[llvm] d7eb924 - [DAG] SimplifyDemandedBits - attempt to use SimplifyMultipleUseDemandedBits for bitcasts from larger element types
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Tue Jul 18 10:38:19 PDT 2023
Author: Simon Pilgrim
Date: 2023-07-18T18:38:03+01:00
New Revision: d7eb9240c0dd4f67f12116e8af4918261e0391b1
URL: https://github.com/llvm/llvm-project/commit/d7eb9240c0dd4f67f12116e8af4918261e0391b1
DIFF: https://github.com/llvm/llvm-project/commit/d7eb9240c0dd4f67f12116e8af4918261e0391b1.diff
LOG: [DAG] SimplifyDemandedBits - attempt to use SimplifyMultipleUseDemandedBits for bitcasts from larger element types
Attempt to avoid multi-use ops if the bitcast doesn't need anything from them.
Added:
Modified:
llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
llvm/test/CodeGen/X86/srem-seteq-illegal-types.ll
llvm/test/CodeGen/X86/vec_setcc.ll
llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-7.ll
Removed:
################################################################################
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 8136b751c2779c..11594da0acb561 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -2575,6 +2575,15 @@ bool TargetLowering::SimplifyDemandedBits(
if (SimplifyDemandedBits(Src, DemandedSrcBits, DemandedSrcElts,
KnownSrcBits, TLO, Depth + 1))
return true;
+
+ // Attempt to avoid multi-use ops if we don't need anything from them.
+ if (!DemandedSrcBits.isAllOnes() || !DemandedSrcElts.isAllOnes()) {
+ if (SDValue DemandedSrc = SimplifyMultipleUseDemandedBits(
+ Src, DemandedSrcBits, DemandedSrcElts, TLO.DAG, Depth + 1)) {
+ SDValue NewOp = TLO.DAG.getBitcast(VT, DemandedSrc);
+ return TLO.CombineTo(Op, NewOp);
+ }
+ }
}
// If this is a bitcast, let computeKnownBits handle it. Only do this on a
diff --git a/llvm/test/CodeGen/X86/srem-seteq-illegal-types.ll b/llvm/test/CodeGen/X86/srem-seteq-illegal-types.ll
index 412128ba7f5107..8b70daee1d60bc 100644
--- a/llvm/test/CodeGen/X86/srem-seteq-illegal-types.ll
+++ b/llvm/test/CodeGen/X86/srem-seteq-illegal-types.ll
@@ -269,7 +269,7 @@ define <3 x i1> @test_srem_vec(<3 x i33> %X) nounwind {
; SSE41-NEXT: pxor %xmm1, %xmm0
; SSE41-NEXT: pcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
; SSE41-NEXT: pxor %xmm1, %xmm2
-; SSE41-NEXT: pextrb $0, %xmm0, %eax
+; SSE41-NEXT: movd %xmm0, %eax
; SSE41-NEXT: pextrb $8, %xmm0, %edx
; SSE41-NEXT: pextrb $0, %xmm2, %ecx
; SSE41-NEXT: # kill: def $al killed $al killed $eax
diff --git a/llvm/test/CodeGen/X86/vec_setcc.ll b/llvm/test/CodeGen/X86/vec_setcc.ll
index faf9dc4b888615..e7232a34f47122 100644
--- a/llvm/test/CodeGen/X86/vec_setcc.ll
+++ b/llvm/test/CodeGen/X86/vec_setcc.ll
@@ -322,7 +322,7 @@ define <3 x i1> @test_setcc_v3i1_v3i16(ptr %a) nounwind {
; SSE41-NEXT: movq {{.*#+}} xmm0 = mem[0],zero
; SSE41-NEXT: pxor %xmm1, %xmm1
; SSE41-NEXT: pcmpeqw %xmm0, %xmm1
-; SSE41-NEXT: pextrb $0, %xmm1, %eax
+; SSE41-NEXT: movd %xmm1, %eax
; SSE41-NEXT: pextrb $2, %xmm1, %edx
; SSE41-NEXT: pextrb $4, %xmm1, %ecx
; SSE41-NEXT: # kill: def $al killed $al killed $eax
@@ -335,7 +335,7 @@ define <3 x i1> @test_setcc_v3i1_v3i16(ptr %a) nounwind {
; AVX-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero
; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpextrb $0, %xmm0, %eax
+; AVX-NEXT: vmovd %xmm0, %eax
; AVX-NEXT: vpextrb $2, %xmm0, %edx
; AVX-NEXT: vpextrb $4, %xmm0, %ecx
; AVX-NEXT: # kill: def $al killed $al killed $eax
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-7.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-7.ll
index dc38021d73fd52..8dbd67be7769ed 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-7.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-7.ll
@@ -2169,10 +2169,10 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
;
; AVX2-SLOW-LABEL: load_i16_stride7_vf16:
; AVX2-SLOW: # %bb.0:
-; AVX2-SLOW-NEXT: vmovdqa (%rdi), %ymm2
-; AVX2-SLOW-NEXT: vmovdqa 32(%rdi), %ymm3
-; AVX2-SLOW-NEXT: vmovdqa 64(%rdi), %ymm0
-; AVX2-SLOW-NEXT: vmovdqa 96(%rdi), %ymm1
+; AVX2-SLOW-NEXT: vmovdqa (%rdi), %ymm0
+; AVX2-SLOW-NEXT: vmovdqa 32(%rdi), %ymm1
+; AVX2-SLOW-NEXT: vmovdqa 64(%rdi), %ymm2
+; AVX2-SLOW-NEXT: vmovdqa 96(%rdi), %ymm3
; AVX2-SLOW-NEXT: vmovdqa 192(%rdi), %ymm9
; AVX2-SLOW-NEXT: vmovdqa 128(%rdi), %ymm5
; AVX2-SLOW-NEXT: vmovdqa 160(%rdi), %ymm6
@@ -2187,22 +2187,22 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm8 = ymm7[0,1,2,1,4,5,6,5]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm8 = ymm2[0,1],ymm3[2],ymm2[3,4,5],ymm3[6],ymm2[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm8 = ymm0[0,1],ymm1[2],ymm0[3,4,5],ymm1[6],ymm0[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm8, %xmm10
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm8 = xmm8[0,1,2,3],xmm10[4],xmm8[5],xmm10[6],xmm8[7]
; AVX2-SLOW-NEXT: vpshufb {{.*#+}} xmm8 = xmm8[0,1,14,15,12,13,10,11,8,9,8,9,8,9,8,9]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm10 = ymm1[0],ymm0[1],ymm1[2,3,4],ymm0[5],ymm1[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm10 = ymm3[0],ymm2[1],ymm3[2,3,4],ymm2[5],ymm3[6,7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm10 = ymm10[0,2,2,3]
; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm10 = ymm10[0,1,14,15,12,13,14,15,2,3,6,7,12,13,2,3,16,17,30,31,u,u,u,u,u,u,u,u,u,u,u,u]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm11 = <255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0,0,0,u,u,u,u,u,u,u,u,u,u,u,u>
; AVX2-SLOW-NEXT: vpblendvb %ymm11, %ymm8, %ymm10, %ymm8
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm8[0,1,2,3,4],ymm4[5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm8 = ymm2[0,1,2],ymm3[3],ymm2[4,5],ymm3[6],ymm2[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm8 = ymm0[0,1,2],ymm1[3],ymm0[4,5],ymm1[6],ymm0[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm8, %xmm10
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm8 = xmm10[0],xmm8[1],xmm10[2,3,4,5],xmm8[6],xmm10[7]
; AVX2-SLOW-NEXT: vpshufb {{.*#+}} xmm8 = xmm8[2,3,0,1,14,15,12,13,10,11,10,11,10,11,10,11]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm10 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm10 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm11 = ymm10[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm10 = ymm10[0,1,2],ymm11[3],ymm10[4,5,6,7,8,9,10],ymm11[11],ymm10[12,13,14,15]
; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm11 = ymm10[2,3,2,3,2,3,2,3,8,9,8,9,6,7,4,5,18,19,18,19,18,19,18,19,24,25,24,25,22,23,20,21]
@@ -2218,12 +2218,12 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm7 = ymm11[0,1,2,3,4,5,6],ymm7[7]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm7 = ymm8[0],ymm7[1,2,3,4,5,6,7],ymm8[8],ymm7[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm8 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm8 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm8, %xmm11
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm8 = xmm8[0],xmm11[1],xmm8[2,3,4,5],xmm11[6],xmm8[7]
; AVX2-SLOW-NEXT: vpshufb {{.*#+}} xmm8 = xmm8[4,5,2,3,0,1,14,15,12,13,12,13,12,13,12,13]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm11 = ymm0[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm12 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm11 = ymm2[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm12 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm11 = ymm12[0,1,2,3],ymm11[4],ymm12[5,6,7,8,9,10,11],ymm11[12],ymm12[13,14,15]
; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm11 = ymm11[4,5,4,5,4,5,4,5,8,9,10,11,8,9,6,7,20,21,20,21,20,21,20,21,24,25,26,27,24,25,22,23]
; AVX2-SLOW-NEXT: vpblendvb %ymm10, %ymm8, %ymm11, %ymm8
@@ -2248,12 +2248,12 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm9 = ymm12[0,1,2,0,4,5,6,4]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm9 = ymm9[0,1,2,3,4,5,7,4,8,9,10,11,12,13,15,12]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm9 = ymm11[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm11 = ymm3[0],ymm2[1],ymm3[2,3,4],ymm2[5],ymm3[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm11 = ymm1[0],ymm0[1],ymm1[2,3,4],ymm0[5],ymm1[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm11, %xmm13
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm11 = xmm13[0],xmm11[1],xmm13[2],xmm11[3],xmm13[4,5,6,7]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm11 = xmm11[3,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm11 = xmm11[0,1,2,3,7,7,7,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm13 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm13 = ymm3[0,1,2],ymm2[3],ymm3[4,5],ymm2[6],ymm3[7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm13 = ymm13[1,3,2,3]
; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm13 = ymm13[6,7,6,7,6,7,6,7,8,9,4,5,10,11,0,1,22,23,22,23,22,23,22,23,24,25,20,21,26,27,16,17]
; AVX2-SLOW-NEXT: vpblendvb %ymm10, %ymm11, %ymm13, %ymm10
@@ -2267,18 +2267,18 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm11 = ymm12[0,1,2,1,4,5,6,5]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5,6],ymm11[7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm11 = ymm0[0,1],ymm1[2],ymm0[3,4,5],ymm1[6],ymm0[7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm12 = ymm11[2,3,0,1]
-; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm11 = ymm11[u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5],ymm12[6],ymm11[7,8,9,10,11,12,13],ymm12[14],ymm11[15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm12 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm12, %xmm13
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm13 = xmm13[3,1,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm12 = xmm12[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm12 = xmm12[0,2,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm12 = xmm12[0],xmm13[0],xmm12[1],xmm13[1],xmm12[2],xmm13[2],xmm12[3],xmm13[3]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm11 = ymm12[0,1],ymm11[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm10 = ymm11[0],ymm10[1,2,3,4,5,6,7],ymm11[8],ymm10[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm11 = ymm2[0,1],ymm3[2],ymm2[3,4,5],ymm3[6],ymm2[7]
+; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm12 = ymm11[u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm10 = ymm12[0],ymm10[1,2,3,4,5,6,7],ymm12[8],ymm10[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm13 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm13, %xmm14
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm14 = xmm14[3,1,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm13 = xmm13[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm13 = xmm13[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm13 = xmm13[0],xmm14[0],xmm13[1],xmm14[1],xmm13[2],xmm14[2],xmm13[3],xmm14[3]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm11 = ymm11[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm11 = ymm12[0,1,2,3,4,5],ymm11[6],ymm12[7,8,9,10,11,12,13],ymm11[14],ymm12[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm11 = xmm13[0,1],xmm11[2,3]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm10 = ymm11[0,1,2,3],ymm10[4,5,6,7]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm11 = ymm6[0,1],ymm5[2],ymm6[3,4],ymm5[5],ymm6[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm11, %xmm12
@@ -2292,43 +2292,43 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm14 = xmm14[0,1,2,3,4,4,7,6]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm14, %ymm0, %ymm14
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm14[5,6,7],ymm11[8,9,10,11,12],ymm14[13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm14 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm14, %xmm15
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm14 = ymm2[0,1,2],ymm3[3],ymm2[4,5],ymm3[6],ymm2[7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm15 = ymm14[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm14 = ymm15[0],ymm14[1,2,3,4,5,6],ymm15[7,8],ymm14[9,10,11,12,13,14],ymm15[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm15 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm15, %xmm4
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm15 = xmm15[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm15 = xmm15[0,2,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm14 = xmm14[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm14 = xmm14[1,3,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm14 = xmm14[0],xmm15[0],xmm14[1],xmm15[1],xmm14[2],xmm15[2],xmm14[3],xmm15[3]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm15 = ymm0[0,1,2],ymm1[3],ymm0[4,5],ymm1[6],ymm0[7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm15[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0],ymm15[1,2,3,4,5,6],ymm4[7,8],ymm15[9,10,11,12,13,14],ymm4[15]
-; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm14[0,1],ymm4[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm11 = ymm4[0],ymm11[1,2,3,4,5,6,7],ymm4[8],ymm11[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm11[4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm5 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm5 = xmm6[0,1,2],xmm5[3],xmm6[4],xmm5[5],xmm6[6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm6 = xmm12[0],xmm13[1],xmm12[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufb {{.*#+}} xmm5 = xmm5[u,u,10,11,8,9,6,7,4,5,u,u,u,u,u,u]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm15 = xmm15[1,3,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm15[0],xmm4[0],xmm15[1],xmm4[1],xmm15[2],xmm4[2],xmm15[3],xmm4[3]
+; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm14 = ymm14[u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm11 = ymm14[0],ymm11[1,2,3,4,5,6,7],ymm14[8],ymm11[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm14[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm11 = ymm4[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm5
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[3],xmm5[4],xmm4[5],xmm5[6,7]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm5 = xmm12[0],xmm13[1],xmm12[2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufb {{.*#+}} xmm4 = xmm4[u,u,10,11,8,9,6,7,4,5,u,u,u,u,u,u]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[0,1,0,3]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,4,7]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm6 = xmm6[0,1,0,3]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,5,4,7]
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm6
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5],ymm2[6],ymm3[7]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm3 = xmm2[0,1,2,3,6,4,6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm2
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,1,2,1]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,7,6,7]
-; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm2 = xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3,4,5,6,7,8],ymm1[9],ymm0[10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm5[1,2,3,4,5,6,7],ymm0[8],ymm5[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7],ymm4[8,9,10,11,12],ymm5[13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm2[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0],ymm3[1],ymm2[2,3,4,5,6,7,8],ymm3[9],ymm2[10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,6,4,6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm0
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,7,6,7]
+; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0],ymm4[1,2,3,4,5,6,7],ymm1[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-SLOW-NEXT: vmovaps %ymm1, (%rsi)
; AVX2-SLOW-NEXT: vmovdqa %ymm7, (%rdx)
@@ -2336,7 +2336,7 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vmovdqa %ymm9, (%r8)
; AVX2-SLOW-NEXT: vmovdqa %ymm10, (%r9)
; AVX2-SLOW-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; AVX2-SLOW-NEXT: vmovdqa %ymm4, (%rax)
+; AVX2-SLOW-NEXT: vmovdqa %ymm11, (%rax)
; AVX2-SLOW-NEXT: movq {{[0-9]+}}(%rsp), %rax
; AVX2-SLOW-NEXT: vmovdqa %ymm0, (%rax)
; AVX2-SLOW-NEXT: vzeroupper
@@ -2352,8 +2352,7 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vmovdqa 160(%rdi), %ymm5
; AVX2-FAST-NEXT: vmovdqa 192(%rdi), %ymm7
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm8 = ymm7[0,1,0,2]
-; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm9 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
-; AVX2-FAST-NEXT: vpshufb %ymm9, %ymm8, %ymm6
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm6 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,26,27]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm10 = ymm5[0,1,2],ymm4[3],ymm5[4,5],ymm4[6],ymm5[7]
; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm11 = [3,6,2,5,3,6,2,5]
; AVX2-FAST-NEXT: # ymm11 = mem[0,1,0,1]
@@ -2389,7 +2388,8 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm8 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,28,29]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm8 = ymm12[0,1,2,3,4,5,6],ymm8[7]
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm8 = ymm10[0],ymm8[1,2,3,4,5,6,7],ymm10[8],ymm8[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm8 = ymm10[0,1,2,3],ymm8[4,5,6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm10[0,1,2,3],ymm8[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm10 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm10, %xmm12
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm10 = xmm10[0],xmm12[1],xmm10[2,3,4,5],xmm12[6],xmm10[7]
@@ -2428,52 +2428,52 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vpblendvb %ymm11, %ymm14, %ymm15, %ymm11
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm12 = ymm11[0],ymm12[1,2,3,4,5,6,7],ymm11[8],ymm12[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm11 = ymm11[0,1,2,3],ymm12[4,5,6,7]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm12 = ymm5[0],ymm4[1],ymm5[2,3,4],ymm4[5],ymm5[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm12, %xmm14
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm12 = xmm14[0],xmm12[1],xmm14[2],xmm12[3],xmm14[4,5,6,7]
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm12 = xmm12[0,1,6,7,4,5,2,3,0,1,14,15,u,u,u,u]
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm12, %ymm0, %ymm12
-; AVX2-FAST-NEXT: vpshufb %ymm9, %ymm13, %ymm9
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm9 = ymm12[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm12 = ymm2[0,1],ymm3[2],ymm2[3,4,5],ymm3[6],ymm2[7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm13 = <0,3,7,2,6,u,u,u>
-; AVX2-FAST-NEXT: vpermd %ymm12, %ymm13, %ymm13
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm12 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm14 = [8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpshufb %xmm14, %xmm12, %xmm15
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm12, %xmm12
-; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm12 = xmm12[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm15 = xmm15[0],xmm12[0],xmm15[1],xmm12[1],xmm15[2],xmm12[2],xmm15[3],xmm12[3]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm12 = <u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31>
-; AVX2-FAST-NEXT: vpshufb %ymm12, %ymm13, %ymm13
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm13 = ymm15[0,1],ymm13[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm9 = ymm13[0],ymm9[1,2,3,4,5,6,7],ymm13[8],ymm9[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm9 = ymm13[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm13 = ymm5[0,1],ymm4[2],ymm5[3,4],ymm4[5],ymm5[6,7]
-; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm15 = [2,5,1,4,2,5,1,4]
-; AVX2-FAST-NEXT: # ymm15 = mem[0,1,0,1]
-; AVX2-FAST-NEXT: vpermd %ymm13, %ymm15, %ymm13
-; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm15 = [0,3,7,0,0,3,7,0]
-; AVX2-FAST-NEXT: # ymm15 = mem[0,1,0,1]
