), %ymm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm2, %ymm1, %ymm3
-; AVX2-FP-NEXT:    vpbroadcastd 324(%rdi), %xmm4
-; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %xmm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm2[0,1,2],xmm4[3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1],xmm4[2,3]
-; AVX2-FP-NEXT:    vpermd %ymm13, %ymm11, %ymm4
-; AVX2-FP-NEXT:    vmovdqa %ymm13, %ymm14
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vmovdqa %ymm8, %ymm13
-; AVX2-FP-NEXT:    vpbroadcastd 436(%rdi), %ymm8
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vpermps %ymm2, %ymm1, %ymm3
+; AVX2-FP-NEXT:    vbroadcastss 324(%rdi), %xmm4
+; AVX2-FP-NEXT:    vmovaps 288(%rdi), %xmm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm2[0,1,2],xmm4[3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm3[0,1],xmm4[2,3]
+; AVX2-FP-NEXT:    vpermps %ymm15, %ymm11, %ymm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    vmovaps %ymm8, %ymm9
+; AVX2-FP-NEXT:    vbroadcastss 436(%rdi), %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm3 = mem[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FP-NEXT:    vpermd %ymm3, %ymm1, %ymm1
-; AVX2-FP-NEXT:    vpbroadcastd 772(%rdi), %xmm4
-; AVX2-FP-NEXT:    vmovdqa 736(%rdi), %xmm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm3[0,1,2],xmm4[3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vpermd %ymm15, %ymm11, %ymm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 884(%rdi), %ymm8
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0],ymm10[1],ymm7[2,3,4],ymm10[5],ymm7[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm1, %ymm1
+; AVX2-FP-NEXT:    vbroadcastss 772(%rdi), %xmm4
+; AVX2-FP-NEXT:    vmovaps 736(%rdi), %xmm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm3[0,1,2],xmm4[3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3]
+; AVX2-FP-NEXT:    vpermps %ymm5, %ymm11, %ymm4
+; AVX2-FP-NEXT:    vmovaps %ymm5, %ymm14
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 884(%rdi), %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm13[0],ymm10[1],ymm13[2,3,4],ymm10[5],ymm13[6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,0,3,3,5,4,7,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
 ; AVX2-FP-NEXT:    vbroadcastss 216(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %xmm10
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm10[0,1,2],xmm0[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,3,2]
-; AVX2-FP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vmovaps 96(%rdi), %xmm10
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm10[0,1,2],xmm0[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,1,3,2]
+; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm4 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextracti128 $1, %ymm4, %xmm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm4[0,1],xmm0[2,3]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm4, %xmm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm4[0,1],xmm0[2,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 544(%rdi), %xmm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm4[0,1,2],xmm6[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,3,2]
-; AVX2-FP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vmovaps 544(%rdi), %xmm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm4[0,1,2],xmm7[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextracti128 $1, %ymm5, %xmm5
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm5[0,1],xmm1[2,3]
-; AVX2-FP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm5[0,1],xmm1[2,3]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm5 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm5 = ymm12[0],mem[1],ymm12[2,3,4],mem[5],ymm12[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-FP-NEXT:    vpbroadcastd 664(%rdi), %ymm7
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 320(%rdi), %xmm12
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm12[0,1,2],xmm2[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,3,2]
-; AVX2-FP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[1,0,3,3,5,4,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
+; AVX2-FP-NEXT:    vbroadcastss 664(%rdi), %ymm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 320(%rdi), %xmm12
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm12[0,1,2],xmm2[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,1,3,2]
+; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextracti128 $1, %ymm5, %xmm5
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm5[0,1],xmm2[2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0],ymm14[1],ymm13[2,3,4],ymm14[5],ymm13[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,3]
-; AVX2-FP-NEXT:    vpbroadcastd 440(%rdi), %ymm7
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm2[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa 768(%rdi), %xmm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,1,3,2]
-; AVX2-FP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm5[0,1],xmm2[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm9[0],ymm15[1],ymm9[2,3,4],ymm15[5],ymm9[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[1,0,3,3,5,4,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
+; AVX2-FP-NEXT:    vbroadcastss 440(%rdi), %ymm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm2[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 768(%rdi), %xmm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,1,3,2]
+; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextracti128 $1, %ymm7, %xmm7
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm7[0,1],xmm3[2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm9[0],ymm15[1],ymm9[2,3,4],ymm15[5],ymm9[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-FP-NEXT:    vpbroadcastd 888(%rdi), %ymm8
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm3[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm7, %xmm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm7[0,1],xmm3[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm6[0],ymm14[1],ymm6[2,3,4],ymm14[5],ymm6[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
+; AVX2-FP-NEXT:    vbroadcastss 888(%rdi), %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm3[0,1,2,3],ymm7[4,5,6,7]
 ; AVX2-FP-NEXT:    vbroadcastss 584(%rdi), %xmm3
 ; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
 ; AVX2-FP-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
 ; AVX2-FP-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FP-NEXT:    vpermd 640(%rdi), %ymm11, %ymm8
+; AVX2-FP-NEXT:    vpermps 640(%rdi), %ymm11, %ymm8
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 528(%rdi), %ymm8
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm8[0,1,2],xmm4[3]
-; AVX2-FP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vbroadcastss 528(%rdi), %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm8[0,1,2],xmm4[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm8 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm8 = ymm8[0],mem[1],ymm8[2,3,4],mem[5],ymm8[6,7]
-; AVX2-FP-NEXT:    vextracti128 $1, %ymm8, %xmm8
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm8[0,1],xmm4[2,3]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm8, %xmm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm8[0,1],xmm4[2,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-FP-NEXT:    vbroadcastss 808(%rdi), %xmm3
 ; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
 ; AVX2-FP-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
 ; AVX2-FP-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FP-NEXT:    vpermd 864(%rdi), %ymm11, %ymm4
+; AVX2-FP-NEXT:    vpermps 864(%rdi), %ymm11, %ymm4
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 752(%rdi), %ymm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm4[0,1,2],xmm2[3]
-; AVX2-FP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vbroadcastss 752(%rdi), %ymm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm4[0,1,2],xmm2[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm4 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm4 = ymm4[0],mem[1],ymm4[2,3,4],mem[5],ymm4[6,7]
-; AVX2-FP-NEXT:    vextracti128 $1, %ymm4, %xmm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm4, %xmm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-FP-NEXT:    vbroadcastss 136(%rdi), %xmm3
 ; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
 ; AVX2-FP-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
 ; AVX2-FP-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FP-NEXT:    vpermd 192(%rdi), %ymm11, %ymm4
+; AVX2-FP-NEXT:    vpermps 192(%rdi), %ymm11, %ymm4
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 80(%rdi), %ymm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm10[3]
-; AVX2-FP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vbroadcastss 80(%rdi), %ymm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm4[0,1,2],xmm10[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm6 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm6 = ymm6[0],mem[1],ymm6[2,3,4],mem[5],ymm6[6,7]
-; AVX2-FP-NEXT:    vextracti128 $1, %ymm6, %xmm6
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm6, %xmm6
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-FP-NEXT:    vbroadcastss 360(%rdi), %xmm4
 ; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
 ; AVX2-FP-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
-; AVX2-FP-NEXT:    vpermd 416(%rdi), %ymm11, %ymm6
+; AVX2-FP-NEXT:    vpermps 416(%rdi), %ymm11, %ymm6
 ; AVX2-FP-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm4
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpbroadcastd 304(%rdi), %ymm6
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm6 = xmm6[0,1,2],xmm12[3]
-; AVX2-FP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-FP-NEXT:    vextracti128 $1, %ymm9, %xmm9
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm6 = xmm9[0,1],xmm6[2,3]
+; AVX2-FP-NEXT:    vbroadcastss 304(%rdi), %ymm6
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm6 = xmm6[0,1,2],xmm12[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm10 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm10 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm10 = ymm10[0],mem[1],ymm10[2,3,4],mem[5],ymm10[6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm10, %xmm10
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm6 = xmm10[0,1],xmm6[2,3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm4[4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%rsi)
@@ -6518,7 +6457,7 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovaps %ymm6, (%rcx)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm6 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%r8)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%r8)
@@ -6530,44 +6469,41 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%r9)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm6, (%r9)
-; AVX2-FP-NEXT:    vmovups (%rsp), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%r9)
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FP-NEXT:    vmovdqa %ymm7, 96(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm5, 32(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm1, 64(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm7, 96(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm5, 32(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm1, 64(%rax)
 ; AVX2-FP-NEXT:    vmovaps %ymm0, (%rax)
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FP-NEXT:    vmovaps %ymm4, 32(%rax)
 ; AVX2-FP-NEXT:    vmovaps %ymm3, (%rax)
 ; AVX2-FP-NEXT:    vmovaps %ymm2, 96(%rax)
 ; AVX2-FP-NEXT:    vmovaps %ymm8, 64(%rax)
-; AVX2-FP-NEXT:    addq $1224, %rsp # imm = 0x4C8
+; AVX2-FP-NEXT:    addq $1192, %rsp # imm = 0x4A8
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
 ; AVX2-FCP-LABEL: load_i32_stride7_vf32:
 ; AVX2-FCP:       # %bb.0:
-; AVX2-FCP-NEXT:    subq $1192, %rsp # imm = 0x4A8
-; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    subq $1224, %rsp # imm = 0x4C8
+; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %ymm9
 ; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %ymm4
 ; AVX2-FCP-NEXT:    vmovdqa 224(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 544(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa 544(%rdi), %ymm12
 ; AVX2-FCP-NEXT:    vmovdqa 480(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vmovdqa 448(%rdi), %ymm13
+; AVX2-FCP-NEXT:    vmovdqa 448(%rdi), %ymm8
 ; AVX2-FCP-NEXT:    vmovdqa (%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpbroadcastq 80(%rdi), %ymm0
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm2, %ymm12
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,7,6,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm9[6],ymm3[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm6
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm14
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
@@ -6580,13 +6516,12 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm7[6],ymm13[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa %ymm7, %ymm8
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1,2,3,4,5],ymm7[6],ymm8[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vpbroadcastq 528(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm12[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 576(%rdi), %xmm2
 ; AVX2-FCP-NEXT:    vmovdqa 608(%rdi), %xmm3
@@ -6598,11 +6533,12 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm5[0,1,2,3,4,5],ymm4[6],ymm5[7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm4, %ymm11
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vpbroadcastq 304(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm9[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm9, (%rsp) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %xmm2
 ; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %xmm3
@@ -6613,16 +6549,17 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 704(%rdi), %ymm14
-; AVX2-FCP-NEXT:    vmovdqa 672(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm14[6],ymm1[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vmovdqa 768(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqa 704(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqa 672(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm6
+; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa %ymm2, %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vmovdqa 768(%rdi), %ymm11
 ; AVX2-FCP-NEXT:    vpbroadcastq 752(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm11[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 800(%rdi), %xmm1
 ; AVX2-FCP-NEXT:    vmovdqa 832(%rdi), %xmm2
@@ -6635,179 +6572,170 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
 ; AVX2-FCP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vmovdqa 160(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqa 608(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 576(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
 ; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm0
-; AVX2-FCP-NEXT:    vmovdqa %ymm12, %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm12[0,1],ymm0[2,3],ymm12[4,5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm0, %ymm12
-; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [5,6,5,6,5,6,5,6]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [1,0,7,7,5,4,7,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm9[0],ymm6[1],ymm9[2,3,4],ymm6[5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-FCP-NEXT:    vmovdqa 512(%rdi), %ymm13
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm13[2,3],ymm12[4,5],ymm13[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0],ymm8[1],ymm7[2,3,4],ymm8[5],ymm7[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 608(%rdi), %ymm15
-; AVX2-FCP-NEXT:    vmovdqa 576(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm15[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm15[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 512(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm5[0,1],ymm10[2,3],ymm5[4,5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm8[0],ymm13[1],ymm8[2,3,4],ymm13[5],ymm8[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm0[5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm0 = [1,0,7,6,5,6,5,6]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %ymm7
+; AVX2-FCP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm4[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, (%rsp) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm9[0,1],ymm3[2,3],ymm9[4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm7[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm7[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 832(%rdi), %ymm6
-; AVX2-FCP-NEXT:    vmovdqa 800(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm6[12,13,14,15],ymm11[0,1,2,3,4,5,6,7,8,9,10,11],ymm6[28,29,30,31],ymm11[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vmovdqa %ymm11, %ymm13
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm2[2,3],ymm9[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0],ymm5[1],ymm4[2,3,4],ymm5[5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4],ymm2[5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 832(%rdi), %ymm9
+; AVX2-FCP-NEXT:    vmovdqa 800(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm9[12,13,14,15],ymm5[0,1,2,3,4,5,6,7,8,9,10,11],ymm9[28,29,30,31],ymm5[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vmovdqa %ymm5, %ymm15
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 736(%rdi), %ymm5
 ; AVX2-FCP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 736(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm8[0,1],ymm4[2,3],ymm8[4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm11[0,1],ymm5[2,3],ymm11[4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm3[0],ymm6[1],ymm3[2,3,4],ymm6[5],ymm3[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm14[0],mem[1],ymm14[2,3,4],mem[5],ymm14[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4],ymm2[5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 160(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm4[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm4[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vmovdqa %ymm4, %ymm8
+; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm7
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm6[0,1],ymm1[2,3],ymm6[4,5],ymm1[6,7]
+; AVX2-FCP-NEXT:    vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = mem[0],ymm14[1],mem[2,3,4],ymm14[5],mem[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm3[5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm0
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 80(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm7[8,9,10,11,12,13,14,15],ymm12[0,1,2,3,4,5,6,7],ymm7[24,25,26,27,28,29,30,31],ymm12[16,17,18,19,20,21,22,23]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 8(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm1[0],ymm14[0],ymm1[2],ymm14[2]
-; AVX2-FCP-NEXT:    vpbroadcastd 204(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm6[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm6[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 8(%rdi), %xmm2
+; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %xmm3
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm2 = ymm8[0],ymm7[0],ymm8[2],ymm7[2]
+; AVX2-FCP-NEXT:    vpbroadcastd 204(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 528(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm5[8,9,10,11,12,13,14,15],ymm10[0,1,2,3,4,5,6,7],ymm5[24,25,26,27,28,29,30,31],ymm10[16,17,18,19,20,21,22,23]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 456(%rdi), %xmm3
-; AVX2-FCP-NEXT:    vmovdqa 480(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0],xmm1[1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm12[8,9,10,11,12,13,14,15],ymm13[0,1,2,3,4,5,6,7],ymm12[24,25,26,27,28,29,30,31],ymm13[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 456(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vmovdqa 480(%rdi), %xmm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0],xmm2[1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm3 = ymm12[0],ymm15[0],ymm12[2],ymm15[2]
-; AVX2-FCP-NEXT:    vpbroadcastd 652(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm11[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm4 = ymm8[0],ymm12[0],ymm8[2],ymm12[2]
+; AVX2-FCP-NEXT:    vpbroadcastd 652(%rdi), %ymm13
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm4[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 752(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm8[8,9,10,11,12,13,14,15],ymm4[0,1,2,3,4,5,6,7],ymm8[24,25,26,27,28,29,30,31],ymm4[16,17,18,19,20,21,22,23]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 680(%rdi), %xmm11
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm4 = ymm11[8,9,10,11,12,13,14,15],ymm5[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm5[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 680(%rdi), %xmm13
 ; AVX2-FCP-NEXT:    vmovdqa 704(%rdi), %xmm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm11 = xmm11[0],xmm10[1],xmm11[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm11[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm11 = ymm13[0],ymm6[0],ymm13[2],ymm6[2]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm13 = xmm13[0],xmm10[1],xmm13[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm13[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm13 = ymm15[0],ymm9[0],ymm15[2],ymm9[2]
 ; AVX2-FCP-NEXT:    vpbroadcastd 876(%rdi), %ymm15
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5,6],ymm15[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm11[5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5,6],ymm15[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm13[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 304(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm11 = ymm9[8,9,10,11,12,13,14,15],ymm4[0,1,2,3,4,5,6,7],ymm9[24,25,26,27,28,29,30,31],ymm4[16,17,18,19,20,21,22,23]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2],ymm0[3],ymm11[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm13 = ymm6[8,9,10,11,12,13,14,15],ymm7[0,1,2,3,4,5,6,7],ymm6[24,25,26,27,28,29,30,31],ymm7[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm0[3],ymm13[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastd 232(%rdi), %xmm15
 ; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm0[1],xmm15[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm15[0,1],ymm11[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm3[0],ymm8[0],ymm3[2],ymm8[2]
-; AVX2-FCP-NEXT:    vpbroadcastd 428(%rdi), %ymm13
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm11[0,1,2,3,4],ymm13[5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm7 = ymm7[0],mem[1],ymm7[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm5[0],ymm4[0],ymm5[2],ymm4[2]
+; AVX2-FCP-NEXT:    vpbroadcastd 428(%rdi), %ymm14
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm15[0,1,2,3,4,5,6],ymm14[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4],ymm14[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd $253, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = mem[0],ymm1[1],mem[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1,2],mem[3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[3,2,2,3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,1,1,0,7,5,5,4]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1],ymm1[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm3 = ymm15[0,2],ymm9[1,3],ymm15[4,6],ymm9[5,7]
+; AVX2-FCP-NEXT:    vbroadcastss 208(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm11[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm3[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $253, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = mem[0],ymm1[1],mem[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0,1,2],mem[3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[3,2,2,3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[3,1,1,0,7,5,5,4]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm7[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm14, %ymm5
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm7 = ymm14[0,2],ymm5[1,3],ymm14[4,6],ymm5[5,7]
-; AVX2-FCP-NEXT:    vbroadcastss 208(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm7[5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm2 = ymm2[0],mem[1],ymm2[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1,2],mem[3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[3,1,1,0,7,5,5,4]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm12, %ymm9
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm12[0,2],ymm13[1,3],ymm12[4,6],ymm13[5,7]
-; AVX2-FCP-NEXT:    vbroadcastss 656(%rdi), %ymm6
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,1,1,0,7,5,5,4]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm8, %ymm13
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm8[0,2],ymm12[1,3],ymm8[4,6],ymm12[5,7]
+; AVX2-FCP-NEXT:    vbroadcastss 656(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd $253, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0],ymm4[1],mem[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm6[0],ymm7[1],ymm6[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],mem[3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,2,2,3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[3,1,1,0,7,5,5,4]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm7
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2],ymm8[1,3],ymm3[4,6],ymm8[5,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm5[0,2],ymm4[1,3],ymm5[4,6],ymm4[5,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm4, %ymm8
 ; AVX2-FCP-NEXT:    vbroadcastss 432(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
@@ -6819,235 +6747,239 @@ define void @load_i32_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,2],ymm12[1,3],ymm10[4,6],ymm12[5,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,2],ymm5[1,3],ymm6[4,6],ymm5[5,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 880(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [4,3,0,0]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm1 = [4,3,0,0]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm1, %ymm0
 ; AVX2-FCP-NEXT:    vpbroadcastd 548(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vmovdqa 512(%rdi), %xmm15
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm15[0,1,2],xmm2[3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3]
+; AVX2-FCP-NEXT:    vmovdqa 512(%rdi), %xmm14
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm14[0,1,2],xmm2[3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3]
 ; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm11 = [0,7,0,7,0,7,0,7]
-; AVX2-FCP-NEXT:    vpermd %ymm9, %ymm11, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm13[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm13, %ymm9
+; AVX2-FCP-NEXT:    vpermd %ymm13, %ymm11, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-FCP-NEXT:    vmovaps %ymm12, %ymm13
 ; AVX2-FCP-NEXT:    vpbroadcastd 660(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpbroadcastd 100(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %xmm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm3[0,1,2],xmm1[3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpbroadcastd 100(%rdi), %xmm2
+; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %xmm0
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm0[0,1,2],xmm2[3]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm3
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; AVX2-FCP-NEXT:    vmovaps %ymm15, %ymm10
+; AVX2-FCP-NEXT:    vpermd %ymm15, %ymm11, %ymm3
+; AVX2-FCP-NEXT:    vmovaps %ymm9, %ymm7
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 212(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3]
-; AVX2-FCP-NEXT:    vpermd %ymm14, %ymm11, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm5, %ymm4
-; AVX2-FCP-NEXT:    vpbroadcastd 212(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm5[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, (%rsp) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vpbroadcastd 324(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %xmm6
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm6[0,1,2],xmm2[3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3]
-; AVX2-FCP-NEXT:    vpermd %ymm7, %ymm11, %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm8, %ymm13
-; AVX2-FCP-NEXT:    vpbroadcastd 436(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm5[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm1, %ymm3
+; AVX2-FCP-NEXT:    vpbroadcastd 324(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %xmm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm2[0,1,2],xmm4[3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm3[0,1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpermd %ymm9, %ymm11, %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm8[6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm8, %ymm15
+; AVX2-FCP-NEXT:    vpbroadcastd 436(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm1
+; AVX2-FCP-NEXT:    vpbroadcastd 772(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vmovdqa 736(%rdi), %xmm3
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm3[0,1,2],xmm4[3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vpermd %ymm6, %ymm11, %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FCP-NEXT:    vmovaps %ymm5, %ymm12
+; AVX2-FCP-NEXT:    vpbroadcastd 884(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vpbroadcastd 772(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vmovdqa 736(%rdi), %xmm8
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm8[0,1,2],xmm1[3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
-; AVX2-FCP-NEXT:    vpermd %ymm10, %ymm11, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm12[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm12, %ymm2
-; AVX2-FCP-NEXT:    vpbroadcastd 884(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm5[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0],ymm14[1],ymm4[2,3,4],ymm14[5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm12 = [1,0,3,3,1,0,7,7]
-; AVX2-FCP-NEXT:    vpermd %ymm0, %ymm12, %ymm0
-; AVX2-FCP-NEXT:    vpbroadcastd 216(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm5[7]
-; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %xmm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm7[0,1,2],xmm3[3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,1,3,2]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm7[0],ymm10[1],ymm7[2,3,4],ymm10[5],ymm7[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm4 = [1,0,3,3,1,0,7,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
+; AVX2-FCP-NEXT:    vbroadcastss 216(%rdi), %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %xmm6
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm6[0,1,2],xmm0[3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,3,2]
+; AVX2-FCP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vextracti128 $1, %ymm7, %xmm7
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm7[0,1],xmm0[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 544(%rdi), %xmm8
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm8[0,1,2],xmm14[3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,3,2]
 ; AVX2-FCP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
 ; AVX2-FCP-NEXT:    vextracti128 $1, %ymm5, %xmm5
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 544(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm0[0,1,2],xmm15[3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,1,3,2]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm5[0,1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = ymm13[0],mem[1],ymm13[2,3,4],mem[5],ymm13[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm4, %ymm5
+; AVX2-FCP-NEXT:    vbroadcastss 664(%rdi), %ymm7
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %xmm13
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm13[0,1,2],xmm2[3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,3,2]
 ; AVX2-FCP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
 ; AVX2-FCP-NEXT:    vextracti128 $1, %ymm5, %xmm5
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm5[0,1],xmm4[2,3]
-; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm5, %ymm12, %ymm5
-; AVX2-FCP-NEXT:    vpbroadcastd 664(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm4[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %xmm9
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm9[0,1,2],xmm6[3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,1,3,2]
-; AVX2-FCP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm5[0,1],xmm2[2,3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm15[0],ymm9[1],ymm15[2,3,4],ymm9[5],ymm15[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm4, %ymm5
+; AVX2-FCP-NEXT:    vbroadcastss 440(%rdi), %ymm7
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm2[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 768(%rdi), %xmm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,1,3,2]
+; AVX2-FCP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vextracti128 $1, %ymm7, %xmm7
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm3 = xmm7[0,1],xmm3[2,3]
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = ymm12[0],mem[1],ymm12[2,3,4],mem[5],ymm12[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm4, %ymm4
+; AVX2-FCP-NEXT:    vbroadcastss 888(%rdi), %ymm7
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm3[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 584(%rdi), %xmm3
+; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
+; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-FCP-NEXT:    vpermd 640(%rdi), %ymm11, %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 528(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm8[3]
+; AVX2-FCP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm8 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm8 = ymm8[0],mem[1],ymm8[2,3,4],mem[5],ymm8[6,7]
+; AVX2-FCP-NEXT:    vextracti128 $1, %ymm8, %xmm8
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm8[0,1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm4[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 808(%rdi), %xmm3
+; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
+; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-FCP-NEXT:    vpermd 864(%rdi), %ymm11, %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 752(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm4[0,1,2],xmm2[3]
+; AVX2-FCP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = ymm4[0],mem[1],ymm4[2,3,4],mem[5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vextracti128 $1, %ymm4, %xmm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 136(%rdi), %xmm3
+; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
+; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-FCP-NEXT:    vpermd 192(%rdi), %ymm11, %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 80(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1,2],xmm6[3]
+; AVX2-FCP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm6[0],mem[1],ymm6[2,3,4],mem[5],ymm6[6,7]
 ; AVX2-FCP-NEXT:    vextracti128 $1, %ymm6, %xmm6
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
-; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm6 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm6 = ymm13[0],mem[1],ymm13[2,3,4],mem[5],ymm13[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm6, %ymm12, %ymm6
-; AVX2-FCP-NEXT:    vpbroadcastd 440(%rdi), %ymm13
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa 768(%rdi), %xmm14
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm6 = xmm14[0,1,2],xmm8[3]
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[0,1,3,2]
-; AVX2-FCP-NEXT:    vpshufd $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm8 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FCP-NEXT:    vextracti128 $1, %ymm8, %xmm8
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm6 = xmm8[0,1],xmm6[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm2[0],ymm10[1],ymm2[2,3,4],ymm10[5],ymm2[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm8, %ymm12, %ymm8
-; AVX2-FCP-NEXT:    vpbroadcastd 888(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6],ymm10[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 584(%rdi), %xmm8
-; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm8, %xmm8 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm8 = xmm8[0],mem[1],xmm8[2,3]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm8, %ymm0, %ymm8
-; AVX2-FCP-NEXT:    vpermd 640(%rdi), %ymm11, %ymm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 528(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm10[0,1,2],xmm0[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 360(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vpermd 416(%rdi), %ymm11, %ymm6
+; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-FCP-NEXT:    vpbroadcastd 304(%rdi), %ymm6
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm6 = xmm6[0,1,2],xmm13[3]
 ; AVX2-FCP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm10 = mem[2,3,2,3,6,7,6,7]
 ; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm10 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm10 = ymm10[0],mem[1],ymm10[2,3,4],mem[5],ymm10[6,7]
 ; AVX2-FCP-NEXT:    vextracti128 $1, %ymm10, %xmm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm10[0,1],xmm0[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm0[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 808(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vpermd 864(%rdi), %ymm11, %ymm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 752(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm10 = xmm10[0,1,2],xmm14[3]
-; AVX2-FCP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm12 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm12 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm12 = ymm12[0],mem[1],ymm12[2,3,4],mem[5],ymm12[6,7]
-; AVX2-FCP-NEXT:    vextracti128 $1, %ymm12, %xmm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm10 = xmm12[0,1],xmm10[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 136(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vpermd 192(%rdi), %ymm11, %ymm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm12[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 80(%rdi), %ymm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm7 = xmm12[0,1,2],xmm7[3]
-; AVX2-FCP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm12 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm12 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm12 = ymm12[0],mem[1],ymm12[2,3,4],mem[5],ymm12[6,7]
-; AVX2-FCP-NEXT:    vextracti128 $1, %ymm12, %xmm12
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm7 = xmm12[0,1],xmm7[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm7[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 360(%rdi), %xmm7
-; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7, %xmm7 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm7 = xmm7[0],mem[1],xmm7[2,3]
-; AVX2-FCP-NEXT:    vpermd 416(%rdi), %ymm11, %ymm11
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm7, %ymm0, %ymm7
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastd 304(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm9 = xmm11[0,1,2],xmm9[3]
-; AVX2-FCP-NEXT:    vpshufd $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vextracti128 $1, %ymm11, %xmm11
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm9 = xmm11[0,1],xmm9[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm9[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm9, 96(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm9, 32(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm9, 64(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm9, (%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 96(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 32(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 64(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm9, (%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 32(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 96(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 64(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, (%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 96(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 32(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 64(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm2, (%r8)
-; AVX2-FCP-NEXT:    vmovdqa %ymm1, 96(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 32(%r9)
-; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, (%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 64(%r9)
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm6 = xmm10[0,1],xmm6[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 96(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 64(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 96(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 64(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 96(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 64(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, (%rcx)
+; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 96(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 64(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, (%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 96(%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 32(%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, (%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm6, 64(%r9)
 ; AVX2-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FCP-NEXT:    vmovdqa %ymm6, 96(%rax)
-; AVX2-FCP-NEXT:    vmovdqa %ymm4, 32(%rax)
-; AVX2-FCP-NEXT:    vmovdqa %ymm5, 64(%rax)
-; AVX2-FCP-NEXT:    vmovdqa %ymm3, (%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm7, 96(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm1, 64(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm0, (%rax)
 ; AVX2-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FCP-NEXT:    vmovdqa %ymm7, 32(%rax)
-; AVX2-FCP-NEXT:    vmovdqa %ymm0, (%rax)
-; AVX2-FCP-NEXT:    vmovdqa %ymm10, 96(%rax)
-; AVX2-FCP-NEXT:    vmovdqa %ymm8, 64(%rax)
-; AVX2-FCP-NEXT:    addq $1192, %rsp # imm = 0x4A8
+; AVX2-FCP-NEXT:    vmovaps %ymm4, 32(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm3, (%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm2, 96(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm8, 64(%rax)
+; AVX2-FCP-NEXT:    addq $1224, %rsp # imm = 0x4C8
 ; AVX2-FCP-NEXT:    vzeroupper
 ; AVX2-FCP-NEXT:    retq
 ;
@@ -11264,13 +11196,14 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ;
 ; AVX2-LABEL: load_i32_stride7_vf64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    subq $2680, %rsp # imm = 0xA78
-; AVX2-NEXT:    vmovdqa 1216(%rdi), %ymm12
-; AVX2-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    subq $2648, %rsp # imm = 0xA58
+; AVX2-NEXT:    vmovdqa 1216(%rdi), %ymm9
+; AVX2-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 1152(%rdi), %ymm4
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 1120(%rdi), %ymm5
-; AVX2-NEXT:    vmovdqa 768(%rdi), %ymm13
-; AVX2-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 768(%rdi), %ymm12
+; AVX2-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 704(%rdi), %ymm6
 ; AVX2-NEXT:    vmovdqa 672(%rdi), %ymm7
 ; AVX2-NEXT:    vmovdqa 320(%rdi), %ymm8
@@ -11297,13 +11230,12 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3,4,5],ymm6[6],ymm7[7]
-; AVX2-NEXT:    vmovdqa %ymm7, %ymm9
 ; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa %ymm6, %ymm8
 ; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vpbroadcastq 752(%rdi), %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm13[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm12[4,5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 800(%rdi), %xmm2
 ; AVX2-NEXT:    vmovdqa 832(%rdi), %xmm3
@@ -11317,11 +11249,9 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm5[0,1,2,3,4,5],ymm4[6],ymm5[7]
 ; AVX2-NEXT:    vmovdqa %ymm5, %ymm6
 ; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa %ymm4, %ymm5
-; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vpbroadcastq 1200(%rdi), %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm12[4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm9[4,5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 1248(%rdi), %xmm2
 ; AVX2-NEXT:    vmovdqa 1280(%rdi), %xmm3
@@ -11332,12 +11262,12 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 1600(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqa 1568(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
-; AVX2-NEXT:    vmovdqa %ymm2, %ymm14
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 1600(%rdi), %ymm13
+; AVX2-NEXT:    vmovdqa 1568(%rdi), %ymm3
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm13[6],ymm3[7]
+; AVX2-NEXT:    vmovdqa %ymm3, %ymm5
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vmovdqa 1664(%rdi), %ymm3
 ; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -11357,11 +11287,11 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpbroadcastq 80(%rdi), %ymm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm3
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqa (%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7]
+; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6],ymm2[7]
 ; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 128(%rdi), %xmm2
@@ -11374,11 +11304,10 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 480(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqa 448(%rdi), %ymm13
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm2[6],ymm13[7]
-; AVX2-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa %ymm2, %ymm12
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 448(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vmovdqa 544(%rdi), %ymm3
 ; AVX2-NEXT:    vmovdqu %ymm3, (%rsp) # 32-byte Spill
@@ -11394,12 +11323,11 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 928(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqa 928(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 896(%rdi), %ymm1
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 896(%rdi), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm1[6],ymm3[7]
-; AVX2-NEXT:    vmovdqa %ymm3, %ymm15
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-NEXT:    vmovdqa 992(%rdi), %ymm3
 ; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -11415,16 +11343,16 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 1376(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 1344(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
+; AVX2-NEXT:    vmovdqa 1376(%rdi), %ymm14
+; AVX2-NEXT:    vmovdqa 1344(%rdi), %ymm15
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm15[0,1,2,3,4,5],ymm14[6],ymm15[7]
+; AVX2-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 1440(%rdi), %ymm7
+; AVX2-NEXT:    vmovdqa 1440(%rdi), %ymm4
 ; AVX2-NEXT:    vpbroadcastq 1424(%rdi), %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqa 1472(%rdi), %xmm1
 ; AVX2-NEXT:    vmovdqa 1504(%rdi), %xmm2
@@ -11443,215 +11371,192 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm1[12,13,14,15],ymm0[0,1,2,3,4,5,6,7,8,9,10,11],ymm1[28,29,30,31],ymm0[16,17,18,19,20,21,22,23,24,25,26,27]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqa 288(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm1[2,3],ymm4[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-NEXT:    vmovdqa 288(%rdi), %ymm12
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm9[0,1],ymm12[2,3],ymm9[4,5],ymm12[6,7]
+; AVX2-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0],ymm11[1],ymm10[2,3,4],ymm11[5],ymm10[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 832(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 800(%rdi), %ymm1
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm0[5,6],ymm2[7]
+; AVX2-NEXT:    vpmovsxbd {{.*#+}} ymm0 = [1,0,7,6,5,6,5,6]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 736(%rdi), %ymm10
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm11[0,1],ymm10[2,3],ymm11[4,5],ymm10[6,7]
-; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0],ymm9[1],ymm8[2,3,4],ymm9[5],ymm8[6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqa 832(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 800(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 1280(%rdi), %ymm2
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-NEXT:    vmovdqa 736(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 1248(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 1184(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm11[0,1],ymm2[2,3],ymm11[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0],ymm7[1],ymm8[2,3,4],ymm7[5],ymm8[6,7]
+; AVX2-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0,1,2,3,4],ymm2[5,6],ymm8[7]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm9[0,1],ymm1[2,3],ymm9[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0],ymm6[1],ymm5[2,3,4],ymm6[5],ymm5[6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqa 1280(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 1248(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 1728(%rdi), %ymm2
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-NEXT:    vmovdqa 1184(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 1696(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0,1],ymm2[2,3],ymm10[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm7 = mem[0],ymm6[1],mem[2,3,4],ymm6[5],mem[6,7]
+; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1,2,3,4],ymm2[5,6],ymm7[7]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 1632(%rdi), %ymm6
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm5[0,1],ymm6[2,3],ymm5[4,5],ymm6[6,7]
-; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm14[0],mem[1],ymm14[2,3,4],mem[5],ymm14[6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqa 1728(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 1696(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 608(%rdi), %ymm2
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-NEXT:    vmovdqa 1632(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 576(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 512(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm2[2,3],ymm6[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0],ymm5[1],ymm13[2,3,4],ymm5[5],ymm13[6,7]
+; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4],ymm2[5,6],ymm5[7]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu (%rsp), %ymm14 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm14[0,1],ymm1[2,3],ymm14[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0],ymm13[1],ymm12[2,3,4],ymm13[5],ymm12[6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqa 608(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 576(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 1056(%rdi), %ymm2
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-NEXT:    vmovdqa 512(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 1024(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 960(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqu (%rsp), %ymm8 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0,1],ymm2[2,3],ymm8[4,5],ymm2[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1],ymm1[2,3],ymm8[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-NEXT:    vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0],ymm15[1],mem[2,3,4],ymm15[5],mem[6,7]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqa 1056(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 1024(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 1504(%rdi), %ymm2
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-NEXT:    vmovdqa 960(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 1472(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 1408(%rdi), %ymm1
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1],ymm2[2,3],ymm13[4,5],ymm2[6,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm1[2,3],ymm7[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = ymm2[0],mem[1],ymm2[2,3,4],mem[5],ymm2[6,7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-NEXT:    vmovdqa 160(%rdi), %ymm2
-; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm3
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-NEXT:    vmovdqa %ymm3, %ymm13
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqa 1504(%rdi), %ymm3
 ; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa %ymm2, %ymm12
+; AVX2-NEXT:    vmovdqa 1472(%rdi), %ymm2
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm3
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-NEXT:    vmovdqa 1408(%rdi), %ymm2
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3],ymm4[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm14[0],ymm15[1],ymm14[2,3,4],ymm15[5],ymm14[6,7]
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vmovdqa 160(%rdi), %ymm3
+; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm5
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm5[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm5[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-NEXT:    vmovdqa %ymm5, %ymm14
+; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa %ymm3, %ymm15
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm4
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm3[2,3],ymm7[4,5],ymm3[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1],ymm4[2,3],ymm7[4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0],ymm2[1],mem[2,3,4],ymm2[5],mem[6,7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-NEXT:    vpermd %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 304(%rdi), %xmm0
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm4[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm4[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm9[8,9,10,11,12,13,14,15],ymm12[0,1,2,3,4,5,6,7],ymm9[24,25,26,27,28,29,30,31],ymm12[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 232(%rdi), %xmm1
-; AVX2-NEXT:    vmovdqa 256(%rdi), %xmm2
-; AVX2-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2,3]
+; AVX2-NEXT:    vmovdqa 256(%rdi), %xmm5
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm5[1],xmm1[2,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = ymm4[0],mem[0],ymm4[2],mem[2]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm1[0],ymm12[0],ymm1[2],ymm12[2]
 ; AVX2-NEXT:    vpbroadcastd 428(%rdi), %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 752(%rdi), %xmm0
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm11[8,9,10,11,12,13,14,15],ymm10[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm10[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm11[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 680(%rdi), %xmm1
 ; AVX2-NEXT:    vmovdqa 704(%rdi), %xmm2
 ; AVX2-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm1 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm1 = ymm11[0],mem[0],ymm11[2],mem[2]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[2],ymm3[2]
 ; AVX2-NEXT:    vpbroadcastd 876(%rdi), %ymm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 1200(%rdi), %xmm0
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm9[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm9[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm10[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm10[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 1128(%rdi), %xmm1
 ; AVX2-NEXT:    vmovdqa 1152(%rdi), %xmm2
@@ -11666,7 +11571,8 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 1648(%rdi), %xmm0
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm5[8,9,10,11,12,13,14,15],ymm6[0,1,2,3,4,5,6,7],ymm5[24,25,26,27,28,29,30,31],ymm6[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm6[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm6[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 1576(%rdi), %xmm1
 ; AVX2-NEXT:    vmovdqa 1600(%rdi), %xmm2
@@ -11680,26 +11586,26 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 80(%rdi), %xmm0
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm7[8,9,10,11,12,13,14,15],ymm3[0,1,2,3,4,5,6,7],ymm7[24,25,26,27,28,29,30,31],ymm3[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm7[8,9,10,11,12,13,14,15],ymm4[0,1,2,3,4,5,6,7],ymm7[24,25,26,27,28,29,30,31],ymm4[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 8(%rdi), %xmm1
-; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm3
-; AVX2-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm3[1],xmm1[2,3]
+; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm4
+; AVX2-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm4[1],xmm1[2,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm13[0],ymm12[0],ymm13[2],ymm12[2]
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm14[0],ymm15[0],ymm14[2],ymm15[2]
 ; AVX2-NEXT:    vpbroadcastd 204(%rdi), %ymm6
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm6[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 528(%rdi), %xmm0
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm14[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm14[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm8[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm8[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 456(%rdi), %xmm1
-; AVX2-NEXT:    vmovdqa 480(%rdi), %xmm3
-; AVX2-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm3[1],xmm1[2,3]
+; AVX2-NEXT:    vmovdqa 480(%rdi), %xmm4
+; AVX2-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm4[1],xmm1[2,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
@@ -11710,23 +11616,23 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 976(%rdi), %xmm0
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm8[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm8[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm1 = ymm13[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm13[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 904(%rdi), %xmm15
-; AVX2-NEXT:    vmovdqa 928(%rdi), %xmm12
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm12[1],xmm15[2,3]
+; AVX2-NEXT:    vmovdqa 928(%rdi), %xmm11
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm11[1],xmm15[2,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm8[0],ymm7[0],ymm8[2],ymm7[2]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm10[0],ymm7[0],ymm10[2],ymm7[2]
 ; AVX2-NEXT:    vpbroadcastd 1100(%rdi), %ymm14
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm14 = ymm15[0,1,2,3,4,5,6],ymm14[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm14[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 1424(%rdi), %xmm0
-; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-NEXT:    vpalignr {{.*#+}} ymm14 = ymm10[8,9,10,11,12,13,14,15],ymm9[0,1,2,3,4,5,6,7],ymm10[24,25,26,27,28,29,30,31],ymm9[16,17,18,19,20,21,22,23]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-NEXT:    vpalignr {{.*#+}} ymm14 = ymm9[8,9,10,11,12,13,14,15],ymm8[0,1,2,3,4,5,6,7],ymm9[24,25,26,27,28,29,30,31],ymm8[16,17,18,19,20,21,22,23]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2],ymm0[3],ymm14[4,5,6,7]
 ; AVX2-NEXT:    vpbroadcastd 1352(%rdi), %xmm15
 ; AVX2-NEXT:    vmovdqa 1376(%rdi), %xmm0
@@ -11739,30 +11645,30 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2,3,4,5,6],ymm13[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm13 = ymm14[0,1,2,3,4],ymm13[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm13 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm13 = ymm3[0],mem[1],ymm3[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm3[0,1,2],mem[3]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm5 = xmm5[3,2,2,3]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[3,1,1,0,7,5,5,4]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1],ymm13[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm4[0,2],ymm15[1,3],ymm4[4,6],ymm15[5,7]
+; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm13 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm13 = ymm4[0],mem[1],ymm4[2,3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1,2],mem[3]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[3,2,2,3]
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,1,1,0,7,5,5,4]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1],ymm13[2,3,4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vshufps {{.*#+}} ymm13 = ymm4[0,2],ymm12[1,3],ymm4[4,6],ymm12[5,7]
+; AVX2-NEXT:    vmovaps %ymm4, %ymm12
 ; AVX2-NEXT:    vbroadcastss 432(%rdi), %ymm14
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5,6],ymm14[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm13[5,6,7]
-; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = ymm3[0],mem[1],ymm3[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm3[0,1,2],mem[3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm13[5,6,7]
+; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = ymm4[0],mem[1],ymm4[2,3,4,5,6,7]
+; AVX2-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm4[0,1,2],mem[3]
 ; AVX2-NEXT:    vshufps {{.*#+}} xmm4 = xmm4[3,2,2,3]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[3,1,1,0,7,5,5,4]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm5[2,3,4,5,6,7]
-; AVX2-NEXT:    vshufps $216, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = ymm11[0,2],mem[1,3],ymm11[4,6],mem[5,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm14[0,2],ymm3[1,3],ymm14[4,6],ymm3[5,7]
 ; AVX2-NEXT:    vbroadcastss 880(%rdi), %ymm13
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm13[7]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm5[5,6,7]
@@ -11775,9 +11681,9 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[3,2,2,3]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[3,1,1,0,7,5,5,4]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm13[0,2],ymm14[1,3],ymm13[4,6],ymm14[5,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vshufps $216, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = ymm4[0,2],mem[1,3],ymm4[4,6],mem[5,7]
 ; AVX2-NEXT:    vbroadcastss 1328(%rdi), %ymm5
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm5[7]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7]
@@ -11789,19 +11695,20 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[3,2,2,3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[3,1,1,0,7,5,5,4]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vshufps $216, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = ymm3[0,2],mem[1,3],ymm3[4,6],mem[5,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-NEXT:    vshufps $216, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = ymm13[0,2],mem[1,3],ymm13[4,6],mem[5,7]
 ; AVX2-NEXT:    vbroadcastss 1776(%rdi), %ymm4
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0],ymm9[1],ymm10[2,3,4,5,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0],ymm8[1],ymm9[2,3,4,5,6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],mem[3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,2,2,3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[3,1,1,0,7,5,5,4]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3,4,5,6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm6[0,2],ymm1[1,3],ymm6[4,6],ymm1[5,7]
+; AVX2-NEXT:    vmovaps %ymm1, %ymm9
 ; AVX2-NEXT:    vbroadcastss 1552(%rdi), %ymm3
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
@@ -11809,13 +11716,13 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm12[0,1,2],mem[3]
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm11[0,1,2],mem[3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[0,2],ymm7[1,3],ymm8[4,6],ymm7[5,7]
-; AVX2-NEXT:    vmovaps %ymm7, %ymm9
-; AVX2-NEXT:    vmovaps %ymm8, %ymm11
+; AVX2-NEXT:    vmovdqa %ymm10, %ymm8
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,2],ymm7[1,3],ymm10[4,6],ymm7[5,7]
+; AVX2-NEXT:    vmovaps %ymm7, %ymm11
 ; AVX2-NEXT:    vbroadcastss 1104(%rdi), %ymm2
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
@@ -11828,9 +11735,9 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,2],ymm8[1,3],ymm6[4,6],ymm8[5,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm4[0,2],ymm7[1,3],ymm4[4,6],ymm7[5,7]
 ; AVX2-NEXT:    vbroadcastss 656(%rdi), %ymm2
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
@@ -11844,125 +11751,128 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2],ymm4[1,3],ymm3[4,6],ymm4[5,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2],ymm10[1,3],ymm3[4,6],ymm10[5,7]
 ; AVX2-NEXT:    vbroadcastss 208(%rdi), %ymm2
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vbroadcastss 100(%rdi), %xmm0
-; AVX2-NEXT:    vmovaps 64(%rdi), %xmm1
-; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
-; AVX2-NEXT:    vmovsd {{.*#+}} xmm7 = [4,3,0,0]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm7, %ymm2
-; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm2[0,1],xmm0[2,3]
-; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [0,7,0,7,0,7,0,7]
-; AVX2-NEXT:    vpermps %ymm3, %ymm0, %ymm3
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-NEXT:    vbroadcastss 212(%rdi), %ymm4
+; AVX2-NEXT:    vmovaps 64(%rdi), %xmm6
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm6[0,1,2],xmm0[3]
+; AVX2-NEXT:    vmovsd {{.*#+}} xmm5 = [4,3,0,0]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm1, %ymm5, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
+; AVX2-NEXT:    vbroadcastsd {{.*#+}} ymm15 = [0,7,0,7,0,7,0,7]
+; AVX2-NEXT:    vpermps %ymm3, %ymm15, %ymm1
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-NEXT:    vbroadcastss 212(%rdi), %ymm2
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vbroadcastss 324(%rdi), %xmm2
+; AVX2-NEXT:    vmovaps 288(%rdi), %xmm1
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0,1,2],xmm2[3]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3]
+; AVX2-NEXT:    vpermps %ymm12, %ymm15, %ymm2
+; AVX2-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = ymm2[0,1,2,3,4,5],mem[6,7]
+; AVX2-NEXT:    vbroadcastss 436(%rdi), %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vbroadcastss 548(%rdi), %xmm3
+; AVX2-NEXT:    vmovaps 512(%rdi), %xmm2
+; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3]
+; AVX2-NEXT:    vpermps %ymm4, %ymm15, %ymm3
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm7[6,7]
+; AVX2-NEXT:    vbroadcastss 660(%rdi), %ymm4
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm2, %ymm7, %ymm3
-; AVX2-NEXT:    vbroadcastss 324(%rdi), %xmm4
-; AVX2-NEXT:    vmovaps 288(%rdi), %xmm2
-; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm2[0,1,2],xmm4[3]
-; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm3[0,1],xmm4[2,3]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vbroadcastss 436(%rdi), %ymm5
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm5[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm3, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm3 = mem[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm3, %ymm7, %ymm4
-; AVX2-NEXT:    vbroadcastss 548(%rdi), %xmm5
-; AVX2-NEXT:    vmovaps 512(%rdi), %xmm3
-; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm3[0,1,2],xmm5[3]
-; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
-; AVX2-NEXT:    vpermps %ymm6, %ymm0, %ymm5
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-NEXT:    vbroadcastss 660(%rdi), %ymm6
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm6[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm4 = mem[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm4, %ymm7, %ymm5
-; AVX2-NEXT:    vbroadcastss 772(%rdi), %xmm6
-; AVX2-NEXT:    vmovaps 736(%rdi), %xmm4
-; AVX2-NEXT:    vblendps {{.*#+}} xmm6 = xmm4[0,1,2],xmm6[3]
-; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1],xmm6[2,3]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm6 # 32-byte Folded Reload
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-NEXT:    vbroadcastss 884(%rdi), %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm5 = mem[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm5, %ymm7, %ymm6
-; AVX2-NEXT:    vbroadcastss 996(%rdi), %xmm8
-; AVX2-NEXT:    vmovaps 960(%rdi), %xmm5
-; AVX2-NEXT:    vblendps {{.*#+}} xmm8 = xmm5[0,1,2],xmm8[3]
-; AVX2-NEXT:    vblendps {{.*#+}} xmm6 = xmm6[0,1],xmm8[2,3]
-; AVX2-NEXT:    vpermps %ymm11, %ymm0, %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vbroadcastss 772(%rdi), %xmm4
+; AVX2-NEXT:    vmovaps 736(%rdi), %xmm3
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm3[0,1,2],xmm4[3]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm4[2,3]
+; AVX2-NEXT:    vpermps %ymm14, %ymm15, %ymm4
+; AVX2-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = ymm4[0,1,2,3,4,5],mem[6,7]
+; AVX2-NEXT:    vbroadcastss 884(%rdi), %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vbroadcastss 996(%rdi), %xmm7
+; AVX2-NEXT:    vmovaps 960(%rdi), %xmm4
+; AVX2-NEXT:    vblendps {{.*#+}} xmm7 = xmm4[0,1,2],xmm7[3]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm7[2,3]
+; AVX2-NEXT:    vpermps %ymm8, %ymm15, %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-NEXT:    vbroadcastss 1108(%rdi), %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vbroadcastss 1220(%rdi), %xmm7
+; AVX2-NEXT:    vmovaps 1184(%rdi), %xmm14
+; AVX2-NEXT:    vblendps {{.*#+}} xmm7 = xmm14[0,1,2],xmm7[3]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm7[2,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-NEXT:    vpermps %ymm10, %ymm15, %ymm7
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-NEXT:    vbroadcastss 1332(%rdi), %ymm8
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm5, %ymm7
+; AVX2-NEXT:    vbroadcastss 1444(%rdi), %xmm8
+; AVX2-NEXT:    vmovaps 1408(%rdi), %xmm0
+; AVX2-NEXT:    vblendps {{.*#+}} xmm8 = xmm0[0,1,2],xmm8[3]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1],xmm8[2,3]
+; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm8 # 32-byte Folded Reload
+; AVX2-NEXT:    vmovaps %ymm9, %ymm11
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-NEXT:    vbroadcastss 1108(%rdi), %ymm9
+; AVX2-NEXT:    vbroadcastss 1556(%rdi), %ymm9
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm6 = mem[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm6, %ymm7, %ymm8
-; AVX2-NEXT:    vbroadcastss 1220(%rdi), %xmm9
-; AVX2-NEXT:    vmovaps 1184(%rdi), %xmm6
-; AVX2-NEXT:    vblendps {{.*#+}} xmm9 = xmm6[0,1,2],xmm9[3]
-; AVX2-NEXT:    vblendps {{.*#+}} xmm8 = xmm8[0,1],xmm9[2,3]
-; AVX2-NEXT:    vpermps %ymm13, %ymm0, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-NEXT:    vbroadcastss 1332(%rdi), %ymm12
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5,6],ymm12[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = mem[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm8, %ymm7, %ymm8
-; AVX2-NEXT:    vbroadcastss 1444(%rdi), %xmm9
-; AVX2-NEXT:    vmovaps 1408(%rdi), %xmm13
-; AVX2-NEXT:    vblendps {{.*#+}} xmm9 = xmm13[0,1,2],xmm9[3]
-; AVX2-NEXT:    vblendps {{.*#+}} xmm8 = xmm8[0,1],xmm9[2,3]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = ymm9[0,1,2,3,4,5],mem[6,7]
-; AVX2-NEXT:    vbroadcastss 1556(%rdi), %ymm12
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5,6],ymm12[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm8 = mem[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-NEXT:    vpermps %ymm8, %ymm7, %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm8[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm7 = mem[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vpermps %ymm7, %ymm5, %ymm7
 ; AVX2-NEXT:    vbroadcastss 1668(%rdi), %xmm8
-; AVX2-NEXT:    vmovaps 1632(%rdi), %xmm14
-; AVX2-NEXT:    vblendps {{.*#+}} xmm8 = xmm14[0,1,2],xmm8[3]
+; AVX2-NEXT:    vmovaps 1632(%rdi), %xmm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm8 = xmm5[0,1,2],xmm8[3]
 ; AVX2-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1],xmm8[2,3]
-; AVX2-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm8 # 32-byte Folded Reload
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-NEXT:    vpermps %ymm13, %ymm15, %ymm8
+; AVX2-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm8 = ymm8[0,1,2,3,4,5],mem[6,7]
 ; AVX2-NEXT:    vbroadcastss 1780(%rdi), %ymm9
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6],ymm9[7]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm8[4,5,6,7]
@@ -11974,355 +11884,356 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
 ; AVX2-NEXT:    vbroadcastss 216(%rdi), %ymm8
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-NEXT:    vmovaps 96(%rdi), %xmm12
-; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm12[0,1,2],xmm1[3]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-NEXT:    vmovaps 96(%rdi), %xmm9
+; AVX2-NEXT:    vblendps {{.*#+}} xmm6 = xmm9[0,1,2],xmm6[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm6 = xmm6[0,1,3,2]
 ; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm8 = mem[1,0,2,3,5,4,6,7]
 ; AVX2-NEXT:    vextractf128 $1, %ymm8, %xmm8
-; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm8[0,1],xmm1[2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm6 = xmm8[0,1],xmm6[2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 320(%rdi), %xmm13
+; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm13[0,1,2],xmm1[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm6 = mem[1,0,2,3,5,4,6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm6, %xmm6
+; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm6[0,1],xmm1[2,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-NEXT:    vblendps $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm6 = mem[0],ymm6[1],mem[2,3,4],ymm6[5],mem[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,0,3,3,5,4,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
+; AVX2-NEXT:    vbroadcastss 440(%rdi), %ymm7
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm6[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 320(%rdi), %xmm1
-; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0,1,2],xmm2[3]
+; AVX2-NEXT:    vmovaps 544(%rdi), %xmm8
+; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm8[0,1,2],xmm2[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = mem[1,0,2,3,5,4,6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm2, %xmm2
+; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm2 = ymm2[0],mem[1],ymm2[2,3,4],mem[5],ymm2[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
+; AVX2-NEXT:    vbroadcastss 664(%rdi), %ymm6
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovaps 768(%rdi), %xmm1
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0,1,2],xmm3[3]
 ; AVX2-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,1,3,2]
-; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm7[0,1],xmm2[2,3]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = ymm15[0],mem[1],ymm15[2,3,4],mem[5],ymm15[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-NEXT:    vbroadcastss 440(%rdi), %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = mem[1,0,2,3,5,4,6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm3, %xmm3
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-NEXT:    vblendps $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm3 = mem[0],ymm3[1],mem[2,3,4],ymm3[5],mem[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[1,0,3,3,5,4,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
+; AVX2-NEXT:    vbroadcastss 888(%rdi), %ymm6
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 544(%rdi), %xmm2
-; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
+; AVX2-NEXT:    vmovaps 992(%rdi), %xmm2
+; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm2[0,1,2],xmm4[3]
 ; AVX2-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,1,3,2]
-; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm7[0,1],xmm3[2,3]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = ymm7[0],mem[1],ymm7[2,3,4],mem[5],ymm7[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-NEXT:    vbroadcastss 664(%rdi), %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 768(%rdi), %xmm3
-; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm3[0,1,2],xmm4[3]
+; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = mem[1,0,2,3,5,4,6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm4, %xmm4
+; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm4[0,1],xmm3[2,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm4 = ymm4[0],mem[1],ymm4[2,3,4],mem[5],ymm4[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[1,0,3,3,5,4,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
+; AVX2-NEXT:    vbroadcastss 1112(%rdi), %ymm6
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm3[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vmovaps 1216(%rdi), %xmm3
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm3[0,1,2],xmm14[3]
 ; AVX2-NEXT:    vshufps {{.*#+}} xmm4 = xmm4[0,1,3,2]
-; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm7[0,1],xmm4[2,3]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = ymm10[0],mem[1],ymm10[2,3,4],mem[5],ymm10[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-NEXT:    vbroadcastss 888(%rdi), %ymm8
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovaps 992(%rdi), %xmm4
-; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm4[0,1,2],xmm5[3]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm5 = xmm5[0,1,3,2]
-; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm7[0,1],xmm5[2,3]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = ymm7[0],mem[1],ymm7[2,3,4],mem[5],ymm7[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-NEXT:    vbroadcastss 1112(%rdi), %ymm10
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm10[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm5[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-NEXT:    vmovaps 1216(%rdi), %xmm15
-; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm15[0,1,2],xmm6[3]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm5 = xmm5[0,1,3,2]
 ; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX2-NEXT:    # ymm6 = mem[1,0,2,3,5,4,6,7]
 ; AVX2-NEXT:    vextractf128 $1, %ymm6, %xmm6
-; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm6 = ymm6[0],mem[1],ymm6[2,3,4],mem[5],ymm6[6,7]
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm12[0],ymm10[1],ymm12[2,3,4],ymm10[5],ymm12[6,7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,0,3,3,5,4,7,7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
-; AVX2-NEXT:    vbroadcastss 1336(%rdi), %ymm7
-; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-NEXT:    vmovaps 1440(%rdi), %xmm5
-; AVX2-NEXT:    vblendps {{.*#+}} xmm6 = xmm5[0,1,2],xmm13[3]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm6 = xmm6[0,1,3,2]
-; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-NEXT:    vblendps {{.*#+}} xmm6 = xmm7[0,1],xmm6[2,3]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-NEXT:    vblendps $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm7 = mem[0],ymm7[1],mem[2,3,4],ymm7[5],mem[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-NEXT:    vbroadcastss 1560(%rdi), %ymm13
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm7 = ymm6[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-NEXT:    vmovaps 1664(%rdi), %xmm6
-; AVX2-NEXT:    vblendps {{.*#+}} xmm13 = xmm6[0,1,2],xmm14[3]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm13 = xmm13[0,1,3,2]
-; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = mem[1,0,2,3,5,4,6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm14, %xmm14
-; AVX2-NEXT:    vblendps {{.*#+}} xmm13 = xmm14[0,1],xmm13[2,3]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
-; AVX2-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[1,0,3,3,5,4,7,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm14 = ymm14[0,1,0,3]
-; AVX2-NEXT:    vbroadcastss 1784(%rdi), %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm14[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm13[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vbroadcastss 136(%rdi), %xmm9
-; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9, %xmm9 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm9 = xmm9[0],mem[1],xmm9[2,3]
-; AVX2-NEXT:    vinsertf128 $1, %xmm9, %ymm0, %ymm9
-; AVX2-NEXT:    vpermps 192(%rdi), %ymm0, %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-NEXT:    vbroadcastss 80(%rdi), %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} xmm12 = xmm14[0,1,2],xmm12[3]
-; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = ymm14[0],mem[1],ymm14[2,3,4],mem[5],ymm14[6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm14, %xmm14
-; AVX2-NEXT:    vblendps {{.*#+}} xmm12 = xmm14[0,1],xmm12[2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vbroadcastss 360(%rdi), %xmm9
-; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9, %xmm9 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm9 = xmm9[0],mem[1],xmm9[2,3]
-; AVX2-NEXT:    vinsertf128 $1, %xmm9, %ymm0, %ymm9
-; AVX2-NEXT:    vpermps 416(%rdi), %ymm0, %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-NEXT:    vbroadcastss 304(%rdi), %ymm14
-; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm14[0,1,2],xmm1[3]
-; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm14 = ymm14[0],mem[1],ymm14[2,3,4],mem[5],ymm14[6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm14, %xmm14
-; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm14[0,1],xmm1[2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm14 = ymm1[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vbroadcastss 584(%rdi), %xmm1
+; AVX2-NEXT:    vbroadcastss 1336(%rdi), %ymm10
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm10[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm6 = ymm4[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-NEXT:    vmovaps 1440(%rdi), %xmm4
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm4[0,1,2],xmm0[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,1,3,2]
+; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm10 = mem[1,0,2,3,5,4,6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm10, %xmm10
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm10[0,1],xmm0[2,3]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm10 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm10 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[1,0,3,3,5,4,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,1,0,3]
+; AVX2-NEXT:    vbroadcastss 1560(%rdi), %ymm12
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-NEXT:    vmovaps 1664(%rdi), %xmm14
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm14[0,1,2],xmm5[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,1,3,2]
+; AVX2-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm5[0,1],xmm0[2,3]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[1,0,3,3,5,4,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
+; AVX2-NEXT:    vbroadcastss 1784(%rdi), %ymm12
+; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-NEXT:    vbroadcastss 136(%rdi), %xmm0
+; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpermps 192(%rdi), %ymm15, %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-NEXT:    vbroadcastss 80(%rdi), %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1,2],xmm9[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm11, %xmm11
+; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm11[0,1],xmm5[2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm11 = ymm5[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vbroadcastss 360(%rdi), %xmm0
+; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpermps 416(%rdi), %ymm15, %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-NEXT:    vbroadcastss 304(%rdi), %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1,2],xmm13[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm13 = mem[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm13 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm13 = ymm13[0],mem[1],ymm13[2,3,4],mem[5],ymm13[6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm13, %xmm13
+; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm13[0,1],xmm5[2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm13 = ymm5[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vbroadcastss 584(%rdi), %xmm0
+; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpermps 640(%rdi), %ymm15, %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-NEXT:    vbroadcastss 528(%rdi), %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1,2],xmm8[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm8 = mem[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm8 = ymm8[0],mem[1],ymm8[2,3,4],mem[5],ymm8[6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm8, %xmm8
+; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm8[0,1],xmm5[2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vbroadcastss 808(%rdi), %xmm0
+; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-NEXT:    vpermps 864(%rdi), %ymm15, %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-NEXT:    vbroadcastss 752(%rdi), %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm5[0,1,2],xmm1[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = mem[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm1 = xmm5[0,1],xmm1[2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-NEXT:    vbroadcastss 1032(%rdi), %xmm1
 ; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm1 = xmm1[0],mem[1],xmm1[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-NEXT:    vpermps 640(%rdi), %ymm0, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-NEXT:    vbroadcastss 528(%rdi), %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm9[0,1,2],xmm2[3]
-; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm9, %xmm9
-; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm9[0,1],xmm2[2,3]
+; AVX2-NEXT:    vpermps 1088(%rdi), %ymm15, %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-NEXT:    vbroadcastss 976(%rdi), %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm5[0,1,2],xmm2[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = mem[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm2 = xmm5[0,1],xmm2[2,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-NEXT:    vbroadcastss 808(%rdi), %xmm2
+; AVX2-NEXT:    vbroadcastss 1256(%rdi), %xmm2
 ; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm2 = xmm2[0],mem[1],xmm2[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-NEXT:    vpermps 864(%rdi), %ymm0, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-NEXT:    vbroadcastss 752(%rdi), %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm9[0,1,2],xmm3[3]
-; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm9, %xmm9
-; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm9[0,1],xmm3[2,3]
+; AVX2-NEXT:    vpermps 1312(%rdi), %ymm15, %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-NEXT:    vbroadcastss 1200(%rdi), %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm5[0,1,2],xmm3[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = mem[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-NEXT:    vbroadcastss 1032(%rdi), %xmm3
+; AVX2-NEXT:    vbroadcastss 1480(%rdi), %xmm3
 ; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
 ; AVX2-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-NEXT:    vpermps 1088(%rdi), %ymm0, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-NEXT:    vbroadcastss 976(%rdi), %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm9[0,1,2],xmm4[3]
-; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm9, %xmm9
-; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm9[0,1],xmm4[2,3]
+; AVX2-NEXT:    vpermps 1536(%rdi), %ymm15, %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-NEXT:    vbroadcastss 1424(%rdi), %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = mem[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm4 = xmm5[0,1],xmm4[2,3]
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-NEXT:    vbroadcastss 1256(%rdi), %xmm4
+; AVX2-NEXT:    vbroadcastss 1704(%rdi), %xmm4
 ; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
+; AVX2-NEXT:    vpermps 1760(%rdi), %ymm15, %ymm5
 ; AVX2-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-NEXT:    vpermps 1312(%rdi), %ymm0, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-NEXT:    vbroadcastss 1200(%rdi), %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} xmm9 = xmm9[0,1,2],xmm15[3]
-; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm15 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm15 = ymm15[0],mem[1],ymm15[2,3,4],mem[5],ymm15[6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm15, %xmm15
-; AVX2-NEXT:    vblendps {{.*#+}} xmm9 = xmm15[0,1],xmm9[2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm9[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-NEXT:    vbroadcastss 1480(%rdi), %xmm9
-; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9, %xmm9 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm9 = xmm9[0],mem[1],xmm9[2,3]
-; AVX2-NEXT:    vinsertf128 $1, %xmm9, %ymm0, %ymm9
-; AVX2-NEXT:    vpermps 1536(%rdi), %ymm0, %ymm15
-; AVX2-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-NEXT:    vbroadcastss 1424(%rdi), %ymm15
-; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm15[0,1,2],xmm5[3]
-; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm15 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm15 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm15 = ymm15[0],mem[1],ymm15[2,3,4],mem[5],ymm15[6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm15, %xmm15
-; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm15[0,1],xmm5[2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-NEXT:    vbroadcastss 1704(%rdi), %xmm9
-; AVX2-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9, %xmm9 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm9 = xmm9[0],mem[1],xmm9[2,3]
-; AVX2-NEXT:    vpermps 1760(%rdi), %ymm0, %ymm0
-; AVX2-NEXT:    vinsertf128 $1, %xmm9, %ymm0, %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-NEXT:    vbroadcastss 1648(%rdi), %ymm9
-; AVX2-NEXT:    vblendps {{.*#+}} xmm6 = xmm9[0,1,2],xmm6[3]
-; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = mem[2,3,2,3,6,7,6,7]
-; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-NEXT:    vextractf128 $1, %ymm9, %xmm9
-; AVX2-NEXT:    vblendps {{.*#+}} xmm6 = xmm9[0,1],xmm6[2,3]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 192(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 128(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 64(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, (%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 224(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 160(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 96(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 32(%rsi)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 192(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 128(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 64(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, (%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 224(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 160(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 96(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 32(%rdx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 192(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 128(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 64(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, (%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 224(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 160(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 96(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 32(%rcx)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, (%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 64(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 128(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 192(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 224(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 160(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 96(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 32(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 224(%r9)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 192(%r9)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 160(%r9)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 128(%r9)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 96(%r9)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 64(%r9)