-; AVX2-FAST-NEXT: vpermd %ymm7, %ymm15, %ymm15
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm15 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25]
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm13 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25,30,31,u,u,u,u,u,u]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm13 = ymm13[0,1,2,3,4],ymm15[5,6,7],ymm13[8,9,10,11,12],ymm15[13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm15 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm14 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} xmm12 = [8,9,4,5,8,9,4,5,8,9,4,5,8,9,4,5]
+; AVX2-FAST-NEXT: vpshufb %xmm12, %xmm14, %xmm15
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm14, %xmm14
+; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm14 = xmm14[3,1,2,3,4,5,6,7]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm14 = xmm15[0],xmm14[0],xmm15[1],xmm14[1],xmm15[2],xmm14[2],xmm15[3],xmm14[3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm15 = ymm2[0,1],ymm3[2],ymm2[3,4,5],ymm3[6],ymm2[7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm9 = <0,3,7,2,6,u,u,u>
+; AVX2-FAST-NEXT: vpermd %ymm15, %ymm9, %ymm9
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm15 = ymm5[0],ymm4[1],ymm5[2,3,4],ymm4[5],ymm5[6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm15, %xmm6
-; AVX2-FAST-NEXT: vpshufb %xmm14, %xmm6, %xmm6
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm14 = xmm15[10,11,6,7,4,5,6,7,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm14[0],xmm6[0],xmm14[1],xmm6[1],xmm14[2],xmm6[2],xmm14[3],xmm6[3]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm6 = xmm6[0],xmm15[1],xmm6[2],xmm15[3],xmm6[4,5,6,7]
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm13 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,26,27]
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm6 = xmm6[0,1,6,7,4,5,2,3,0,1,14,15,u,u,u,u]
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm6
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31>
+; AVX2-FAST-NEXT: vpshufb %ymm13, %ymm9, %ymm9
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm14 = xmm14[0,1],xmm9[2,3]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm6 = ymm9[0],ymm6[1,2,3,4,5,6,7],ymm9[8],ymm6[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm9 = ymm14[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm5[0,1],ymm4[2],ymm5[3,4],ymm4[5],ymm5[6,7]
+; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm14 = [2,5,1,4,2,5,1,4]
+; AVX2-FAST-NEXT: # ymm14 = mem[0,1,0,1]
+; AVX2-FAST-NEXT: vpermd %ymm6, %ymm14, %ymm6
+; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm14 = [0,3,7,0,0,3,7,0]
+; AVX2-FAST-NEXT: # ymm14 = mem[0,1,0,1]
+; AVX2-FAST-NEXT: vpermd %ymm7, %ymm14, %ymm14
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm14 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25]
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm6 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25,30,31,u,u,u,u,u,u]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm14[5,6,7],ymm6[8,9,10,11,12],ymm14[13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm14 = ymm2[0,1,2],ymm3[3],ymm2[4,5],ymm3[6],ymm2[7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm15 = <0,4,7,3,6,u,u,u>
; AVX2-FAST-NEXT: vpermd %ymm14, %ymm15, %ymm14
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm14 = ymm14[u,u,u,u,u,u,u,u,2,3,4,5,10,11,12,13,18,19,18,19,18,19,18,19,18,19,20,21,26,27,28,29]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm6[0,1],ymm14[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm13 = ymm6[0],ymm13[1,2,3,4,5,6,7],ymm6[8],ymm13[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm13[4,5,6,7]
-; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm13 = [0,4,7,0,0,4,7,0]
-; AVX2-FAST-NEXT: # ymm13 = mem[0,1,0,1]
-; AVX2-FAST-NEXT: vpermd %ymm7, %ymm13, %ymm7
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm15 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm15, %xmm8
+; AVX2-FAST-NEXT: vpshufb %xmm12, %xmm8, %xmm8
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm12 = xmm15[10,11,6,7,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm8 = xmm12[0],xmm8[0],xmm12[1],xmm8[1],xmm12[2],xmm8[2],xmm12[3],xmm8[3]
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm12 = ymm14[u,u,u,u,u,u,u,u,2,3,4,5,10,11,12,13,18,19,18,19,18,19,18,19,18,19,20,21,26,27,28,29]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm6 = ymm12[0],ymm6[1,2,3,4,5,6,7],ymm12[8],ymm6[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm8 = xmm8[0,1],xmm12[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm8[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm8 = [0,4,7,0,0,4,7,0]
+; AVX2-FAST-NEXT: # ymm8 = mem[0,1,0,1]
+; AVX2-FAST-NEXT: vpermd %ymm7, %ymm8, %ymm7
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3],ymm5[4,5],ymm4[6,7]
; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm5 = [2,6,1,5,2,6,1,5]
; AVX2-FAST-NEXT: # ymm5 = mem[0,1,0,1]
@@ -2484,18 +2484,19 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6,7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm3 = <1,4,0,3,7,u,u,u>
; AVX2-FAST-NEXT: vpermd %ymm2, %ymm3, %ymm2
-; AVX2-FAST-NEXT: vpshufb %ymm12, %ymm2, %ymm2
+; AVX2-FAST-NEXT: vpshufb %ymm13, %ymm2, %ymm2
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[u,u,u,u,u,u,u,u,10,11,6,7,4,5,6,7]
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[u,u,u,u,u,u,u,u,10,11,6,7,u,u,u,u]
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,4,6,7]
; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm4[1,2,3,4,5,6,7],ymm0[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm4[1,2,3,4,5,6,7],ymm2[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FAST-NEXT: vmovaps %ymm1, (%rsi)
-; AVX2-FAST-NEXT: vmovdqa %ymm8, (%rdx)
+; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm1, (%rdx)
; AVX2-FAST-NEXT: vmovdqa %ymm10, (%rcx)
; AVX2-FAST-NEXT: vmovdqa %ymm11, (%r8)
; AVX2-FAST-NEXT: vmovdqa %ymm9, (%r9)
@@ -2508,63 +2509,64 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
;
; AVX2-FAST-PERLANE-LABEL: load_i16_stride7_vf16:
; AVX2-FAST-PERLANE: # %bb.0:
-; AVX2-FAST-PERLANE-NEXT: vmovdqa (%rdi), %ymm2
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 32(%rdi), %ymm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 64(%rdi), %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 96(%rdi), %ymm1
+; AVX2-FAST-PERLANE-NEXT: vmovdqa (%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 32(%rdi), %ymm1
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 64(%rdi), %ymm2
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 96(%rdi), %ymm3
; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %ymm10
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 128(%rdi), %ymm4
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 128(%rdi), %ymm5
; AVX2-FAST-PERLANE-NEXT: vmovdqa 160(%rdi), %ymm6
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm6[0,1,2],ymm4[3],ymm6[4,5],ymm4[6],ymm6[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1,2],ymm5[3],ymm6[4,5],ymm5[6],ymm6[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm7
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm7 = xmm7[u,u,u,u,u,u,u,u,8,9,10,11,6,7,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,6,4,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm5 = xmm5[4],xmm7[4],xmm5[5],xmm7[5],xmm5[6],xmm7[6],xmm5[7],xmm7[7]
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
+; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,6,4,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm4[4],xmm7[4],xmm4[5],xmm7[5],xmm4[6],xmm7[6],xmm4[7],xmm7[7]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm7 = ymm10[0,1,0,2]
-; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm9 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm9, %ymm7, %ymm8
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm8 = ymm2[0,1],ymm3[2],ymm2[3,4,5],ymm3[6],ymm2[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm8, %xmm11
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm8 = xmm8[0,1,2,3],xmm11[4],xmm8[5],xmm11[6],xmm8[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm8 = xmm8[0,1,14,15,12,13,10,11,8,9,8,9,8,9,8,9]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm11 = ymm1[0],ymm0[1],ymm1[2,3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm8 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm8, %ymm7, %ymm9
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm9[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm9 = ymm0[0,1],ymm1[2],ymm0[3,4,5],ymm1[6],ymm0[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm9, %xmm11
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm9 = xmm9[0,1,2,3],xmm11[4],xmm9[5],xmm11[6],xmm9[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm9 = xmm9[0,1,14,15,12,13,10,11,8,9,8,9,8,9,8,9]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm11 = ymm3[0],ymm2[1],ymm3[2,3,4],ymm2[5],ymm3[6,7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm11 = ymm11[0,1,14,15,12,13,14,15,2,3,6,7,12,13,2,3,16,17,30,31,u,u,u,u,u,u,u,u,u,u,u,u]
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm12 = <255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0,0,0,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm12, %ymm8, %ymm11, %ymm8
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm8[0,1,2,3,4],ymm5[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm8 = ymm2[0,1,2],ymm3[3],ymm2[4,5],ymm3[6],ymm2[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm8, %xmm11
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm8 = xmm11[0],xmm8[1],xmm11[2,3,4,5],xmm8[6],xmm11[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm8 = xmm8[2,3,0,1,14,15,12,13,10,11,10,11,10,11,10,11]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm11 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm12, %ymm9, %ymm11, %ymm9
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm9[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm9 = ymm0[0,1,2],ymm1[3],ymm0[4,5],ymm1[6],ymm0[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm9, %xmm11
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm9 = xmm11[0],xmm9[1],xmm11[2,3,4,5],xmm9[6],xmm11[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm9 = xmm9[2,3,0,1,14,15,12,13,10,11,10,11,10,11,10,11]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm11 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm12 = ymm11[2,3,0,1]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3],ymm11[4,5,6,7,8,9,10],ymm12[11],ymm11[12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm12 = ymm11[2,3,2,3,2,3,2,3,8,9,8,9,6,7,4,5,18,19,18,19,18,19,18,19,24,25,24,25,22,23,20,21]
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm11 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm8, %ymm12, %ymm8
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm12 = ymm4[0,1],ymm6[2],ymm4[3,4,5],ymm6[6],ymm4[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm9, %ymm12, %ymm9
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm12 = ymm5[0,1],ymm6[2],ymm5[3,4,5],ymm6[6],ymm5[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm12, %xmm13
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm12 = xmm12[0,1,2,3],xmm13[4],xmm12[5],xmm13[6],xmm12[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm12 = xmm12[0,1,0,1,14,15,12,13,10,11,8,9,u,u,u,u]
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm12, %ymm0, %ymm12
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm7 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,28,29]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm7 = ymm12[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm7 = ymm8[0],ymm7[1,2,3,4,5,6,7],ymm8[8],ymm7[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm7 = ymm8[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm8 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm8, %xmm12
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm8 = xmm8[0],xmm12[1],xmm8[2,3,4,5],xmm12[6],xmm8[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm8 = xmm8[4,5,2,3,0,1,14,15,12,13,12,13,12,13,12,13]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm12 = ymm0[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm13 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm7 = ymm9[0],ymm7[1,2,3,4,5,6,7],ymm9[8],ymm7[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm9[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm9 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm9, %xmm12
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm9 = xmm9[0],xmm12[1],xmm9[2,3,4,5],xmm12[6],xmm9[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm9 = xmm9[4,5,2,3,0,1,14,15,12,13,12,13,12,13,12,13]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm12 = ymm2[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm13 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm12 = ymm13[0,1,2,3],ymm12[4],ymm13[5,6,7,8,9,10,11],ymm12[12],ymm13[13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm12 = ymm12[4,5,4,5,4,5,4,5,8,9,10,11,8,9,6,7,20,21,20,21,20,21,20,21,24,25,26,27,24,25,22,23]
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm8, %ymm12, %ymm8
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm12 = ymm4[0,1,2],ymm6[3],ymm4[4,5],ymm6[6],ymm4[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm9, %ymm12, %ymm9
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm12 = ymm5[0,1,2],ymm6[3],ymm5[4,5],ymm6[6],ymm5[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm12, %xmm13
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm12 = xmm13[0],xmm12[1],xmm13[2,3,4,5],xmm12[6],xmm13[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm12 = xmm12[0,1,2,3,0,1,14,15,12,13,10,11,u,u,u,u]
@@ -2572,98 +2574,99 @@ define void @load_i16_stride7_vf16(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm13 = ymm10[0,1,1,2]
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm13 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,30,31]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm12 = ymm12[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm12 = ymm8[0],ymm12[1,2,3,4,5,6,7],ymm8[8],ymm12[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3],ymm12[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm12 = ymm6[0],ymm4[1],ymm6[2,3],ymm4[4],ymm6[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm12 = ymm9[0],ymm12[1,2,3,4,5,6,7],ymm9[8],ymm12[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3],ymm12[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm12 = ymm6[0],ymm5[1],ymm6[2,3],ymm5[4],ymm6[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm12, %xmm13
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm12 = xmm12[0],xmm13[1],xmm12[2,3,4,5],xmm13[6],xmm12[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm12 = xmm12[0,1,4,5,2,3,0,1,14,15,12,13,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm12, %ymm0, %ymm12
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm13 = ymm10[0,1,1,3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm10 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,24,25]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm10 = ymm12[0,1,2,3,4,5,6],ymm10[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm12 = ymm3[0],ymm2[1],ymm3[2,3,4],ymm2[5],ymm3[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm12, %xmm14
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm12 = xmm14[0],xmm12[1],xmm14[2],xmm12[3],xmm14[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm12 = xmm12[6,7,4,5,2,3,0,1,14,15,14,15,14,15,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm14 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm12, %ymm0, %ymm13
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm12 = ymm10[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm10 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,24,25]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm10 = ymm13[0,1,2,3,4,5,6],ymm10[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm13 = ymm1[0],ymm0[1],ymm1[2,3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm13, %xmm14
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm13 = xmm14[0],xmm13[1],xmm14[2],xmm13[3],xmm14[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm13 = xmm13[6,7,4,5,2,3,0,1,14,15,14,15,14,15,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm14 = ymm3[0,1,2],ymm2[3],ymm3[4,5],ymm2[6],ymm3[7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm14 = ymm14[1,3,2,3]
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm14 = ymm14[6,7,6,7,6,7,6,7,8,9,4,5,10,11,0,1,22,23,22,23,22,23,22,23,24,25,20,21,26,27,16,17]
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm12, %ymm14, %ymm11
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm13, %ymm14, %ymm11
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm10 = ymm11[0],ymm10[1,2,3,4,5,6,7],ymm11[8],ymm10[9,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm10 = ymm11[0,1,2,3],ymm10[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm11 = ymm6[0],ymm4[1],ymm6[2,3,4],ymm4[5],ymm6[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm11, %xmm12
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm11 = xmm12[0],xmm11[1],xmm12[2],xmm11[3],xmm12[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm11 = xmm11[0,1,6,7,4,5,2,3,0,1,14,15,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm11, %ymm0, %ymm11
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm9, %ymm13, %ymm9
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm9 = ymm11[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm11 = ymm0[0,1],ymm1[2],ymm0[3,4,5],ymm1[6],ymm0[7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm12 = ymm11[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm11 = ymm11[u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5],ymm12[6],ymm11[7,8,9,10,11,12,13],ymm12[14],ymm11[15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm12 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm13 = [8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm12, %xmm14
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm12, %xmm12
-; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm12 = xmm12[3,1,2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm12 = xmm14[0],xmm12[0],xmm14[1],xmm12[1],xmm14[2],xmm12[2],xmm14[3],xmm12[3]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm11 = ymm12[0,1],ymm11[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm9 = ymm11[0],ymm9[1,2,3,4,5,6,7],ymm11[8],ymm9[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm9 = ymm11[0,1,2,3],ymm9[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm11 = ymm6[0,1],ymm4[2],ymm6[3,4],ymm4[5],ymm6[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm11, %xmm12
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm11 = xmm12[0,1],xmm11[2],xmm12[3],xmm11[4],xmm12[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm11 = xmm11[u,u,8,9,6,7,4,5,2,3,u,u,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm11, %ymm0, %ymm11
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 208(%rdi), %xmm12
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %xmm14
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm15 = xmm14[0,1,2,3,4,5],xmm12[6],xmm14[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm15 = xmm15[u,u,u,u,u,u,u,u,u,u,0,1,14,15,12,13]
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm15, %ymm0, %ymm15
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm11 = ymm11[0,1,2,3,4],ymm15[5,6,7],ymm11[8,9,10,11,12],ymm15[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm15 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm15, %xmm5
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm5, %xmm5