-; AVX2-NEXT:    vmovups (%rsp), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 32(%r9)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, (%r9)
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-NEXT:    vbroadcastss 1648(%rdi), %ymm5
+; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1,2],xmm14[3]
+; AVX2-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm14 = mem[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
+; AVX2-NEXT:    # ymm14 = ymm14[0],mem[1],ymm14[2,3,4],mem[5],ymm14[6,7]
+; AVX2-NEXT:    vextractf128 $1, %ymm14, %xmm14
+; AVX2-NEXT:    vblendps {{.*#+}} xmm5 = xmm14[0,1],xmm5[2,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 192(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 128(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 64(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, (%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 224(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 160(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 96(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 32(%rsi)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 192(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 128(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 64(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, (%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 224(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 160(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 96(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 32(%rdx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 192(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 128(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 64(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, (%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 224(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 160(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 96(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 32(%rcx)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, (%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 64(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 128(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 192(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 224(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 160(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 96(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 32(%r8)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 224(%r9)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 192(%r9)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 160(%r9)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 128(%r9)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 96(%r9)
+; AVX2-NEXT:    vmovups (%rsp), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 64(%r9)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 32(%r9)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, (%r9)
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-NEXT:    vmovaps %ymm13, 224(%rax)
-; AVX2-NEXT:    vmovaps %ymm7, 192(%rax)
-; AVX2-NEXT:    vmovaps %ymm8, 160(%rax)
-; AVX2-NEXT:    vmovaps %ymm10, 128(%rax)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 96(%rax)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 64(%rax)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, 32(%rax)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm6, (%rax)
+; AVX2-NEXT:    vmovaps %ymm12, 224(%rax)
+; AVX2-NEXT:    vmovaps %ymm10, 192(%rax)
+; AVX2-NEXT:    vmovaps %ymm6, 160(%rax)
+; AVX2-NEXT:    vmovaps %ymm7, 128(%rax)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 96(%rax)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 64(%rax)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, 32(%rax)
+; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm5, (%rax)
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-NEXT:    vmovaps %ymm0, 224(%rax)
-; AVX2-NEXT:    vmovaps %ymm5, 192(%rax)
-; AVX2-NEXT:    vmovaps %ymm4, 160(%rax)
-; AVX2-NEXT:    vmovaps %ymm3, 128(%rax)
-; AVX2-NEXT:    vmovaps %ymm2, 96(%rax)
-; AVX2-NEXT:    vmovaps %ymm1, 64(%rax)
-; AVX2-NEXT:    vmovaps %ymm14, 32(%rax)
-; AVX2-NEXT:    vmovaps %ymm12, (%rax)
-; AVX2-NEXT:    addq $2680, %rsp # imm = 0xA78
+; AVX2-NEXT:    vmovaps %ymm4, 224(%rax)
+; AVX2-NEXT:    vmovaps %ymm3, 192(%rax)
+; AVX2-NEXT:    vmovaps %ymm2, 160(%rax)
+; AVX2-NEXT:    vmovaps %ymm1, 128(%rax)
+; AVX2-NEXT:    vmovaps %ymm0, 96(%rax)
+; AVX2-NEXT:    vmovaps %ymm8, 64(%rax)
+; AVX2-NEXT:    vmovaps %ymm13, 32(%rax)
+; AVX2-NEXT:    vmovaps %ymm11, (%rax)
+; AVX2-NEXT:    addq $2648, %rsp # imm = 0xA58
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: load_i32_stride7_vf64:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    subq $2680, %rsp # imm = 0xA78
-; AVX2-FP-NEXT:    vmovdqa 1216(%rdi), %ymm12
-; AVX2-FP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    subq $2648, %rsp # imm = 0xA58
+; AVX2-FP-NEXT:    vmovdqa 1216(%rdi), %ymm9
+; AVX2-FP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 1152(%rdi), %ymm4
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 1120(%rdi), %ymm5
-; AVX2-FP-NEXT:    vmovdqa 768(%rdi), %ymm13
-; AVX2-FP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 768(%rdi), %ymm12
+; AVX2-FP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 704(%rdi), %ymm6
 ; AVX2-FP-NEXT:    vmovdqa 672(%rdi), %ymm7
 ; AVX2-FP-NEXT:    vmovdqa 320(%rdi), %ymm8
@@ -12349,13 +12260,12 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3,4,5],ymm6[6],ymm7[7]
-; AVX2-FP-NEXT:    vmovdqa %ymm7, %ymm9
 ; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa %ymm6, %ymm8
 ; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vpbroadcastq 752(%rdi), %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm13[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm12[4,5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 800(%rdi), %xmm2
 ; AVX2-FP-NEXT:    vmovdqa 832(%rdi), %xmm3
@@ -12369,11 +12279,9 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm5[0,1,2,3,4,5],ymm4[6],ymm5[7]
 ; AVX2-FP-NEXT:    vmovdqa %ymm5, %ymm6
 ; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa %ymm4, %ymm5
-; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vpbroadcastq 1200(%rdi), %ymm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm12[4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm9[4,5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 1248(%rdi), %xmm2
 ; AVX2-FP-NEXT:    vmovdqa 1280(%rdi), %xmm3
@@ -12384,12 +12292,12 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 1600(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqa 1568(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
-; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm14
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 1600(%rdi), %ymm13
+; AVX2-FP-NEXT:    vmovdqa 1568(%rdi), %ymm3
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm13[6],ymm3[7]
+; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm5
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vmovdqa 1664(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -12409,11 +12317,11 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpbroadcastq 80(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm3
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7]
+; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6],ymm2[7]
 ; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %xmm2
@@ -12426,11 +12334,10 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 480(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqa 448(%rdi), %ymm13
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm13[0,1,2,3,4,5],ymm2[6],ymm13[7]
-; AVX2-FP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm12
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 448(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vmovdqa 544(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vmovdqu %ymm3, (%rsp) # 32-byte Spill
@@ -12446,12 +12353,11 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 928(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqa 928(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 896(%rdi), %ymm1
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 896(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm1[6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm15
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FP-NEXT:    vmovdqa 992(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -12467,16 +12373,16 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 1376(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 1344(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
+; AVX2-FP-NEXT:    vmovdqa 1376(%rdi), %ymm14
+; AVX2-FP-NEXT:    vmovdqa 1344(%rdi), %ymm15
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm15[0,1,2,3,4,5],ymm14[6],ymm15[7]
+; AVX2-FP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 1440(%rdi), %ymm7
+; AVX2-FP-NEXT:    vmovdqa 1440(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vpbroadcastq 1424(%rdi), %ymm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqa 1472(%rdi), %xmm1
 ; AVX2-FP-NEXT:    vmovdqa 1504(%rdi), %xmm2
@@ -12495,215 +12401,192 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm1[12,13,14,15],ymm0[0,1,2,3,4,5,6,7,8,9,10,11],ymm1[28,29,30,31],ymm0[16,17,18,19,20,21,22,23,24,25,26,27]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm1[2,3],ymm4[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FP-NEXT:    vmovdqa 288(%rdi), %ymm12
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm9[0,1],ymm12[2,3],ymm9[4,5],ymm12[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0],ymm11[1],ymm10[2,3,4],ymm11[5],ymm10[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 832(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 800(%rdi), %ymm1
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm0[5,6],ymm2[7]
+; AVX2-FP-NEXT:    vpmovsxbd {{.*#+}} ymm0 = [1,0,7,6,5,6,5,6]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 736(%rdi), %ymm10
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm11[0,1],ymm10[2,3],ymm11[4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0],ymm9[1],ymm8[2,3,4],ymm9[5],ymm8[6,7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqa 832(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 800(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 1280(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovdqa 736(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 1248(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 1184(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm11[0,1],ymm2[2,3],ymm11[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0],ymm7[1],ymm8[2,3,4],ymm7[5],ymm8[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0,1,2,3,4],ymm2[5,6],ymm8[7]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm9[0,1],ymm1[2,3],ymm9[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0],ymm6[1],ymm5[2,3,4],ymm6[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqa 1280(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 1248(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 1728(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovdqa 1184(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 1696(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0,1],ymm2[2,3],ymm10[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm7 = mem[0],ymm6[1],mem[2,3,4],ymm6[5],mem[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1,2,3,4],ymm2[5,6],ymm7[7]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 1632(%rdi), %ymm6
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm5[0,1],ymm6[2,3],ymm5[4,5],ymm6[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm14[0],mem[1],ymm14[2,3,4],mem[5],ymm14[6,7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqa 1728(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 1696(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 608(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovdqa 1632(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 576(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 512(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm2[2,3],ymm6[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0],ymm5[1],ymm13[2,3,4],ymm5[5],ymm13[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4],ymm2[5,6],ymm5[7]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu (%rsp), %ymm14 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm14[0,1],ymm1[2,3],ymm14[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0],ymm13[1],ymm12[2,3,4],ymm13[5],ymm12[6,7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqa 608(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 576(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 1056(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovdqa 512(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 1024(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 960(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqu (%rsp), %ymm8 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0,1],ymm2[2,3],ymm8[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1],ymm1[2,3],ymm8[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FP-NEXT:    vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0],ymm15[1],mem[2,3,4],ymm15[5],mem[6,7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqa 1056(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 1024(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 1504(%rdi), %ymm2
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovdqa 960(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 1472(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 1408(%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm13[0,1],ymm2[2,3],ymm13[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm1[2,3],ymm7[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = ymm2[0],mem[1],ymm2[2,3,4],mem[5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqa 1504(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 1472(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FP-NEXT:    vmovdqa 160(%rdi), %ymm2
-; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm3
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm13
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovdqa 1408(%rdi), %ymm2
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3],ymm4[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm14[0],ymm15[1],ymm14[2,3,4],ymm15[5],ymm14[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vmovdqa 160(%rdi), %ymm3
+; AVX2-FP-NEXT:    vmovdqa 128(%rdi), %ymm5
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm5[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm5[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FP-NEXT:    vmovdqa %ymm5, %ymm14
+; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa %ymm3, %ymm15
 ; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa %ymm2, %ymm12
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm3
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm3[2,3],ymm7[4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1],ymm4[2,3],ymm7[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,2,2,3,5,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0],ymm2[1],mem[2,3,4],ymm2[5],mem[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,3,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-FP-NEXT:    vpermd %ymm2, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 304(%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm4[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm4[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm9[8,9,10,11,12,13,14,15],ymm12[0,1,2,3,4,5,6,7],ymm9[24,25,26,27,28,29,30,31],ymm12[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 232(%rdi), %xmm1
-; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %xmm2
-; AVX2-FP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2,3]
+; AVX2-FP-NEXT:    vmovdqa 256(%rdi), %xmm5
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm5[1],xmm1[2,3]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = ymm4[0],mem[0],ymm4[2],mem[2]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm1[0],ymm12[0],ymm1[2],ymm12[2]
 ; AVX2-FP-NEXT:    vpbroadcastd 428(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 752(%rdi), %xmm0
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm11[8,9,10,11,12,13,14,15],ymm10[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm10[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm11[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 680(%rdi), %xmm1
 ; AVX2-FP-NEXT:    vmovdqa 704(%rdi), %xmm2
 ; AVX2-FP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2,3]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FP-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm1 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm1 = ymm11[0],mem[0],ymm11[2],mem[2]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[2],ymm3[2]
 ; AVX2-FP-NEXT:    vpbroadcastd 876(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 1200(%rdi), %xmm0
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm9[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm9[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm10[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm10[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 1128(%rdi), %xmm1
 ; AVX2-FP-NEXT:    vmovdqa 1152(%rdi), %xmm2
@@ -12718,7 +12601,8 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 1648(%rdi), %xmm0
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm5[8,9,10,11,12,13,14,15],ymm6[0,1,2,3,4,5,6,7],ymm5[24,25,26,27,28,29,30,31],ymm6[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm6[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm6[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 1576(%rdi), %xmm1
 ; AVX2-FP-NEXT:    vmovdqa 1600(%rdi), %xmm2
@@ -12732,26 +12616,26 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 80(%rdi), %xmm0
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm7[8,9,10,11,12,13,14,15],ymm3[0,1,2,3,4,5,6,7],ymm7[24,25,26,27,28,29,30,31],ymm3[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm7[8,9,10,11,12,13,14,15],ymm4[0,1,2,3,4,5,6,7],ymm7[24,25,26,27,28,29,30,31],ymm4[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 8(%rdi), %xmm1
-; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %xmm3
-; AVX2-FP-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm3[1],xmm1[2,3]
+; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %xmm4
+; AVX2-FP-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm4[1],xmm1[2,3]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm13[0],ymm12[0],ymm13[2],ymm12[2]
+; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm14[0],ymm15[0],ymm14[2],ymm15[2]
 ; AVX2-FP-NEXT:    vpbroadcastd 204(%rdi), %ymm6
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm6[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 528(%rdi), %xmm0
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm14[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm14[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm8[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm8[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 456(%rdi), %xmm1
-; AVX2-FP-NEXT:    vmovdqa 480(%rdi), %xmm3
-; AVX2-FP-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm3[1],xmm1[2,3]
+; AVX2-FP-NEXT:    vmovdqa 480(%rdi), %xmm4
+; AVX2-FP-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm4[1],xmm1[2,3]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-FP-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
@@ -12762,23 +12646,23 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 976(%rdi), %xmm0
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm8[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm8[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm13[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm13[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 904(%rdi), %xmm15
-; AVX2-FP-NEXT:    vmovdqa 928(%rdi), %xmm12
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm12[1],xmm15[2,3]
+; AVX2-FP-NEXT:    vmovdqa 928(%rdi), %xmm11
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm11[1],xmm15[2,3]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm8[0],ymm7[0],ymm8[2],ymm7[2]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FP-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm10[0],ymm7[0],ymm10[2],ymm7[2]
 ; AVX2-FP-NEXT:    vpbroadcastd 1100(%rdi), %ymm14
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm15[0,1,2,3,4,5,6],ymm14[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm14[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 1424(%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm14 = ymm10[8,9,10,11,12,13,14,15],ymm9[0,1,2,3,4,5,6,7],ymm10[24,25,26,27,28,29,30,31],ymm9[16,17,18,19,20,21,22,23]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FP-NEXT:    vpalignr {{.*#+}} ymm14 = ymm9[8,9,10,11,12,13,14,15],ymm8[0,1,2,3,4,5,6,7],ymm9[24,25,26,27,28,29,30,31],ymm8[16,17,18,19,20,21,22,23]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2],ymm0[3],ymm14[4,5,6,7]
 ; AVX2-FP-NEXT:    vpbroadcastd 1352(%rdi), %xmm15
 ; AVX2-FP-NEXT:    vmovdqa 1376(%rdi), %xmm0
@@ -12791,30 +12675,30 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2,3,4,5,6],ymm13[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm14[0,1,2,3,4],ymm13[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm13 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm13 = ymm3[0],mem[1],ymm3[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm3[0,1,2],mem[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm5 = xmm5[3,2,2,3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[3,1,1,0,7,5,5,4]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1],ymm13[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm4[0,2],ymm15[1,3],ymm4[4,6],ymm15[5,7]
+; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm13 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm13 = ymm4[0],mem[1],ymm4[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1,2],mem[3]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[3,2,2,3]
+; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,1,1,0,7,5,5,4]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1],ymm13[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm13 = ymm4[0,2],ymm12[1,3],ymm4[4,6],ymm12[5,7]
+; AVX2-FP-NEXT:    vmovaps %ymm4, %ymm12
 ; AVX2-FP-NEXT:    vbroadcastss 432(%rdi), %ymm14
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5,6],ymm14[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm13[5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = ymm3[0],mem[1],ymm3[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm3[0,1,2],mem[3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm13[5,6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = ymm4[0],mem[1],ymm4[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm4[0,1,2],mem[3]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm4 = xmm4[3,2,2,3]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[3,1,1,0,7,5,5,4]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm5[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps $216, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = ymm11[0,2],mem[1,3],ymm11[4,6],mem[5,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm14[0,2],ymm3[1,3],ymm14[4,6],ymm3[5,7]
 ; AVX2-FP-NEXT:    vbroadcastss 880(%rdi), %ymm13
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm13[7]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm5[5,6,7]
@@ -12827,9 +12711,9 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[3,2,2,3]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[3,1,1,0,7,5,5,4]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm13[0,2],ymm14[1,3],ymm13[4,6],ymm14[5,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vshufps $216, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = ymm4[0,2],mem[1,3],ymm4[4,6],mem[5,7]
 ; AVX2-FP-NEXT:    vbroadcastss 1328(%rdi), %ymm5
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm5[7]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7]
@@ -12841,19 +12725,20 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[3,2,2,3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[3,1,1,0,7,5,5,4]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps $216, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = ymm3[0,2],mem[1,3],ymm3[4,6],mem[5,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FP-NEXT:    vshufps $216, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = ymm13[0,2],mem[1,3],ymm13[4,6],mem[5,7]
 ; AVX2-FP-NEXT:    vbroadcastss 1776(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0],ymm9[1],ymm10[2,3,4,5,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0],ymm8[1],ymm9[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],mem[3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,2,2,3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[3,1,1,0,7,5,5,4]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm6[0,2],ymm1[1,3],ymm6[4,6],ymm1[5,7]
+; AVX2-FP-NEXT:    vmovaps %ymm1, %ymm9
 ; AVX2-FP-NEXT:    vbroadcastss 1552(%rdi), %ymm3
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
@@ -12861,13 +12746,13 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm12[0,1,2],mem[3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm11[0,1,2],mem[3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm8[0,2],ymm7[1,3],ymm8[4,6],ymm7[5,7]
-; AVX2-FP-NEXT:    vmovaps %ymm7, %ymm9
-; AVX2-FP-NEXT:    vmovaps %ymm8, %ymm11
+; AVX2-FP-NEXT:    vmovdqa %ymm10, %ymm8
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,2],ymm7[1,3],ymm10[4,6],ymm7[5,7]
+; AVX2-FP-NEXT:    vmovaps %ymm7, %ymm11
 ; AVX2-FP-NEXT:    vbroadcastss 1104(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
@@ -12880,9 +12765,9 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,2],ymm8[1,3],ymm6[4,6],ymm8[5,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm4[0,2],ymm7[1,3],ymm4[4,6],ymm7[5,7]
 ; AVX2-FP-NEXT:    vbroadcastss 656(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
@@ -12896,125 +12781,128 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2],ymm4[1,3],ymm3[4,6],ymm4[5,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2],ymm10[1,3],ymm3[4,6],ymm10[5,7]
 ; AVX2-FP-NEXT:    vbroadcastss 208(%rdi), %ymm2
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vbroadcastss 100(%rdi), %xmm0
-; AVX2-FP-NEXT:    vmovaps 64(%rdi), %xmm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[3]
-; AVX2-FP-NEXT:    vmovsd {{.*#+}} xmm7 = [4,3,0,0]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm7, %ymm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm2[0,1],xmm0[2,3]
-; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [0,7,0,7,0,7,0,7]
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm0, %ymm3
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 212(%rdi), %ymm4
+; AVX2-FP-NEXT:    vmovaps 64(%rdi), %xmm6
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm6[0,1,2],xmm0[3]
+; AVX2-FP-NEXT:    vmovsd {{.*#+}} xmm5 = [4,3,0,0]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm1, %ymm5, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
+; AVX2-FP-NEXT:    vbroadcastsd {{.*#+}} ymm15 = [0,7,0,7,0,7,0,7]
+; AVX2-FP-NEXT:    vpermps %ymm3, %ymm15, %ymm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 212(%rdi), %ymm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FP-NEXT:    vbroadcastss 324(%rdi), %xmm2
+; AVX2-FP-NEXT:    vmovaps 288(%rdi), %xmm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0,1,2],xmm2[3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3]
+; AVX2-FP-NEXT:    vpermps %ymm12, %ymm15, %ymm2
+; AVX2-FP-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = ymm2[0,1,2,3,4,5],mem[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 436(%rdi), %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FP-NEXT:    vbroadcastss 548(%rdi), %xmm3
+; AVX2-FP-NEXT:    vmovaps 512(%rdi), %xmm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3]
+; AVX2-FP-NEXT:    vpermps %ymm4, %ymm15, %ymm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm7[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 660(%rdi), %ymm4
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm2, %ymm7, %ymm3
-; AVX2-FP-NEXT:    vbroadcastss 324(%rdi), %xmm4
-; AVX2-FP-NEXT:    vmovaps 288(%rdi), %xmm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm2[0,1,2],xmm4[3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm3[0,1],xmm4[2,3]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 436(%rdi), %ymm5
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm5[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm3, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm3 = mem[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm3, %ymm7, %ymm4
-; AVX2-FP-NEXT:    vbroadcastss 548(%rdi), %xmm5
-; AVX2-FP-NEXT:    vmovaps 512(%rdi), %xmm3
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm3[0,1,2],xmm5[3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
-; AVX2-FP-NEXT:    vpermps %ymm6, %ymm0, %ymm5
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 660(%rdi), %ymm6
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm6[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm4 = mem[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm4, %ymm7, %ymm5
-; AVX2-FP-NEXT:    vbroadcastss 772(%rdi), %xmm6
-; AVX2-FP-NEXT:    vmovaps 736(%rdi), %xmm4
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm6 = xmm4[0,1,2],xmm6[3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1],xmm6[2,3]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm6 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FP-NEXT:    vbroadcastss 772(%rdi), %xmm4
+; AVX2-FP-NEXT:    vmovaps 736(%rdi), %xmm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm3[0,1,2],xmm4[3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm4[2,3]
+; AVX2-FP-NEXT:    vpermps %ymm14, %ymm15, %ymm4
+; AVX2-FP-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = ymm4[0,1,2,3,4,5],mem[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 884(%rdi), %ymm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FP-NEXT:    vbroadcastss 996(%rdi), %xmm7
+; AVX2-FP-NEXT:    vmovaps 960(%rdi), %xmm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm7 = xmm4[0,1,2],xmm7[3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm7[2,3]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm15, %ymm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 1108(%rdi), %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm5, %ymm0
+; AVX2-FP-NEXT:    vbroadcastss 1220(%rdi), %xmm7
+; AVX2-FP-NEXT:    vmovaps 1184(%rdi), %xmm14
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm7 = xmm14[0,1,2],xmm7[3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0,1],xmm7[2,3]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 884(%rdi), %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm5 = mem[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm5, %ymm7, %ymm6
-; AVX2-FP-NEXT:    vbroadcastss 996(%rdi), %xmm8
-; AVX2-FP-NEXT:    vmovaps 960(%rdi), %xmm5
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm8 = xmm5[0,1,2],xmm8[3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm6 = xmm6[0,1],xmm8[2,3]
-; AVX2-FP-NEXT:    vpermps %ymm11, %ymm0, %ymm8
+; AVX2-FP-NEXT:    vpermps %ymm10, %ymm15, %ymm7
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 1332(%rdi), %ymm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm0 = mem[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm0, %ymm5, %ymm7
+; AVX2-FP-NEXT:    vbroadcastss 1444(%rdi), %xmm8
+; AVX2-FP-NEXT:    vmovaps 1408(%rdi), %xmm0
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm8 = xmm0[0,1,2],xmm8[3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1],xmm8[2,3]
+; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm8 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    vmovaps %ymm9, %ymm11
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 1108(%rdi), %ymm9
+; AVX2-FP-NEXT:    vbroadcastss 1556(%rdi), %ymm9
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm6 = mem[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm6, %ymm7, %ymm8
-; AVX2-FP-NEXT:    vbroadcastss 1220(%rdi), %xmm9
-; AVX2-FP-NEXT:    vmovaps 1184(%rdi), %xmm6
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm9 = xmm6[0,1,2],xmm9[3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm8 = xmm8[0,1],xmm9[2,3]
-; AVX2-FP-NEXT:    vpermps %ymm13, %ymm0, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 1332(%rdi), %ymm12
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5,6],ymm12[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = mem[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm8, %ymm7, %ymm8
-; AVX2-FP-NEXT:    vbroadcastss 1444(%rdi), %xmm9
-; AVX2-FP-NEXT:    vmovaps 1408(%rdi), %xmm13
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm9 = xmm13[0,1,2],xmm9[3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm8 = xmm8[0,1],xmm9[2,3]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = ymm9[0,1,2,3,4,5],mem[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 1556(%rdi), %ymm12
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5,6],ymm12[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm8 = mem[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-FP-NEXT:    vpermps %ymm8, %ymm7, %ymm7
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm8[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm7 = mem[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FP-NEXT:    vpermps %ymm7, %ymm5, %ymm7
 ; AVX2-FP-NEXT:    vbroadcastss 1668(%rdi), %xmm8
-; AVX2-FP-NEXT:    vmovaps 1632(%rdi), %xmm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm8 = xmm14[0,1,2],xmm8[3]
+; AVX2-FP-NEXT:    vmovaps 1632(%rdi), %xmm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm8 = xmm5[0,1,2],xmm8[3]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm7 = xmm7[0,1],xmm8[2,3]
-; AVX2-FP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm8 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-FP-NEXT:    vpermps %ymm13, %ymm15, %ymm8
+; AVX2-FP-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm8 = ymm8[0,1,2,3,4,5],mem[6,7]
 ; AVX2-FP-NEXT:    vbroadcastss 1780(%rdi), %ymm9
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6],ymm9[7]
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm8[4,5,6,7]
@@ -13026,367 +12914,370 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
 ; AVX2-FP-NEXT:    vbroadcastss 216(%rdi), %ymm8
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vmovaps 96(%rdi), %xmm12
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm12[0,1,2],xmm1[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-FP-NEXT:    vmovaps 96(%rdi), %xmm9
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm6 = xmm9[0,1,2],xmm6[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm6 = xmm6[0,1,3,2]
 ; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm8 = mem[1,0,2,3,5,4,6,7]
 ; AVX2-FP-NEXT:    vextractf128 $1, %ymm8, %xmm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm8[0,1],xmm1[2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 320(%rdi), %xmm1
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0,1,2],xmm2[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,1,3,2]
-; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm7[0,1],xmm2[2,3]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = ymm15[0],mem[1],ymm15[2,3,4],mem[5],ymm15[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-FP-NEXT:    vbroadcastss 440(%rdi), %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 544(%rdi), %xmm2
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,1,3,2]
-; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm7[0,1],xmm3[2,3]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = ymm7[0],mem[1],ymm7[2,3,4],mem[5],ymm7[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-FP-NEXT:    vbroadcastss 664(%rdi), %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 768(%rdi), %xmm3
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm3[0,1,2],xmm4[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm4 = xmm4[0,1,3,2]
-; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm7[0,1],xmm4[2,3]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = ymm10[0],mem[1],ymm10[2,3,4],mem[5],ymm10[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-FP-NEXT:    vbroadcastss 888(%rdi), %ymm8
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovaps 992(%rdi), %xmm4
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm4[0,1,2],xmm5[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm5 = xmm5[0,1,3,2]
-; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm7[0,1],xmm5[2,3]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = ymm7[0],mem[1],ymm7[2,3,4],mem[5],ymm7[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-FP-NEXT:    vbroadcastss 1112(%rdi), %ymm10
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm10[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm5[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 1216(%rdi), %xmm15
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm15[0,1,2],xmm6[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm5 = xmm5[0,1,3,2]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm6 = xmm8[0,1],xmm6[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 320(%rdi), %xmm13
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm13[0,1,2],xmm1[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
 ; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX2-FP-NEXT:    # ymm6 = mem[1,0,2,3,5,4,6,7]
 ; AVX2-FP-NEXT:    vextractf128 $1, %ymm6, %xmm6
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm6[0,1],xmm1[2,3]
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm6 = ymm6[0],mem[1],ymm6[2,3,4],mem[5],ymm6[6,7]
+; AVX2-FP-NEXT:    vblendps $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm6 = mem[0],ymm6[1],mem[2,3,4],ymm6[5],mem[6,7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,0,3,3,5,4,7,7]
 ; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
-; AVX2-FP-NEXT:    vbroadcastss 1336(%rdi), %ymm7
+; AVX2-FP-NEXT:    vbroadcastss 440(%rdi), %ymm7
 ; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 1440(%rdi), %xmm5
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm6 = xmm5[0,1,2],xmm13[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm6 = xmm6[0,1,3,2]
-; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm6 = xmm7[0,1],xmm6[2,3]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FP-NEXT:    vblendps $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm7 = mem[0],ymm7[1],mem[2,3,4],ymm7[5],mem[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm7 = ymm7[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm7 = ymm7[0,1,0,3]
-; AVX2-FP-NEXT:    vbroadcastss 1560(%rdi), %ymm13
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm6[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FP-NEXT:    vmovaps 1664(%rdi), %xmm6
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm13 = xmm6[0,1,2],xmm14[3]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm13 = xmm13[0,1,3,2]
-; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm14, %xmm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm13 = xmm14[0,1],xmm13[2,3]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm14 = ymm14[1,0,3,3,5,4,7,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm14 = ymm14[0,1,0,3]
-; AVX2-FP-NEXT:    vbroadcastss 1784(%rdi), %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm14[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm13[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastss 136(%rdi), %xmm9
-; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9, %xmm9 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm9 = xmm9[0],mem[1],xmm9[2,3]
-; AVX2-FP-NEXT:    vinsertf128 $1, %xmm9, %ymm0, %ymm9
-; AVX2-FP-NEXT:    vpermps 192(%rdi), %ymm0, %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 80(%rdi), %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm12 = xmm14[0,1,2],xmm12[3]
-; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = ymm14[0],mem[1],ymm14[2,3,4],mem[5],ymm14[6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm14, %xmm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm12 = xmm14[0,1],xmm12[2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm12[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastss 360(%rdi), %xmm9
-; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9, %xmm9 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm9 = xmm9[0],mem[1],xmm9[2,3]
-; AVX2-FP-NEXT:    vinsertf128 $1, %xmm9, %ymm0, %ymm9
-; AVX2-FP-NEXT:    vpermps 416(%rdi), %ymm0, %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm14[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 304(%rdi), %ymm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm14[0,1,2],xmm1[3]
-; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm14 = ymm14[0],mem[1],ymm14[2,3,4],mem[5],ymm14[6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm14, %xmm14
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm14[0,1],xmm1[2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm14 = ymm1[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastss 584(%rdi), %xmm1
-; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm1 = xmm1[0],mem[1],xmm1[2,3]
-; AVX2-FP-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FP-NEXT:    vpermps 640(%rdi), %ymm0, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 528(%rdi), %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm9[0,1,2],xmm2[3]
-; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm9, %xmm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm9[0,1],xmm2[2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastss 808(%rdi), %xmm2
-; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm2 = xmm2[0],mem[1],xmm2[2,3]
-; AVX2-FP-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FP-NEXT:    vpermps 864(%rdi), %ymm0, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 752(%rdi), %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm9[0,1,2],xmm3[3]
-; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm9, %xmm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm9[0,1],xmm3[2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastss 1032(%rdi), %xmm3
-; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
-; AVX2-FP-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FP-NEXT:    vpermps 1088(%rdi), %ymm0, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 976(%rdi), %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm9[0,1,2],xmm4[3]
-; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm9, %xmm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm9[0,1],xmm4[2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastss 1256(%rdi), %xmm4
-; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
-; AVX2-FP-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-FP-NEXT:    vpermps 1312(%rdi), %ymm0, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 1200(%rdi), %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm9 = xmm9[0,1,2],xmm15[3]
-; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm15 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm15 = ymm15[0],mem[1],ymm15[2,3,4],mem[5],ymm15[6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm15, %xmm15
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm9 = xmm15[0,1],xmm9[2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm9[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastss 1480(%rdi), %xmm9
-; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9, %xmm9 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm9 = xmm9[0],mem[1],xmm9[2,3]
-; AVX2-FP-NEXT:    vinsertf128 $1, %xmm9, %ymm0, %ymm9
-; AVX2-FP-NEXT:    vpermps 1536(%rdi), %ymm0, %ymm15
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 1424(%rdi), %ymm15
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm15[0,1,2],xmm5[3]
-; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm15 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm15 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm15 = ymm15[0],mem[1],ymm15[2,3,4],mem[5],ymm15[6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm15, %xmm15
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm15[0,1],xmm5[2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FP-NEXT:    vbroadcastss 1704(%rdi), %xmm9
-; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9, %xmm9 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm9 = xmm9[0],mem[1],xmm9[2,3]
-; AVX2-FP-NEXT:    vpermps 1760(%rdi), %ymm0, %ymm0
-; AVX2-FP-NEXT:    vinsertf128 $1, %xmm9, %ymm0, %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm9[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vbroadcastss 1648(%rdi), %ymm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm6 = xmm9[0,1,2],xmm6[3]
-; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
-; AVX2-FP-NEXT:    # ymm9 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
-; AVX2-FP-NEXT:    vextractf128 $1, %ymm9, %xmm9
-; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm6 = xmm9[0,1],xmm6[2,3]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 128(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, (%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 224(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 160(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%rsi)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 128(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, (%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 224(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 160(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%rdx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 128(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, (%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 224(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 160(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%rcx)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, (%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 128(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 224(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 160(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%r8)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 224(%r9)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 192(%r9)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 160(%r9)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 128(%r9)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%r9)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%r9)