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm13 = xmm15[10,11,6,7,4,5,6,7,u,u,u,u,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm5 = xmm13[0],xmm5[0],xmm13[1],xmm5[1],xmm13[2],xmm5[2],xmm13[3],xmm5[3]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm13 = ymm0[0,1,2],ymm1[3],ymm0[4,5],ymm1[6],ymm0[7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm15 = ymm13[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm13 = ymm15[0],ymm13[1,2,3,4,5,6],ymm15[7,8],ymm13[9,10,11,12,13,14],ymm15[15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm13 = ymm13[u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm5[0,1],ymm13[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm11 = ymm5[0],ymm11[1,2,3,4,5,6,7],ymm5[8],ymm11[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm11[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm4[2,3],ymm6[4,5],ymm4[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm6[0,1,2],xmm4[3],xmm6[4],xmm4[5],xmm6[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm6 = xmm12[0],xmm14[1],xmm12[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm4 = xmm4[u,u,10,11,8,9,6,7,4,5,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm13 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} xmm11 = [8,9,4,5,8,9,4,5,8,9,4,5,8,9,4,5]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm11, %xmm13, %xmm14
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm13, %xmm13
+; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm13 = xmm13[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm13 = xmm14[0],xmm13[0],xmm14[1],xmm13[1],xmm14[2],xmm13[2],xmm14[3],xmm13[3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm14 = ymm2[0,1],ymm3[2],ymm2[3,4,5],ymm3[6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm15 = ymm14[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm14 = ymm14[u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm15 = ymm14[0,1,2,3,4,5],ymm15[6],ymm14[7,8,9,10,11,12,13],ymm15[14],ymm14[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm13 = xmm13[0,1],xmm15[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm15 = ymm6[0],ymm5[1],ymm6[2,3,4],ymm5[5],ymm6[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm15, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0],xmm15[1],xmm4[2],xmm15[3],xmm4[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm4 = xmm4[0,1,6,7,4,5,2,3,0,1,14,15,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm8, %ymm12, %ymm8
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm14[0],ymm4[1,2,3,4,5,6,7],ymm14[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm8 = ymm13[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm5[2],ymm6[3,4],ymm5[5],ymm6[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm12
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm12[0,1],xmm4[2],xmm12[3],xmm4[4],xmm12[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm4 = xmm4[u,u,8,9,6,7,4,5,2,3,u,u,u,u,u,u]
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 208(%rdi), %xmm12
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %xmm13
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm14 = xmm13[0,1,2,3,4,5],xmm12[6],xmm13[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm14 = xmm14[u,u,u,u,u,u,u,u,u,u,0,1,14,15,12,13]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm14, %ymm0, %ymm14
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm14[5,6,7],ymm4[8,9,10,11,12],ymm14[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm14 = ymm2[0,1,2],ymm3[3],ymm2[4,5],ymm3[6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm15 = ymm14[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm14 = ymm15[0],ymm14[1,2,3,4,5,6],ymm15[7,8],ymm14[9,10,11,12,13,14],ymm15[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm15 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm15, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm11, %xmm7, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm11 = xmm15[10,11,6,7,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm7 = xmm11[0],xmm7[0],xmm11[1],xmm7[1],xmm11[2],xmm7[2],xmm11[3],xmm7[3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm11 = ymm14[u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm11[0],ymm4[1,2,3,4,5,6,7],ymm11[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm7 = xmm7[0,1],xmm11[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm7[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm6
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm5 = xmm6[0,1,2],xmm5[3],xmm6[4],xmm5[5],xmm6[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm6 = xmm12[0],xmm13[1],xmm12[2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm5 = xmm5[u,u,10,11,8,9,6,7,4,5,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm6 = xmm6[u,u,u,u,u,u,u,u,u,u,2,3,0,1,14,15]
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm6[5,6,7],ymm4[8,9,10,11,12],ymm6[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5],ymm2[6],ymm3[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm3 = xmm3[u,u,u,u,u,u,u,u,10,11,6,7,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,6,4,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm2 = xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3,4,5,6,7,8],ymm1[9],ymm0[10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm4[1,2,3,4,5,6,7],ymm0[8],ymm4[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm2[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0],ymm3[1],ymm2[2,3,4,5,6,7,8],ymm3[9],ymm2[10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[u,u,u,u,u,u,u,u,10,11,6,7,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,4,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0],ymm5[1,2,3,4,5,6,7],ymm1[8],ymm5[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vmovaps %ymm1, (%rsi)
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm7, (%rdx)
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm8, (%rcx)
+; AVX2-FAST-PERLANE-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovaps %ymm1, (%rdx)
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm9, (%rcx)
; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm10, (%r8)
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm9, (%r9)
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm8, (%r9)
; AVX2-FAST-PERLANE-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm5, (%rax)
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm4, (%rax)
; AVX2-FAST-PERLANE-NEXT: movq {{[0-9]+}}(%rsp), %rax
; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm0, (%rax)
; AVX2-FAST-PERLANE-NEXT: vzeroupper
@@ -4657,69 +4660,67 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
;
; AVX2-SLOW-LABEL: load_i16_stride7_vf32:
; AVX2-SLOW: # %bb.0:
-; AVX2-SLOW-NEXT: subq $520, %rsp # imm = 0x208
-; AVX2-SLOW-NEXT: vmovdqa 288(%rdi), %ymm13
-; AVX2-SLOW-NEXT: vmovdqa 320(%rdi), %ymm8
-; AVX2-SLOW-NEXT: vmovdqa 256(%rdi), %ymm10
-; AVX2-SLOW-NEXT: vmovdqa 224(%rdi), %ymm11
-; AVX2-SLOW-NEXT: vmovdqa (%rdi), %ymm3
+; AVX2-SLOW-NEXT: subq $488, %rsp # imm = 0x1E8
+; AVX2-SLOW-NEXT: vmovdqa 288(%rdi), %ymm11
+; AVX2-SLOW-NEXT: vmovdqa 320(%rdi), %ymm10
+; AVX2-SLOW-NEXT: vmovdqa 256(%rdi), %ymm9
+; AVX2-SLOW-NEXT: vmovdqa 224(%rdi), %ymm13
+; AVX2-SLOW-NEXT: vmovdqa (%rdi), %ymm12
; AVX2-SLOW-NEXT: vmovdqa 32(%rdi), %ymm4
-; AVX2-SLOW-NEXT: vmovdqa 64(%rdi), %ymm9
+; AVX2-SLOW-NEXT: vmovdqa 64(%rdi), %ymm8
; AVX2-SLOW-NEXT: vmovdqa 96(%rdi), %ymm1
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm9[1],ymm1[2,3,4],ymm9[5],ymm1[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm8[1],ymm1[2,3,4],ymm8[5],ymm1[6,7]
; AVX2-SLOW-NEXT: vmovdqa %ymm1, %ymm7
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,2,2,3]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm0 = [0,1,14,15,12,13,14,15,2,3,6,7,12,13,2,3,16,17,30,31,28,29,30,31,18,19,22,23,28,29,18,19]
; AVX2-SLOW-NEXT: vpshufb %ymm0, %ymm1, %ymm2
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1],ymm4[2],ymm3[3,4,5],ymm4[6],ymm3[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm12[0,1],ymm4[2],ymm12[3,4,5],ymm4[6],ymm12[7]
; AVX2-SLOW-NEXT: vmovdqa %ymm4, %ymm5
-; AVX2-SLOW-NEXT: vmovdqa %ymm3, %ymm6
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm3
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4],xmm1[5],xmm3[6],xmm1[7]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,14,15,12,13,10,11,8,9,8,9,8,9,8,9]
; AVX2-SLOW-NEXT: vpshufb %xmm3, %xmm1, %xmm4
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-SLOW-NEXT: vpblendvb %ymm1, %ymm4, %ymm2, %ymm14
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm11[0,1],ymm10[2],ymm11[3,4,5],ymm10[6],ymm11[7]
+; AVX2-SLOW-NEXT: vpblendvb %ymm1, %ymm4, %ymm2, %ymm15
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm13[0,1],ymm9[2],ymm13[3,4,5],ymm9[6],ymm13[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm4
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm4[4],xmm2[5],xmm4[6],xmm2[7]
; AVX2-SLOW-NEXT: vpshufb %xmm3, %xmm2, %xmm2
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0],ymm13[1],ymm8[2,3,4],ymm13[5],ymm8[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm10[0],ymm11[1],ymm10[2,3,4],ymm11[5],ymm10[6,7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
; AVX2-SLOW-NEXT: vpshufb %ymm0, %ymm3, %ymm0
; AVX2-SLOW-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1],ymm13[2],ymm8[3,4],ymm13[5],ymm8[6,7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm8, %ymm12
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm11[2],ymm10[3,4],ymm11[5],ymm10[6,7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm0[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3],ymm0[4,5,6,7,8,9,10],ymm2[11],ymm0[12,13,14,15]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm2 = [2,3,2,3,2,3,2,3,8,9,8,9,6,7,4,5,18,19,18,19,18,19,18,19,24,25,24,25,22,23,20,21]
; AVX2-SLOW-NEXT: vpshufb %ymm2, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm11[0,1,2],ymm10[3],ymm11[4,5],ymm10[6],ymm11[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm13[0,1,2],ymm9[3],ymm13[4,5],ymm9[6],ymm13[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2,3,4,5],xmm3[6],xmm4[7]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm4 = [2,3,0,1,14,15,12,13,10,11,10,11,10,11,10,11]
; AVX2-SLOW-NEXT: vpshufb %xmm4, %xmm3, %xmm3
; AVX2-SLOW-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm9[2],ymm7[3,4],ymm9[5],ymm7[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm8[2],ymm7[3,4],ymm8[5],ymm7[6,7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm0[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3],ymm0[4,5,6,7,8,9,10],ymm3[11],ymm0[12,13,14,15]
; AVX2-SLOW-NEXT: vpshufb %ymm2, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm6[0,1,2],ymm5[3],ymm6[4,5],ymm5[6],ymm6[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm12[0,1,2],ymm5[3],ymm12[4,5],ymm5[6],ymm12[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0],xmm2[1],xmm3[2,3,4,5],xmm2[6],xmm3[7]
; AVX2-SLOW-NEXT: vpshufb %xmm4, %xmm2, %xmm2
; AVX2-SLOW-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm0 = ymm9[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm7[0,1],ymm9[2,3],ymm7[4,5],ymm9[6,7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm9, %ymm8
-; AVX2-SLOW-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm0 = ymm8[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm7[0,1],ymm8[2,3],ymm7[4,5],ymm8[6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa %ymm7, %ymm14
; AVX2-SLOW-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4],ymm2[5,6,7,8,9,10,11],ymm0[12],ymm2[13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0],ymm6[1],ymm5[2,3],ymm6[4],ymm5[5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0],ymm12[1],ymm5[2,3],ymm12[4],ymm5[5,6,7]
+; AVX2-SLOW-NEXT: vmovdqa %ymm5, %ymm7
; AVX2-SLOW-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3,4,5],xmm3[6],xmm2[7]
@@ -4729,39 +4730,42 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpshufb %xmm4, %xmm2, %xmm2
; AVX2-SLOW-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0],ymm11[1],ymm10[2,3],ymm11[4],ymm10[5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm10, (%rsp) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa %ymm9, %ymm5
+; AVX2-SLOW-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0],ymm13[1],ymm9[2,3],ymm13[4],ymm9[5,6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2,3,4,5],xmm2[6],xmm0[7]
; AVX2-SLOW-NEXT: vpshufb %xmm4, %xmm0, %xmm0
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm13[2,3],ymm12[4,5],ymm13[6,7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm13[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm10[0,1],ymm11[2,3],ymm10[4,5],ymm11[6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm11[2,3,0,1]
+; AVX2-SLOW-NEXT: vmovdqu %ymm11, (%rsp) # 32-byte Spill
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4],ymm2[5,6,7,8,9,10,11],ymm4[12],ymm2[13,14,15]
; AVX2-SLOW-NEXT: vpshufb %ymm3, %ymm2, %ymm2
; AVX2-SLOW-NEXT: vpblendvb %ymm1, %ymm0, %ymm2, %ymm9
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0],ymm6[1],ymm5[2,3,4],ymm6[5],ymm5[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0],ymm12[1],ymm7[2,3,4],ymm12[5],ymm7[6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2],xmm0[3],xmm2[4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm7[0,1,2],ymm8[3],ymm7[4,5],ymm8[6],ymm7[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm14[0,1,2],ymm8[3],ymm14[4,5],ymm8[6],ymm14[7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm2[1,3,2,3]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm3 = [6,7,6,7,6,7,6,7,8,9,4,5,10,11,0,1,22,23,22,23,22,23,22,23,24,25,20,21,26,27,16,17]
; AVX2-SLOW-NEXT: vpshufb %ymm3, %ymm2, %ymm2
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,7,7,7]
; AVX2-SLOW-NEXT: vpblendvb %ymm1, %ymm0, %ymm2, %ymm8
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0],ymm11[1],ymm10[2,3,4],ymm11[5],ymm10[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0],ymm13[1],ymm5[2,3,4],ymm13[5],ymm5[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2],xmm0[3],xmm2[4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm12[0,1,2],ymm13[3],ymm12[4,5],ymm13[6],ymm12[7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm13, %ymm6
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm10[0,1,2],ymm11[3],ymm10[4,5],ymm11[6],ymm10[7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm2[1,3,2,3]
; AVX2-SLOW-NEXT: vpshufb %ymm3, %ymm2, %ymm2
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[3,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,7,7,7]
; AVX2-SLOW-NEXT: vpblendvb %ymm1, %ymm0, %ymm2, %ymm7
-; AVX2-SLOW-NEXT: vmovdqa 128(%rdi), %ymm5
-; AVX2-SLOW-NEXT: vmovdqa 160(%rdi), %ymm15
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1,2],ymm5[3],ymm15[4,5],ymm5[6],ymm15[7]
+; AVX2-SLOW-NEXT: vmovdqa 128(%rdi), %ymm13
+; AVX2-SLOW-NEXT: vmovdqa 160(%rdi), %ymm14
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0,1,2],ymm13[3],ymm14[4,5],ymm13[6],ymm14[7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,6,4,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
@@ -4773,11 +4777,11 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm3 = ymm2[0,1,2,1,4,5,6,5]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1,2,3,4],ymm0[5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa 352(%rdi), %ymm14
-; AVX2-SLOW-NEXT: vmovdqa 384(%rdi), %ymm13
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1,2],ymm14[3],ymm13[4,5],ymm14[6],ymm13[7]
+; AVX2-SLOW-NEXT: vmovdqa 352(%rdi), %ymm6
+; AVX2-SLOW-NEXT: vmovdqa 384(%rdi), %ymm5
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm6[3],ymm5[4,5],ymm6[6],ymm5[7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,6,4,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
@@ -4786,17 +4790,17 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm1
; AVX2-SLOW-NEXT: vmovdqa 416(%rdi), %ymm3
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm0 = ymm3[0,1,0,2]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm10 = ymm0[0,1,2,1,4,5,6,5]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm10[7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm15 = ymm0[0,1,2,1,4,5,6,5]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm15 = ymm15[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm15[7]
; AVX2-SLOW-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
; AVX2-SLOW-NEXT: # ymm1 = mem[0,1,2,3,4],ymm1[5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm14[0,1],ymm13[2],ymm14[3,4,5],ymm13[6],ymm14[7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm10
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm10[4],xmm1[5],xmm10[6],xmm1[7]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm10 = [0,1,0,1,14,15,12,13,10,11,8,9,12,13,14,15]
-; AVX2-SLOW-NEXT: vpshufb %xmm10, %xmm1, %xmm1
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm6[0,1],ymm5[2],ymm6[3,4,5],ymm5[6],ymm6[7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm15
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm15[4],xmm1[5],xmm15[6],xmm1[7]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm15 = [0,1,0,1,14,15,12,13,10,11,8,9,12,13,14,15]
+; AVX2-SLOW-NEXT: vpshufb %xmm15, %xmm1, %xmm1
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,1,1,3,4,5,5,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,5,6,8,9,10,11,12,13,13,14]
@@ -4805,10 +4809,10 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1],ymm15[2],ymm5[3,4,5],ymm15[6],ymm5[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1],ymm14[2],ymm13[3,4,5],ymm14[6],ymm13[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4],xmm0[5],xmm1[6],xmm0[7]
-; AVX2-SLOW-NEXT: vpshufb %xmm10, %xmm0, %xmm0
+; AVX2-SLOW-NEXT: vpshufb %xmm15, %xmm0, %xmm0
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm1 = ymm2[0,1,1,3,4,5,5,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,5,6,8,9,10,11,12,13,13,14]
@@ -4817,7 +4821,7 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm15[3],ymm5[4,5],ymm15[6],ymm5[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1,2],ymm14[3],ymm13[4,5],ymm14[6],ymm13[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3,4,5],xmm0[6],xmm1[7]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm1 = [0,1,2,3,0,1,14,15,12,13,10,11,12,13,14,15]
@@ -4831,7 +4835,7 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0,1,2],ymm13[3],ymm14[4,5],ymm13[6],ymm14[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1,2],ymm5[3],ymm6[4,5],ymm5[6],ymm6[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5],xmm0[6],xmm2[7]
; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm0, %xmm0
@@ -4843,8 +4847,7 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm9[0],ymm0[1,2,3,4,5,6,7],ymm9[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0],ymm5[1],ymm15[2,3],ymm5[4],ymm15[5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0],ymm13[1],ymm14[2,3],ymm13[4],ymm14[5,6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5],xmm1[6],xmm0[7]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,1,0,4,5,6,7]
@@ -4857,7 +4860,7 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm8[0],ymm0[1,2,3,4,5,6,7],ymm8[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0],ymm14[1],ymm13[2,3],ymm14[4],ymm13[5,6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0],ymm6[1],ymm5[2,3],ymm6[4],ymm5[5,6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2,3,4,5],xmm2[6],xmm0[7]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,1,0,4,5,6,7]
@@ -4870,171 +4873,169 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm7[0],ymm0[1,2,3,4,5,6,7],ymm7[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa %ymm6, %ymm8
-; AVX2-SLOW-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1],ymm12[2],ymm6[3,4,5],ymm12[6],ymm6[7]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm3 = <u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27>
-; AVX2-SLOW-NEXT: vpshufb %ymm3, %ymm0, %ymm4
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm4[0,1,2,3,4,5],ymm0[6],ymm4[7,8,9,10,11,12,13],ymm0[14],ymm4[15]