-; AVX2-FP-NEXT:    vmovups (%rsp), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%r9)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, (%r9)
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 544(%rdi), %xmm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm8[0,1,2],xmm2[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm2, %xmm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm2 = ymm2[0],mem[1],ymm2[2,3,4],mem[5],ymm2[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[1,0,3,3,5,4,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[0,1,0,3]
+; AVX2-FP-NEXT:    vbroadcastss 664(%rdi), %ymm6
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 768(%rdi), %xmm1
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0,1,2],xmm3[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,1,3,2]
+; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm3, %xmm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm3 = mem[0],ymm3[1],mem[2,3,4],ymm3[5],mem[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[1,0,3,3,5,4,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,3]
+; AVX2-FP-NEXT:    vbroadcastss 888(%rdi), %ymm6
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovaps 992(%rdi), %xmm2
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm2[0,1,2],xmm4[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,1,3,2]
+; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm4, %xmm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm4[0,1],xmm3[2,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm4 = ymm4[0],mem[1],ymm4[2,3,4],mem[5],ymm4[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[1,0,3,3,5,4,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm4 = ymm4[0,1,0,3]
+; AVX2-FP-NEXT:    vbroadcastss 1112(%rdi), %ymm6
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm7 = ymm3[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 1216(%rdi), %xmm3
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm3[0,1,2],xmm14[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm4 = xmm4[0,1,3,2]
+; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm6 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm6, %xmm6
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm12[0],ymm10[1],ymm12[2,3,4],ymm10[5],ymm12[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm6 = ymm6[1,0,3,3,5,4,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm6 = ymm6[0,1,0,3]
+; AVX2-FP-NEXT:    vbroadcastss 1336(%rdi), %ymm10
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm10[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm6 = ymm4[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 1440(%rdi), %xmm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm4[0,1,2],xmm0[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,1,3,2]
+; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm10 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm10, %xmm10
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm10[0,1],xmm0[2,3]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm10 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm10 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm10 = ymm10[1,0,3,3,5,4,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm10 = ymm10[0,1,0,3]
+; AVX2-FP-NEXT:    vbroadcastss 1560(%rdi), %ymm12
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm10 = ymm0[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-FP-NEXT:    vmovaps 1664(%rdi), %xmm14
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm14[0,1,2],xmm5[3]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,1,3,2]
+; AVX2-FP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm0 = xmm5[0,1],xmm0[2,3]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[1,0,3,3,5,4,7,7]
+; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[0,1,0,3]
+; AVX2-FP-NEXT:    vbroadcastss 1784(%rdi), %ymm12
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastss 136(%rdi), %xmm0
+; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-FP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpermps 192(%rdi), %ymm15, %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 80(%rdi), %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1,2],xmm9[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm11, %xmm11
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm11[0,1],xmm5[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm11 = ymm5[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastss 360(%rdi), %xmm0
+; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-FP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpermps 416(%rdi), %ymm15, %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 304(%rdi), %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1,2],xmm13[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm13 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm13 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm13 = ymm13[0],mem[1],ymm13[2,3,4],mem[5],ymm13[6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm13, %xmm13
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm13[0,1],xmm5[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm13 = ymm5[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastss 584(%rdi), %xmm0
+; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-FP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpermps 640(%rdi), %ymm15, %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 528(%rdi), %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1,2],xmm8[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm8 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm8 = ymm8[0],mem[1],ymm8[2,3,4],mem[5],ymm8[6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm8, %xmm8
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm8[0,1],xmm5[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm5[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastss 808(%rdi), %xmm0
+; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-FP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FP-NEXT:    vpermps 864(%rdi), %ymm15, %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 752(%rdi), %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm5[0,1,2],xmm1[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm1 = xmm5[0,1],xmm1[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastss 1032(%rdi), %xmm1
+; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    # xmm1 = xmm1[0],mem[1],xmm1[2,3]
+; AVX2-FP-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FP-NEXT:    vpermps 1088(%rdi), %ymm15, %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 976(%rdi), %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm5[0,1,2],xmm2[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm2 = xmm5[0,1],xmm2[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastss 1256(%rdi), %xmm2
+; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    # xmm2 = xmm2[0],mem[1],xmm2[2,3]
+; AVX2-FP-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-FP-NEXT:    vpermps 1312(%rdi), %ymm15, %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 1200(%rdi), %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm5[0,1,2],xmm3[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastss 1480(%rdi), %xmm3
+; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
+; AVX2-FP-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-FP-NEXT:    vpermps 1536(%rdi), %ymm15, %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 1424(%rdi), %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm5, %xmm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm4 = xmm5[0,1],xmm4[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FP-NEXT:    vbroadcastss 1704(%rdi), %xmm4
+; AVX2-FP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
+; AVX2-FP-NEXT:    vpermps 1760(%rdi), %ymm15, %ymm5
+; AVX2-FP-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vbroadcastss 1648(%rdi), %ymm5
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1,2],xmm14[3]
+; AVX2-FP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm14 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
+; AVX2-FP-NEXT:    # ymm14 = ymm14[0],mem[1],ymm14[2,3,4],mem[5],ymm14[6,7]
+; AVX2-FP-NEXT:    vextractf128 $1, %ymm14, %xmm14
+; AVX2-FP-NEXT:    vblendps {{.*#+}} xmm5 = xmm14[0,1],xmm5[2,3]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 192(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 128(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 64(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, (%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 224(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 160(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 96(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 32(%rsi)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 192(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 128(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 64(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, (%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 224(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 160(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 96(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 32(%rdx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 192(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 128(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 64(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, (%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 224(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 160(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 96(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 32(%rcx)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, (%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 64(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 128(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 192(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 224(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 160(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 96(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 32(%r8)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 224(%r9)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 192(%r9)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 160(%r9)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 128(%r9)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 96(%r9)
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 64(%r9)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 32(%r9)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, (%r9)
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FP-NEXT:    vmovaps %ymm13, 224(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm7, 192(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm8, 160(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm10, 128(%rax)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 96(%rax)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 64(%rax)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, 32(%rax)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm6, (%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm12, 224(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm10, 192(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm6, 160(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm7, 128(%rax)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 96(%rax)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 64(%rax)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, 32(%rax)
+; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm5, (%rax)
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FP-NEXT:    vmovaps %ymm0, 224(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm5, 192(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm4, 160(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm3, 128(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm2, 96(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm1, 64(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm14, 32(%rax)
-; AVX2-FP-NEXT:    vmovaps %ymm12, (%rax)
-; AVX2-FP-NEXT:    addq $2680, %rsp # imm = 0xA78
+; AVX2-FP-NEXT:    vmovaps %ymm4, 224(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm3, 192(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm2, 160(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm1, 128(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm0, 96(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm8, 64(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm13, 32(%rax)
+; AVX2-FP-NEXT:    vmovaps %ymm11, (%rax)
+; AVX2-FP-NEXT:    addq $2648, %rsp # imm = 0xA58
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
 ; AVX2-FCP-LABEL: load_i32_stride7_vf64:
 ; AVX2-FCP:       # %bb.0:
-; AVX2-FCP-NEXT:    subq $2680, %rsp # imm = 0xA78
-; AVX2-FCP-NEXT:    vmovdqa 1216(%rdi), %ymm13
+; AVX2-FCP-NEXT:    subq $2648, %rsp # imm = 0xA58
+; AVX2-FCP-NEXT:    vmovdqa 1216(%rdi), %ymm9
+; AVX2-FCP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 1152(%rdi), %ymm4
 ; AVX2-FCP-NEXT:    vmovdqa 1120(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vmovdqa 768(%rdi), %ymm8
-; AVX2-FCP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa 768(%rdi), %ymm13
 ; AVX2-FCP-NEXT:    vmovdqa 704(%rdi), %ymm6
 ; AVX2-FCP-NEXT:    vmovdqa 672(%rdi), %ymm7
-; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %ymm12
+; AVX2-FCP-NEXT:    vmovdqa 320(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 224(%rdi), %ymm10
+; AVX2-FCP-NEXT:    vmovdqa 224(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} xmm0 = [0,7,6,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm10[0,1,2,3,4,5],ymm2[6],ymm10[7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa %ymm2, %ymm11
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm11
+; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa %ymm2, %ymm10
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vpbroadcastq 304(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm12[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm8[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %xmm2
 ; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %xmm3
@@ -13399,11 +13290,12 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3,4,5],ymm6[6],ymm7[7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa %ymm6, %ymm9
+; AVX2-FCP-NEXT:    vmovdqa %ymm6, %ymm8
 ; AVX2-FCP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vpbroadcastq 752(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm8[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm13[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm13, %ymm6
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 800(%rdi), %xmm2
 ; AVX2-FCP-NEXT:    vmovdqa 832(%rdi), %xmm3
@@ -13415,14 +13307,10 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm5[0,1,2,3,4,5],ymm4[6],ymm5[7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm5, %ymm6
-; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa %ymm4, %ymm8
 ; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vpbroadcastq 1200(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm13[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm13, %ymm15
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm9[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 1248(%rdi), %xmm2
 ; AVX2-FCP-NEXT:    vmovdqa 1280(%rdi), %xmm3
@@ -13433,13 +13321,12 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 1600(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqa 1600(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 1568(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm13
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5],ymm1[6],ymm3[7]
+; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm5
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa %ymm2, %ymm5
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vmovdqa 1664(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -13456,14 +13343,14 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 96(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpbroadcastq 80(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm2, %ymm14
-; AVX2-FCP-NEXT:    vmovdqa (%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqa (%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7]
+; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %xmm2
@@ -13475,11 +13362,11 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 480(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 448(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqa 480(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6],ymm2[7]
+; AVX2-FCP-NEXT:    vmovdqa 448(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
 ; AVX2-FCP-NEXT:    vmovdqa 544(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
@@ -13501,10 +13388,10 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vmovdqa 992(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa 992(%rdi), %ymm12
 ; AVX2-FCP-NEXT:    vpbroadcastq 976(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm12[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm12, (%rsp) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 1024(%rdi), %xmm2
 ; AVX2-FCP-NEXT:    vmovdqa 1056(%rdi), %xmm3
@@ -13515,16 +13402,16 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 1376(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 1344(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 1376(%rdi), %ymm15
+; AVX2-FCP-NEXT:    vmovdqa 1344(%rdi), %ymm14
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm14[0,1,2,3,4,5],ymm15[6],ymm14[7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vmovdqa 1440(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa 1440(%rdi), %ymm9
 ; AVX2-FCP-NEXT:    vpbroadcastq 1424(%rdi), %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm9[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqa 1472(%rdi), %xmm1
 ; AVX2-FCP-NEXT:    vmovdqa 1504(%rdi), %xmm2
@@ -13536,197 +13423,184 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm0 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %ymm2
-; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %ymm1
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm3
+; AVX2-FCP-NEXT:    vmovdqa 384(%rdi), %ymm1
 ; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm2[12,13,14,15],ymm1[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm1[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vmovdqa 352(%rdi), %ymm0
+; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm1[12,13,14,15],ymm0[0,1,2,3,4,5,6,7,8,9,10,11],ymm1[28,29,30,31],ymm0[16,17,18,19,20,21,22,23,24,25,26,27]
 ; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FCP-NEXT:    vmovdqa 288(%rdi), %ymm0
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm12[0,1],ymm0[2,3],ymm12[4,5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vpbroadcastq {{.*#+}} ymm0 = [5,6,5,6,5,6,5,6]
-; AVX2-FCP-NEXT:    vpermd %ymm1, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [1,0,7,7,5,4,7,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm11[0],ymm10[1],ymm11[2,3,4],ymm10[5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm13[0,1],ymm0[2,3],ymm13[4,5],ymm0[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm10[0],ymm11[1],ymm10[2,3,4],ymm11[5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm0[5,6],ymm2[7]
+; AVX2-FCP-NEXT:    vpmovsxbd {{.*#+}} ymm0 = [1,0,7,6,5,6,5,6]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 832(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 832(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 800(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm4[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 736(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm11[0,1],ymm3[2,3],ymm11[4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm9[0],ymm7[1],ymm9[2,3,4],ymm7[5],ymm9[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm2[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm2[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 736(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 1280(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 1248(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm2[2,3],ymm6[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm8[0],ymm7[1],ymm8[2,3,4],ymm7[5],ymm8[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm4[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 1184(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 1280(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa %ymm15, %ymm10
-; AVX2-FCP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm15[0,1],ymm3[2,3],ymm15[4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm8[0],ymm6[1],ymm8[2,3,4],ymm6[5],ymm8[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 1248(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 1728(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 1696(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 1184(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm8[0,1],ymm2[2,3],ymm8[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = ymm4[0],mem[1],ymm4[2,3,4],mem[5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1,2,3,4],ymm2[5,6],ymm7[7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 1728(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm4[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 1632(%rdi), %ymm15
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm6[0,1],ymm15[2,3],ymm6[4,5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0],ymm13[1],ymm5[2,3,4],ymm13[5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 1696(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 160(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 1632(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1],ymm2[2,3],ymm7[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = mem[0],ymm5[1],mem[2,3,4],ymm5[5],mem[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4],ymm2[5,6],ymm5[7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 608(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm4[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqa %ymm14, %ymm8
-; AVX2-FCP-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm14[0,1],ymm4[2,3],ymm14[4,5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm4, %ymm14
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0],ymm4[1],mem[2,3,4],ymm4[5],mem[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 576(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 608(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 576(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vmovdqu %ymm3, (%rsp) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm4[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 512(%rdi), %ymm13
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm9[0,1],ymm13[2,3],ymm9[4,5],ymm13[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0],ymm4[1],mem[2,3,4],ymm4[5],mem[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 512(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 1056(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 1024(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm11[0,1],ymm2[2,3],ymm11[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm4[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 960(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 1056(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm7[0,1],ymm3[2,3],ymm7[4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = ymm4[0],mem[1],ymm4[2,3,4],mem[5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqa 1024(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm2 = mem[2,2,2,2]
-; AVX2-FCP-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vmovdqa 1504(%rdi), %ymm4
-; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqa 1472(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 960(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm2[2,3],ymm12[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 1504(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm3 = ymm4[12,13,14,15],ymm3[0,1,2,3,4,5,6,7,8,9,10,11],ymm4[28,29,30,31],ymm3[16,17,18,19,20,21,22,23,24,25,26,27]
-; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,2,0]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FCP-NEXT:    vmovdqa 1408(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovdqa 1472(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm2[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm2[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 1408(%rdi), %ymm2
+; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm2[2,3],ymm9[4,5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm15[0],ymm14[1],ymm15[2,3,4],ymm14[5],ymm15[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3],ymm5[4,5],ymm3[6,7]
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = mem[2,2,2,2]
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FCP-NEXT:    vmovdqa 160(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vmovdqa 128(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm2 = ymm3[12,13,14,15],ymm5[0,1,2,3,4,5,6,7,8,9,10,11],ymm3[28,29,30,31],ymm5[16,17,18,19,20,21,22,23,24,25,26,27]
+; AVX2-FCP-NEXT:    vmovdqa %ymm5, %ymm14
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovdqa %ymm3, %ymm15
+; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,2,0]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
+; AVX2-FCP-NEXT:    vmovdqa 64(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm4[2,3],ymm12[4,5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpermd %ymm3, %ymm1, %ymm1
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3,4],ymm2[5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vpermd %ymm2, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 304(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm12[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm12[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm13[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm13[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastd 232(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vmovdqa 256(%rdi), %xmm5
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm5[1],xmm1[2,3]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm12[0],ymm4[0],ymm12[2],ymm4[2]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm10[0],ymm9[0],ymm10[2],ymm9[2]
 ; AVX2-FCP-NEXT:    vpbroadcastd 428(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 752(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm11[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm6[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm6[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastd 680(%rdi), %xmm1
 ; AVX2-FCP-NEXT:    vmovdqa 704(%rdi), %xmm2
@@ -13734,30 +13608,31 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2,3]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[2],ymm3[2]
+; AVX2-FCP-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = ymm3[0],mem[0],ymm3[2],mem[2]
 ; AVX2-FCP-NEXT:    vpbroadcastd 876(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 1200(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm10[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm10[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm8[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm8[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastd 1128(%rdi), %xmm1
 ; AVX2-FCP-NEXT:    vmovdqa 1152(%rdi), %xmm2
 ; AVX2-FCP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2,3]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = ymm1[0],mem[0],ymm1[2],mem[2]
+; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm1[0],ymm8[0],ymm1[2],ymm8[2]
 ; AVX2-FCP-NEXT:    vpbroadcastd 1324(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 1648(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm6[8,9,10,11,12,13,14,15],ymm15[0,1,2,3,4,5,6,7],ymm6[24,25,26,27,28,29,30,31],ymm15[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm7[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm7[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastd 1576(%rdi), %xmm1
 ; AVX2-FCP-NEXT:    vmovdqa 1600(%rdi), %xmm2
@@ -13771,29 +13646,28 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 80(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm8[8,9,10,11,12,13,14,15],ymm14[0,1,2,3,4,5,6,7],ymm8[24,25,26,27,28,29,30,31],ymm14[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm12[8,9,10,11,12,13,14,15],ymm4[0,1,2,3,4,5,6,7],ymm12[24,25,26,27,28,29,30,31],ymm4[16,17,18,19,20,21,22,23]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastd 8(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %xmm6
-; AVX2-FCP-NEXT:    vmovdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm6[1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vmovdqa 32(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm4[1],xmm1[2,3]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = ymm1[0],mem[0],ymm1[2],mem[2]
+; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm1 = ymm14[0],ymm15[0],ymm14[2],ymm15[2]
 ; AVX2-FCP-NEXT:    vpbroadcastd 204(%rdi), %ymm6
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm6[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 528(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm9[8,9,10,11,12,13,14,15],ymm13[0,1,2,3,4,5,6,7],ymm9[24,25,26,27,28,29,30,31],ymm13[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm11[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm11[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastd 456(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vmovdqa 480(%rdi), %xmm6
-; AVX2-FCP-NEXT:    vmovdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm6[1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vmovdqa 480(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm4[1],xmm1[2,3]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm1 = ymm1[0],mem[0],ymm1[2],mem[2]
 ; AVX2-FCP-NEXT:    vpbroadcastd 652(%rdi), %ymm15
@@ -13802,48 +13676,48 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 976(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm1 = ymm7[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],ymm7[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpalignr $8, (%rsp), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm1 = mem[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],mem[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastd 904(%rdi), %xmm15
-; AVX2-FCP-NEXT:    vmovdqa 928(%rdi), %xmm11
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm11[1],xmm15[2,3]
+; AVX2-FCP-NEXT:    vmovdqa 928(%rdi), %xmm12
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm12[1],xmm15[2,3]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm10[0],ymm7[0],ymm10[2],ymm7[2]
-; AVX2-FCP-NEXT:    vpbroadcastd 1100(%rdi), %ymm13
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm13[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpunpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm15 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm15 = ymm1[0],mem[0],ymm1[2],mem[2]
+; AVX2-FCP-NEXT:    vpbroadcastd 1100(%rdi), %ymm14
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm15[0,1,2,3,4,5,6],ymm14[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm14[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovdqa 1424(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vmovdqa %ymm5, %ymm9
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm13 = ymm5[8,9,10,11,12,13,14,15],ymm8[0,1,2,3,4,5,6,7],ymm5[24,25,26,27,28,29,30,31],ymm8[16,17,18,19,20,21,22,23]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2],ymm0[3],ymm13[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpalignr {{.*#+}} ymm14 = ymm7[8,9,10,11,12,13,14,15],ymm6[0,1,2,3,4,5,6,7],ymm7[24,25,26,27,28,29,30,31],ymm6[16,17,18,19,20,21,22,23]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2],ymm0[3],ymm14[4,5,6,7]
 ; AVX2-FCP-NEXT:    vpbroadcastd 1352(%rdi), %xmm15
 ; AVX2-FCP-NEXT:    vmovdqa 1376(%rdi), %xmm0
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm15 = xmm15[0],xmm0[1],xmm15[2,3]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm15[0,1],ymm14[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm6[0],ymm1[0],ymm6[2],ymm1[2]
-; AVX2-FCP-NEXT:    vpbroadcastd 1548(%rdi), %ymm14
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm15[0,1,2,3,4,5,6],ymm14[7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm13[0,1,2,3,4],ymm14[5,6,7]
+; AVX2-FCP-NEXT:    vpunpcklqdq {{.*#+}} ymm15 = ymm11[0],ymm1[0],ymm11[2],ymm1[2]
+; AVX2-FCP-NEXT:    vpbroadcastd 1548(%rdi), %ymm13
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm15[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm13 = ymm14[0,1,2,3,4],ymm13[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm13 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm13 = ymm5[0],mem[1],ymm5[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1,2],mem[3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm5 = xmm5[3,2,2,3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm13 = ymm13[3,1,1,0,7,5,5,4]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1],ymm13[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm13 = ymm12[0,2],ymm4[1,3],ymm12[4,6],ymm4[5,7]
+; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm13 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm13 = ymm4[0],mem[1],ymm4[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1,2],mem[3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[3,2,2,3]
+; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[3,1,1,0,7,5,5,4]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1],ymm13[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm13 = ymm10[0,2],ymm9[1,3],ymm10[4,6],ymm9[5,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 432(%rdi), %ymm14
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5,6],ymm14[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm13[5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm13[5,6,7]
+; AVX2-FCP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm5 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm5 = ymm4[0],mem[1],ymm4[2,3,4,5,6,7]
@@ -13852,11 +13726,10 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm4 = xmm4[3,2,2,3]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[3,1,1,0,7,5,5,4]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm5[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm5 = ymm15[0,2],ymm3[1,3],ymm15[4,6],ymm3[5,7]
-; AVX2-FCP-NEXT:    vmovaps %ymm3, %ymm13
-; AVX2-FCP-NEXT:    vbroadcastss 880(%rdi), %ymm12
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm5 = ymm3[0,2],ymm10[1,3],ymm3[4,6],ymm10[5,7]
+; AVX2-FCP-NEXT:    vbroadcastss 880(%rdi), %ymm13
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm13[7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4],ymm5[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
@@ -13868,8 +13741,8 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm4[3,1,1,0,7,5,5,4]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm14[0,2],ymm12[1,3],ymm14[4,6],ymm12[5,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm4 = ymm14[0,2],ymm8[1,3],ymm14[4,6],ymm8[5,7]
+; AVX2-FCP-NEXT:    vmovaps %ymm8, %ymm13
 ; AVX2-FCP-NEXT:    vbroadcastss 1328(%rdi), %ymm5
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm5[7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7]
@@ -13888,26 +13761,26 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0],ymm8[1],ymm9[2,3,4,5,6,7]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0],ymm6[1],ymm7[2,3,4,5,6,7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],mem[3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,2,2,3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[3,1,1,0,7,5,5,4]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm6[0,2],ymm1[1,3],ymm6[4,6],ymm1[5,7]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm2 = ymm11[0,2],ymm1[1,3],ymm11[4,6],ymm1[5,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 1552(%rdi), %ymm3
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
 ; AVX2-FCP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovdqu (%rsp), %ymm0 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vpblendd $2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm0 = ymm0[0],mem[1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm11[0,1,2],mem[3]
+; AVX2-FCP-NEXT:    vpblendd {{.*#+}} xmm1 = xmm12[0,1,2],mem[3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-FCP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovdqa %ymm10, %ymm8
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm10[0,2],ymm7[1,3],ymm10[4,6],ymm7[5,7]
-; AVX2-FCP-NEXT:    vmovaps %ymm7, %ymm10
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm11[0,2],ymm8[1,3],ymm11[4,6],ymm8[5,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 1104(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
@@ -13920,7 +13793,7 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm6[0,2],ymm7[1,3],ymm6[4,6],ymm7[5,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 656(%rdi), %ymm2
@@ -13935,285 +13808,327 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[3,2,2,3]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[3,1,1,0,7,5,5,4]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm3[0,2],ymm5[1,3],ymm3[4,6],ymm5[5,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} ymm1 = ymm4[0,2],ymm12[1,3],ymm4[4,6],ymm12[5,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 208(%rdi), %ymm2
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vbroadcastss 100(%rdi), %xmm0
-; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %xmm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm0 = xmm9[0,1,2],xmm0[3]
-; AVX2-FCP-NEXT:    vmovsd {{.*#+}} xmm4 = [4,3,0,0]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3]
-; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm0 = [0,7,0,7,0,7,0,7]
-; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 212(%rdi), %ymm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vmovaps 64(%rdi), %xmm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0,1,2],xmm0[3]
+; AVX2-FCP-NEXT:    vmovsd {{.*#+}} xmm0 = [4,3,0,0]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vbroadcastsd {{.*#+}} ymm15 = [0,7,0,7,0,7,0,7]
+; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm15, %ymm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 212(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm4[7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
-; AVX2-FCP-NEXT:    vbroadcastss 324(%rdi), %xmm3
-; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm3 = ymm3[0,1,2,3,4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm0, %ymm2
+; AVX2-FCP-NEXT:    vbroadcastss 324(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vmovaps 288(%rdi), %xmm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm4 = xmm1[0,1,2],xmm4[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm2 = xmm2[0,1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm9[6,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 436(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm5[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = mem[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm0, %ymm4
 ; AVX2-FCP-NEXT:    vbroadcastss 548(%rdi), %xmm5
-; AVX2-FCP-NEXT:    vmovaps 512(%rdi), %xmm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm3[0,1,2],xmm5[3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1],xmm5[2,3]
-; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm0, %ymm5
+; AVX2-FCP-NEXT:    vmovaps 512(%rdi), %xmm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm2[0,1,2],xmm5[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm15, %ymm5
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5],ymm7[6,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 660(%rdi), %ymm6
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm6[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = mem[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm0, %ymm5
 ; AVX2-FCP-NEXT:    vbroadcastss 772(%rdi), %xmm6
-; AVX2-FCP-NEXT:    vmovaps 736(%rdi), %xmm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm6 = xmm5[0,1,2],xmm6[3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1],xmm6[2,3]
-; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm0, %ymm6
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-FCP-NEXT:    vmovaps 736(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm6 = xmm4[0,1,2],xmm6[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1],xmm6[2,3]
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5],ymm10[6,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 884(%rdi), %ymm7
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = ymm1[0,1,2,3],mem[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm5, (%rsp) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = mem[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm0, %ymm6
 ; AVX2-FCP-NEXT:    vbroadcastss 996(%rdi), %xmm7
-; AVX2-FCP-NEXT:    vmovaps 960(%rdi), %xmm6
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm7 = xmm6[0,1,2],xmm7[3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1],xmm7[2,3]
-; AVX2-FCP-NEXT:    vpermps %ymm8, %ymm0, %ymm7
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FCP-NEXT:    vmovaps 960(%rdi), %xmm5
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm7 = xmm5[0,1,2],xmm7[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm6 = xmm6[0,1],xmm7[2,3]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm15, %ymm7
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm8[6,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 1108(%rdi), %ymm8
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
-; AVX2-FCP-NEXT:    vbroadcastss 1220(%rdi), %xmm8
-; AVX2-FCP-NEXT:    vmovaps 1184(%rdi), %xmm7
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm8 = xmm7[0,1,2],xmm8[3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1],xmm8[2,3]
-; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm0, %ymm8
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm12[6,7]
-; AVX2-FCP-NEXT:    vmovaps %ymm12, %ymm15
-; AVX2-FCP-NEXT:    vbroadcastss 1332(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6],ymm10[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
-; AVX2-FCP-NEXT:    vbroadcastss 1444(%rdi), %xmm10
-; AVX2-FCP-NEXT:    vmovaps 1408(%rdi), %xmm8
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm10 = xmm8[0,1,2],xmm10[3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1],xmm10[2,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm0, %ymm10
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = mem[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm0, %ymm6
+; AVX2-FCP-NEXT:    vbroadcastss 1220(%rdi), %xmm7
+; AVX2-FCP-NEXT:    vmovaps 1184(%rdi), %xmm10
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm7 = xmm10[0,1,2],xmm7[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm6 = xmm6[0,1],xmm7[2,3]
+; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm15, %ymm7
+; AVX2-FCP-NEXT:    vmovaps %ymm13, %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm13[6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 1332(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = mem[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm0, %ymm6
+; AVX2-FCP-NEXT:    vbroadcastss 1444(%rdi), %xmm7
+; AVX2-FCP-NEXT:    vmovaps 1408(%rdi), %xmm14
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm7 = xmm14[0,1,2],xmm7[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm6 = xmm6[0,1],xmm7[2,3]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5],ymm13[6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 1556(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5,6],ymm11[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm13, %ymm15, %ymm7
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 1556(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $240, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm6[0,1,2,3],mem[4,5,6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm0, %ymm6
+; AVX2-FCP-NEXT:    vbroadcastss 1668(%rdi), %xmm7
+; AVX2-FCP-NEXT:    vmovaps 1632(%rdi), %xmm0
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm7 = xmm0[0,1,2],xmm7[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm6 = xmm6[0,1],xmm7[2,3]
+; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm7 = ymm7[0,1,2,3,4,5],mem[6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 1780(%rdi), %ymm8
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm6 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm6 = ymm12[0],mem[1],ymm12[2,3,4],mem[5],ymm12[6,7]
+; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm12 = [1,0,3,3,1,0,7,7]
+; AVX2-FCP-NEXT:    vpermps %ymm6, %ymm12, %ymm6
+; AVX2-FCP-NEXT:    vbroadcastss 216(%rdi), %ymm7
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm7[7]
+; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %xmm7
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm7[0,1,2],xmm3[3]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,1,3,2]
+; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm8 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vextractf128 $1, %ymm8, %xmm8
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm8[0,1],xmm3[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %xmm8
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm8[0,1,2],xmm1[3]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vextractf128 $1, %ymm3, %xmm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0,1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm3 = ymm3[0],mem[1],ymm3[2,3,4],mem[5],ymm3[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm3, %ymm12, %ymm3
+; AVX2-FCP-NEXT:    vbroadcastss 440(%rdi), %ymm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm3[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $15, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm1 = mem[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm1, %ymm4, %ymm1
-; AVX2-FCP-NEXT:    vbroadcastss 1668(%rdi), %xmm4
-; AVX2-FCP-NEXT:    vmovaps 1632(%rdi), %xmm12
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm4 = xmm12[0,1,2],xmm4[3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3]
-; AVX2-FCP-NEXT:    vpermps {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    vblendps $192, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = ymm4[0,1,2,3,4,5],mem[6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 1780(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm10[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps 544(%rdi), %xmm6
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm6[0,1,2],xmm2[3]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vextractf128 $1, %ymm2, %xmm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm2[0],mem[1],ymm2[2,3,4],mem[5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm12, %ymm2
+; AVX2-FCP-NEXT:    vbroadcastss 664(%rdi), %ymm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = ymm1[0],mem[1],ymm1[2,3,4],mem[5],ymm1[6,7]
-; AVX2-FCP-NEXT:    vmovaps {{.*#+}} ymm1 = [1,0,3,3,1,0,7,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vbroadcastss 216(%rdi), %ymm10
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm10[7]
-; AVX2-FCP-NEXT:    vmovaps 96(%rdi), %xmm10
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm9 = xmm10[0,1,2],xmm9[3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm9 = xmm9[0,1,3,2]
-; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm9 = xmm11[0,1],xmm9[2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm9[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 320(%rdi), %xmm4
-; AVX2-FCP-NEXT:    vmovaps %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm2 = xmm4[0,1,2],xmm2[3]
+; AVX2-FCP-NEXT:    vmovaps 768(%rdi), %xmm3
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm3[0,1,2],xmm4[3]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,1,3,2]
+; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vextractf128 $1, %ymm2, %xmm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm2 = ymm2[0],mem[1],ymm2[2,3,4],mem[5],ymm2[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm2, %ymm12, %ymm2
+; AVX2-FCP-NEXT:    vbroadcastss 888(%rdi), %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FCP-NEXT:    vmovaps 992(%rdi), %xmm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm2 = xmm1[0,1,2],xmm5[3]
 ; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[0,1,3,2]
 ; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm4 = mem[1,0,2,3,5,4,6,7]
 ; AVX2-FCP-NEXT:    vextractf128 $1, %ymm4, %xmm4
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3]
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[0],ymm4[1],mem[2,3,4],ymm4[5],mem[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vbroadcastss 440(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm9[7]
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm4 = ymm4[0],mem[1],ymm4[2,3,4],mem[5],ymm4[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm12, %ymm4
+; AVX2-FCP-NEXT:    vbroadcastss 1112(%rdi), %ymm5
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm5[7]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4,5,6,7]
 ; AVX2-FCP-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 544(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm2[0,1,2],xmm3[3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,1,3,2]
-; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FCP-NEXT:    vextractf128 $1, %ymm4, %xmm4
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm4[0,1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $34, (%rsp), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm4 = ymm4[0],mem[1],ymm4[2,3,4],mem[5],ymm4[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm4, %ymm1, %ymm4
-; AVX2-FCP-NEXT:    vbroadcastss 664(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm3, (%rsp) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 768(%rdi), %xmm4
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm4[0,1,2],xmm5[3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,1,3,2]
-; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FCP-NEXT:    vextractf128 $1, %ymm5, %xmm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm1, %ymm5
-; AVX2-FCP-NEXT:    vbroadcastss 888(%rdi), %ymm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FCP-NEXT:    vmovaps 992(%rdi), %xmm9
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm9[0,1,2],xmm6[3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm3 = xmm3[0,1,3,2]
+; AVX2-FCP-NEXT:    vmovaps 1216(%rdi), %xmm2
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm4 = xmm2[0,1,2],xmm10[3]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm4 = xmm4[0,1,3,2]
 ; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm5 = mem[1,0,2,3,5,4,6,7]
 ; AVX2-FCP-NEXT:    vextractf128 $1, %ymm5, %xmm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm5 = ymm5[0],mem[1],ymm5[2,3,4],mem[5],ymm5[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm1, %ymm5
-; AVX2-FCP-NEXT:    vbroadcastss 1112(%rdi), %ymm6
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm6[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm6 = ymm3[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 1216(%rdi), %xmm3
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm3[0,1,2],xmm7[3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm5 = xmm5[0,1,3,2]
-; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm7 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FCP-NEXT:    vextractf128 $1, %ymm7, %xmm7
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm7[0,1],xmm5[2,3]
-; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm7 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm7 = ymm15[0],mem[1],ymm15[2,3,4],mem[5],ymm15[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm7, %ymm1, %ymm7
-; AVX2-FCP-NEXT:    vbroadcastss 1336(%rdi), %ymm15
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm15[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm5[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 1440(%rdi), %xmm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm8 = xmm5[0,1,2],xmm8[3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm8 = xmm8[0,1,3,2]
-; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm15 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FCP-NEXT:    vextractf128 $1, %ymm15, %xmm15
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm8 = xmm15[0,1],xmm8[2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm15 = ymm13[0],ymm14[1],ymm13[2,3,4],ymm14[5],ymm13[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm15, %ymm1, %ymm15
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm4 = xmm5[0,1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm5 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm5 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm5, %ymm12, %ymm5
+; AVX2-FCP-NEXT:    vbroadcastss 1336(%rdi), %ymm10
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm10[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm5 = ymm4[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps 1440(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm10 = xmm4[0,1,2],xmm14[3]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm10 = xmm10[0,1,3,2]
+; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm14 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vextractf128 $1, %ymm14, %xmm14
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm10 = xmm14[0,1],xmm10[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm9[0],ymm13[1],ymm9[2,3,4],ymm13[5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm14, %ymm12, %ymm14
 ; AVX2-FCP-NEXT:    vbroadcastss 1560(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm15[0,1,2,3,4,5,6],ymm11[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3],ymm11[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovaps 1664(%rdi), %xmm15
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm11 = xmm15[0,1,2],xmm12[3]
-; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm11 = xmm11[0,1,3,2]
-; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm12 = mem[1,0,2,3,5,4,6,7]
-; AVX2-FCP-NEXT:    vextractf128 $1, %ymm12, %xmm12
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm11 = xmm12[0,1],xmm11[2,3]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm12 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm12 = ymm12[0],mem[1],ymm12[2,3,4],mem[5],ymm12[6,7]
-; AVX2-FCP-NEXT:    vpermps %ymm12, %ymm1, %ymm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm14[0,1,2,3,4,5,6],ymm11[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovaps 1664(%rdi), %xmm14
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm0 = xmm14[0,1,2],xmm0[3]
+; AVX2-FCP-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,1,3,2]
+; AVX2-FCP-NEXT:    vpermilps $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm11 = mem[1,0,2,3,5,4,6,7]
+; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm0 = xmm11[0,1],xmm0[2,3]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm11 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm11 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
+; AVX2-FCP-NEXT:    vpermps %ymm11, %ymm12, %ymm11
 ; AVX2-FCP-NEXT:    vbroadcastss 1784(%rdi), %ymm12
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm12[7]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm11[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 136(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm1 = xmm1[0],mem[1],xmm1[2,3]
-; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vpermps 192(%rdi), %ymm0, %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5,6],ymm12[7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm12 = ymm0[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 136(%rdi), %xmm0
+; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpermps 192(%rdi), %ymm15, %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm11[6,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 80(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm10 = xmm11[0,1,2],xmm10[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm7 = xmm11[0,1,2],xmm7[3]
 ; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
 ; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm10 = xmm11[0,1],xmm10[2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm10 = ymm10[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 360(%rdi), %xmm1
-; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm1 = xmm1[0],mem[1],xmm1[2,3]
-; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vpermps 416(%rdi), %ymm0, %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm7 = xmm11[0,1],xmm7[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 360(%rdi), %xmm0
+; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpermps 416(%rdi), %ymm15, %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm11[6,7]
 ; AVX2-FCP-NEXT:    vbroadcastss 304(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vblendps $8, {{[-0-9]+}}(%r{{[sb]}}p), %xmm11, %xmm11 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm11 = xmm11[0,1,2],mem[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm11 = xmm11[0,1,2],xmm8[3]
 ; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm13 = mem[2,3,2,3,6,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm13 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm13 = ymm13[0],mem[1],ymm13[2,3,4],mem[5],ymm13[6,7]
 ; AVX2-FCP-NEXT:    vextractf128 $1, %ymm13, %xmm13
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm11 = xmm13[0,1],xmm11[2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm13 = ymm11[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 584(%rdi), %xmm1
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm13 = ymm11[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 584(%rdi), %xmm0
+; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpermps 640(%rdi), %ymm15, %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 528(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm8 = xmm11[0,1,2],xmm6[3]
+; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm8 = xmm11[0,1],xmm8[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 808(%rdi), %xmm0
+; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpermps 864(%rdi), %ymm15, %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 752(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm11[0,1,2],xmm3[3]
+; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm11[0,1],xmm3[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 1032(%rdi), %xmm0
+; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm0 = xmm0[0],mem[1],xmm0[2,3]
+; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FCP-NEXT:    vpermps 1088(%rdi), %ymm15, %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 976(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm11[0,1,2],xmm1[3]
+; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
+; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
+; AVX2-FCP-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm1 = xmm11[0,1],xmm1[2,3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 1256(%rdi), %xmm1
 ; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
 ; AVX2-FCP-NEXT:    # xmm1 = xmm1[0],mem[1],xmm1[2,3]
 ; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FCP-NEXT:    vpermps 640(%rdi), %ymm0, %ymm11
+; AVX2-FCP-NEXT:    vpermps 1312(%rdi), %ymm15, %ymm11
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 528(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vbroadcastss 1200(%rdi), %ymm11
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm2 = xmm11[0,1,2],xmm2[3]
 ; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
@@ -14222,13 +14137,13 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm2 = xmm11[0,1],xmm2[2,3]
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 808(%rdi), %xmm2
+; AVX2-FCP-NEXT:    vbroadcastss 1480(%rdi), %xmm2
 ; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
 ; AVX2-FCP-NEXT:    # xmm2 = xmm2[0],mem[1],xmm2[2,3]
 ; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vpermps 864(%rdi), %ymm0, %ymm11
+; AVX2-FCP-NEXT:    vpermps 1536(%rdi), %ymm15, %ymm11
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 752(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vbroadcastss 1424(%rdi), %ymm11
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm4 = xmm11[0,1,2],xmm4[3]
 ; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
@@ -14236,170 +14151,126 @@ define void @load_i32_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
 ; AVX2-FCP-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
 ; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm4 = xmm11[0,1],xmm4[2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 1032(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm2 = xmm2[0],mem[1],xmm2[2,3]
-; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vpermps 1088(%rdi), %ymm0, %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 976(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm11 = xmm11[0,1,2],xmm9[3]
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 1704(%rdi), %xmm4
+; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
+; AVX2-FCP-NEXT:    # xmm4 = xmm4[0],mem[1],xmm4[2,3]
+; AVX2-FCP-NEXT:    vpermps 1760(%rdi), %ymm15, %ymm11
+; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm11[6,7]
+; AVX2-FCP-NEXT:    vbroadcastss 1648(%rdi), %ymm11
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm11 = xmm11[0,1,2],xmm14[3]
 ; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm14 = mem[2,3,2,3,6,7,6,7]
 ; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm14 # 32-byte Folded Reload
 ; AVX2-FCP-NEXT:    # ymm14 = ymm14[0],mem[1],ymm14[2,3,4],mem[5],ymm14[6,7]
 ; AVX2-FCP-NEXT:    vextractf128 $1, %ymm14, %xmm14
 ; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm11 = xmm14[0,1],xmm11[2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm14 = ymm11[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 1256(%rdi), %xmm2
-; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm2 = xmm2[0],mem[1],xmm2[2,3]
-; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FCP-NEXT:    vpermps 1312(%rdi), %ymm0, %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 1200(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm11[0,1,2],xmm3[3]
-; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm3 = xmm11[0,1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 1480(%rdi), %xmm3
-; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm3 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm3 = xmm3[0],mem[1],xmm3[2,3]
-; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FCP-NEXT:    vpermps 1536(%rdi), %ymm0, %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 1424(%rdi), %ymm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm11[0,1,2],xmm5[3]
-; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm11[0,1],xmm5[2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm3 = ymm5[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 1704(%rdi), %xmm5
-; AVX2-FCP-NEXT:    vblendps $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5, %xmm5 # 16-byte Folded Reload
-; AVX2-FCP-NEXT:    # xmm5 = xmm5[0],mem[1],xmm5[2,3]
-; AVX2-FCP-NEXT:    vpermps 1760(%rdi), %ymm0, %ymm0
-; AVX2-FCP-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm5[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FCP-NEXT:    vbroadcastss 1648(%rdi), %ymm5
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm5[0,1,2],xmm15[3]
-; AVX2-FCP-NEXT:    vpermilps $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = mem[2,3,2,3,6,7,6,7]
-; AVX2-FCP-NEXT:    vblendps $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
-; AVX2-FCP-NEXT:    # ymm11 = ymm11[0],mem[1],ymm11[2,3,4],mem[5],ymm11[6,7]
-; AVX2-FCP-NEXT:    vextractf128 $1, %ymm11, %xmm11
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} xmm5 = xmm11[0,1],xmm5[2,3]
-; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm0 = ymm5[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 192(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, (%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 224(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 160(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%rsi)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 192(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, (%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 224(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 160(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%rdx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 192(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, (%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 224(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 160(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%rcx)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, (%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 192(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 224(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 160(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%r8)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 224(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 192(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 160(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%r9)
-; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FCP-NEXT:    vmovaps %ymm5, (%r9)
+; AVX2-FCP-NEXT:    vblendps {{.*#+}} ymm4 = ymm11[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 192(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 128(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 64(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, (%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 224(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 160(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 96(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 32(%rsi)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 192(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 128(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 64(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, (%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 224(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 160(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 96(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm11, 32(%rdx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 192(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 128(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 64(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, (%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 224(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 160(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 96(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 32(%rcx)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, (%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 64(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 128(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 192(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 224(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 160(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 96(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 32(%r8)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 224(%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 192(%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 160(%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 128(%r9)
+; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 96(%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 64(%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, 32(%r9)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm9, (%r9)
 ; AVX2-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FCP-NEXT:    vmovaps %ymm12, 224(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm8, 192(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm7, 160(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm6, 128(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm10, 192(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 160(%rax)
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovaps %ymm5, 128(%rax)
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm5, 96(%rax)
-; AVX2-FCP-NEXT:    vmovups (%rsp), %ymm5 # 32-byte Reload
+; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm5, 64(%rax)
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm5, 32(%rax)
 ; AVX2-FCP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
 ; AVX2-FCP-NEXT:    vmovaps %ymm5, (%rax)
 ; AVX2-FCP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FCP-NEXT:    vmovaps %ymm0, 224(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm3, 192(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm2, 160(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm14, 128(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm4, 96(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm1, 64(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm4, 224(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm2, 192(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm1, 160(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm0, 128(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm3, 96(%rax)
+; AVX2-FCP-NEXT:    vmovaps %ymm8, 64(%rax)
 ; AVX2-FCP-NEXT:    vmovaps %ymm13, 32(%rax)
-; AVX2-FCP-NEXT:    vmovaps %ymm10, (%rax)
-; AVX2-FCP-NEXT:    addq $2680, %rsp # imm = 0xA78
+; AVX2-FCP-NEXT:    vmovaps %ymm7, (%rax)
+; AVX2-FCP-NEXT:    addq $2648, %rsp # imm = 0xA58
 ; AVX2-FCP-NEXT:    vzeroupper
 ; AVX2-FCP-NEXT:    retq
 ;
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-8.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-8.ll
index 32825f291e98b6..2b268af107f6b3 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-8.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-8.ll
@@ -1939,137 +1939,87 @@ define void @store_i16_stride8_vf16(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.ve
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm1
+; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-NEXT:    vmovdqa (%rsi), %ymm1
 ; AVX2-NEXT:    vmovdqa (%rax), %xmm5
-; AVX2-NEXT:    vmovdqa (%r10), %xmm7
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm7[0],xmm5[0],xmm7[1],xmm5[1],xmm7[2],xmm5[2],xmm7[3],xmm5[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm2[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
+; AVX2-NEXT:    vmovdqa (%r10), %xmm8
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm8[0],xmm5[0],xmm8[1],xmm5[1],xmm8[2],xmm5[2],xmm8[3],xmm5[3]
 ; AVX2-NEXT:    vmovdqa (%r9), %xmm10
 ; AVX2-NEXT:    vmovdqa (%r8), %xmm11
 ; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm6[0],zero,xmm6[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6],ymm0[7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm6[0],xmm3[0],xmm6[1],xmm3[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
 ; AVX2-NEXT:    vmovdqa (%rcx), %xmm12
 ; AVX2-NEXT:    vmovdqa (%rdx), %xmm13
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm8[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm0[0,1,1,3]
-; AVX2-NEXT:    vmovdqa (%rsi), %xmm15
-; AVX2-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm0[0],xmm15[0],xmm0[1],xmm15[1],xmm0[2],xmm15[2],xmm0[3],xmm15[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm14 = xmm9[0],zero,xmm9[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0],ymm3[1],ymm14[2,3,4],ymm3[5],ymm14[6,7]
-; AVX2-NEXT:    vmovdqa (%rdx), %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm14[0,1],ymm4[2,3],ymm14[4,5],ymm4[6,7]
-; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa (%rcx), %ymm4
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2],ymm2[3],ymm6[4,5,6],ymm2[7]
-; AVX2-NEXT:    vmovdqa (%r8), %ymm6
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm9 = xmm9[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0],ymm8[1],ymm9[2,3,4],ymm8[5],ymm9[6,7]
-; AVX2-NEXT:    vmovdqa (%r9), %ymm8
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm2[2,3],ymm9[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
+; AVX2-NEXT:    vmovdqa (%rsi), %xmm14
+; AVX2-NEXT:    vmovdqa (%rdi), %xmm15
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm15[0],xmm14[0],xmm15[1],xmm14[1],xmm15[2],xmm14[2],xmm15[3],xmm14[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm9[0],xmm7[0],xmm9[1],xmm7[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3],ymm4[4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa (%r10), %ymm9
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm7 = xmm7[4],xmm5[4],xmm7[5],xmm5[5],xmm7[6],xmm5[6],xmm7[7],xmm5[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm5 = xmm7[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,0,2,1]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm10 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm11 = xmm10[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm11[0,1,2],ymm5[3],ymm11[4,5,6],ymm5[7]
-; AVX2-NEXT:    vmovdqa (%rax), %ymm14
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm11 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm12 = xmm11[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,1,1,3]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm15[4],xmm0[5],xmm15[5],xmm0[6],xmm15[6],xmm0[7],xmm15[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm13 = xmm0[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm13[0],ymm12[1],ymm13[2,3,4],ymm12[5],ymm13[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm5[2,3],ymm12[4,5],ymm5[6,7]
+; AVX2-NEXT:    vmovdqa (%rdx), %ymm4
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm3 = xmm6[2],xmm3[2],xmm6[3],xmm3[3]
+; AVX2-NEXT:    vmovdqa (%rcx), %ymm6
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm9 = xmm9[2],xmm7[2],xmm9[3],xmm7[3]
+; AVX2-NEXT:    vmovdqa (%r8), %ymm7
+; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm3[2,3],ymm9[4,5],ymm3[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm7 = xmm7[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm10[0],zero,xmm10[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1,2],ymm7[3],ymm10[4,5,6],ymm7[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm10 = xmm11[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm10[1],ymm0[2,3,4],ymm10[5],ymm0[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm0[0,1],ymm7[2,3],ymm0[4,5],ymm7[6,7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm10 = ymm9[0],ymm14[0],ymm9[1],ymm14[1],ymm9[2],ymm14[2],ymm9[3],ymm14[3],ymm9[8],ymm14[8],ymm9[9],ymm14[9],ymm9[10],ymm14[10],ymm9[11],ymm14[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm10[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm12 = ymm6[0],ymm8[0],ymm6[1],ymm8[1],ymm6[2],ymm8[2],ymm6[3],ymm8[3],ymm6[8],ymm8[8],ymm6[9],ymm8[9],ymm6[10],ymm8[10],ymm6[11],ymm8[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm11 = ymm12[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2],ymm0[3],ymm11[4,5,6],ymm0[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm13 = ymm3[0],ymm4[0],ymm3[1],ymm4[1],ymm3[2],ymm4[2],ymm3[3],ymm4[3],ymm3[8],ymm4[8],ymm3[9],ymm4[9],ymm3[10],ymm4[10],ymm3[11],ymm4[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm13[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm0[2,1,3,3]
-; AVX2-NEXT:    vmovdqa (%rsi), %ymm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[2,1,3,3,6,5,7,7]
+; AVX2-NEXT:    vmovdqa (%r9), %ymm9
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm8[4],xmm5[4],xmm8[5],xmm5[5],xmm8[6],xmm5[6],xmm8[7],xmm5[7]
+; AVX2-NEXT:    vmovdqa (%r10), %ymm5
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm11 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
+; AVX2-NEXT:    vmovdqa (%rax), %ymm10
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm13 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm12 = xmm11[2],xmm8[2],xmm11[3],xmm8[3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,0,2,1]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm14 = xmm15[4],xmm14[4],xmm15[5],xmm14[5],xmm15[6],xmm14[6],xmm15[7],xmm14[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm15 = xmm14[2],xmm13[2],xmm14[3],xmm13[3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm15[0,1],ymm12[2,3],ymm15[4,5],ymm12[6,7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm8 = xmm11[0],xmm8[0],xmm11[1],xmm8[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm11 = xmm14[0],xmm13[0],xmm14[1],xmm13[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm11[0,1],ymm8[2,3],ymm11[4,5],ymm8[6,7]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm11 = ymm5[0],ymm10[0],ymm5[1],ymm10[1],ymm5[2],ymm10[2],ymm5[3],ymm10[3],ymm5[8],ymm10[8],ymm5[9],ymm10[9],ymm5[10],ymm10[10],ymm5[11],ymm10[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm13 = ymm7[0],ymm9[0],ymm7[1],ymm9[1],ymm7[2],ymm9[2],ymm7[3],ymm9[3],ymm7[8],ymm9[8],ymm7[9],ymm9[9],ymm7[10],ymm9[10],ymm7[11],ymm9[11]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm14 = ymm13[2],ymm11[2],ymm13[3],ymm11[3],ymm13[6],ymm11[6],ymm13[7],ymm11[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,2,2,3]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm15 = ymm4[0],ymm6[0],ymm4[1],ymm6[1],ymm4[2],ymm6[2],ymm4[3],ymm6[3],ymm4[8],ymm6[8],ymm4[9],ymm6[9],ymm4[10],ymm6[10],ymm4[11],ymm6[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm3[2],ymm15[2],ymm3[3],ymm15[3],ymm3[6],ymm15[6],ymm3[7],ymm15[7]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0],ymm15[1],ymm2[2,3,4],ymm15[5],ymm2[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm2[0,1],ymm11[2,3],ymm2[4,5],ymm11[6,7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm9[4],ymm14[4],ymm9[5],ymm14[5],ymm9[6],ymm14[6],ymm9[7],ymm14[7],ymm9[12],ymm14[12],ymm9[13],ymm14[13],ymm9[14],ymm14[14],ymm9[15],ymm14[15]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm6[4],ymm8[4],ymm6[5],ymm8[5],ymm6[6],ymm8[6],ymm6[7],ymm8[7],ymm6[12],ymm8[12],ymm6[13],ymm8[13],ymm6[14],ymm8[14],ymm6[15],ymm8[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm8 = ymm2[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm9 = ymm6[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1,2],ymm8[3],ymm9[4,5,6],ymm8[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm3[4],ymm4[4],ymm3[5],ymm4[5],ymm3[6],ymm4[6],ymm3[7],ymm4[7],ymm3[12],ymm4[12],ymm3[13],ymm4[13],ymm3[14],ymm4[14],ymm3[15],ymm4[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm3[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm9 = ymm0[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm9[0],ymm4[1],ymm9[2,3,4],ymm4[5],ymm9[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1],ymm8[2,3],ymm4[4,5],ymm8[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[0,1,1,3,4,5,5,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm14[2,3],ymm2[4,5],ymm14[6,7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm5[4],ymm10[4],ymm5[5],ymm10[5],ymm5[6],ymm10[6],ymm5[7],ymm10[7],ymm5[12],ymm10[12],ymm5[13],ymm10[13],ymm5[14],ymm10[14],ymm5[15],ymm10[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm7 = ymm7[4],ymm9[4],ymm7[5],ymm9[5],ymm7[6],ymm9[6],ymm7[7],ymm9[7],ymm7[12],ymm9[12],ymm7[13],ymm9[13],ymm7[14],ymm9[14],ymm7[15],ymm9[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm4[4],ymm6[4],ymm4[5],ymm6[5],ymm4[6],ymm6[6],ymm4[7],ymm6[7],ymm4[12],ymm6[12],ymm4[13],ymm6[13],ymm4[14],ymm6[14],ymm4[15],ymm6[15]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm6 = ymm7[2],ymm5[2],ymm7[3],ymm5[3],ymm7[6],ymm5[6],ymm7[7],ymm5[7]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2],ymm2[3],ymm6[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm3[1],ymm0[2,3,4],ymm3[5],ymm0[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm10[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm12[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm13[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm5[0,1,1,3,4,5,5,7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm9 = ymm1[2],ymm4[2],ymm1[3],ymm4[3],ymm1[6],ymm4[6],ymm1[7],ymm4[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm9[0,1],ymm6[2,3],ymm9[4,5],ymm6[6,7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm7[0],ymm5[0],ymm7[1],ymm5[1],ymm7[4],ymm5[4],ymm7[5],ymm5[5]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm1[0],ymm4[0],ymm1[1],ymm4[1],ymm1[4],ymm4[4],ymm1[5],ymm4[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm5[0,2,2,3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2,3,4],ymm3[5],ymm1[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm4[2,3],ymm1[4,5],ymm4[6,7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm4 = ymm13[0],ymm11[0],ymm13[1],ymm11[1],ymm13[4],ymm11[4],ymm13[5],ymm11[5]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm3[0],ymm15[0],ymm3[1],ymm15[1],ymm3[4],ymm15[4],ymm3[5],ymm15[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm4[2,3],ymm0[4,5],ymm4[6,7]
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-NEXT:    vmovdqa %ymm1, 128(%rax)
-; AVX2-NEXT:    vmovdqa %ymm0, 192(%rax)
-; AVX2-NEXT:    vmovdqa %ymm4, 224(%rax)
-; AVX2-NEXT:    vmovdqa %ymm11, 160(%rax)
-; AVX2-NEXT:    vmovdqa %ymm7, 64(%rax)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 96(%rax)
+; AVX2-NEXT:    vmovdqa %ymm0, 128(%rax)
+; AVX2-NEXT:    vmovdqa %ymm1, 192(%rax)
+; AVX2-NEXT:    vmovdqa %ymm6, 224(%rax)
+; AVX2-NEXT:    vmovdqa %ymm2, 160(%rax)
+; AVX2-NEXT:    vmovdqa %ymm8, 64(%rax)
+; AVX2-NEXT:    vmovdqa %ymm12, 96(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 32(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
@@ -2081,137 +2031,87 @@ define void @store_i16_stride8_vf16(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.ve
 ; AVX2-FP:       # %bb.0:
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm1
+; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FP-NEXT:    vmovdqa (%rsi), %ymm1
 ; AVX2-FP-NEXT:    vmovdqa (%rax), %xmm5
-; AVX2-FP-NEXT:    vmovdqa (%r10), %xmm7
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm7[0],xmm5[0],xmm7[1],xmm5[1],xmm7[2],xmm5[2],xmm7[3],xmm5[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm2[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
+; AVX2-FP-NEXT:    vmovdqa (%r10), %xmm8
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm8[0],xmm5[0],xmm8[1],xmm5[1],xmm8[2],xmm5[2],xmm8[3],xmm5[3]
 ; AVX2-FP-NEXT:    vmovdqa (%r9), %xmm10
 ; AVX2-FP-NEXT:    vmovdqa (%r8), %xmm11
 ; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm6[0],zero,xmm6[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6],ymm0[7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm6[0],xmm3[0],xmm6[1],xmm3[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
 ; AVX2-FP-NEXT:    vmovdqa (%rcx), %xmm12
 ; AVX2-FP-NEXT:    vmovdqa (%rdx), %xmm13
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm8[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm0[0,1,1,3]
-; AVX2-FP-NEXT:    vmovdqa (%rsi), %xmm15
-; AVX2-FP-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm0[0],xmm15[0],xmm0[1],xmm15[1],xmm0[2],xmm15[2],xmm0[3],xmm15[3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm14 = xmm9[0],zero,xmm9[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0],ymm3[1],ymm14[2,3,4],ymm3[5],ymm14[6,7]
-; AVX2-FP-NEXT:    vmovdqa (%rdx), %ymm3
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm14[0,1],ymm4[2,3],ymm14[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa (%rcx), %ymm4
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2],ymm2[3],ymm6[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vmovdqa (%r8), %ymm6
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm9 = xmm9[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0],ymm8[1],ymm9[2,3,4],ymm8[5],ymm9[6,7]
-; AVX2-FP-NEXT:    vmovdqa (%r9), %ymm8
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm2[2,3],ymm9[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
+; AVX2-FP-NEXT:    vmovdqa (%rsi), %xmm14
+; AVX2-FP-NEXT:    vmovdqa (%rdi), %xmm15
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm15[0],xmm14[0],xmm15[1],xmm14[1],xmm15[2],xmm14[2],xmm15[3],xmm14[3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm9[0],xmm7[0],xmm9[1],xmm7[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3],ymm4[4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa (%r10), %ymm9
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm7 = xmm7[4],xmm5[4],xmm7[5],xmm5[5],xmm7[6],xmm5[6],xmm7[7],xmm5[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm5 = xmm7[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,0,2,1]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm10 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm11 = xmm10[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm11[0,1,2],ymm5[3],ymm11[4,5,6],ymm5[7]
-; AVX2-FP-NEXT:    vmovdqa (%rax), %ymm14
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm11 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm12 = xmm11[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,1,1,3]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm15[4],xmm0[5],xmm15[5],xmm0[6],xmm15[6],xmm0[7],xmm15[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm13 = xmm0[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm13[0],ymm12[1],ymm13[2,3,4],ymm12[5],ymm13[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm5[2,3],ymm12[4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vmovdqa (%rdx), %ymm4
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm3 = xmm6[2],xmm3[2],xmm6[3],xmm3[3]
+; AVX2-FP-NEXT:    vmovdqa (%rcx), %ymm6
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm9 = xmm9[2],xmm7[2],xmm9[3],xmm7[3]
+; AVX2-FP-NEXT:    vmovdqa (%r8), %ymm7
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm3[2,3],ymm9[4,5],ymm3[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm7 = xmm7[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm10[0],zero,xmm10[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm10[0,1,2],ymm7[3],ymm10[4,5,6],ymm7[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm10 = xmm11[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm10[1],ymm0[2,3,4],ymm10[5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm0[0,1],ymm7[2,3],ymm0[4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm10 = ymm9[0],ymm14[0],ymm9[1],ymm14[1],ymm9[2],ymm14[2],ymm9[3],ymm14[3],ymm9[8],ymm14[8],ymm9[9],ymm14[9],ymm9[10],ymm14[10],ymm9[11],ymm14[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm10[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm12 = ymm6[0],ymm8[0],ymm6[1],ymm8[1],ymm6[2],ymm8[2],ymm6[3],ymm8[3],ymm6[8],ymm8[8],ymm6[9],ymm8[9],ymm6[10],ymm8[10],ymm6[11],ymm8[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm11 = ymm12[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm11[0,1,2],ymm0[3],ymm11[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm13 = ymm3[0],ymm4[0],ymm3[1],ymm4[1],ymm3[2],ymm4[2],ymm3[3],ymm4[3],ymm3[8],ymm4[8],ymm3[9],ymm4[9],ymm3[10],ymm4[10],ymm3[11],ymm4[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm13[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm0[2,1,3,3]
-; AVX2-FP-NEXT:    vmovdqa (%rsi), %ymm0
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[2,1,3,3,6,5,7,7]
+; AVX2-FP-NEXT:    vmovdqa (%r9), %ymm9
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm8[4],xmm5[4],xmm8[5],xmm5[5],xmm8[6],xmm5[6],xmm8[7],xmm5[7]
+; AVX2-FP-NEXT:    vmovdqa (%r10), %ymm5
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm11 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
+; AVX2-FP-NEXT:    vmovdqa (%rax), %ymm10
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm13 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm12 = xmm11[2],xmm8[2],xmm11[3],xmm8[3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,0,2,1]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm14 = xmm15[4],xmm14[4],xmm15[5],xmm14[5],xmm15[6],xmm14[6],xmm15[7],xmm14[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm15 = xmm14[2],xmm13[2],xmm14[3],xmm13[3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm15[0,1],ymm12[2,3],ymm15[4,5],ymm12[6,7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm8 = xmm11[0],xmm8[0],xmm11[1],xmm8[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm11 = xmm14[0],xmm13[0],xmm14[1],xmm13[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm11[0,1],ymm8[2,3],ymm11[4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm11 = ymm5[0],ymm10[0],ymm5[1],ymm10[1],ymm5[2],ymm10[2],ymm5[3],ymm10[3],ymm5[8],ymm10[8],ymm5[9],ymm10[9],ymm5[10],ymm10[10],ymm5[11],ymm10[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm13 = ymm7[0],ymm9[0],ymm7[1],ymm9[1],ymm7[2],ymm9[2],ymm7[3],ymm9[3],ymm7[8],ymm9[8],ymm7[9],ymm9[9],ymm7[10],ymm9[10],ymm7[11],ymm9[11]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm14 = ymm13[2],ymm11[2],ymm13[3],ymm11[3],ymm13[6],ymm11[6],ymm13[7],ymm11[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,2,2,3]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm15 = ymm4[0],ymm6[0],ymm4[1],ymm6[1],ymm4[2],ymm6[2],ymm4[3],ymm6[3],ymm4[8],ymm6[8],ymm4[9],ymm6[9],ymm4[10],ymm6[10],ymm4[11],ymm6[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm3[2],ymm15[2],ymm3[3],ymm15[3],ymm3[6],ymm15[6],ymm3[7],ymm15[7]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0],ymm15[1],ymm2[2,3,4],ymm15[5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm2[0,1],ymm11[2,3],ymm2[4,5],ymm11[6,7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm9[4],ymm14[4],ymm9[5],ymm14[5],ymm9[6],ymm14[6],ymm9[7],ymm14[7],ymm9[12],ymm14[12],ymm9[13],ymm14[13],ymm9[14],ymm14[14],ymm9[15],ymm14[15]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm6[4],ymm8[4],ymm6[5],ymm8[5],ymm6[6],ymm8[6],ymm6[7],ymm8[7],ymm6[12],ymm8[12],ymm6[13],ymm8[13],ymm6[14],ymm8[14],ymm6[15],ymm8[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm8 = ymm2[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm9 = ymm6[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1,2],ymm8[3],ymm9[4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm3[4],ymm4[4],ymm3[5],ymm4[5],ymm3[6],ymm4[6],ymm3[7],ymm4[7],ymm3[12],ymm4[12],ymm3[13],ymm4[13],ymm3[14],ymm4[14],ymm3[15],ymm4[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm3[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm9 = ymm0[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm9[0],ymm4[1],ymm9[2,3,4],ymm4[5],ymm9[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1],ymm8[2,3],ymm4[4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[0,1,1,3,4,5,5,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm14[2,3],ymm2[4,5],ymm14[6,7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm5[4],ymm10[4],ymm5[5],ymm10[5],ymm5[6],ymm10[6],ymm5[7],ymm10[7],ymm5[12],ymm10[12],ymm5[13],ymm10[13],ymm5[14],ymm10[14],ymm5[15],ymm10[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm7 = ymm7[4],ymm9[4],ymm7[5],ymm9[5],ymm7[6],ymm9[6],ymm7[7],ymm9[7],ymm7[12],ymm9[12],ymm7[13],ymm9[13],ymm7[14],ymm9[14],ymm7[15],ymm9[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm4[4],ymm6[4],ymm4[5],ymm6[5],ymm4[6],ymm6[6],ymm4[7],ymm6[7],ymm4[12],ymm6[12],ymm4[13],ymm6[13],ymm4[14],ymm6[14],ymm4[15],ymm6[15]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm6 = ymm7[2],ymm5[2],ymm7[3],ymm5[3],ymm7[6],ymm5[6],ymm7[7],ymm5[7]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2],ymm2[3],ymm6[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm3[1],ymm0[2,3,4],ymm3[5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm10[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm12[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm13[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm5[0,1,1,3,4,5,5,7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm9 = ymm1[2],ymm4[2],ymm1[3],ymm4[3],ymm1[6],ymm4[6],ymm1[7],ymm4[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm9[0,1],ymm6[2,3],ymm9[4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm7[0],ymm5[0],ymm7[1],ymm5[1],ymm7[4],ymm5[4],ymm7[5],ymm5[5]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm1[0],ymm4[0],ymm1[1],ymm4[1],ymm1[4],ymm4[4],ymm1[5],ymm4[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm5[0,2,2,3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2,3,4],ymm3[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3],ymm1[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm4[2,3],ymm1[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm4 = ymm13[0],ymm11[0],ymm13[1],ymm11[1],ymm13[4],ymm11[4],ymm13[5],ymm11[5]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm3[0],ymm15[0],ymm3[1],ymm15[1],ymm3[4],ymm15[4],ymm3[5],ymm15[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm4[2,3],ymm0[4,5],ymm4[6,7]