-; AVX2-SLOW-NEXT: vmovdqu (%rsp), %ymm9 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm9[0,1],ymm11[2],ymm9[3,4],ymm11[5],ymm9[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm7
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm7 = xmm7[3,1,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm4[0],xmm7[0],xmm4[1],xmm7[1],xmm4[2],xmm7[2],xmm4[3],xmm7[3]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm13[0],ymm14[1],ymm13[2,3,4],ymm14[5],ymm13[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm7
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm7[0],xmm4[1],xmm7[2],xmm4[3],xmm7[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm7 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
-; AVX2-SLOW-NEXT: vpshufb %xmm7, %xmm4, %xmm4
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[0,1,2,1,4,5,6,5]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0],ymm2[1,2,3,4,5,6,7],ymm0[8],ymm2[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0],ymm5[1],ymm15[2,3,4],ymm5[5],ymm15[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm2
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2],xmm0[3],xmm2[4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm7, %xmm0, %xmm0
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0],ymm13[1],ymm14[2,3,4],ymm13[5],ymm14[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm3
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0],xmm0[1],xmm3[2],xmm0[3],xmm3[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
+; AVX2-SLOW-NEXT: vpshufb %xmm3, %xmm0, %xmm0
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,1,2,1,4,5,6,5]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm1 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm1 = mem[0,1],ymm10[2],mem[3,4,5],ymm10[6],mem[7]
-; AVX2-SLOW-NEXT: vpshufb %ymm3, %ymm1, %ymm2
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm1[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6],ymm2[7,8,9,10,11,12,13],ymm1[14],ymm2[15]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm5[2],ymm6[3,4],ymm5[5],ymm6[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm3
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm3 = xmm3[3,1,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm11[0,1],ymm12[2],ymm11[3,4],ymm12[5],ymm11[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm4
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm4 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm4 = mem[0,1],ymm9[2],mem[3,4,5],ymm9[6],mem[7]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm7 = <u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27>
+; AVX2-SLOW-NEXT: vpshufb %ymm7, %ymm4, %ymm8
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm8[0],ymm0[1,2,3,4,5,6,7],ymm8[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm4[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm8[0,1,2,3,4,5],ymm4[6],ymm8[7,8,9,10,11,12,13],ymm4[14],ymm8[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm4[2,3]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1,2],ymm12[3],ymm8[4,5],ymm12[6],ymm8[7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7,8],ymm0[9,10,11,12,13,14],ymm1[15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm9[0,1],ymm11[2,3],ymm9[4,5],ymm11[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[1,3,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm3 = <u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29>
-; AVX2-SLOW-NEXT: vpshufb %ymm3, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm13[0,1],ymm14[2],ymm13[3,4],ymm14[5],ymm13[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2],xmm2[3],xmm1[4],xmm2[5,6,7]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm4 = [8,9,8,9,6,7,4,5,2,3,2,3,2,3,2,3]
-; AVX2-SLOW-NEXT: vpshufb %xmm4, %xmm1, %xmm1
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm7
-; AVX2-SLOW-NEXT: vmovdqa 432(%rdi), %xmm2
-; AVX2-SLOW-NEXT: vmovdqa 416(%rdi), %xmm1
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm8 = xmm1[0,1,2,3,4,5],xmm2[6],xmm1[7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm8 = xmm8[0,1,0,3]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,4,4,7,6]
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm8, %ymm0, %ymm8
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm8[5,6,7],ymm7[8,9,10,11,12],ymm8[13,14,15]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm7 = ymm0[0],ymm7[1,2,3,4,5,6,7],ymm0[8],ymm7[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm15[2],ymm12[3,4],ymm15[5],ymm12[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-SLOW-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqu (%rsp), %ymm8 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm8[0,1],ymm10[2],ymm8[3,4,5],ymm10[6],ymm8[7]
+; AVX2-SLOW-NEXT: vpshufb %ymm7, %ymm1, %ymm4
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm1[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm4[0,1,2,3,4,5],ymm1[6],ymm4[7,8,9,10,11,12,13],ymm1[14],ymm4[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm5[0],ymm6[1],ymm5[2,3,4],ymm6[5],ymm5[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm7
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm7[0],xmm1[1],xmm7[2],xmm1[3],xmm7[4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm3, %xmm1, %xmm1
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[0,1,2,1,4,5,6,5]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm4[0],ymm1[1,2,3,4,5,6,7],ymm4[8],ymm1[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm9[2],ymm15[3,4],ymm9[5],ymm15[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm7
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm7[0,1],xmm0[2],xmm7[3],xmm0[4],xmm7[5,6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm4, %xmm0, %xmm0
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1,2],ymm10[3],ymm8[4,5],ymm10[6],ymm8[7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7,8],ymm0[9,10,11,12,13,14],ymm1[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1],ymm6[2],ymm5[3,4],ymm6[5],ymm5[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2],xmm1[3],xmm0[4],xmm1[5,6,7]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm3 = [8,9,8,9,6,7,4,5,2,3,2,3,2,3,2,3]
+; AVX2-SLOW-NEXT: vpshufb %xmm3, %xmm0, %xmm0
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm4
-; AVX2-SLOW-NEXT: vmovdqa 208(%rdi), %xmm12
-; AVX2-SLOW-NEXT: vmovdqa 192(%rdi), %xmm0
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm7 = xmm0[0,1,2,3,4,5],xmm12[6],xmm0[7]
+; AVX2-SLOW-NEXT: vmovdqa 432(%rdi), %xmm1
+; AVX2-SLOW-NEXT: vmovdqa 416(%rdi), %xmm0
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm7 = xmm0[0,1,2,3,4,5],xmm1[6],xmm0[7]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm7 = xmm7[0,1,0,3]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,4,4,7,6]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm7, %ymm0, %ymm7
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm7[5,6,7],ymm4[8,9,10,11,12],ymm7[13,14,15]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm7 = ymm11[0,1,2],ymm10[3],ymm11[4,5],ymm10[6],ymm11[7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm8 = ymm7[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm7 = ymm8[0],ymm7[1,2,3,4,5,6],ymm8[7,8],ymm7[9,10,11,12,13,14],ymm8[15]
-; AVX2-SLOW-NEXT: vpshufb %ymm3, %ymm7, %ymm3
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm7 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm7 = ymm12[0,1],ymm15[2,3],ymm12[4,5],ymm15[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm7, %xmm8
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm8 = xmm8[2,1,2,3]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm8 = xmm8[0,2,2,3,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm7 = xmm7[2,1,2,3]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm7 = xmm7[1,3,2,3,4,5,6,7]
; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm7 = xmm7[0],xmm8[0],xmm7[1],xmm8[1],xmm7[2],xmm8[2],xmm7[3],xmm8[3]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm7[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm3[0],ymm4[1,2,3,4,5,6,7],ymm3[8],ymm4[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm8 = ymm3[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm3 = ymm3[0],mem[1],ymm3[2,3],mem[4],ymm3[5,6,7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm3[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0],ymm4[1],ymm3[2,3,4,5,6,7,8],ymm4[9],ymm3[10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vmovdqu (%rsp), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm4 = ymm4[0,1,2],mem[3],ymm4[4,5],mem[6],ymm4[7]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm7 = xmm4[0,1,2,3,6,4,6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm4
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[0,1,2,1]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,5,7,6,7]
-; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm7[4],xmm4[4],xmm7[5],xmm4[5],xmm7[6],xmm4[6],xmm7[7],xmm4[7]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm7 = <u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31>
-; AVX2-SLOW-NEXT: vpshufb %ymm7, %ymm3, %ymm3
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm13[0,1],ymm14[2,3],ymm13[4,5],ymm14[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm5
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[3],xmm5[4],xmm4[5],xmm5[6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm2 = [8,9,10,11,8,9,6,7,4,5,4,5,4,5,4,5]
-; AVX2-SLOW-NEXT: vpshufb %xmm2, %xmm4, %xmm4
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,3]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,4,7]
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm4[0,1,2,3,4],ymm1[5,6,7],ymm4[8,9,10,11,12],ymm1[13,14,15]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm3[0],ymm1[1,2,3,4,5,6,7],ymm3[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm15[0,1],ymm9[2,3],ymm15[4,5],ymm9[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0,1,2],xmm3[3],xmm4[4],xmm3[5],xmm4[6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm12[0],xmm0[1],xmm12[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29>