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FP-NEXT:    vmovdqa %ymm1, 128(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm0, 192(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm4, 224(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm11, 160(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm7, 64(%rax)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 96(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm0, 128(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm1, 192(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm6, 224(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm2, 160(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm8, 64(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm12, 96(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
@@ -3558,289 +3458,189 @@ define void @store_i16_stride8_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.ve
 ;
 ; AVX2-LABEL: store_i16_stride8_vf32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    subq $264, %rsp # imm = 0x108
+; AVX2-NEXT:    subq $232, %rsp
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX2-NEXT:    vmovdqa (%rax), %xmm1
+; AVX2-NEXT:    vmovdqa (%rax), %xmm0
+; AVX2-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-NEXT:    vmovdqa 32(%rax), %xmm10
+; AVX2-NEXT:    vmovdqa (%r10), %xmm1
 ; AVX2-NEXT:    vmovdqa %xmm1, (%rsp) # 16-byte Spill
-; AVX2-NEXT:    vmovdqa 32(%rax), %xmm11
-; AVX2-NEXT:    vmovdqa (%r10), %xmm0
+; AVX2-NEXT:    vmovdqa 32(%r10), %xmm9
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-NEXT:    vmovdqa (%r9), %xmm0
 ; AVX2-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-NEXT:    vmovdqa (%r9), %xmm3
-; AVX2-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-NEXT:    vmovdqa (%r8), %xmm2
 ; AVX2-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm3[0,1,2],ymm1[3],ymm3[4,5,6],ymm1[7]
-; AVX2-NEXT:    vmovdqa (%rcx), %xmm1
-; AVX2-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
+; AVX2-NEXT:    vmovdqa (%rcx), %xmm4
+; AVX2-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-NEXT:    vmovdqa (%rdx), %xmm3
 ; AVX2-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm15 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm15[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
-; AVX2-NEXT:    vmovdqa (%rsi), %xmm1
-; AVX2-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
+; AVX2-NEXT:    vmovdqa (%rsi), %xmm12
 ; AVX2-NEXT:    vmovdqa (%rdi), %xmm13
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm13[0],xmm1[0],xmm13[1],xmm1[1],xmm13[2],xmm1[2],xmm13[3],xmm1[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm14 = xmm1[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0],ymm3[1],ymm14[2,3,4],ymm3[5],ymm14[6,7]
-; AVX2-NEXT:    vmovdqa 32(%r10), %xmm9
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm14[0,1],ymm5[2,3],ymm14[4,5],ymm5[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 32(%r9), %xmm10
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 32(%r8), %xmm14
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm15[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3,4],ymm2[5],ymm1[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm14 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm15 = xmm14[2],xmm3[2],xmm14[3],xmm3[3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2,3],ymm15[4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm9[4],xmm11[4],xmm9[5],xmm11[5],xmm9[6],xmm11[6],xmm9[7],xmm11[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm4[0,0,1,1]
+; AVX2-NEXT:    vmovdqa 32(%r9), %xmm8
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; AVX2-NEXT:    vmovdqa 32(%r8), %xmm15
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm14[4],xmm10[4],xmm14[5],xmm10[5],xmm14[6],xmm10[6],xmm14[7],xmm10[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm3[0,1,2],ymm1[3],ymm3[4,5,6],ymm1[7]
-; AVX2-NEXT:    vmovdqa 32(%rcx), %xmm15
-; AVX2-NEXT:    vmovdqa 32(%rdx), %xmm8
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm8[4],xmm15[4],xmm8[5],xmm15[5],xmm8[6],xmm15[6],xmm8[7],xmm15[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
-; AVX2-NEXT:    vmovdqa 32(%rsi), %xmm7
-; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm6
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm6[4],xmm7[4],xmm6[5],xmm7[5],xmm6[6],xmm7[6],xmm6[7],xmm7[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm12 = xmm0[0],zero,xmm0[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm12[0],ymm3[1],ymm12[2,3,4],ymm3[5],ymm12[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm5[2,3],ymm3[4,5],ymm5[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm4[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3],ymm2[4,5,6],ymm3[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm14[0],xmm3[0],xmm14[1],xmm3[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm9[4],xmm10[4],xmm9[5],xmm10[5],xmm9[6],xmm10[6],xmm9[7],xmm10[7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm14 = xmm15[4],xmm8[4],xmm15[5],xmm8[5],xmm15[6],xmm8[6],xmm15[7],xmm8[7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm14[0],xmm3[0],xmm14[1],xmm3[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm2[0,0,2,1]
+; AVX2-NEXT:    vmovdqa 32(%rcx), %xmm6
+; AVX2-NEXT:    vmovdqa 32(%rdx), %xmm5
+; AVX2-NEXT:    vmovdqa 32(%rsi), %xmm4
+; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm2
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm5[4],xmm6[4],xmm5[5],xmm6[5],xmm5[6],xmm6[6],xmm5[7],xmm6[7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm11 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm11[0,1],ymm7[2,3],ymm11[4,5],ymm7[6,7]
+; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm3 = xmm14[2],xmm3[2],xmm14[3],xmm3[3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm3[0,0,2,1]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3,4],ymm1[5],ymm0[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm9[0],xmm11[0],xmm9[1],xmm11[1],xmm9[2],xmm11[2],xmm9[3],xmm11[3]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm14[0],xmm10[0],xmm14[1],xmm10[1],xmm14[2],xmm10[2],xmm14[3],xmm10[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm8[0],xmm15[0],xmm8[1],xmm15[1],xmm8[2],xmm15[2],xmm8[3],xmm15[3]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm6[0],xmm7[0],xmm6[1],xmm7[1],xmm6[2],xmm7[2],xmm6[3],xmm7[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm6 = xmm5[0],zero,xmm5[1],zero
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm15[0],xmm8[0],xmm15[1],xmm8[1],xmm15[2],xmm8[2],xmm15[3],xmm8[3]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm2[0],xmm5[0],xmm2[1],xmm5[1]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0],ymm4[1],ymm6[2,3,4],ymm4[5],ymm6[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3],ymm4[4,5],ymm2[6,7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm6[0,1],ymm4[2,3],ymm6[4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm2[2],xmm5[2],xmm2[3],xmm5[3]
+; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm14
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm3[2,2,3,3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm5[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
-; AVX2-NEXT:    vmovdqa 32(%r8), %ymm14
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 32(%r9), %ymm15
-; AVX2-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; AVX2-NEXT:    vpunpckhwd (%rsp), %xmm0, %xmm1 # 16-byte Folded Reload
+; AVX2-NEXT:    vmovdqa 32(%r8), %ymm15
+; AVX2-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX2-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm1 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm1 = xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; AVX2-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm2 = xmm2[4],mem[4],xmm2[5],mem[5],xmm2[6],mem[6],xmm2[7],mem[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa 32(%r10), %ymm5
+; AVX2-NEXT:    vmovdqa 32(%r9), %ymm4
+; AVX2-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX2-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm2 # 16-byte Folded Reload
+; AVX2-NEXT:    # xmm2 = xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7]
+; AVX2-NEXT:    vmovdqa 32(%r10), %ymm6
 ; AVX2-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; AVX2-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm7 # 16-byte Folded Reload
 ; AVX2-NEXT:    # xmm7 = xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm6 = xmm7[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,3]
-; AVX2-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm13, %xmm8 # 16-byte Folded Reload
-; AVX2-NEXT:    # xmm8 = xmm13[4],mem[4],xmm13[5],mem[5],xmm13[6],mem[6],xmm13[7],mem[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm9 = xmm8[0],zero,xmm8[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm9[0],ymm6[1],ymm9[2,3,4],ymm6[5],ymm9[6,7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm8 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm9 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm10 = xmm9[0],xmm7[0],xmm9[1],xmm7[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm8[2,3],ymm10[4,5],ymm8[6,7]
+; AVX2-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 32(%rax), %ymm10
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1],ymm3[2,3],ymm6[4,5],ymm3[6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm7[2,2,3,3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm9[2],xmm7[2],xmm9[3],xmm7[3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm8[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3,4],ymm2[5],ymm3[6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm5[4],ymm10[4],ymm5[5],ymm10[5],ymm5[6],ymm10[6],ymm5[7],ymm10[7],ymm5[12],ymm10[12],ymm5[13],ymm10[13],ymm5[14],ymm10[14],ymm5[15],ymm10[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm1[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm14[4],ymm15[4],ymm14[5],ymm15[5],ymm14[6],ymm15[6],ymm14[7],ymm15[7],ymm14[12],ymm15[12],ymm14[13],ymm15[13],ymm14[14],ymm15[14],ymm14[15],ymm15[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm8 = ymm3[0,1,1,3,4,5,5,7]
+; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm6[4],ymm10[4],ymm6[5],ymm10[5],ymm6[6],ymm10[6],ymm6[7],ymm10[7],ymm6[12],ymm10[12],ymm6[13],ymm10[13],ymm6[14],ymm10[14],ymm6[15],ymm10[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm15[4],ymm4[4],ymm15[5],ymm4[5],ymm15[6],ymm4[6],ymm15[7],ymm4[7],ymm15[12],ymm4[12],ymm15[13],ymm4[13],ymm15[14],ymm4[14],ymm15[15],ymm4[15]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm8 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[4],ymm3[4],ymm2[5],ymm3[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm8[0,1,2],ymm2[3],ymm8[4,5,6],ymm2[7]
-; AVX2-NEXT:    vmovdqa 32(%rdx), %ymm11
-; AVX2-NEXT:    vmovdqa 32(%rcx), %ymm12
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm9 = ymm11[4],ymm12[4],ymm11[5],ymm12[5],ymm11[6],ymm12[6],ymm11[7],ymm12[7],ymm11[12],ymm12[12],ymm11[13],ymm12[13],ymm11[14],ymm12[14],ymm11[15],ymm12[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm8 = ymm9[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,3]
-; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm13
-; AVX2-NEXT:    vmovdqa 32(%rsi), %ymm2
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm13[4],ymm2[4],ymm13[5],ymm2[5],ymm13[6],ymm2[6],ymm13[7],ymm2[7],ymm13[12],ymm2[12],ymm13[13],ymm2[13],ymm13[14],ymm2[14],ymm13[15],ymm2[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm6 = ymm0[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0],ymm8[1],ymm6[2,3,4],ymm8[5],ymm6[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm4[2,3],ymm6[4,5],ymm4[6,7]
-; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3],ymm3[4,5,6],ymm1[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm9[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,1,3,3,6,5,7,7]
+; AVX2-NEXT:    vmovdqa 32(%rsi), %ymm11
+; AVX2-NEXT:    vmovdqa 32(%rdx), %ymm13
+; AVX2-NEXT:    vmovdqa 32(%rcx), %ymm9
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm13[4],ymm9[4],ymm13[5],ymm9[5],ymm13[6],ymm9[6],ymm13[7],ymm9[7],ymm13[12],ymm9[12],ymm13[13],ymm9[13],ymm13[14],ymm9[14],ymm13[15],ymm9[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm14[4],ymm11[4],ymm14[5],ymm11[5],ymm14[6],ymm11[6],ymm14[7],ymm11[7],ymm14[12],ymm11[12],ymm14[13],ymm11[13],ymm14[14],ymm11[14],ymm14[15],ymm11[15]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1],ymm8[2,3],ymm5[4,5],ymm8[6,7]
+; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[6],ymm3[6],ymm2[7],ymm3[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm2[0,2,2,3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm3[1],ymm0[2,3,4],ymm3[5],ymm0[6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm5[0],ymm10[0],ymm5[1],ymm10[1],ymm5[2],ymm10[2],ymm5[3],ymm10[3],ymm5[8],ymm10[8],ymm5[9],ymm10[9],ymm5[10],ymm10[10],ymm5[11],ymm10[11]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm14[0],ymm15[0],ymm14[1],ymm15[1],ymm14[2],ymm15[2],ymm14[3],ymm15[3],ymm14[8],ymm15[8],ymm14[9],ymm15[9],ymm14[10],ymm15[10],ymm14[11],ymm15[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm0[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm1[0,1,1,3,4,5,5,7]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm6[0],ymm10[0],ymm6[1],ymm10[1],ymm6[2],ymm10[2],ymm6[3],ymm10[3],ymm6[8],ymm10[8],ymm6[9],ymm10[9],ymm6[10],ymm10[10],ymm6[11],ymm10[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm15[0],ymm4[0],ymm15[1],ymm4[1],ymm15[2],ymm4[2],ymm15[3],ymm4[3],ymm15[8],ymm4[8],ymm15[9],ymm4[9],ymm15[10],ymm4[10],ymm15[11],ymm4[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm13[0],ymm9[0],ymm13[1],ymm9[1],ymm13[2],ymm9[2],ymm13[3],ymm9[3],ymm13[8],ymm9[8],ymm13[9],ymm9[9],ymm13[10],ymm9[10],ymm13[11],ymm9[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm4 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[4],ymm0[4],ymm2[5],ymm0[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3],ymm4[4,5,6],ymm3[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm11[0],ymm12[0],ymm11[1],ymm12[1],ymm11[2],ymm12[2],ymm11[3],ymm12[3],ymm11[8],ymm12[8],ymm11[9],ymm12[9],ymm11[10],ymm12[10],ymm11[11],ymm12[11]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm13[0],ymm2[0],ymm13[1],ymm2[1],ymm13[2],ymm2[2],ymm13[3],ymm2[3],ymm13[8],ymm2[8],ymm13[9],ymm2[9],ymm13[10],ymm2[10],ymm13[11],ymm2[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm4[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm6 = ymm2[0,1,1,3,4,5,5,7]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm14[0],ymm11[0],ymm14[1],ymm11[1],ymm14[2],ymm11[2],ymm14[3],ymm11[3],ymm14[8],ymm11[8],ymm14[9],ymm11[9],ymm14[10],ymm11[10],ymm14[11],ymm11[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm6 = ymm5[0],ymm3[0],ymm5[1],ymm3[1],ymm5[4],ymm3[4],ymm5[5],ymm3[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0],ymm5[1],ymm6[2,3,4],ymm5[5],ymm6[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3],ymm5[4,5],ymm3[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa (%r10), %ymm12
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa (%rax), %ymm13
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm4[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm12[4],ymm13[4],ymm12[5],ymm13[5],ymm12[6],ymm13[6],ymm12[7],ymm13[7],ymm12[12],ymm13[12],ymm12[13],ymm13[13],ymm12[14],ymm13[14],ymm12[15],ymm13[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm1[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm6[0,1],ymm4[2,3],ymm6[4,5],ymm4[6,7]
 ; AVX2-NEXT:    vmovdqa (%r8), %ymm8
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[6],ymm0[6],ymm2[7],ymm0[7]
 ; AVX2-NEXT:    vmovdqa (%r9), %ymm7
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm8[4],ymm7[4],ymm8[5],ymm7[5],ymm8[6],ymm7[6],ymm8[7],ymm7[7],ymm8[12],ymm7[12],ymm8[13],ymm7[13],ymm8[14],ymm7[14],ymm8[15],ymm7[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm4[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm0[3],ymm5[4,5,6],ymm0[7]
-; AVX2-NEXT:    vmovdqa (%rdx), %ymm9
-; AVX2-NEXT:    vmovdqa (%rcx), %ymm6
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm14 = ymm9[4],ymm6[4],ymm9[5],ymm6[5],ymm9[6],ymm6[6],ymm9[7],ymm6[7],ymm9[12],ymm6[12],ymm9[13],ymm6[13],ymm9[14],ymm6[14],ymm9[15],ymm6[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm15 = ymm14[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm3
-; AVX2-NEXT:    vmovdqa (%rsi), %ymm2
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm3[4],ymm2[4],ymm3[5],ymm2[5],ymm3[6],ymm2[6],ymm3[7],ymm2[7],ymm3[12],ymm2[12],ymm3[13],ymm2[13],ymm3[14],ymm2[14],ymm3[15],ymm2[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm10 = ymm0[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0],ymm15[1],ymm10[2,3,4],ymm15[5],ymm10[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm10[0,1],ymm5[2,3],ymm10[4,5],ymm5[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2],ymm1[3],ymm4[4,5,6],ymm1[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm14[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,1,3,3,6,5,7,7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm3 = ymm5[2],ymm3[2],ymm5[3],ymm3[3],ymm5[6],ymm3[6],ymm5[7],ymm3[7]
+; AVX2-NEXT:    vmovdqa (%r10), %ymm5
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm3[0,1],ymm0[2,3],ymm3[4,5],ymm0[6,7]
+; AVX2-NEXT:    vmovdqa (%rax), %ymm4
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm5[4],ymm4[4],ymm5[5],ymm4[5],ymm5[6],ymm4[6],ymm5[7],ymm4[7],ymm5[12],ymm4[12],ymm5[13],ymm4[13],ymm5[14],ymm4[14],ymm5[15],ymm4[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm8[4],ymm7[4],ymm8[5],ymm7[5],ymm8[6],ymm7[6],ymm8[7],ymm7[7],ymm8[12],ymm7[12],ymm8[13],ymm7[13],ymm8[14],ymm7[14],ymm8[15],ymm7[15]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm9 = ymm6[0],ymm3[0],ymm6[1],ymm3[1],ymm6[4],ymm3[4],ymm6[5],ymm3[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
+; AVX2-NEXT:    vmovdqa (%rdi), %ymm13
+; AVX2-NEXT:    vmovdqa (%rsi), %ymm14
+; AVX2-NEXT:    vmovdqa (%rdx), %ymm15
+; AVX2-NEXT:    vmovdqa (%rcx), %ymm2
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm15[4],ymm2[4],ymm15[5],ymm2[5],ymm15[6],ymm2[6],ymm15[7],ymm2[7],ymm15[12],ymm2[12],ymm15[13],ymm2[13],ymm15[14],ymm2[14],ymm15[15],ymm2[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm13[4],ymm14[4],ymm13[5],ymm14[5],ymm13[6],ymm14[6],ymm13[7],ymm14[7],ymm13[12],ymm14[12],ymm13[13],ymm14[13],ymm13[14],ymm14[14],ymm13[15],ymm14[15]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm12 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm12[0,1],ymm9[2,3],ymm12[4,5],ymm9[6,7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm3 = ymm6[2],ymm3[2],ymm6[3],ymm3[3],ymm6[6],ymm3[6],ymm6[7],ymm3[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm3[0,2,2,3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm4[1],ymm0[2,3,4],ymm4[5],ymm0[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm12[0],ymm13[0],ymm12[1],ymm13[1],ymm12[2],ymm13[2],ymm12[3],ymm13[3],ymm12[8],ymm13[8],ymm12[9],ymm13[9],ymm12[10],ymm13[10],ymm12[11],ymm13[11]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm8[0],ymm7[0],ymm8[1],ymm7[1],ymm8[2],ymm7[2],ymm8[3],ymm7[3],ymm8[8],ymm7[8],ymm8[9],ymm7[9],ymm8[10],ymm7[10],ymm8[11],ymm7[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm0[0,0,2,1,4,4,6,5]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm5[0],ymm4[0],ymm5[1],ymm4[1],ymm5[2],ymm4[2],ymm5[3],ymm4[3],ymm5[8],ymm4[8],ymm5[9],ymm4[9],ymm5[10],ymm4[10],ymm5[11],ymm4[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm8[0],ymm7[0],ymm8[1],ymm7[1],ymm8[2],ymm7[2],ymm8[3],ymm7[3],ymm8[8],ymm7[8],ymm8[9],ymm7[9],ymm8[10],ymm7[10],ymm8[11],ymm7[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm15[0],ymm2[0],ymm15[1],ymm2[1],ymm15[2],ymm2[2],ymm15[3],ymm2[3],ymm15[8],ymm2[8],ymm15[9],ymm2[9],ymm15[10],ymm2[10],ymm15[11],ymm2[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm13[0],ymm14[0],ymm13[1],ymm14[1],ymm13[2],ymm14[2],ymm13[3],ymm14[3],ymm13[8],ymm14[8],ymm13[9],ymm14[9],ymm13[10],ymm14[10],ymm13[11],ymm14[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[4],ymm1[4],ymm3[5],ymm1[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm4[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm7[0,1,2],ymm5[3],ymm7[4,5,6],ymm5[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm9[0],ymm6[0],ymm9[1],ymm6[1],ymm9[2],ymm6[2],ymm9[3],ymm6[3],ymm9[8],ymm6[8],ymm9[9],ymm6[9],ymm9[10],ymm6[10],ymm9[11],ymm6[11]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm3[0],ymm2[0],ymm3[1],ymm2[1],ymm3[2],ymm2[2],ymm3[3],ymm2[3],ymm3[8],ymm2[8],ymm3[9],ymm2[9],ymm3[10],ymm2[10],ymm3[11],ymm2[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm6[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm2[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm7[0],ymm3[1],ymm7[2,3,4],ymm3[5],ymm7[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm5[2,3],ymm3[4,5],ymm5[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6],ymm0[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm6[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,1,3,3,6,5,7,7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm6 = ymm4[0],ymm2[0],ymm4[1],ymm2[1],ymm4[4],ymm2[4],ymm4[5],ymm2[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm1 = ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[6],ymm1[6],ymm3[7],ymm1[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm4[2],ymm2[2],ymm4[3],ymm2[3],ymm4[6],ymm2[6],ymm4[7],ymm2[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0],ymm4[1],ymm2[2,3,4],ymm4[5],ymm2[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3],ymm2[4,5],ymm0[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-NEXT:    vmovdqa %ymm0, 160(%rax)
-; AVX2-NEXT:    vmovdqa %ymm3, 128(%rax)
-; AVX2-NEXT:    vmovdqa %ymm1, 224(%rax)
-; AVX2-NEXT:    vmovdqa %ymm15, 192(%rax)
+; AVX2-NEXT:    vmovdqa %ymm1, 160(%rax)
+; AVX2-NEXT:    vmovdqa %ymm5, 128(%rax)
+; AVX2-NEXT:    vmovdqa %ymm0, 224(%rax)
+; AVX2-NEXT:    vmovdqa %ymm9, 192(%rax)
 ; AVX2-NEXT:    vmovdqa %ymm11, 416(%rax)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 384(%rax)
+; AVX2-NEXT:    vmovdqa %ymm10, 384(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 480(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 448(%rax)
-; AVX2-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 96(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm0, 96(%rax)
+; AVX2-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 64(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 288(%rax)
@@ -3854,295 +3654,195 @@ define void @store_i16_stride8_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.ve
 ; AVX2-NEXT:    vmovaps %ymm0, (%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 32(%rax)
-; AVX2-NEXT:    addq $264, %rsp # imm = 0x108
+; AVX2-NEXT:    addq $232, %rsp
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: store_i16_stride8_vf32:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    subq $264, %rsp # imm = 0x108
+; AVX2-FP-NEXT:    subq $232, %rsp
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX2-FP-NEXT:    vmovdqa (%rax), %xmm1
+; AVX2-FP-NEXT:    vmovdqa (%rax), %xmm0
+; AVX2-FP-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 32(%rax), %xmm10
+; AVX2-FP-NEXT:    vmovdqa (%r10), %xmm1
 ; AVX2-FP-NEXT:    vmovdqa %xmm1, (%rsp) # 16-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 32(%rax), %xmm11
-; AVX2-FP-NEXT:    vmovdqa (%r10), %xmm0
+; AVX2-FP-NEXT:    vmovdqa 32(%r10), %xmm9
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-FP-NEXT:    vmovdqa (%r9), %xmm0
 ; AVX2-FP-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-FP-NEXT:    vmovdqa (%r9), %xmm3
-; AVX2-FP-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa (%r8), %xmm2
 ; AVX2-FP-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm3[0,1,2],ymm1[3],ymm3[4,5,6],ymm1[7]
-; AVX2-FP-NEXT:    vmovdqa (%rcx), %xmm1
-; AVX2-FP-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
+; AVX2-FP-NEXT:    vmovdqa (%rcx), %xmm4
+; AVX2-FP-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa (%rdx), %xmm3
 ; AVX2-FP-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm15 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm15[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
-; AVX2-FP-NEXT:    vmovdqa (%rsi), %xmm1
-; AVX2-FP-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
+; AVX2-FP-NEXT:    vmovdqa (%rsi), %xmm12
 ; AVX2-FP-NEXT:    vmovdqa (%rdi), %xmm13
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm13[0],xmm1[0],xmm13[1],xmm1[1],xmm13[2],xmm1[2],xmm13[3],xmm1[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm14 = xmm1[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0],ymm3[1],ymm14[2,3,4],ymm3[5],ymm14[6,7]
-; AVX2-FP-NEXT:    vmovdqa 32(%r10), %xmm9
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm14[0,1],ymm5[2,3],ymm14[4,5],ymm5[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 32(%r9), %xmm10
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 32(%r8), %xmm14
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm15[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3,4],ymm2[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm14 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm15 = xmm14[2],xmm3[2],xmm14[3],xmm3[3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm0[2,3],ymm15[4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm9[4],xmm11[4],xmm9[5],xmm11[5],xmm9[6],xmm11[6],xmm9[7],xmm11[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm4[0,0,1,1]
+; AVX2-FP-NEXT:    vmovdqa 32(%r9), %xmm8
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; AVX2-FP-NEXT:    vmovdqa 32(%r8), %xmm15
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm14[4],xmm10[4],xmm14[5],xmm10[5],xmm14[6],xmm10[6],xmm14[7],xmm10[7]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm3[0,1,2],ymm1[3],ymm3[4,5,6],ymm1[7]
-; AVX2-FP-NEXT:    vmovdqa 32(%rcx), %xmm15
-; AVX2-FP-NEXT:    vmovdqa 32(%rdx), %xmm8
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm8[4],xmm15[4],xmm8[5],xmm15[5],xmm8[6],xmm15[6],xmm8[7],xmm15[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
-; AVX2-FP-NEXT:    vmovdqa 32(%rsi), %xmm7
-; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %xmm6
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm6[4],xmm7[4],xmm6[5],xmm7[5],xmm6[6],xmm7[6],xmm6[7],xmm7[7]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm12 = xmm0[0],zero,xmm0[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm12[0],ymm3[1],ymm12[2,3,4],ymm3[5],ymm12[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm5[2,3],ymm3[4,5],ymm5[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm4[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3],ymm2[4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm14[0],xmm3[0],xmm14[1],xmm3[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm9[4],xmm10[4],xmm9[5],xmm10[5],xmm9[6],xmm10[6],xmm9[7],xmm10[7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm14 = xmm15[4],xmm8[4],xmm15[5],xmm8[5],xmm15[6],xmm8[6],xmm15[7],xmm8[7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm14[0],xmm3[0],xmm14[1],xmm3[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm2[0,0,2,1]
+; AVX2-FP-NEXT:    vmovdqa 32(%rcx), %xmm6
+; AVX2-FP-NEXT:    vmovdqa 32(%rdx), %xmm5
+; AVX2-FP-NEXT:    vmovdqa 32(%rsi), %xmm4
+; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %xmm2
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm5[4],xmm6[4],xmm5[5],xmm6[5],xmm5[6],xmm6[6],xmm5[7],xmm6[7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm11 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm11[0,1],ymm7[2,3],ymm11[4,5],ymm7[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm3 = xmm14[2],xmm3[2],xmm14[3],xmm3[3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm3[0,0,2,1]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3,4],ymm1[5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm9[0],xmm11[0],xmm9[1],xmm11[1],xmm9[2],xmm11[2],xmm9[3],xmm11[3]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm14[0],xmm10[0],xmm14[1],xmm10[1],xmm14[2],xmm10[2],xmm14[3],xmm10[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm8[0],xmm15[0],xmm8[1],xmm15[1],xmm8[2],xmm15[2],xmm8[3],xmm15[3]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm6[0],xmm7[0],xmm6[1],xmm7[1],xmm6[2],xmm7[2],xmm6[3],xmm7[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm6 = xmm5[0],zero,xmm5[1],zero
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm15[0],xmm8[0],xmm15[1],xmm8[1],xmm15[2],xmm8[2],xmm15[3],xmm8[3]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm2[0],xmm5[0],xmm2[1],xmm5[1]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0],ymm4[1],ymm6[2,3,4],ymm4[5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3],ymm4[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm6[0,1],ymm4[2,3],ymm6[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm2[2],xmm5[2],xmm2[3],xmm5[3]
+; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm14
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm3[2,2,3,3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm5[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
-; AVX2-FP-NEXT:    vmovdqa 32(%r8), %ymm14
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 32(%r9), %ymm15
-; AVX2-FP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; AVX2-FP-NEXT:    vpunpckhwd (%rsp), %xmm0, %xmm1 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    vmovdqa 32(%r8), %ymm15
+; AVX2-FP-NEXT:    vmovdqa (%rsp), %xmm0 # 16-byte Reload
+; AVX2-FP-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm1 # 16-byte Folded Reload
 ; AVX2-FP-NEXT:    # xmm1 = xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-FP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; AVX2-FP-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm2 = xmm2[4],mem[4],xmm2[5],mem[5],xmm2[6],mem[6],xmm2[7],mem[7]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2],ymm0[3],ymm3[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa 32(%r10), %ymm5
+; AVX2-FP-NEXT:    vmovdqa 32(%r9), %ymm4
+; AVX2-FP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX2-FP-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm2 # 16-byte Folded Reload
+; AVX2-FP-NEXT:    # xmm2 = xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7]
+; AVX2-FP-NEXT:    vmovdqa 32(%r10), %ymm6
 ; AVX2-FP-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; AVX2-FP-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm7 # 16-byte Folded Reload
 ; AVX2-FP-NEXT:    # xmm7 = xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm6 = xmm7[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,3]
-; AVX2-FP-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm13, %xmm8 # 16-byte Folded Reload
-; AVX2-FP-NEXT:    # xmm8 = xmm13[4],mem[4],xmm13[5],mem[5],xmm13[6],mem[6],xmm13[7],mem[7]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm9 = xmm8[0],zero,xmm8[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm9[0],ymm6[1],ymm9[2,3,4],ymm6[5],ymm9[6,7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm8 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm9 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm10 = xmm9[0],xmm7[0],xmm9[1],xmm7[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm8[2,3],ymm10[4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 32(%rax), %ymm10
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1],ymm3[2,3],ymm6[4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm7[2,2,3,3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm9[2],xmm7[2],xmm9[3],xmm7[3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm8[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3,4],ymm2[5],ymm3[6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm5[4],ymm10[4],ymm5[5],ymm10[5],ymm5[6],ymm10[6],ymm5[7],ymm10[7],ymm5[12],ymm10[12],ymm5[13],ymm10[13],ymm5[14],ymm10[14],ymm5[15],ymm10[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm1[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm14[4],ymm15[4],ymm14[5],ymm15[5],ymm14[6],ymm15[6],ymm14[7],ymm15[7],ymm14[12],ymm15[12],ymm14[13],ymm15[13],ymm14[14],ymm15[14],ymm14[15],ymm15[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm8 = ymm3[0,1,1,3,4,5,5,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm6[4],ymm10[4],ymm6[5],ymm10[5],ymm6[6],ymm10[6],ymm6[7],ymm10[7],ymm6[12],ymm10[12],ymm6[13],ymm10[13],ymm6[14],ymm10[14],ymm6[15],ymm10[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm15[4],ymm4[4],ymm15[5],ymm4[5],ymm15[6],ymm4[6],ymm15[7],ymm4[7],ymm15[12],ymm4[12],ymm15[13],ymm4[13],ymm15[14],ymm4[14],ymm15[15],ymm4[15]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm8 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[4],ymm3[4],ymm2[5],ymm3[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm8[0,1,2],ymm2[3],ymm8[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vmovdqa 32(%rdx), %ymm11
-; AVX2-FP-NEXT:    vmovdqa 32(%rcx), %ymm12
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm9 = ymm11[4],ymm12[4],ymm11[5],ymm12[5],ymm11[6],ymm12[6],ymm11[7],ymm12[7],ymm11[12],ymm12[12],ymm11[13],ymm12[13],ymm11[14],ymm12[14],ymm11[15],ymm12[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm8 = ymm9[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,3]
-; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm13
-; AVX2-FP-NEXT:    vmovdqa 32(%rsi), %ymm2
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm13[4],ymm2[4],ymm13[5],ymm2[5],ymm13[6],ymm2[6],ymm13[7],ymm2[7],ymm13[12],ymm2[12],ymm13[13],ymm2[13],ymm13[14],ymm2[14],ymm13[15],ymm2[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm6 = ymm0[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0],ymm8[1],ymm6[2,3,4],ymm8[5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm4[2,3],ymm6[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3],ymm3[4,5,6],ymm1[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm9[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,1,3,3,6,5,7,7]
+; AVX2-FP-NEXT:    vmovdqa 32(%rsi), %ymm11
+; AVX2-FP-NEXT:    vmovdqa 32(%rdx), %ymm13
+; AVX2-FP-NEXT:    vmovdqa 32(%rcx), %ymm9
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm13[4],ymm9[4],ymm13[5],ymm9[5],ymm13[6],ymm9[6],ymm13[7],ymm9[7],ymm13[12],ymm9[12],ymm13[13],ymm9[13],ymm13[14],ymm9[14],ymm13[15],ymm9[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm14[4],ymm11[4],ymm14[5],ymm11[5],ymm14[6],ymm11[6],ymm14[7],ymm11[7],ymm14[12],ymm11[12],ymm14[13],ymm11[13],ymm14[14],ymm11[14],ymm14[15],ymm11[15]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1],ymm8[2,3],ymm5[4,5],ymm8[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[6],ymm3[6],ymm2[7],ymm3[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm2[0,2,2,3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm3[1],ymm0[2,3,4],ymm3[5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm5[0],ymm10[0],ymm5[1],ymm10[1],ymm5[2],ymm10[2],ymm5[3],ymm10[3],ymm5[8],ymm10[8],ymm5[9],ymm10[9],ymm5[10],ymm10[10],ymm5[11],ymm10[11]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm14[0],ymm15[0],ymm14[1],ymm15[1],ymm14[2],ymm15[2],ymm14[3],ymm15[3],ymm14[8],ymm15[8],ymm14[9],ymm15[9],ymm14[10],ymm15[10],ymm14[11],ymm15[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm0[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm1[0,1,1,3,4,5,5,7]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm6[0],ymm10[0],ymm6[1],ymm10[1],ymm6[2],ymm10[2],ymm6[3],ymm10[3],ymm6[8],ymm10[8],ymm6[9],ymm10[9],ymm6[10],ymm10[10],ymm6[11],ymm10[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm15[0],ymm4[0],ymm15[1],ymm4[1],ymm15[2],ymm4[2],ymm15[3],ymm4[3],ymm15[8],ymm4[8],ymm15[9],ymm4[9],ymm15[10],ymm4[10],ymm15[11],ymm4[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm13[0],ymm9[0],ymm13[1],ymm9[1],ymm13[2],ymm9[2],ymm13[3],ymm9[3],ymm13[8],ymm9[8],ymm13[9],ymm9[9],ymm13[10],ymm9[10],ymm13[11],ymm9[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm4 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[4],ymm0[4],ymm2[5],ymm0[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1,2],ymm3[3],ymm4[4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm11[0],ymm12[0],ymm11[1],ymm12[1],ymm11[2],ymm12[2],ymm11[3],ymm12[3],ymm11[8],ymm12[8],ymm11[9],ymm12[9],ymm11[10],ymm12[10],ymm11[11],ymm12[11]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm13[0],ymm2[0],ymm13[1],ymm2[1],ymm13[2],ymm2[2],ymm13[3],ymm2[3],ymm13[8],ymm2[8],ymm13[9],ymm2[9],ymm13[10],ymm2[10],ymm13[11],ymm2[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm4[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm6 = ymm2[0,1,1,3,4,5,5,7]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm14[0],ymm11[0],ymm14[1],ymm11[1],ymm14[2],ymm11[2],ymm14[3],ymm11[3],ymm14[8],ymm11[8],ymm14[9],ymm11[9],ymm14[10],ymm11[10],ymm14[11],ymm11[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm6 = ymm5[0],ymm3[0],ymm5[1],ymm3[1],ymm5[4],ymm3[4],ymm5[5],ymm3[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0],ymm5[1],ymm6[2,3,4],ymm5[5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3],ymm5[4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa (%r10), %ymm12
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa (%rax), %ymm13
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm4[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm12[4],ymm13[4],ymm12[5],ymm13[5],ymm12[6],ymm13[6],ymm12[7],ymm13[7],ymm12[12],ymm13[12],ymm12[13],ymm13[13],ymm12[14],ymm13[14],ymm12[15],ymm13[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm1[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm6[0,1],ymm4[2,3],ymm6[4,5],ymm4[6,7]
 ; AVX2-FP-NEXT:    vmovdqa (%r8), %ymm8
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[6],ymm0[6],ymm2[7],ymm0[7]
 ; AVX2-FP-NEXT:    vmovdqa (%r9), %ymm7
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm8[4],ymm7[4],ymm8[5],ymm7[5],ymm8[6],ymm7[6],ymm8[7],ymm7[7],ymm8[12],ymm7[12],ymm8[13],ymm7[13],ymm8[14],ymm7[14],ymm8[15],ymm7[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm4[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm0[3],ymm5[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vmovdqa (%rdx), %ymm9
-; AVX2-FP-NEXT:    vmovdqa (%rcx), %ymm6
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm14 = ymm9[4],ymm6[4],ymm9[5],ymm6[5],ymm9[6],ymm6[6],ymm9[7],ymm6[7],ymm9[12],ymm6[12],ymm9[13],ymm6[13],ymm9[14],ymm6[14],ymm9[15],ymm6[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm15 = ymm14[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
-; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm3
-; AVX2-FP-NEXT:    vmovdqa (%rsi), %ymm2
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm3[4],ymm2[4],ymm3[5],ymm2[5],ymm3[6],ymm2[6],ymm3[7],ymm2[7],ymm3[12],ymm2[12],ymm3[13],ymm2[13],ymm3[14],ymm2[14],ymm3[15],ymm2[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm10 = ymm0[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0],ymm15[1],ymm10[2,3,4],ymm15[5],ymm10[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm10[0,1],ymm5[2,3],ymm10[4,5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1,2],ymm1[3],ymm4[4,5,6],ymm1[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm14[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,1,3,3,6,5,7,7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm3 = ymm5[2],ymm3[2],ymm5[3],ymm3[3],ymm5[6],ymm3[6],ymm5[7],ymm3[7]
+; AVX2-FP-NEXT:    vmovdqa (%r10), %ymm5
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm3[0,1],ymm0[2,3],ymm3[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vmovdqa (%rax), %ymm4
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm5[4],ymm4[4],ymm5[5],ymm4[5],ymm5[6],ymm4[6],ymm5[7],ymm4[7],ymm5[12],ymm4[12],ymm5[13],ymm4[13],ymm5[14],ymm4[14],ymm5[15],ymm4[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm8[4],ymm7[4],ymm8[5],ymm7[5],ymm8[6],ymm7[6],ymm8[7],ymm7[7],ymm8[12],ymm7[12],ymm8[13],ymm7[13],ymm8[14],ymm7[14],ymm8[15],ymm7[15]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm9 = ymm6[0],ymm3[0],ymm6[1],ymm3[1],ymm6[4],ymm3[4],ymm6[5],ymm3[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
+; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm13
+; AVX2-FP-NEXT:    vmovdqa (%rsi), %ymm14
+; AVX2-FP-NEXT:    vmovdqa (%rdx), %ymm15
+; AVX2-FP-NEXT:    vmovdqa (%rcx), %ymm2
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm15[4],ymm2[4],ymm15[5],ymm2[5],ymm15[6],ymm2[6],ymm15[7],ymm2[7],ymm15[12],ymm2[12],ymm15[13],ymm2[13],ymm15[14],ymm2[14],ymm15[15],ymm2[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm13[4],ymm14[4],ymm13[5],ymm14[5],ymm13[6],ymm14[6],ymm13[7],ymm14[7],ymm13[12],ymm14[12],ymm13[13],ymm14[13],ymm13[14],ymm14[14],ymm13[15],ymm14[15]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm12 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm12[0,1],ymm9[2,3],ymm12[4,5],ymm9[6,7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm3 = ymm6[2],ymm3[2],ymm6[3],ymm3[3],ymm6[6],ymm3[6],ymm6[7],ymm3[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm3[0,2,2,3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm4[1],ymm0[2,3,4],ymm4[5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm12[0],ymm13[0],ymm12[1],ymm13[1],ymm12[2],ymm13[2],ymm12[3],ymm13[3],ymm12[8],ymm13[8],ymm12[9],ymm13[9],ymm12[10],ymm13[10],ymm12[11],ymm13[11]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm8[0],ymm7[0],ymm8[1],ymm7[1],ymm8[2],ymm7[2],ymm8[3],ymm7[3],ymm8[8],ymm7[8],ymm8[9],ymm7[9],ymm8[10],ymm7[10],ymm8[11],ymm7[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm0[0,0,2,1,4,4,6,5]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm5[0],ymm4[0],ymm5[1],ymm4[1],ymm5[2],ymm4[2],ymm5[3],ymm4[3],ymm5[8],ymm4[8],ymm5[9],ymm4[9],ymm5[10],ymm4[10],ymm5[11],ymm4[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm8[0],ymm7[0],ymm8[1],ymm7[1],ymm8[2],ymm7[2],ymm8[3],ymm7[3],ymm8[8],ymm7[8],ymm8[9],ymm7[9],ymm8[10],ymm7[10],ymm8[11],ymm7[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm15[0],ymm2[0],ymm15[1],ymm2[1],ymm15[2],ymm2[2],ymm15[3],ymm2[3],ymm15[8],ymm2[8],ymm15[9],ymm2[9],ymm15[10],ymm2[10],ymm15[11],ymm2[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm13[0],ymm14[0],ymm13[1],ymm14[1],ymm13[2],ymm14[2],ymm13[3],ymm14[3],ymm13[8],ymm14[8],ymm13[9],ymm14[9],ymm13[10],ymm14[10],ymm13[11],ymm14[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[4],ymm1[4],ymm3[5],ymm1[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm4[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm7[0,1,2],ymm5[3],ymm7[4,5,6],ymm5[7]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm9[0],ymm6[0],ymm9[1],ymm6[1],ymm9[2],ymm6[2],ymm9[3],ymm6[3],ymm9[8],ymm6[8],ymm9[9],ymm6[9],ymm9[10],ymm6[10],ymm9[11],ymm6[11]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm3[0],ymm2[0],ymm3[1],ymm2[1],ymm3[2],ymm2[2],ymm3[3],ymm2[3],ymm3[8],ymm2[8],ymm3[9],ymm2[9],ymm3[10],ymm2[10],ymm3[11],ymm2[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm6[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm2[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm7[0],ymm3[1],ymm7[2,3,4],ymm3[5],ymm7[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm5[2,3],ymm3[4,5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm6[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,1,3,3,6,5,7,7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm6 = ymm4[0],ymm2[0],ymm4[1],ymm2[1],ymm4[4],ymm2[4],ymm4[5],ymm2[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm1 = ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[6],ymm1[6],ymm3[7],ymm1[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm4[2],ymm2[2],ymm4[3],ymm2[3],ymm4[6],ymm2[6],ymm4[7],ymm2[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0],ymm4[1],ymm2[2,3,4],ymm4[5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3],ymm2[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FP-NEXT:    vmovdqa %ymm0, 160(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm3, 128(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm1, 224(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm15, 192(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm1, 160(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm5, 128(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm0, 224(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm9, 192(%rax)
 ; AVX2-FP-NEXT:    vmovdqa %ymm11, 416(%rax)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 384(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm10, 384(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 480(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 448(%rax)
-; AVX2-FP-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 96(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm0, 96(%rax)
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 64(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 288(%rax)
@@ -4156,7 +3856,7 @@ define void @store_i16_stride8_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.ve
 ; AVX2-FP-NEXT:    vmovaps %ymm0, (%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rax)
-; AVX2-FP-NEXT:    addq $264, %rsp # imm = 0x108
+; AVX2-FP-NEXT:    addq $232, %rsp
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
@@ -6964,533 +6664,339 @@ define void @store_i16_stride8_vf64(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.ve
 ;
 ; AVX2-LABEL: store_i16_stride8_vf64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    subq $744, %rsp # imm = 0x2E8
+; AVX2-NEXT:    subq $712, %rsp # imm = 0x2C8
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX2-NEXT:    vmovdqa (%rax), %xmm6
+; AVX2-NEXT:    vmovdqa (%rax), %xmm5
 ; AVX2-NEXT:    vmovdqa 32(%rax), %xmm0
-; AVX2-NEXT:    vmovdqa (%r10), %xmm7
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm7[0],xmm6[0],xmm7[1],xmm6[1],xmm7[2],xmm6[2],xmm7[3],xmm6[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm3[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
+; AVX2-NEXT:    vmovdqa (%r10), %xmm6
+; AVX2-NEXT:    vmovdqa 32(%r10), %xmm1
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3]
 ; AVX2-NEXT:    vmovdqa (%r9), %xmm8
 ; AVX2-NEXT:    vmovdqa (%r8), %xmm9
 ; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm4[0],zero,xmm4[1],zero
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-NEXT:    vmovdqa (%rcx), %xmm10
 ; AVX2-NEXT:    vmovdqa (%rdx), %xmm11
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm5[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
 ; AVX2-NEXT:    vmovdqa (%rsi), %xmm12
 ; AVX2-NEXT:    vmovdqa (%rdi), %xmm13
 ; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm14 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm15 = xmm14[0],zero,xmm14[1],zero
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm15 = xmm14[0],xmm7[0],xmm14[1],xmm7[1]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm15[0],ymm2[1],ymm15[2,3,4],ymm2[5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqa 32(%r10), %xmm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm15[0,1],ymm1[2,3],ymm15[4,5],ymm1[6,7]
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 32(%r9), %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm4[0,1,2],ymm3[3],ymm4[4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqa 32(%r8), %xmm3
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm5[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm5 = xmm14[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm4[1],ymm5[2,3,4],ymm4[5],ymm5[6,7]
-; AVX2-NEXT:    vmovdqa 32(%rcx), %xmm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1],ymm15[2,3],ymm5[4,5],ymm15[6,7]
-; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 32(%rdx), %xmm5
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm7[4],xmm6[4],xmm7[5],xmm6[5],xmm7[6],xmm6[6],xmm7[7],xmm6[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm7 = xmm6[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm9 = xmm8[0],zero,xmm8[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2],ymm7[3],ymm9[4,5,6],ymm7[7]
-; AVX2-NEXT:    vmovdqa 32(%rsi), %xmm7
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1],ymm2[2,3],ymm15[4,5],ymm2[6,7]
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 32(%r9), %xmm2
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm15 = xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX2-NEXT:    vmovdqa 32(%r8), %xmm4
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm7 = xmm14[2],xmm7[2],xmm14[3],xmm7[3]
+; AVX2-NEXT:    vmovdqa 32(%rcx), %xmm3
+; AVX2-NEXT:    vpermq {{.*#+}} ymm14 = ymm15[0,0,2,1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1],ymm14[2,3],ymm7[4,5],ymm14[6,7]
+; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 32(%rdx), %xmm7
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
+; AVX2-NEXT:    vmovdqa 32(%rsi), %xmm5
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm9 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
+; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm8
 ; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm10 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm11 = xmm10[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,1,3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm11 = xmm9[0],xmm6[0],xmm9[1],xmm6[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,2,1]
 ; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm12 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm13 = xmm12[0],zero,xmm12[1],zero
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm13 = xmm12[0],xmm10[0],xmm12[1],xmm10[1]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm13[0],ymm11[1],ymm13[2,3,4],ymm11[5],ymm13[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm11[0,1],ymm9[2,3],ymm11[4,5],ymm9[6,7]
-; AVX2-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,2,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm13[0,1],ymm11[2,3],ymm13[4,5],ymm11[6,7]
+; AVX2-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm9[2],xmm6[2],xmm9[3],xmm6[3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm8[0,1,2],ymm6[3],ymm8[4,5,6],ymm6[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm8 = xmm10[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm9 = xmm12[2,2,3,3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm9 = xmm12[2],xmm10[2],xmm12[3],xmm10[3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0],ymm8[1],ymm9[2,3,4],ymm8[5],ymm9[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm8[0,1],ymm6[2,3],ymm8[4,5],ymm6[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm9[0,1],ymm6[2,3],ymm9[4,5],ymm6[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm8 = xmm6[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm10 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm9 = xmm10[0],zero,xmm10[1],zero
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm4[0],xmm2[0],xmm4[1],xmm2[1],xmm4[2],xmm2[2],xmm4[3],xmm2[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm10 = xmm9[0],xmm6[0],xmm9[1],xmm6[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,0,2,1]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm7[0],xmm3[0],xmm7[1],xmm3[1],xmm7[2],xmm3[2],xmm7[3],xmm3[3]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm12 = xmm8[0],xmm5[0],xmm8[1],xmm5[1],xmm8[2],xmm5[2],xmm8[3],xmm5[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm13 = xmm12[0],xmm11[0],xmm12[1],xmm11[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm13[0,1],ymm10[2,3],ymm13[4,5],ymm10[6,7]
+; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm9 = xmm9[2],xmm6[2],xmm9[3],xmm6[3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm10 = xmm12[2],xmm11[2],xmm12[3],xmm11[3]
+; AVX2-NEXT:    vmovdqa 64(%rax), %xmm6
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1,2],ymm8[3],ymm9[4,5,6],ymm8[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm9 = xmm11[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm9[0,1,1,3]
-; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm9
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm9[0],xmm7[0],xmm9[1],xmm7[1],xmm9[2],xmm7[2],xmm9[3],xmm7[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm14 = xmm13[0],zero,xmm13[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm14[0],ymm12[1],ymm14[2,3,4],ymm12[5],ymm14[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm12[0,1],ymm8[2,3],ymm12[4,5],ymm8[6,7]
-; AVX2-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm8 = xmm10[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2],ymm6[3],ymm8[4,5,6],ymm6[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm6 = xmm11[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm10 = xmm13[2,2,3,3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0],ymm6[1],ymm10[2,3,4],ymm6[5],ymm10[6,7]
-; AVX2-NEXT:    vmovdqa 64(%rax), %xmm6
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm10[0,1],ymm8[2,3],ymm10[4,5],ymm8[6,7]
-; AVX2-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 64(%r10), %xmm8
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm2[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm3[4],xmm1[4],xmm3[5],xmm1[5],xmm3[6],xmm1[6],xmm3[7],xmm1[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm3[0],zero,xmm3[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0,1],ymm9[2,3],ymm10[4,5],ymm9[6,7]
+; AVX2-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vmovdqa 64(%r10), %xmm9
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm10 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; AVX2-NEXT:    vmovdqa 64(%r9), %xmm0
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm4[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm9[4],xmm7[4],xmm9[5],xmm7[5],xmm9[6],xmm7[6],xmm9[7],xmm7[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm7 = xmm5[0],zero,xmm5[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0],ymm1[1],ymm7[2,3,4],ymm1[5],ymm7[6,7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm4[4],xmm2[4],xmm4[5],xmm2[5],xmm4[6],xmm2[6],xmm4[7],xmm2[7]
 ; AVX2-NEXT:    vmovdqa 64(%r8), %xmm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1],ymm10[2,3],ymm7[4,5],ymm10[6,7]
-; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm7[4],xmm3[4],xmm7[5],xmm3[5],xmm7[6],xmm3[6],xmm7[7],xmm3[7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm4[0],xmm10[0],xmm4[1],xmm10[1]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm4[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm5[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3,4],ymm3[5],ymm4[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm8[4],xmm5[4],xmm8[5],xmm5[5],xmm8[6],xmm5[6],xmm8[7],xmm5[7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm7 = xmm5[0],xmm3[0],xmm5[1],xmm3[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1],ymm2[2,3],ymm7[4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm8[0],xmm6[0],xmm8[1],xmm6[1],xmm8[2],xmm6[2],xmm8[3],xmm6[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm7[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm9[0],zero,xmm9[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
 ; AVX2-NEXT:    vmovdqa 64(%rcx), %xmm2
-; AVX2-NEXT:    vmovdqa 64(%rdx), %xmm3
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm11[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm4[0,1,1,3]
-; AVX2-NEXT:    vmovdqa 64(%rsi), %xmm4
-; AVX2-NEXT:    vmovdqa 64(%rdi), %xmm5
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm14 = xmm13[0],zero,xmm13[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm14[0],ymm12[1],ymm14[2,3,4],ymm12[5],ymm14[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm12[0,1],ymm10[2,3],ymm12[4,5],ymm10[6,7]
-; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm7 = xmm7[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm9 = xmm9[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm9[0,1,2],ymm7[3],ymm9[4,5,6],ymm7[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm9 = xmm11[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm10 = xmm13[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0],ymm9[1],ymm10[2,3,4],ymm9[5],ymm10[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm9[0,1],ymm7[2,3],ymm9[4,5],ymm7[6,7]
-; AVX2-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm8[4],xmm6[4],xmm8[5],xmm6[5],xmm8[6],xmm6[6],xmm8[7],xmm6[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm6[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm7 = xmm0[0],zero,xmm0[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2],ymm1[3],ymm7[4,5,6],ymm1[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm2[0,0,1,1]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm4 = xmm4[2],xmm10[2],xmm4[3],xmm10[3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm3 = xmm5[2],xmm3[2],xmm5[3],xmm3[3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm9[0],xmm6[0],xmm9[1],xmm6[1],xmm9[2],xmm6[2],xmm9[3],xmm6[3]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,0,2,1]
+; AVX2-NEXT:    vmovdqa 64(%rdx), %xmm7
+; AVX2-NEXT:    vmovdqa 64(%rsi), %xmm8
+; AVX2-NEXT:    vmovdqa 64(%rdi), %xmm10
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm7[0],xmm2[0],xmm7[1],xmm2[1],xmm7[2],xmm2[2],xmm7[3],xmm2[3]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm12 = xmm10[0],xmm8[0],xmm10[1],xmm8[1],xmm10[2],xmm8[2],xmm10[3],xmm8[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm13 = xmm12[0],xmm11[0],xmm12[1],xmm11[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1],ymm5[2,3],ymm13[4,5],ymm5[6,7]
+; AVX2-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm3 = xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm4 = xmm12[2],xmm11[2],xmm12[3],xmm11[3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm3[0],zero,xmm3[1],zero
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3],ymm4[4,5],ymm3[6,7]
+; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm9[4],xmm6[4],xmm9[5],xmm6[5],xmm9[6],xmm6[6],xmm9[7],xmm6[7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm7[4],xmm2[4],xmm7[5],xmm2[5],xmm7[6],xmm2[6],xmm7[7],xmm2[7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm10[4],xmm8[4],xmm10[5],xmm8[5],xmm10[6],xmm8[6],xmm10[7],xmm8[7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0],ymm4[1],ymm5[2,3,4],ymm4[5],ymm5[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm1[2,3],ymm4[4,5],ymm1[6,7]
-; AVX2-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm6[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm2[2,3],ymm5[4,5],ymm2[6,7]
+; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm3[2],xmm0[3],xmm3[3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3],ymm0[4,5,6],ymm1[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[2,2,3,3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm4[2],xmm1[2],xmm4[3],xmm1[3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm3[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-NEXT:    vmovdqa 96(%rax), %xmm0
 ; AVX2-NEXT:    vmovdqa 96(%r10), %xmm1
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm8[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm2[0,0,2,1]
-; AVX2-NEXT:    vmovdqa 96(%r9), %xmm2
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-NEXT:    vmovdqa 96(%r9), %xmm3
 ; AVX2-NEXT:    vmovdqa 96(%r8), %xmm4
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm4[0],xmm2[0],xmm4[1],xmm2[1],xmm4[2],xmm2[2],xmm4[3],xmm2[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm9[0],zero,xmm9[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm5[0,1,2],ymm3[3],ymm5[4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqa 96(%rcx), %xmm3
-; AVX2-NEXT:    vmovdqa 96(%rdx), %xmm5
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm5[0],xmm3[0],xmm5[1],xmm3[1],xmm5[2],xmm3[2],xmm5[3],xmm3[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm6 = xmm11[0,0,1,1]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm6[0,1,1,3]
-; AVX2-NEXT:    vmovdqa 96(%rsi), %xmm6
-; AVX2-NEXT:    vmovdqa 96(%rdi), %xmm7
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm7[0],xmm6[0],xmm7[1],xmm6[1],xmm7[2],xmm6[2],xmm7[3],xmm6[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm14 = xmm13[0],zero,xmm13[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm14[0],ymm12[1],ymm14[2,3,4],ymm12[5],ymm14[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm12[0,1],ymm10[2,3],ymm12[4,5],ymm10[6,7]
-; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm9 = xmm9[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1,2],ymm8[3],ymm9[4,5,6],ymm8[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm9 = xmm11[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm10 = xmm13[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0],ymm9[1],ymm10[2,3,4],ymm9[5],ymm10[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1],ymm8[2,3],ymm9[4,5],ymm8[6,7]
-; AVX2-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm4[4],xmm2[4],xmm4[5],xmm2[5],xmm4[6],xmm2[6],xmm4[7],xmm2[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[0,0,1,1]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm5[0],xmm2[0],xmm5[1],xmm2[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
+; AVX2-NEXT:    vmovdqa 96(%rcx), %xmm7
+; AVX2-NEXT:    vmovdqa 96(%rdx), %xmm8
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm8[0],xmm7[0],xmm8[1],xmm7[1],xmm8[2],xmm7[2],xmm8[3],xmm7[3]
+; AVX2-NEXT:    vmovdqa 96(%rsi), %xmm10
+; AVX2-NEXT:    vmovdqa 96(%rdi), %xmm11
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm12 = xmm13[0],xmm9[0],xmm13[1],xmm9[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm12[0,1],ymm6[2,3],ymm12[4,5],ymm6[6,7]
+; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm5[2],xmm2[2],xmm5[3],xmm2[3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm13[2],xmm9[2],xmm13[3],xmm9[3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm5[4],xmm3[4],xmm5[5],xmm3[5],xmm5[6],xmm3[6],xmm5[7],xmm3[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm7[4],xmm6[4],xmm7[5],xmm6[5],xmm7[6],xmm6[6],xmm7[7],xmm6[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm5 = xmm3[0,0,1,1]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,1,3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm6 = xmm4[0],zero,xmm4[1],zero
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0],ymm5[1],ymm6[2,3,4],ymm5[5],ymm6[6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm2[2,3],ymm5[4,5],ymm2[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm8[4],xmm7[4],xmm8[5],xmm7[5],xmm8[6],xmm7[6],xmm8[7],xmm7[7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,1,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3],ymm5[4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm3[2],xmm2[2],xmm3[3],xmm2[3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm3[2,2,3,3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm4[2,2,3,3]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa (%r10), %ymm0
-; AVX2-NEXT:    vmovdqa (%rax), %ymm1
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT:    vmovdqa (%r8), %ymm4
-; AVX2-NEXT:    vmovdqa (%r9), %ymm5
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm4[0],ymm5[0],ymm4[1],ymm5[1],ymm4[2],ymm5[2],ymm4[3],ymm5[3],ymm4[8],ymm5[8],ymm4[9],ymm5[9],ymm4[10],ymm5[10],ymm4[11],ymm5[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm6[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm7[0,1,2],ymm3[3],ymm7[4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqa (%rdx), %ymm7
-; AVX2-NEXT:    vmovdqa (%rcx), %ymm8
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm9 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[8],ymm8[8],ymm7[9],ymm8[9],ymm7[10],ymm8[10],ymm7[11],ymm8[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm10 = ymm9[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,3]
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm11
-; AVX2-NEXT:    vmovdqa (%rsi), %ymm12
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm13 = ymm11[0],ymm12[0],ymm11[1],ymm12[1],ymm11[2],ymm12[2],ymm11[3],ymm12[3],ymm11[8],ymm12[8],ymm11[9],ymm12[9],ymm11[10],ymm12[10],ymm11[11],ymm12[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm14 = ymm13[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm14[0],ymm10[1],ymm14[2,3,4],ymm10[5],ymm14[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm10[0,1],ymm3[2,3],ymm10[4,5],ymm3[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm6[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm9[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm6 = ymm13[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm6[0],ymm3[1],ymm6[2,3,4],ymm3[5],ymm6[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm4[4],ymm5[4],ymm4[5],ymm5[5],ymm4[6],ymm5[6],ymm4[7],ymm5[7],ymm4[12],ymm5[12],ymm4[13],ymm5[13],ymm4[14],ymm5[14],ymm4[15],ymm5[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm0[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm1[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
+; AVX2-NEXT:    vmovdqa (%r8), %ymm11
+; AVX2-NEXT:    vmovdqa (%r9), %ymm1
+; AVX2-NEXT:    vmovdqa (%r10), %ymm2
+; AVX2-NEXT:    vmovdqa (%rax), %ymm3
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[8],ymm3[8],ymm2[9],ymm3[9],ymm2[10],ymm3[10],ymm2[11],ymm3[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm11[0],ymm1[0],ymm11[1],ymm1[1],ymm11[2],ymm1[2],ymm11[3],ymm1[3],ymm11[8],ymm1[8],ymm11[9],ymm1[9],ymm11[10],ymm1[10],ymm11[11],ymm1[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm6 = ymm5[0],ymm4[0],ymm5[1],ymm4[1],ymm5[4],ymm4[4],ymm5[5],ymm4[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm6[0,2,2,3]
+; AVX2-NEXT:    vmovdqa (%rdi), %ymm7
+; AVX2-NEXT:    vmovdqa (%rsi), %ymm8
+; AVX2-NEXT:    vmovdqa (%rdx), %ymm9
+; AVX2-NEXT:    vmovdqa (%rcx), %ymm6
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm10 = ymm9[0],ymm6[0],ymm9[1],ymm6[1],ymm9[2],ymm6[2],ymm9[3],ymm6[3],ymm9[8],ymm6[8],ymm9[9],ymm6[9],ymm9[10],ymm6[10],ymm9[11],ymm6[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[8],ymm8[8],ymm7[9],ymm8[9],ymm7[10],ymm8[10],ymm7[11],ymm8[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm13 = ymm0[0],ymm10[0],ymm0[1],ymm10[1],ymm0[4],ymm10[4],ymm0[5],ymm10[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm13[0,1],ymm12[2,3],ymm13[4,5],ymm12[6,7]
+; AVX2-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm4 = ymm5[2],ymm4[2],ymm5[3],ymm4[3],ymm5[6],ymm4[6],ymm5[7],ymm4[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm10[2],ymm0[3],ymm10[3],ymm0[6],ymm10[6],ymm0[7],ymm10[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm4[2,3],ymm0[4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[12],ymm3[12],ymm2[13],ymm3[13],ymm2[14],ymm3[14],ymm2[15],ymm3[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm11[4],ymm1[4],ymm11[5],ymm1[5],ymm11[6],ymm1[6],ymm11[7],ymm1[7],ymm11[12],ymm1[12],ymm11[13],ymm1[13],ymm11[14],ymm1[14],ymm11[15],ymm1[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm9[4],ymm6[4],ymm9[5],ymm6[5],ymm9[6],ymm6[6],ymm9[7],ymm6[7],ymm9[12],ymm6[12],ymm9[13],ymm6[13],ymm9[14],ymm6[14],ymm9[15],ymm6[15]
 ; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm7[4],ymm8[4],ymm7[5],ymm8[5],ymm7[6],ymm8[6],ymm7[7],ymm8[7],ymm7[12],ymm8[12],ymm7[13],ymm8[13],ymm7[14],ymm8[14],ymm7[15],ymm8[15]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm11[4],ymm12[4],ymm11[5],ymm12[5],ymm11[6],ymm12[6],ymm11[7],ymm12[7],ymm11[12],ymm12[12],ymm11[13],ymm12[13],ymm11[14],ymm12[14],ymm11[15],ymm12[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm3[0,0,2,1,4,4,6,5]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm4 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[4],ymm0[4],ymm1[5],ymm0[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm3[0],ymm2[0],ymm3[1],ymm2[1],ymm3[4],ymm2[4],ymm3[5],ymm2[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm6 = ymm4[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0],ymm5[1],ymm6[2,3,4],ymm5[5],ymm6[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm2[2,3],ymm5[4,5],ymm2[6,7]
-; AVX2-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3],ymm5[4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[6],ymm0[6],ymm1[7],ymm0[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm1 = ymm3[2],ymm2[2],ymm3[3],ymm2[3],ymm3[6],ymm2[6],ymm3[7],ymm2[7]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[0,2,2,3,4,6,6,7]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm4[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 32(%r10), %ymm13
-; AVX2-NEXT:    vmovdqa 32(%rax), %ymm11
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm13[0],ymm11[0],ymm13[1],ymm11[1],ymm13[2],ymm11[2],ymm13[3],ymm11[3],ymm13[8],ymm11[8],ymm13[9],ymm11[9],ymm13[10],ymm11[10],ymm13[11],ymm11[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT:    vmovdqa 32(%r8), %ymm4
-; AVX2-NEXT:    vmovdqa 32(%r9), %ymm5
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm4[0],ymm5[0],ymm4[1],ymm5[1],ymm4[2],ymm5[2],ymm4[3],ymm5[3],ymm4[8],ymm5[8],ymm4[9],ymm5[9],ymm4[10],ymm5[10],ymm4[11],ymm5[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm6[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm7[0,1,2],ymm3[3],ymm7[4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqa 32(%rdx), %ymm7
-; AVX2-NEXT:    vmovdqa 32(%rcx), %ymm8
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm9 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[8],ymm8[8],ymm7[9],ymm8[9],ymm7[10],ymm8[10],ymm7[11],ymm8[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm10 = ymm9[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,3]
-; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-NEXT:    vmovdqa 32(%rsi), %ymm3
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[8],ymm3[8],ymm1[9],ymm3[9],ymm1[10],ymm3[10],ymm1[11],ymm3[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm15 = ymm0[0,1,1,3,4,5,5,7]
+; AVX2-NEXT:    vmovdqa 32(%r8), %ymm9
+; AVX2-NEXT:    vmovdqa 32(%r9), %ymm8
+; AVX2-NEXT:    vmovdqa 32(%r10), %ymm7
+; AVX2-NEXT:    vmovdqa 32(%rax), %ymm3
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm7[0],ymm3[0],ymm7[1],ymm3[1],ymm7[2],ymm3[2],ymm7[3],ymm3[3],ymm7[8],ymm3[8],ymm7[9],ymm3[9],ymm7[10],ymm3[10],ymm7[11],ymm3[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm9[0],ymm8[0],ymm9[1],ymm8[1],ymm9[2],ymm8[2],ymm9[3],ymm8[3],ymm9[8],ymm8[8],ymm9[9],ymm8[9],ymm9[10],ymm8[10],ymm9[11],ymm8[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm6 = ymm5[0],ymm4[0],ymm5[1],ymm4[1],ymm5[4],ymm4[4],ymm5[5],ymm4[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm10 = ymm6[0,2,2,3]
+; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm11
+; AVX2-NEXT:    vmovdqa 32(%rsi), %ymm13
+; AVX2-NEXT:    vmovdqa 32(%rdx), %ymm2
+; AVX2-NEXT:    vmovdqa 32(%rcx), %ymm1
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm2[0],ymm1[0],ymm2[1],ymm1[1],ymm2[2],ymm1[2],ymm2[3],ymm1[3],ymm2[8],ymm1[8],ymm2[9],ymm1[9],ymm2[10],ymm1[10],ymm2[11],ymm1[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm11[0],ymm13[0],ymm11[1],ymm13[1],ymm11[2],ymm13[2],ymm11[3],ymm13[3],ymm11[8],ymm13[8],ymm11[9],ymm13[9],ymm11[10],ymm13[10],ymm11[11],ymm13[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm15 = ymm0[0],ymm6[0],ymm0[1],ymm6[1],ymm0[4],ymm6[4],ymm0[5],ymm6[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm15[0],ymm10[1],ymm15[2,3,4],ymm10[5],ymm15[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1],ymm12[2,3],ymm10[4,5],ymm12[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm15[0,1],ymm10[2,3],ymm15[4,5],ymm10[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2],ymm2[3],ymm6[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm6 = ymm9[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,1,3,3,6,5,7,7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm4 = ymm5[2],ymm4[2],ymm5[3],ymm4[3],ymm5[6],ymm4[6],ymm5[7],ymm4[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm6[2],ymm0[3],ymm6[3],ymm0[6],ymm6[6],ymm0[7],ymm6[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm6[1],ymm0[2,3,4],ymm6[5],ymm0[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm13[4],ymm11[4],ymm13[5],ymm11[5],ymm13[6],ymm11[6],ymm13[7],ymm11[7],ymm13[12],ymm11[12],ymm13[13],ymm11[13],ymm13[14],ymm11[14],ymm13[15],ymm11[15]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm4[4],ymm5[4],ymm4[5],ymm5[5],ymm4[6],ymm5[6],ymm4[7],ymm5[7],ymm4[12],ymm5[12],ymm4[13],ymm5[13],ymm4[14],ymm5[14],ymm4[15],ymm5[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm0[0,0,2,1,4,4,6,5]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm4[2,3],ymm0[4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm7[4],ymm3[4],ymm7[5],ymm3[5],ymm7[6],ymm3[6],ymm7[7],ymm3[7],ymm7[12],ymm3[12],ymm7[13],ymm3[13],ymm7[14],ymm3[14],ymm7[15],ymm3[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm9[4],ymm8[4],ymm9[5],ymm8[5],ymm9[6],ymm8[6],ymm9[7],ymm8[7],ymm9[12],ymm8[12],ymm9[13],ymm8[13],ymm9[14],ymm8[14],ymm9[15],ymm8[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm2[4],ymm1[4],ymm2[5],ymm1[5],ymm2[6],ymm1[6],ymm2[7],ymm1[7],ymm2[12],ymm1[12],ymm2[13],ymm1[13],ymm2[14],ymm1[14],ymm2[15],ymm1[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm11[4],ymm13[4],ymm11[5],ymm13[5],ymm11[6],ymm13[6],ymm11[7],ymm13[7],ymm11[12],ymm13[12],ymm11[13],ymm13[13],ymm11[14],ymm13[14],ymm11[15],ymm13[15]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm4 = ymm3[0],ymm0[0],ymm3[1],ymm0[1],ymm3[4],ymm0[4],ymm3[5],ymm0[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm2[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1,2],ymm4[3],ymm5[4,5,6],ymm4[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm7[4],ymm8[4],ymm7[5],ymm8[5],ymm7[6],ymm8[6],ymm7[7],ymm8[7],ymm7[12],ymm8[12],ymm7[13],ymm8[13],ymm7[14],ymm8[14],ymm7[15],ymm8[15]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm1[4],ymm3[4],ymm1[5],ymm3[5],ymm1[6],ymm3[6],ymm1[7],ymm3[7],ymm1[12],ymm3[12],ymm1[13],ymm3[13],ymm1[14],ymm3[14],ymm1[15],ymm3[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm5[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm6 = ymm1[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm6[0],ymm3[1],ymm6[2,3,4],ymm3[5],ymm6[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, (%rsp) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm2[0],ymm1[0],ymm2[1],ymm1[1],ymm2[4],ymm1[4],ymm2[5],ymm1[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3],ymm5[4,5],ymm4[6,7]
+; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm3[2],ymm0[2],ymm3[3],ymm0[3],ymm3[6],ymm0[6],ymm3[7],ymm0[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm1 = ymm2[2],ymm1[2],ymm2[3],ymm1[3],ymm2[6],ymm1[6],ymm2[7],ymm1[7]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6],ymm0[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,3,3,6,5,7,7]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3,4],ymm2[5],ymm1[6,7]
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vmovdqa 64(%r10), %ymm10
-; AVX2-NEXT:    vmovdqa 64(%rax), %ymm9
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm10[0],ymm9[0],ymm10[1],ymm9[1],ymm10[2],ymm9[2],ymm10[3],ymm9[3],ymm10[8],ymm9[8],ymm10[9],ymm9[9],ymm10[10],ymm9[10],ymm10[11],ymm9[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT:    vmovdqa 64(%r8), %ymm8
-; AVX2-NEXT:    vmovdqa 64(%r9), %ymm5
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm8[0],ymm5[0],ymm8[1],ymm5[1],ymm8[2],ymm5[2],ymm8[3],ymm5[3],ymm8[8],ymm5[8],ymm8[9],ymm5[9],ymm8[10],ymm5[10],ymm8[11],ymm5[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm6[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm7[0,1,2],ymm3[3],ymm7[4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqa 64(%rdx), %ymm11
-; AVX2-NEXT:    vmovdqa 64(%rcx), %ymm13
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm15 = ymm11[0],ymm13[0],ymm11[1],ymm13[1],ymm11[2],ymm13[2],ymm11[3],ymm13[3],ymm11[8],ymm13[8],ymm11[9],ymm13[9],ymm11[10],ymm13[10],ymm11[11],ymm13[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm7 = ymm15[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,3]
-; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm3
-; AVX2-NEXT:    vmovdqa 64(%rsi), %ymm1
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[8],ymm1[8],ymm3[9],ymm1[9],ymm3[10],ymm1[10],ymm3[11],ymm1[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm14 = ymm0[0,1,1,3,4,5,5,7]
+; AVX2-NEXT:    vmovdqa 64(%r8), %ymm5
+; AVX2-NEXT:    vmovdqa 64(%r9), %ymm4
+; AVX2-NEXT:    vmovdqa 64(%r10), %ymm7
+; AVX2-NEXT:    vmovdqa 64(%rax), %ymm8
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[8],ymm8[8],ymm7[9],ymm8[9],ymm7[10],ymm8[10],ymm7[11],ymm8[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm9 = ymm5[0],ymm4[0],ymm5[1],ymm4[1],ymm5[2],ymm4[2],ymm5[3],ymm4[3],ymm5[8],ymm4[8],ymm5[9],ymm4[9],ymm5[10],ymm4[10],ymm5[11],ymm4[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm3 = ymm9[0],ymm2[0],ymm9[1],ymm2[1],ymm9[4],ymm2[4],ymm9[5],ymm2[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm3[0,2,2,3]
+; AVX2-NEXT:    vmovdqa 64(%rdi), %ymm11
+; AVX2-NEXT:    vmovdqa 64(%rsi), %ymm13
+; AVX2-NEXT:    vmovdqa 64(%rdx), %ymm15
+; AVX2-NEXT:    vmovdqa 64(%rcx), %ymm3
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm15[0],ymm3[0],ymm15[1],ymm3[1],ymm15[2],ymm3[2],ymm15[3],ymm3[3],ymm15[8],ymm3[8],ymm15[9],ymm3[9],ymm15[10],ymm3[10],ymm15[11],ymm3[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm11[0],ymm13[0],ymm11[1],ymm13[1],ymm11[2],ymm13[2],ymm11[3],ymm13[3],ymm11[8],ymm13[8],ymm11[9],ymm13[9],ymm11[10],ymm13[10],ymm11[11],ymm13[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm14 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm14[0],ymm7[1],ymm14[2,3,4],ymm7[5],ymm14[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm7[0,1],ymm4[2,3],ymm7[4,5],ymm4[6,7]
-; AVX2-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm6[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm15[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,1,3,3,6,5,7,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm14[0,1],ymm6[2,3],ymm14[4,5],ymm6[6,7]
+; AVX2-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm9[2],ymm2[2],ymm9[3],ymm2[3],ymm9[6],ymm2[6],ymm9[7],ymm2[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm2[0,2,2,3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm4[1],ymm0[2,3,4],ymm4[5],ymm0[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm10[4],ymm9[4],ymm10[5],ymm9[5],ymm10[6],ymm9[6],ymm10[7],ymm9[7],ymm10[12],ymm9[12],ymm10[13],ymm9[13],ymm10[14],ymm9[14],ymm10[15],ymm9[15]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm8[4],ymm5[4],ymm8[5],ymm5[5],ymm8[6],ymm5[6],ymm8[7],ymm5[7],ymm8[12],ymm5[12],ymm8[13],ymm5[13],ymm8[14],ymm5[14],ymm8[15],ymm5[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm0[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm2[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1,2],ymm4[3],ymm5[4,5,6],ymm4[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm8 = ymm11[4],ymm13[4],ymm11[5],ymm13[5],ymm11[6],ymm13[6],ymm11[7],ymm13[7],ymm11[12],ymm13[12],ymm11[13],ymm13[13],ymm11[14],ymm13[14],ymm11[15],ymm13[15]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm3[4],ymm1[4],ymm3[5],ymm1[5],ymm3[6],ymm1[6],ymm3[7],ymm1[7],ymm3[12],ymm1[12],ymm3[13],ymm1[13],ymm3[14],ymm1[14],ymm3[15],ymm1[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm8[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm1[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm5[0],ymm3[1],ymm5[2,3,4],ymm3[5],ymm5[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]
-; AVX2-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm7[4],ymm8[4],ymm7[5],ymm8[5],ymm7[6],ymm8[6],ymm7[7],ymm8[7],ymm7[12],ymm8[12],ymm7[13],ymm8[13],ymm7[14],ymm8[14],ymm7[15],ymm8[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm5[4],ymm4[4],ymm5[5],ymm4[5],ymm5[6],ymm4[6],ymm5[7],ymm4[7],ymm5[12],ymm4[12],ymm5[13],ymm4[13],ymm5[14],ymm4[14],ymm5[15],ymm4[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm15[4],ymm3[4],ymm15[5],ymm3[5],ymm15[6],ymm3[6],ymm15[7],ymm3[7],ymm15[12],ymm3[12],ymm15[13],ymm3[13],ymm15[14],ymm3[14],ymm15[15],ymm3[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm11[4],ymm13[4],ymm11[5],ymm13[5],ymm11[6],ymm13[6],ymm11[7],ymm13[7],ymm11[12],ymm13[12],ymm11[13],ymm13[13],ymm11[14],ymm13[14],ymm11[15],ymm13[15]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm4[0],ymm0[0],ymm4[1],ymm0[1],ymm4[4],ymm0[4],ymm4[5],ymm0[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm7 = ymm5[0],ymm3[0],ymm5[1],ymm3[1],ymm5[4],ymm3[4],ymm5[5],ymm3[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm7[0,1],ymm1[2,3],ymm7[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm4[2],ymm0[2],ymm4[3],ymm0[3],ymm4[6],ymm0[6],ymm4[7],ymm0[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm3 = ymm5[2],ymm3[2],ymm5[3],ymm3[3],ymm5[6],ymm3[6],ymm5[7],ymm3[7]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6],ymm0[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm8[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3,4],ymm2[5],ymm1[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm7 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-NEXT:    vmovdqa 96(%r10), %ymm6
-; AVX2-NEXT:    vmovdqa 96(%rax), %ymm5
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm6[0],ymm5[0],ymm6[1],ymm5[1],ymm6[2],ymm5[2],ymm6[3],ymm5[3],ymm6[8],ymm5[8],ymm6[9],ymm5[9],ymm6[10],ymm5[10],ymm6[11],ymm5[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-NEXT:    vmovdqa 96(%r8), %ymm8
-; AVX2-NEXT:    vmovdqa 96(%r9), %ymm9
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm10 = ymm8[0],ymm9[0],ymm8[1],ymm9[1],ymm8[2],ymm9[2],ymm8[3],ymm9[3],ymm8[8],ymm9[8],ymm8[9],ymm9[9],ymm8[10],ymm9[10],ymm8[11],ymm9[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm11 = ymm10[0,1,1,3,4,5,5,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm6 = ymm3[0,1],ymm0[2,3],ymm3[4,5],ymm0[6,7]
+; AVX2-NEXT:    vmovdqa 96(%r8), %ymm3
+; AVX2-NEXT:    vmovdqa 96(%r9), %ymm4
+; AVX2-NEXT:    vmovdqa 96(%r10), %ymm5
+; AVX2-NEXT:    vmovdqa 96(%rax), %ymm7
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm8 = ymm5[0],ymm7[0],ymm5[1],ymm7[1],ymm5[2],ymm7[2],ymm5[3],ymm7[3],ymm5[8],ymm7[8],ymm5[9],ymm7[9],ymm5[10],ymm7[10],ymm5[11],ymm7[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm9 = ymm3[0],ymm4[0],ymm3[1],ymm4[1],ymm3[2],ymm4[2],ymm3[3],ymm4[3],ymm3[8],ymm4[8],ymm3[9],ymm4[9],ymm3[10],ymm4[10],ymm3[11],ymm4[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm11 = ymm9[0],ymm8[0],ymm9[1],ymm8[1],ymm9[4],ymm8[4],ymm9[5],ymm8[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm4 = ymm11[0,1,2],ymm3[3],ymm11[4,5,6],ymm3[7]
-; AVX2-NEXT:    vmovdqa 96(%rdx), %ymm11
-; AVX2-NEXT:    vmovdqa 96(%rcx), %ymm13
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm14 = ymm11[0],ymm13[0],ymm11[1],ymm13[1],ymm11[2],ymm13[2],ymm11[3],ymm13[3],ymm11[8],ymm13[8],ymm11[9],ymm13[9],ymm11[10],ymm13[10],ymm11[11],ymm13[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm15 = ymm14[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
-; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm3
-; AVX2-NEXT:    vmovdqa 96(%rsi), %ymm1
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[8],ymm1[8],ymm3[9],ymm1[9],ymm3[10],ymm1[10],ymm3[11],ymm1[11]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm12 = ymm0[0,1,1,3,4,5,5,7]
+; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm13
+; AVX2-NEXT:    vmovdqa 96(%rsi), %ymm14
+; AVX2-NEXT:    vmovdqa 96(%rdx), %ymm15
+; AVX2-NEXT:    vmovdqa 96(%rcx), %ymm2
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm15[0],ymm2[0],ymm15[1],ymm2[1],ymm15[2],ymm2[2],ymm15[3],ymm2[3],ymm15[8],ymm2[8],ymm15[9],ymm2[9],ymm15[10],ymm2[10],ymm15[11],ymm2[11]
+; AVX2-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm13[0],ymm14[0],ymm13[1],ymm14[1],ymm13[2],ymm14[2],ymm13[3],ymm14[3],ymm13[8],ymm14[8],ymm13[9],ymm14[9],ymm13[10],ymm14[10],ymm13[11],ymm14[11]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm12 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0],ymm15[1],ymm12[2,3,4],ymm15[5],ymm12[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm15 = ymm12[0,1],ymm4[2,3],ymm12[4,5],ymm4[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm10[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6],ymm2[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm14[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,1,3,3,6,5,7,7]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm12[0,1],ymm11[2,3],ymm12[4,5],ymm11[6,7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm8 = ymm9[2],ymm8[2],ymm9[3],ymm8[3],ymm9[6],ymm8[6],ymm9[7],ymm8[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm8[0,2,2,3]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm4[1],ymm0[2,3,4],ymm4[5],ymm0[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm6[4],ymm5[4],ymm6[5],ymm5[5],ymm6[6],ymm5[6],ymm6[7],ymm5[7],ymm6[12],ymm5[12],ymm6[13],ymm5[13],ymm6[14],ymm5[14],ymm6[15],ymm5[15]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm8[4],ymm9[4],ymm8[5],ymm9[5],ymm8[6],ymm9[6],ymm8[7],ymm9[7],ymm8[12],ymm9[12],ymm8[13],ymm9[13],ymm8[14],ymm9[14],ymm8[15],ymm9[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm5 = ymm0[0,0,2,1,4,4,6,5]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm5[4],ymm7[4],ymm5[5],ymm7[5],ymm5[6],ymm7[6],ymm5[7],ymm7[7],ymm5[12],ymm7[12],ymm5[13],ymm7[13],ymm5[14],ymm7[14],ymm5[15],ymm7[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm3[4],ymm4[4],ymm3[5],ymm4[5],ymm3[6],ymm4[6],ymm3[7],ymm4[7],ymm3[12],ymm4[12],ymm3[13],ymm4[13],ymm3[14],ymm4[14],ymm3[15],ymm4[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm15[4],ymm2[4],ymm15[5],ymm2[5],ymm15[6],ymm2[6],ymm15[7],ymm2[7],ymm15[12],ymm2[12],ymm15[13],ymm2[13],ymm15[14],ymm2[14],ymm15[15],ymm2[15]
+; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm13[4],ymm14[4],ymm13[5],ymm14[5],ymm13[6],ymm14[6],ymm13[7],ymm14[7],ymm13[12],ymm14[12],ymm13[13],ymm14[13],ymm13[14],ymm14[14],ymm13[15],ymm14[15]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[4],ymm1[4],ymm3[5],ymm1[5]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm6 = ymm4[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0,1,2],ymm5[3],ymm6[4,5,6],ymm5[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm11[4],ymm13[4],ymm11[5],ymm13[5],ymm11[6],ymm13[6],ymm11[7],ymm13[7],ymm11[12],ymm13[12],ymm11[13],ymm13[13],ymm11[14],ymm13[14],ymm11[15],ymm13[15]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm3[4],ymm1[4],ymm3[5],ymm1[5],ymm3[6],ymm1[6],ymm3[7],ymm1[7],ymm3[12],ymm1[12],ymm3[13],ymm1[13],ymm3[14],ymm1[14],ymm3[15],ymm1[15]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm3 = ymm6[0,0,2,1,4,4,6,5]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm8 = ymm1[0,1,1,3,4,5,5,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm8[0],ymm3[1],ymm8[2,3,4],ymm3[5],ymm8[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm5[2,3],ymm3[4,5],ymm5[6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6],ymm0[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm4 = ymm6[0,2,2,3,4,6,6,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,3,3,6,5,7,7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm4[1],ymm1[2,3,4],ymm4[5],ymm1[6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm7 = ymm4[0],ymm2[0],ymm4[1],ymm2[1],ymm4[4],ymm2[4],ymm4[5],ymm2[5]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm7[0,1],ymm5[2,3],ymm7[4,5],ymm5[6,7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm1 = ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[6],ymm1[6],ymm3[7],ymm1[7]
+; AVX2-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm4[2],ymm2[2],ymm4[3],ymm2[3],ymm4[6],ymm2[6],ymm4[7],ymm2[7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
 ; AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-NEXT:    vmovdqa %ymm0, 992(%rax)
-; AVX2-NEXT:    vmovdqa %ymm3, 960(%rax)
-; AVX2-NEXT:    vmovdqa %ymm2, 928(%rax)
-; AVX2-NEXT:    vmovdqa %ymm15, 896(%rax)
-; AVX2-NEXT:    vmovdqa %ymm7, 736(%rax)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 704(%rax)
+; AVX2-NEXT:    vmovdqa %ymm1, 992(%rax)
+; AVX2-NEXT:    vmovdqa %ymm5, 960(%rax)
+; AVX2-NEXT:    vmovdqa %ymm0, 928(%rax)
+; AVX2-NEXT:    vmovdqa %ymm11, 896(%rax)
+; AVX2-NEXT:    vmovdqa %ymm6, 736(%rax)
+; AVX2-NEXT:    vmovdqa %ymm10, 704(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 672(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 640(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 480(%rax)
-; AVX2-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 448(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-NEXT:    vmovaps %ymm0, 448(%rax)
+; AVX2-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 416(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, 384(%rax)
@@ -7534,539 +7040,345 @@ define void @store_i16_stride8_vf64(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.ve
 ; AVX2-NEXT:    vmovaps %ymm0, 32(%rax)
 ; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-NEXT:    vmovaps %ymm0, (%rax)
-; AVX2-NEXT:    addq $744, %rsp # imm = 0x2E8
+; AVX2-NEXT:    addq $712, %rsp # imm = 0x2C8
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: store_i16_stride8_vf64:
 ; AVX2-FP:       # %bb.0:
-; AVX2-FP-NEXT:    subq $744, %rsp # imm = 0x2E8
+; AVX2-FP-NEXT:    subq $712, %rsp # imm = 0x2C8
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX2-FP-NEXT:    vmovdqa (%rax), %xmm6
+; AVX2-FP-NEXT:    vmovdqa (%rax), %xmm5
 ; AVX2-FP-NEXT:    vmovdqa 32(%rax), %xmm0
-; AVX2-FP-NEXT:    vmovdqa (%r10), %xmm7
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm7[0],xmm6[0],xmm7[1],xmm6[1],xmm7[2],xmm6[2],xmm7[3],xmm6[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm3[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
+; AVX2-FP-NEXT:    vmovdqa (%r10), %xmm6
+; AVX2-FP-NEXT:    vmovdqa 32(%r10), %xmm1
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3]
 ; AVX2-FP-NEXT:    vmovdqa (%r9), %xmm8
 ; AVX2-FP-NEXT:    vmovdqa (%r8), %xmm9
 ; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm4[0],zero,xmm4[1],zero
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5,6],ymm1[7]
 ; AVX2-FP-NEXT:    vmovdqa (%rcx), %xmm10
 ; AVX2-FP-NEXT:    vmovdqa (%rdx), %xmm11
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm5[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
 ; AVX2-FP-NEXT:    vmovdqa (%rsi), %xmm12
 ; AVX2-FP-NEXT:    vmovdqa (%rdi), %xmm13
 ; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm14 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm15 = xmm14[0],zero,xmm14[1],zero
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm15 = xmm14[0],xmm7[0],xmm14[1],xmm7[1]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm15[0],ymm2[1],ymm15[2,3,4],ymm2[5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqa 32(%r10), %xmm2
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm15[0,1],ymm1[2,3],ymm15[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 32(%r9), %xmm1
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm4[0,1,2],ymm3[3],ymm4[4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqa 32(%r8), %xmm3
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm4 = xmm5[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm5 = xmm14[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0],ymm4[1],ymm5[2,3,4],ymm4[5],ymm5[6,7]
-; AVX2-FP-NEXT:    vmovdqa 32(%rcx), %xmm4
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1],ymm15[2,3],ymm5[4,5],ymm15[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 32(%rdx), %xmm5
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm7[4],xmm6[4],xmm7[5],xmm6[5],xmm7[6],xmm6[6],xmm7[7],xmm6[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm7 = xmm6[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm9 = xmm8[0],zero,xmm8[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2],ymm7[3],ymm9[4,5,6],ymm7[7]
-; AVX2-FP-NEXT:    vmovdqa 32(%rsi), %xmm7
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm15[0,1],ymm2[2,3],ymm15[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 32(%r9), %xmm2
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm15 = xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX2-FP-NEXT:    vmovdqa 32(%r8), %xmm4
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm7 = xmm14[2],xmm7[2],xmm14[3],xmm7[3]
+; AVX2-FP-NEXT:    vmovdqa 32(%rcx), %xmm3
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm14 = ymm15[0,0,2,1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1],ymm14[2,3],ymm7[4,5],ymm14[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 32(%rdx), %xmm7
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
+; AVX2-FP-NEXT:    vmovdqa 32(%rsi), %xmm5
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm9 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
+; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %xmm8
 ; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm10 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm11 = xmm10[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,1,3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm11 = xmm9[0],xmm6[0],xmm9[1],xmm6[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,2,1]
 ; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm12 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm13 = xmm12[0],zero,xmm12[1],zero
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm13 = xmm12[0],xmm10[0],xmm12[1],xmm10[1]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm13[0],ymm11[1],ymm13[2,3,4],ymm11[5],ymm13[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm11[0,1],ymm9[2,3],ymm11[4,5],ymm9[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,2,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm13[0,1],ymm11[2,3],ymm13[4,5],ymm11[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm6 = xmm9[2],xmm6[2],xmm9[3],xmm6[3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm8[0,1,2],ymm6[3],ymm8[4,5,6],ymm6[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm8 = xmm10[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm9 = xmm12[2,2,3,3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm9 = xmm12[2],xmm10[2],xmm12[3],xmm10[3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0],ymm8[1],ymm9[2,3,4],ymm8[5],ymm9[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm8[0,1],ymm6[2,3],ymm8[4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm9[0,1],ymm6[2,3],ymm9[4,5],ymm6[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm8 = xmm6[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm10 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm9 = xmm10[0],zero,xmm10[1],zero
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm4[0],xmm2[0],xmm4[1],xmm2[1],xmm4[2],xmm2[2],xmm4[3],xmm2[3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm10 = xmm9[0],xmm6[0],xmm9[1],xmm6[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,0,2,1]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm7[0],xmm3[0],xmm7[1],xmm3[1],xmm7[2],xmm3[2],xmm7[3],xmm3[3]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm12 = xmm8[0],xmm5[0],xmm8[1],xmm5[1],xmm8[2],xmm5[2],xmm8[3],xmm5[3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm13 = xmm12[0],xmm11[0],xmm12[1],xmm11[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm13[0,1],ymm10[2,3],ymm13[4,5],ymm10[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm9 = xmm9[2],xmm6[2],xmm9[3],xmm6[3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm10 = xmm12[2],xmm11[2],xmm12[3],xmm11[3]
+; AVX2-FP-NEXT:    vmovdqa 64(%rax), %xmm6
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1,2],ymm8[3],ymm9[4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm9 = xmm11[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm9[0,1,1,3]
-; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %xmm9
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm9[0],xmm7[0],xmm9[1],xmm7[1],xmm9[2],xmm7[2],xmm9[3],xmm7[3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm14 = xmm13[0],zero,xmm13[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm14[0],ymm12[1],ymm14[2,3,4],ymm12[5],ymm14[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm12[0,1],ymm8[2,3],ymm12[4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm8 = xmm10[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2],ymm6[3],ymm8[4,5,6],ymm6[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm6 = xmm11[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm10 = xmm13[2,2,3,3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0],ymm6[1],ymm10[2,3,4],ymm6[5],ymm10[6,7]
-; AVX2-FP-NEXT:    vmovdqa 64(%rax), %xmm6
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm10[0,1],ymm8[2,3],ymm10[4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 64(%r10), %xmm8
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm2[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm3[4],xmm1[4],xmm3[5],xmm1[5],xmm3[6],xmm1[6],xmm3[7],xmm1[7]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm3[0],zero,xmm3[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0,1],ymm9[2,3],ymm10[4,5],ymm9[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vmovdqa 64(%r10), %xmm9
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm10 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; AVX2-FP-NEXT:    vmovdqa 64(%r9), %xmm0
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm4[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm9[4],xmm7[4],xmm9[5],xmm7[5],xmm9[6],xmm7[6],xmm9[7],xmm7[7]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm7 = xmm5[0],zero,xmm5[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0],ymm1[1],ymm7[2,3,4],ymm1[5],ymm7[6,7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm4[4],xmm2[4],xmm4[5],xmm2[5],xmm4[6],xmm2[6],xmm4[7],xmm2[7]
 ; AVX2-FP-NEXT:    vmovdqa 64(%r8), %xmm1
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1],ymm10[2,3],ymm7[4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm7[4],xmm3[4],xmm7[5],xmm3[5],xmm7[6],xmm3[6],xmm7[7],xmm3[7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm4[0],xmm10[0],xmm4[1],xmm10[1]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm3 = xmm4[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm4 = xmm5[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3,4],ymm3[5],ymm4[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm8[4],xmm5[4],xmm8[5],xmm5[5],xmm8[6],xmm5[6],xmm8[7],xmm5[7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm7 = xmm5[0],xmm3[0],xmm5[1],xmm3[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm7[0,1],ymm2[2,3],ymm7[4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm8[0],xmm6[0],xmm8[1],xmm6[1],xmm8[2],xmm6[2],xmm8[3],xmm6[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm7[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm9[0],zero,xmm9[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
 ; AVX2-FP-NEXT:    vmovdqa 64(%rcx), %xmm2
-; AVX2-FP-NEXT:    vmovdqa 64(%rdx), %xmm3
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm4 = xmm11[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm4[0,1,1,3]
-; AVX2-FP-NEXT:    vmovdqa 64(%rsi), %xmm4
-; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %xmm5
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm14 = xmm13[0],zero,xmm13[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm14[0],ymm12[1],ymm14[2,3,4],ymm12[5],ymm14[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm12[0,1],ymm10[2,3],ymm12[4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm7 = xmm7[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm9 = xmm9[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm9[0,1,2],ymm7[3],ymm9[4,5,6],ymm7[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm9 = xmm11[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm10 = xmm13[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0],ymm9[1],ymm10[2,3,4],ymm9[5],ymm10[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm9[0,1],ymm7[2,3],ymm9[4,5],ymm7[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm8[4],xmm6[4],xmm8[5],xmm6[5],xmm8[6],xmm6[6],xmm8[7],xmm6[7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm6[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm7 = xmm0[0],zero,xmm0[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2],ymm1[3],ymm7[4,5,6],ymm1[7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm4 = xmm2[0,0,1,1]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm4 = xmm4[2],xmm10[2],xmm4[3],xmm10[3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm3 = xmm5[2],xmm3[2],xmm5[3],xmm3[3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm9[0],xmm6[0],xmm9[1],xmm6[1],xmm9[2],xmm6[2],xmm9[3],xmm6[3]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,0,2,1]
+; AVX2-FP-NEXT:    vmovdqa 64(%rdx), %xmm7
+; AVX2-FP-NEXT:    vmovdqa 64(%rsi), %xmm8
+; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %xmm10
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm7[0],xmm2[0],xmm7[1],xmm2[1],xmm7[2],xmm2[2],xmm7[3],xmm2[3]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm12 = xmm10[0],xmm8[0],xmm10[1],xmm8[1],xmm10[2],xmm8[2],xmm10[3],xmm8[3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm13 = xmm12[0],xmm11[0],xmm12[1],xmm11[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm13[0,1],ymm5[2,3],ymm13[4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm3 = xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm4 = xmm12[2],xmm11[2],xmm12[3],xmm11[3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm3[0],zero,xmm3[1],zero
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3],ymm4[4,5],ymm3[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm9[4],xmm6[4],xmm9[5],xmm6[5],xmm9[6],xmm6[6],xmm9[7],xmm6[7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm7[4],xmm2[4],xmm7[5],xmm2[5],xmm7[6],xmm2[6],xmm7[7],xmm2[7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm10[4],xmm8[4],xmm10[5],xmm8[5],xmm10[6],xmm8[6],xmm10[7],xmm8[7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0],ymm4[1],ymm5[2,3,4],ymm4[5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm1[2,3],ymm4[4,5],ymm1[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm6[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm2[2,3],ymm5[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm3[2],xmm0[3],xmm3[3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3],ymm0[4,5,6],ymm1[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[2,2,3,3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm4[2],xmm1[2],xmm4[3],xmm1[3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm3[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FP-NEXT:    vmovdqa 96(%rax), %xmm0
 ; AVX2-FP-NEXT:    vmovdqa 96(%r10), %xmm1
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm8[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm2[0,0,2,1]
-; AVX2-FP-NEXT:    vmovdqa 96(%r9), %xmm2
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-FP-NEXT:    vmovdqa 96(%r9), %xmm3
 ; AVX2-FP-NEXT:    vmovdqa 96(%r8), %xmm4
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm4[0],xmm2[0],xmm4[1],xmm2[1],xmm4[2],xmm2[2],xmm4[3],xmm2[3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm9[0],zero,xmm9[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm5[0,1,2],ymm3[3],ymm5[4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqa 96(%rcx), %xmm3
-; AVX2-FP-NEXT:    vmovdqa 96(%rdx), %xmm5
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm5[0],xmm3[0],xmm5[1],xmm3[1],xmm5[2],xmm3[2],xmm5[3],xmm3[3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm6 = xmm11[0,0,1,1]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm6[0,1,1,3]
-; AVX2-FP-NEXT:    vmovdqa 96(%rsi), %xmm6
-; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %xmm7
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm7[0],xmm6[0],xmm7[1],xmm6[1],xmm7[2],xmm6[2],xmm7[3],xmm6[3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm14 = xmm13[0],zero,xmm13[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm14[0],ymm12[1],ymm14[2,3,4],ymm12[5],ymm14[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm12[0,1],ymm10[2,3],ymm12[4,5],ymm10[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm9 = xmm9[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1,2],ymm8[3],ymm9[4,5,6],ymm8[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm9 = xmm11[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm10 = xmm13[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm9 = ymm10[0],ymm9[1],ymm10[2,3,4],ymm9[5],ymm10[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm8 = ymm9[0,1],ymm8[2,3],ymm9[4,5],ymm8[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm4[4],xmm2[4],xmm4[5],xmm2[5],xmm4[6],xmm2[6],xmm4[7],xmm2[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[0,0,1,1]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm5[0],xmm2[0],xmm5[1],xmm2[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
+; AVX2-FP-NEXT:    vmovdqa 96(%rcx), %xmm7
+; AVX2-FP-NEXT:    vmovdqa 96(%rdx), %xmm8
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm8[0],xmm7[0],xmm8[1],xmm7[1],xmm8[2],xmm7[2],xmm8[3],xmm7[3]
+; AVX2-FP-NEXT:    vmovdqa 96(%rsi), %xmm10
+; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %xmm11
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm12 = xmm13[0],xmm9[0],xmm13[1],xmm9[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm12[0,1],ymm6[2,3],ymm12[4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm5[2],xmm2[2],xmm5[3],xmm2[3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm13[2],xmm9[2],xmm13[3],xmm9[3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm5[4],xmm3[4],xmm5[5],xmm3[5],xmm5[6],xmm3[6],xmm5[7],xmm3[7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm7[4],xmm6[4],xmm7[5],xmm6[5],xmm7[6],xmm6[6],xmm7[7],xmm6[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm5 = xmm3[0,0,1,1]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,1,3]
-; AVX2-FP-NEXT:    vpmovzxdq {{.*#+}} xmm6 = xmm4[0],zero,xmm4[1],zero
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0],ymm5[1],ymm6[2,3,4],ymm5[5],ymm6[6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm2[2,3],ymm5[4,5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm8[4],xmm7[4],xmm8[5],xmm7[5],xmm8[6],xmm7[6],xmm8[7],xmm7[7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,1,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3],ymm5[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} xmm1 = xmm3[2],xmm2[2],xmm3[3],xmm2[3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm1 = xmm3[2,2,3,3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} xmm2 = xmm4[2,2,3,3]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa (%r10), %ymm0
-; AVX2-FP-NEXT:    vmovdqa (%rax), %ymm1
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FP-NEXT:    vmovdqa (%r8), %ymm4
-; AVX2-FP-NEXT:    vmovdqa (%r9), %ymm5
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm4[0],ymm5[0],ymm4[1],ymm5[1],ymm4[2],ymm5[2],ymm4[3],ymm5[3],ymm4[8],ymm5[8],ymm4[9],ymm5[9],ymm4[10],ymm5[10],ymm4[11],ymm5[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm6[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm7[0,1,2],ymm3[3],ymm7[4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqa (%rdx), %ymm7
-; AVX2-FP-NEXT:    vmovdqa (%rcx), %ymm8
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm9 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[8],ymm8[8],ymm7[9],ymm8[9],ymm7[10],ymm8[10],ymm7[11],ymm8[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm10 = ymm9[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,3]
-; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm11
-; AVX2-FP-NEXT:    vmovdqa (%rsi), %ymm12
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm13 = ymm11[0],ymm12[0],ymm11[1],ymm12[1],ymm11[2],ymm12[2],ymm11[3],ymm12[3],ymm11[8],ymm12[8],ymm11[9],ymm12[9],ymm11[10],ymm12[10],ymm11[11],ymm12[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm14 = ymm13[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm14[0],ymm10[1],ymm14[2,3,4],ymm10[5],ymm14[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm10[0,1],ymm3[2,3],ymm10[4,5],ymm3[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm6[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm9[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm6 = ymm13[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm6[0],ymm3[1],ymm6[2,3,4],ymm3[5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm4[4],ymm5[4],ymm4[5],ymm5[5],ymm4[6],ymm5[6],ymm4[7],ymm5[7],ymm4[12],ymm5[12],ymm4[13],ymm5[13],ymm4[14],ymm5[14],ymm4[15],ymm5[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm0[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm1[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5,6],ymm2[7]
+; AVX2-FP-NEXT:    vmovdqa (%r8), %ymm11
+; AVX2-FP-NEXT:    vmovdqa (%r9), %ymm1
+; AVX2-FP-NEXT:    vmovdqa (%r10), %ymm2
+; AVX2-FP-NEXT:    vmovdqa (%rax), %ymm3
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[8],ymm3[8],ymm2[9],ymm3[9],ymm2[10],ymm3[10],ymm2[11],ymm3[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm11[0],ymm1[0],ymm11[1],ymm1[1],ymm11[2],ymm1[2],ymm11[3],ymm1[3],ymm11[8],ymm1[8],ymm11[9],ymm1[9],ymm11[10],ymm1[10],ymm11[11],ymm1[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm6 = ymm5[0],ymm4[0],ymm5[1],ymm4[1],ymm5[4],ymm4[4],ymm5[5],ymm4[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm6[0,2,2,3]
+; AVX2-FP-NEXT:    vmovdqa (%rdi), %ymm7
+; AVX2-FP-NEXT:    vmovdqa (%rsi), %ymm8
+; AVX2-FP-NEXT:    vmovdqa (%rdx), %ymm9
+; AVX2-FP-NEXT:    vmovdqa (%rcx), %ymm6
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm10 = ymm9[0],ymm6[0],ymm9[1],ymm6[1],ymm9[2],ymm6[2],ymm9[3],ymm6[3],ymm9[8],ymm6[8],ymm9[9],ymm6[9],ymm9[10],ymm6[10],ymm9[11],ymm6[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[8],ymm8[8],ymm7[9],ymm8[9],ymm7[10],ymm8[10],ymm7[11],ymm8[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm13 = ymm0[0],ymm10[0],ymm0[1],ymm10[1],ymm0[4],ymm10[4],ymm0[5],ymm10[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm13[0,1],ymm12[2,3],ymm13[4,5],ymm12[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm4 = ymm5[2],ymm4[2],ymm5[3],ymm4[3],ymm5[6],ymm4[6],ymm5[7],ymm4[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm10[2],ymm0[3],ymm10[3],ymm0[6],ymm10[6],ymm0[7],ymm10[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm4[2,3],ymm0[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[12],ymm3[12],ymm2[13],ymm3[13],ymm2[14],ymm3[14],ymm2[15],ymm3[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm11[4],ymm1[4],ymm11[5],ymm1[5],ymm11[6],ymm1[6],ymm11[7],ymm1[7],ymm11[12],ymm1[12],ymm11[13],ymm1[13],ymm11[14],ymm1[14],ymm11[15],ymm1[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm9[4],ymm6[4],ymm9[5],ymm6[5],ymm9[6],ymm6[6],ymm9[7],ymm6[7],ymm9[12],ymm6[12],ymm9[13],ymm6[13],ymm9[14],ymm6[14],ymm9[15],ymm6[15]
 ; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm7[4],ymm8[4],ymm7[5],ymm8[5],ymm7[6],ymm8[6],ymm7[7],ymm8[7],ymm7[12],ymm8[12],ymm7[13],ymm8[13],ymm7[14],ymm8[14],ymm7[15],ymm8[15]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm11[4],ymm12[4],ymm11[5],ymm12[5],ymm11[6],ymm12[6],ymm11[7],ymm12[7],ymm11[12],ymm12[12],ymm11[13],ymm12[13],ymm11[14],ymm12[14],ymm11[15],ymm12[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm3[0,0,2,1,4,4,6,5]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm4 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[4],ymm0[4],ymm1[5],ymm0[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm3[0],ymm2[0],ymm3[1],ymm2[1],ymm3[4],ymm2[4],ymm3[5],ymm2[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm6 = ymm4[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0],ymm5[1],ymm6[2,3,4],ymm5[5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm2[2,3],ymm5[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3],ymm5[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[6],ymm0[6],ymm1[7],ymm0[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm1 = ymm3[2],ymm2[2],ymm3[3],ymm2[3],ymm3[6],ymm2[6],ymm3[7],ymm2[7]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[0,2,2,3,4,6,6,7]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm4[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 32(%r10), %ymm13
-; AVX2-FP-NEXT:    vmovdqa 32(%rax), %ymm11
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm13[0],ymm11[0],ymm13[1],ymm11[1],ymm13[2],ymm11[2],ymm13[3],ymm11[3],ymm13[8],ymm11[8],ymm13[9],ymm11[9],ymm13[10],ymm11[10],ymm13[11],ymm11[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FP-NEXT:    vmovdqa 32(%r8), %ymm4
-; AVX2-FP-NEXT:    vmovdqa 32(%r9), %ymm5
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm4[0],ymm5[0],ymm4[1],ymm5[1],ymm4[2],ymm5[2],ymm4[3],ymm5[3],ymm4[8],ymm5[8],ymm4[9],ymm5[9],ymm4[10],ymm5[10],ymm4[11],ymm5[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm6[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm7[0,1,2],ymm3[3],ymm7[4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqa 32(%rdx), %ymm7
-; AVX2-FP-NEXT:    vmovdqa 32(%rcx), %ymm8
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm9 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[8],ymm8[8],ymm7[9],ymm8[9],ymm7[10],ymm8[10],ymm7[11],ymm8[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm10 = ymm9[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,3]
-; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-FP-NEXT:    vmovdqa 32(%rsi), %ymm3
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[8],ymm3[8],ymm1[9],ymm3[9],ymm1[10],ymm3[10],ymm1[11],ymm3[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm15 = ymm0[0,1,1,3,4,5,5,7]
+; AVX2-FP-NEXT:    vmovdqa 32(%r8), %ymm9
+; AVX2-FP-NEXT:    vmovdqa 32(%r9), %ymm8
+; AVX2-FP-NEXT:    vmovdqa 32(%r10), %ymm7
+; AVX2-FP-NEXT:    vmovdqa 32(%rax), %ymm3
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm7[0],ymm3[0],ymm7[1],ymm3[1],ymm7[2],ymm3[2],ymm7[3],ymm3[3],ymm7[8],ymm3[8],ymm7[9],ymm3[9],ymm7[10],ymm3[10],ymm7[11],ymm3[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm9[0],ymm8[0],ymm9[1],ymm8[1],ymm9[2],ymm8[2],ymm9[3],ymm8[3],ymm9[8],ymm8[8],ymm9[9],ymm8[9],ymm9[10],ymm8[10],ymm9[11],ymm8[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm6 = ymm5[0],ymm4[0],ymm5[1],ymm4[1],ymm5[4],ymm4[4],ymm5[5],ymm4[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm10 = ymm6[0,2,2,3]
+; AVX2-FP-NEXT:    vmovdqa 32(%rdi), %ymm11
+; AVX2-FP-NEXT:    vmovdqa 32(%rsi), %ymm13
+; AVX2-FP-NEXT:    vmovdqa 32(%rdx), %ymm2
+; AVX2-FP-NEXT:    vmovdqa 32(%rcx), %ymm1
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm2[0],ymm1[0],ymm2[1],ymm1[1],ymm2[2],ymm1[2],ymm2[3],ymm1[3],ymm2[8],ymm1[8],ymm2[9],ymm1[9],ymm2[10],ymm1[10],ymm2[11],ymm1[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm11[0],ymm13[0],ymm11[1],ymm13[1],ymm11[2],ymm13[2],ymm11[3],ymm13[3],ymm11[8],ymm13[8],ymm11[9],ymm13[9],ymm11[10],ymm13[10],ymm11[11],ymm13[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm15 = ymm0[0],ymm6[0],ymm0[1],ymm6[1],ymm0[4],ymm6[4],ymm0[5],ymm6[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm15[0],ymm10[1],ymm15[2,3,4],ymm10[5],ymm15[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1],ymm12[2,3],ymm10[4,5],ymm12[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm15[0,1],ymm10[2,3],ymm15[4,5],ymm10[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm6[0,1,2],ymm2[3],ymm6[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm6 = ymm9[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,1,3,3,6,5,7,7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm4 = ymm5[2],ymm4[2],ymm5[3],ymm4[3],ymm5[6],ymm4[6],ymm5[7],ymm4[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm6[2],ymm0[3],ymm6[3],ymm0[6],ymm6[6],ymm0[7],ymm6[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm6[1],ymm0[2,3,4],ymm6[5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm13[4],ymm11[4],ymm13[5],ymm11[5],ymm13[6],ymm11[6],ymm13[7],ymm11[7],ymm13[12],ymm11[12],ymm13[13],ymm11[13],ymm13[14],ymm11[14],ymm13[15],ymm11[15]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm4[4],ymm5[4],ymm4[5],ymm5[5],ymm4[6],ymm5[6],ymm4[7],ymm5[7],ymm4[12],ymm5[12],ymm4[13],ymm5[13],ymm4[14],ymm5[14],ymm4[15],ymm5[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm0[0,0,2,1,4,4,6,5]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm4[2,3],ymm0[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm7[4],ymm3[4],ymm7[5],ymm3[5],ymm7[6],ymm3[6],ymm7[7],ymm3[7],ymm7[12],ymm3[12],ymm7[13],ymm3[13],ymm7[14],ymm3[14],ymm7[15],ymm3[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm9[4],ymm8[4],ymm9[5],ymm8[5],ymm9[6],ymm8[6],ymm9[7],ymm8[7],ymm9[12],ymm8[12],ymm9[13],ymm8[13],ymm9[14],ymm8[14],ymm9[15],ymm8[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm2[4],ymm1[4],ymm2[5],ymm1[5],ymm2[6],ymm1[6],ymm2[7],ymm1[7],ymm2[12],ymm1[12],ymm2[13],ymm1[13],ymm2[14],ymm1[14],ymm2[15],ymm1[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm11[4],ymm13[4],ymm11[5],ymm13[5],ymm11[6],ymm13[6],ymm11[7],ymm13[7],ymm11[12],ymm13[12],ymm11[13],ymm13[13],ymm11[14],ymm13[14],ymm11[15],ymm13[15]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm4 = ymm3[0],ymm0[0],ymm3[1],ymm0[1],ymm3[4],ymm0[4],ymm3[5],ymm0[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm2[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1,2],ymm4[3],ymm5[4,5,6],ymm4[7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm7[4],ymm8[4],ymm7[5],ymm8[5],ymm7[6],ymm8[6],ymm7[7],ymm8[7],ymm7[12],ymm8[12],ymm7[13],ymm8[13],ymm7[14],ymm8[14],ymm7[15],ymm8[15]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm1[4],ymm3[4],ymm1[5],ymm3[5],ymm1[6],ymm3[6],ymm1[7],ymm3[7],ymm1[12],ymm3[12],ymm1[13],ymm3[13],ymm1[14],ymm3[14],ymm1[15],ymm3[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm5[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm6 = ymm1[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm6[0],ymm3[1],ymm6[2,3,4],ymm3[5],ymm6[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, (%rsp) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm2[0],ymm1[0],ymm2[1],ymm1[1],ymm2[4],ymm1[4],ymm2[5],ymm1[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3],ymm5[4,5],ymm4[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm3[2],ymm0[2],ymm3[3],ymm0[3],ymm3[6],ymm0[6],ymm3[7],ymm0[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm1 = ymm2[2],ymm1[2],ymm2[3],ymm1[3],ymm2[6],ymm1[6],ymm2[7],ymm1[7]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,3,3,6,5,7,7]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3,4],ymm2[5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vmovdqa 64(%r10), %ymm10
-; AVX2-FP-NEXT:    vmovdqa 64(%rax), %ymm9
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm10[0],ymm9[0],ymm10[1],ymm9[1],ymm10[2],ymm9[2],ymm10[3],ymm9[3],ymm10[8],ymm9[8],ymm10[9],ymm9[9],ymm10[10],ymm9[10],ymm10[11],ymm9[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FP-NEXT:    vmovdqa 64(%r8), %ymm8
-; AVX2-FP-NEXT:    vmovdqa 64(%r9), %ymm5
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm8[0],ymm5[0],ymm8[1],ymm5[1],ymm8[2],ymm5[2],ymm8[3],ymm5[3],ymm8[8],ymm5[8],ymm8[9],ymm5[9],ymm8[10],ymm5[10],ymm8[11],ymm5[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm6[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm7[0,1,2],ymm3[3],ymm7[4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqa 64(%rdx), %ymm11
-; AVX2-FP-NEXT:    vmovdqa 64(%rcx), %ymm13
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm15 = ymm11[0],ymm13[0],ymm11[1],ymm13[1],ymm11[2],ymm13[2],ymm11[3],ymm13[3],ymm11[8],ymm13[8],ymm11[9],ymm13[9],ymm11[10],ymm13[10],ymm11[11],ymm13[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm7 = ymm15[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,3]
-; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm3
-; AVX2-FP-NEXT:    vmovdqa 64(%rsi), %ymm1
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[8],ymm1[8],ymm3[9],ymm1[9],ymm3[10],ymm1[10],ymm3[11],ymm1[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm14 = ymm0[0,1,1,3,4,5,5,7]
+; AVX2-FP-NEXT:    vmovdqa 64(%r8), %ymm5
+; AVX2-FP-NEXT:    vmovdqa 64(%r9), %ymm4
+; AVX2-FP-NEXT:    vmovdqa 64(%r10), %ymm7
+; AVX2-FP-NEXT:    vmovdqa 64(%rax), %ymm8
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[8],ymm8[8],ymm7[9],ymm8[9],ymm7[10],ymm8[10],ymm7[11],ymm8[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm9 = ymm5[0],ymm4[0],ymm5[1],ymm4[1],ymm5[2],ymm4[2],ymm5[3],ymm4[3],ymm5[8],ymm4[8],ymm5[9],ymm4[9],ymm5[10],ymm4[10],ymm5[11],ymm4[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm3 = ymm9[0],ymm2[0],ymm9[1],ymm2[1],ymm9[4],ymm2[4],ymm9[5],ymm2[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm3[0,2,2,3]
+; AVX2-FP-NEXT:    vmovdqa 64(%rdi), %ymm11
+; AVX2-FP-NEXT:    vmovdqa 64(%rsi), %ymm13
+; AVX2-FP-NEXT:    vmovdqa 64(%rdx), %ymm15
+; AVX2-FP-NEXT:    vmovdqa 64(%rcx), %ymm3
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm15[0],ymm3[0],ymm15[1],ymm3[1],ymm15[2],ymm3[2],ymm15[3],ymm3[3],ymm15[8],ymm3[8],ymm15[9],ymm3[9],ymm15[10],ymm3[10],ymm15[11],ymm3[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm11[0],ymm13[0],ymm11[1],ymm13[1],ymm11[2],ymm13[2],ymm11[3],ymm13[3],ymm11[8],ymm13[8],ymm11[9],ymm13[9],ymm11[10],ymm13[10],ymm11[11],ymm13[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm14 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm14[0],ymm7[1],ymm14[2,3,4],ymm7[5],ymm14[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm7[0,1],ymm4[2,3],ymm7[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm6[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm15[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,1,3,3,6,5,7,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm14[0,1],ymm6[2,3],ymm14[4,5],ymm6[6,7]
+; AVX2-FP-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm9[2],ymm2[2],ymm9[3],ymm2[3],ymm9[6],ymm2[6],ymm9[7],ymm2[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm2[0,2,2,3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm4[1],ymm0[2,3,4],ymm4[5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm10[4],ymm9[4],ymm10[5],ymm9[5],ymm10[6],ymm9[6],ymm10[7],ymm9[7],ymm10[12],ymm9[12],ymm10[13],ymm9[13],ymm10[14],ymm9[14],ymm10[15],ymm9[15]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm8[4],ymm5[4],ymm8[5],ymm5[5],ymm8[6],ymm5[6],ymm8[7],ymm5[7],ymm8[12],ymm5[12],ymm8[13],ymm5[13],ymm8[14],ymm5[14],ymm8[15],ymm5[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm0[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm2[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1,2],ymm4[3],ymm5[4,5,6],ymm4[7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm8 = ymm11[4],ymm13[4],ymm11[5],ymm13[5],ymm11[6],ymm13[6],ymm11[7],ymm13[7],ymm11[12],ymm13[12],ymm11[13],ymm13[13],ymm11[14],ymm13[14],ymm11[15],ymm13[15]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm3[4],ymm1[4],ymm3[5],ymm1[5],ymm3[6],ymm1[6],ymm3[7],ymm1[7],ymm3[12],ymm1[12],ymm3[13],ymm1[13],ymm3[14],ymm1[14],ymm3[15],ymm1[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm8[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm1[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm5[0],ymm3[1],ymm5[2,3,4],ymm3[5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm7[4],ymm8[4],ymm7[5],ymm8[5],ymm7[6],ymm8[6],ymm7[7],ymm8[7],ymm7[12],ymm8[12],ymm7[13],ymm8[13],ymm7[14],ymm8[14],ymm7[15],ymm8[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm5[4],ymm4[4],ymm5[5],ymm4[5],ymm5[6],ymm4[6],ymm5[7],ymm4[7],ymm5[12],ymm4[12],ymm5[13],ymm4[13],ymm5[14],ymm4[14],ymm5[15],ymm4[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm15[4],ymm3[4],ymm15[5],ymm3[5],ymm15[6],ymm3[6],ymm15[7],ymm3[7],ymm15[12],ymm3[12],ymm15[13],ymm3[13],ymm15[14],ymm3[14],ymm15[15],ymm3[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm11[4],ymm13[4],ymm11[5],ymm13[5],ymm11[6],ymm13[6],ymm11[7],ymm13[7],ymm11[12],ymm13[12],ymm11[13],ymm13[13],ymm11[14],ymm13[14],ymm11[15],ymm13[15]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm4[0],ymm0[0],ymm4[1],ymm0[1],ymm4[4],ymm0[4],ymm4[5],ymm0[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm7 = ymm5[0],ymm3[0],ymm5[1],ymm3[1],ymm5[4],ymm3[4],ymm5[5],ymm3[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm10 = ymm7[0,1],ymm1[2,3],ymm7[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm4[2],ymm0[2],ymm4[3],ymm0[3],ymm4[6],ymm0[6],ymm4[7],ymm0[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm3 = ymm5[2],ymm3[2],ymm5[3],ymm3[3],ymm5[6],ymm3[6],ymm5[7],ymm3[7]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm8[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3,4],ymm2[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm7 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
-; AVX2-FP-NEXT:    vmovdqa 96(%r10), %ymm6
-; AVX2-FP-NEXT:    vmovdqa 96(%rax), %ymm5
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm6[0],ymm5[0],ymm6[1],ymm5[1],ymm6[2],ymm5[2],ymm6[3],ymm5[3],ymm6[8],ymm5[8],ymm6[9],ymm5[9],ymm6[10],ymm5[10],ymm6[11],ymm5[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FP-NEXT:    vmovdqa 96(%r8), %ymm8
-; AVX2-FP-NEXT:    vmovdqa 96(%r9), %ymm9
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm10 = ymm8[0],ymm9[0],ymm8[1],ymm9[1],ymm8[2],ymm9[2],ymm8[3],ymm9[3],ymm8[8],ymm9[8],ymm8[9],ymm9[9],ymm8[10],ymm9[10],ymm8[11],ymm9[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm11 = ymm10[0,1,1,3,4,5,5,7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm6 = ymm3[0,1],ymm0[2,3],ymm3[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vmovdqa 96(%r8), %ymm3
+; AVX2-FP-NEXT:    vmovdqa 96(%r9), %ymm4
+; AVX2-FP-NEXT:    vmovdqa 96(%r10), %ymm5
+; AVX2-FP-NEXT:    vmovdqa 96(%rax), %ymm7
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm8 = ymm5[0],ymm7[0],ymm5[1],ymm7[1],ymm5[2],ymm7[2],ymm5[3],ymm7[3],ymm5[8],ymm7[8],ymm5[9],ymm7[9],ymm5[10],ymm7[10],ymm5[11],ymm7[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm9 = ymm3[0],ymm4[0],ymm3[1],ymm4[1],ymm3[2],ymm4[2],ymm3[3],ymm4[3],ymm3[8],ymm4[8],ymm3[9],ymm4[9],ymm3[10],ymm4[10],ymm3[11],ymm4[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm11 = ymm9[0],ymm8[0],ymm9[1],ymm8[1],ymm9[4],ymm8[4],ymm9[5],ymm8[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm4 = ymm11[0,1,2],ymm3[3],ymm11[4,5,6],ymm3[7]
-; AVX2-FP-NEXT:    vmovdqa 96(%rdx), %ymm11
-; AVX2-FP-NEXT:    vmovdqa 96(%rcx), %ymm13
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm14 = ymm11[0],ymm13[0],ymm11[1],ymm13[1],ymm11[2],ymm13[2],ymm11[3],ymm13[3],ymm11[8],ymm13[8],ymm11[9],ymm13[9],ymm11[10],ymm13[10],ymm11[11],ymm13[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm15 = ymm14[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
-; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm3
-; AVX2-FP-NEXT:    vmovdqa 96(%rsi), %ymm1
-; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[8],ymm1[8],ymm3[9],ymm1[9],ymm3[10],ymm1[10],ymm3[11],ymm1[11]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm12 = ymm0[0,1,1,3,4,5,5,7]
+; AVX2-FP-NEXT:    vmovdqa 96(%rdi), %ymm13
+; AVX2-FP-NEXT:    vmovdqa 96(%rsi), %ymm14
+; AVX2-FP-NEXT:    vmovdqa 96(%rdx), %ymm15
+; AVX2-FP-NEXT:    vmovdqa 96(%rcx), %ymm2
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm15[0],ymm2[0],ymm15[1],ymm2[1],ymm15[2],ymm2[2],ymm15[3],ymm2[3],ymm15[8],ymm2[8],ymm15[9],ymm2[9],ymm15[10],ymm2[10],ymm15[11],ymm2[11]
+; AVX2-FP-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm13[0],ymm14[0],ymm13[1],ymm14[1],ymm13[2],ymm14[2],ymm13[3],ymm14[3],ymm13[8],ymm14[8],ymm13[9],ymm14[9],ymm13[10],ymm14[10],ymm13[11],ymm14[11]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm12 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm12 = ymm12[0],ymm15[1],ymm12[2,3,4],ymm15[5],ymm12[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm15 = ymm12[0,1],ymm4[2,3],ymm12[4,5],ymm4[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm10[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1,2],ymm2[3],ymm4[4,5,6],ymm2[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm14[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,1,3,3,6,5,7,7]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm11 = ymm12[0,1],ymm11[2,3],ymm12[4,5],ymm11[6,7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm8 = ymm9[2],ymm8[2],ymm9[3],ymm8[3],ymm9[6],ymm8[6],ymm9[7],ymm8[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm8[0,2,2,3]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm4[1],ymm0[2,3,4],ymm4[5],ymm0[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm2 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm6[4],ymm5[4],ymm6[5],ymm5[5],ymm6[6],ymm5[6],ymm6[7],ymm5[7],ymm6[12],ymm5[12],ymm6[13],ymm5[13],ymm6[14],ymm5[14],ymm6[15],ymm5[15]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm8[4],ymm9[4],ymm8[5],ymm9[5],ymm8[6],ymm9[6],ymm8[7],ymm9[7],ymm8[12],ymm9[12],ymm8[13],ymm9[13],ymm8[14],ymm9[14],ymm8[15],ymm9[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm5 = ymm0[0,0,2,1,4,4,6,5]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm5[4],ymm7[4],ymm5[5],ymm7[5],ymm5[6],ymm7[6],ymm5[7],ymm7[7],ymm5[12],ymm7[12],ymm5[13],ymm7[13],ymm5[14],ymm7[14],ymm5[15],ymm7[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm3[4],ymm4[4],ymm3[5],ymm4[5],ymm3[6],ymm4[6],ymm3[7],ymm4[7],ymm3[12],ymm4[12],ymm3[13],ymm4[13],ymm3[14],ymm4[14],ymm3[15],ymm4[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm15[4],ymm2[4],ymm15[5],ymm2[5],ymm15[6],ymm2[6],ymm15[7],ymm2[7],ymm15[12],ymm2[12],ymm15[13],ymm2[13],ymm15[14],ymm2[14],ymm15[15],ymm2[15]
+; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm13[4],ymm14[4],ymm13[5],ymm14[5],ymm13[6],ymm14[6],ymm13[7],ymm14[7],ymm13[12],ymm14[12],ymm13[13],ymm14[13],ymm13[14],ymm14[14],ymm13[15],ymm14[15]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm5 = ymm3[0],ymm1[0],ymm3[1],ymm1[1],ymm3[4],ymm1[4],ymm3[5],ymm1[5]
 ; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm6 = ymm4[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm6[0,1,2],ymm5[3],ymm6[4,5,6],ymm5[7]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm11[4],ymm13[4],ymm11[5],ymm13[5],ymm11[6],ymm13[6],ymm11[7],ymm13[7],ymm11[12],ymm13[12],ymm11[13],ymm13[13],ymm11[14],ymm13[14],ymm11[15],ymm13[15]
-; AVX2-FP-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm3[4],ymm1[4],ymm3[5],ymm1[5],ymm3[6],ymm1[6],ymm3[7],ymm1[7],ymm3[12],ymm1[12],ymm3[13],ymm1[13],ymm3[14],ymm1[14],ymm3[15],ymm1[15]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm3 = ymm6[0,0,2,1,4,4,6,5]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm8 = ymm1[0,1,1,3,4,5,5,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm8[0],ymm3[1],ymm8[2,3,4],ymm3[5],ymm8[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm5[2,3],ymm3[4,5],ymm5[6,7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm0[3],ymm4[4,5,6],ymm0[7]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm4 = ymm6[0,2,2,3,4,6,6,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-FP-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,3,3,6,5,7,7]
-; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm4[1],ymm1[2,3,4],ymm4[5],ymm1[6,7]
-; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-FP-NEXT:    vpunpckldq {{.*#+}} ymm7 = ymm4[0],ymm2[0],ymm4[1],ymm2[1],ymm4[4],ymm2[4],ymm4[5],ymm2[5]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm5 = ymm7[0,1],ymm5[2,3],ymm7[4,5],ymm5[6,7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm1 = ymm3[2],ymm1[2],ymm3[3],ymm1[3],ymm3[6],ymm1[6],ymm3[7],ymm1[7]
+; AVX2-FP-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm4[2],ymm2[2],ymm4[3],ymm2[3],ymm4[6],ymm2[6],ymm4[7],ymm2[7]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-FP-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-FP-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
 ; AVX2-FP-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FP-NEXT:    vmovdqa %ymm0, 992(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm3, 960(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm2, 928(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm15, 896(%rax)
-; AVX2-FP-NEXT:    vmovdqa %ymm7, 736(%rax)
-; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 704(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm1, 992(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm5, 960(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm0, 928(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm11, 896(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm6, 736(%rax)
+; AVX2-FP-NEXT:    vmovdqa %ymm10, 704(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 672(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 640(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 480(%rax)
-; AVX2-FP-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
-; AVX2-FP-NEXT:    vmovaps %ymm0, 448(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FP-NEXT:    vmovaps %ymm0, 448(%rax)
+; AVX2-FP-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 416(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 384(%rax)
@@ -8110,7 +7422,7 @@ define void @store_i16_stride8_vf64(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.ve
 ; AVX2-FP-NEXT:    vmovaps %ymm0, 32(%rax)
 ; AVX2-FP-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FP-NEXT:    vmovaps %ymm0, (%rax)
-; AVX2-FP-NEXT:    addq $744, %rsp # imm = 0x2E8
+; AVX2-FP-NEXT:    addq $712, %rsp # imm = 0x2C8
 ; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-6.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-6.ll
index 0cd72be39557a1..e4f616ed730ebf 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-6.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-6.ll
@@ -388,12 +388,11 @@ define void @store_i32_stride6_vf4(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX-NEXT:    vunpcklps {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
 ; AVX-NEXT:    vblendps {{.*#+}} ymm4 = ymm6[0,1],ymm4[2,3],ymm6[4,5,6,7]
 ; AVX-NEXT:    vunpckhpd {{.*#+}} ymm5 = ymm8[1],ymm9[1],ymm8[3],ymm9[3]
-; AVX-NEXT:    vshufps {{.*#+}} ymm5 = ymm5[0,2,3,1,4,6,7,5]
 ; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
 ; AVX-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[3,3],ymm2[3,3],ymm0[7,7],ymm2[7,7]
-; AVX-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2,1,3,4,6,5,7]
-; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm5[0,1],ymm0[2,3,4,5],ymm5[6,7]
+; AVX-NEXT:    vblendps {{.*#+}} ymm0 = ymm5[0],ymm0[1],ymm5[2],ymm0[3,4],ymm5[5],ymm0[6],ymm5[7]
+; AVX-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,2,1,3,4,6,7,5]
 ; AVX-NEXT:    vmovaps %ymm0, 64(%rax)
 ; AVX-NEXT:    vmovaps %ymm4, 32(%rax)
 ; AVX-NEXT:    vmovaps %ymm10, (%rax)
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-7.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-7.ll
index 45a76599d3e9d7..13930bc2c67407 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-7.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-7.ll
@@ -521,12 +521,10 @@ define void @store_i32_stride7_vf4(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX2-NEXT:    vinsertf128 $1, (%rsi), %ymm4, %ymm4
 ; AVX2-NEXT:    vinsertf128 $1, %xmm0, %ymm5, %ymm6
 ; AVX2-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm7
-; AVX2-NEXT:    vbroadcastf128 {{.*#+}} ymm8 = [2,6,0,3,2,6,0,3]
-; AVX2-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX2-NEXT:    vpermps %ymm6, %ymm8, %ymm8
-; AVX2-NEXT:    vshufps {{.*#+}} ymm9 = ymm4[3,3,3,3,7,7,7,7]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm9 = ymm9[0,1,0,2]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm9[5,6],ymm8[7]
+; AVX2-NEXT:    vshufps {{.*#+}} ymm8 = ymm4[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm8 = ymm6[0],ymm8[1],ymm6[2,3],ymm8[4],ymm6[5,6,7]
+; AVX2-NEXT:    vmovaps {{.*#+}} ymm9 = [2,6,u,u,u,1,4,3]
+; AVX2-NEXT:    vpermps %ymm8, %ymm9, %ymm8
 ; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm9
 ; AVX2-NEXT:    vunpckhps {{.*#+}} ymm9 = ymm9[2],ymm3[2],ymm9[3],ymm3[3],ymm9[6],ymm3[6],ymm9[7],ymm3[7]
 ; AVX2-NEXT:    vshufps {{.*#+}} ymm9 = ymm9[0,1,0,1,4,5,4,5]
@@ -573,12 +571,10 @@ define void @store_i32_stride7_vf4(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX2-FP-NEXT:    vinsertf128 $1, (%rsi), %ymm4, %ymm4
 ; AVX2-FP-NEXT:    vinsertf128 $1, %xmm0, %ymm5, %ymm6
 ; AVX2-FP-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm7
-; AVX2-FP-NEXT:    vbroadcastf128 {{.*#+}} ymm8 = [2,6,0,3,2,6,0,3]
-; AVX2-FP-NEXT:    # ymm8 = mem[0,1,0,1]
-; AVX2-FP-NEXT:    vpermps %ymm6, %ymm8, %ymm8
-; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm9 = ymm4[3,3,3,3,7,7,7,7]
-; AVX2-FP-NEXT:    vpermpd {{.*#+}} ymm9 = ymm9[0,1,0,2]
-; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm8[0,1,2,3,4],ymm9[5,6],ymm8[7]
+; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm8 = ymm4[3,3,3,3,7,7,7,7]
+; AVX2-FP-NEXT:    vblendps {{.*#+}} ymm8 = ymm6[0],ymm8[1],ymm6[2,3],ymm8[4],ymm6[5,6,7]
+; AVX2-FP-NEXT:    vmovaps {{.*#+}} ymm9 = [2,6,u,u,u,1,4,3]
+; AVX2-FP-NEXT:    vpermps %ymm8, %ymm9, %ymm8
 ; AVX2-FP-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm9
 ; AVX2-FP-NEXT:    vunpckhps {{.*#+}} ymm9 = ymm9[2],ymm3[2],ymm9[3],ymm3[3],ymm9[6],ymm3[6],ymm9[7],ymm3[7]
 ; AVX2-FP-NEXT:    vshufps {{.*#+}} ymm9 = ymm9[0,1,0,1,4,5,4,5]
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll
index 4859a8e0eaaa51..81ce14132c8799 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll
@@ -308,16 +308,14 @@ define <4 x float> @combine_vpermilvar_4f32_as_insertps(<4 x float> %a0) {
 define <8 x i32> @combine_blend_of_permutes_v8i32(<4 x i64> %a0, <4 x i64> %a1) {
 ; AVX1-LABEL: combine_blend_of_permutes_v8i32:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,0,1]
-; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[2,3,0,1]
-; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3,4],ymm0[5,6],ymm1[7]
 ; AVX1-NEXT:    ret{{[l|q]}}
 ;
 ; AVX2-LABEL: combine_blend_of_permutes_v8i32:
 ; AVX2:       # %bb.0:
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
-; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[2,3,0,1]
-; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3,4],ymm0[5,6],ymm1[7]
 ; AVX2-NEXT:    ret{{[l|q]}}
 ;
 ; AVX512-LABEL: combine_blend_of_permutes_v8i32:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-sse41.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-sse41.ll
index 8d213d25774338..50be3c5bc6c065 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-sse41.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-sse41.ll
@@ -25,23 +25,20 @@ define <16 x i8> @combine_vpshufb_as_movzx(<16 x i8> %a0) {
 define <4 x i32> @combine_blend_of_permutes_v4i32(<2 x i64> %a0, <2 x i64> %a1) {
 ; SSE-LABEL: combine_blend_of_permutes_v4i32:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE-NEXT:    pblendw {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3],xmm2[4,5],xmm0[6,7]
+; SSE-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_blend_of_permutes_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX1-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; AVX1-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: combine_blend_of_permutes_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX2-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; AVX2-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: combine_blend_of_permutes_v4i32:



More information about the llvm-commits mailing list