+; AVX2-SLOW-NEXT: vpshufb %ymm8, %ymm2, %ymm2
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm2[0],ymm4[1,2,3,4,5,6,7],ymm2[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm2 = xmm7[0,1],xmm2[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm14[0,1],ymm13[2],ymm14[3,4],ymm13[5],ymm14[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm4
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm4[0,1],xmm2[2],xmm4[3],xmm2[4],xmm4[5,6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm3, %xmm2, %xmm2
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm3
+; AVX2-SLOW-NEXT: vmovdqa 208(%rdi), %xmm12
+; AVX2-SLOW-NEXT: vmovdqa 192(%rdi), %xmm2
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm7 = xmm2[0,1,2,3,4,5],xmm12[6],xmm2[7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm7 = xmm7[0,1,0,3]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,4,4,7,6]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm7, %ymm0, %ymm7
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm7[5,6,7],ymm3[8,9,10,11,12],ymm7[13,14,15]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm7 = ymm10[0,1,2],ymm9[3],ymm10[4,5],ymm9[6],ymm10[7]
+; AVX2-SLOW-NEXT: vmovdqa %ymm9, %ymm15
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm9 = ymm7[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm7 = ymm9[0],ymm7[1,2,3,4,5,6],ymm9[7,8],ymm7[9,10,11,12,13,14],ymm9[15]
+; AVX2-SLOW-NEXT: vpshufb %ymm8, %ymm7, %ymm7
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm8 = ymm11[0,1],ymm4[2,3],ymm11[4,5],ymm4[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm8, %xmm9
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm9 = xmm9[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm9 = xmm9[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm8 = xmm8[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm8 = xmm8[1,3,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm8 = xmm8[0],xmm9[0],xmm8[1],xmm9[1],xmm8[2],xmm9[2],xmm8[3],xmm9[3]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm7[0],ymm3[1,2,3,4,5,6,7],ymm7[8],ymm3[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm7 = xmm8[0,1],xmm7[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm9 = ymm7[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu (%rsp), %ymm3 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm3 = mem[0],ymm3[1],mem[2,3],ymm3[4],mem[5,6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm7 = ymm3[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0],ymm7[1],ymm3[2,3,4,5,6,7,8],ymm7[9],ymm3[10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm5 = ymm5[0,1],ymm6[2,3],ymm5[4,5],ymm6[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm6
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm5 = xmm6[0,1,2],xmm5[3],xmm6[4],xmm5[5],xmm6[6,7]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm1 = [8,9,10,11,8,9,6,7,4,5,4,5,4,5,4,5]
+; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm5, %xmm5
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,3]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,4,7]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0,1,2,3,4],ymm0[5,6,7],ymm2[8,9,10,11,12],ymm0[13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm10[0],ymm11[1],ymm10[2,3],ymm11[4],ymm10[5,6,7]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm5[0,1,2,3,4],ymm0[5,6,7],ymm5[8,9,10,11,12],ymm0[13,14,15]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm5 = ymm5[0,1,2],mem[3],ymm5[4,5],mem[6],ymm5[7]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm6 = xmm5[0,1,2,3,6,4,6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm5
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[0,1,2,1]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,7,6,7]
+; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm5 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31>
+; AVX2-SLOW-NEXT: vpshufb %ymm6, %ymm3, %ymm3
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3,4,5,6,7],ymm3[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm14[0,1],ymm13[2,3],ymm14[4,5],ymm13[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm5
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm5[0,1,2],xmm3[3],xmm5[4],xmm3[5],xmm5[6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm3, %xmm1
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm12[0],xmm2[1],xmm12[2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,3]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,7]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7],ymm1[8,9,10,11,12],ymm2[13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm15[0],ymm10[1],ymm15[2,3],ymm10[4],ymm15[5,6,7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm2[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0],ymm3[1],ymm2[2,3,4,5,6,7,8],ymm3[9],ymm2[10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpshufb %ymm7, %ymm2, %ymm2
-; AVX2-SLOW-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm3 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm3 = ymm6[0,1,2],mem[3],ymm6[4,5],mem[6],ymm6[7]
+; AVX2-SLOW-NEXT: vpshufb %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm11[0,1,2],ymm4[3],ymm11[4,5],ymm4[6],ymm11[7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm4 = xmm3[0,1,2,3,6,4,6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm3
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[0,1,2,1]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,7,6,7]
; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-SLOW-NEXT: vmovaps %ymm2, 32(%rsi)
; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
@@ -5058,79 +5059,78 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: movq {{[0-9]+}}(%rsp), %rax
; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-SLOW-NEXT: vmovaps %ymm2, 32(%rax)
-; AVX2-SLOW-NEXT: vmovdqa %ymm8, (%rax)
+; AVX2-SLOW-NEXT: vmovdqa %ymm9, (%rax)
; AVX2-SLOW-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; AVX2-SLOW-NEXT: vmovdqa %ymm1, 32(%rax)
-; AVX2-SLOW-NEXT: vmovdqa %ymm0, (%rax)
-; AVX2-SLOW-NEXT: addq $520, %rsp # imm = 0x208
+; AVX2-SLOW-NEXT: vmovdqa %ymm0, 32(%rax)
+; AVX2-SLOW-NEXT: vmovdqa %ymm1, (%rax)
+; AVX2-SLOW-NEXT: addq $488, %rsp # imm = 0x1E8
; AVX2-SLOW-NEXT: vzeroupper
; AVX2-SLOW-NEXT: retq
;
; AVX2-FAST-LABEL: load_i16_stride7_vf32:
; AVX2-FAST: # %bb.0:
; AVX2-FAST-NEXT: subq $584, %rsp # imm = 0x248
-; AVX2-FAST-NEXT: vmovdqa 288(%rdi), %ymm5
-; AVX2-FAST-NEXT: vmovdqa 320(%rdi), %ymm6
-; AVX2-FAST-NEXT: vmovdqa 256(%rdi), %ymm7
-; AVX2-FAST-NEXT: vmovdqa 224(%rdi), %ymm8
+; AVX2-FAST-NEXT: vmovdqa 288(%rdi), %ymm7
+; AVX2-FAST-NEXT: vmovdqa 320(%rdi), %ymm8
+; AVX2-FAST-NEXT: vmovdqa 256(%rdi), %ymm5
+; AVX2-FAST-NEXT: vmovdqa 224(%rdi), %ymm6
; AVX2-FAST-NEXT: vmovdqa (%rdi), %ymm3
; AVX2-FAST-NEXT: vmovdqa 32(%rdi), %ymm4
; AVX2-FAST-NEXT: vmovdqa 64(%rdi), %ymm1
; AVX2-FAST-NEXT: vmovdqa 96(%rdi), %ymm2
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
-; AVX2-FAST-NEXT: vmovdqa %ymm2, %ymm10
-; AVX2-FAST-NEXT: vmovdqa %ymm1, %ymm12
+; AVX2-FAST-NEXT: vmovdqa %ymm2, %ymm13
+; AVX2-FAST-NEXT: vmovdqa %ymm1, %ymm14
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = [0,1,14,15,12,13,14,15,2,3,6,7,12,13,2,3,16,17,30,31,28,29,30,31,18,19,22,23,28,29,18,19]
; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm0, %ymm2
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm4[2],ymm3[3,4,5],ymm4[6],ymm3[7]
-; AVX2-FAST-NEXT: vmovdqa %ymm4, %ymm9
-; AVX2-FAST-NEXT: vmovdqa %ymm3, %ymm11
+; AVX2-FAST-NEXT: vmovdqa %ymm4, %ymm11
+; AVX2-FAST-NEXT: vmovdqa %ymm3, %ymm12
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm3
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4],xmm0[5],xmm3[6],xmm0[7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,14,15,12,13,10,11,8,9,8,9,8,9,8,9]
; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm0, %xmm4
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm0 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm4, %ymm2, %ymm2
-; AVX2-FAST-NEXT: vmovdqu %ymm2, (%rsp) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm8[0,1],ymm7[2],ymm8[3,4,5],ymm7[6],ymm8[7]
+; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm4, %ymm2, %ymm15
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm5[2],ymm6[3,4,5],ymm5[6],ymm6[7]
+; AVX2-FAST-NEXT: vmovdqa %ymm6, %ymm9
+; AVX2-FAST-NEXT: vmovdqa %ymm5, %ymm10
; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm4
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm4[4],xmm2[5],xmm4[6],xmm2[7]
; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm2, %xmm2
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0],ymm5[1],ymm6[2,3,4],ymm5[5],ymm6[6,7]
-; AVX2-FAST-NEXT: vmovdqa %ymm5, %ymm13
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0],ymm7[1],ymm8[2,3,4],ymm7[5],ymm8[6,7]
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm3, %ymm1
; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm2, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm10[0,1],ymm12[2],ymm10[3,4],ymm12[5],ymm10[6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm1, (%rsp) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm13[0,1],ymm14[2],ymm13[3,4],ymm14[5],ymm13[6,7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm2 = <2,5,1,u,4,u,u,u>
; AVX2-FAST-NEXT: vpermd %ymm1, %ymm2, %ymm1
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm3 = [2,3,2,3,2,3,2,3,8,9,0,1,6,7,8,9,18,19,18,19,18,19,18,19,24,25,16,17,22,23,24,25]
; AVX2-FAST-NEXT: vpshufb %ymm3, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0,1,2],ymm9[3],ymm11[4,5],ymm9[6],ymm11[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm12[0,1,2],ymm11[3],ymm12[4,5],ymm11[6],ymm12[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2,3,4,5],xmm4[6],xmm5[7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm5 = [2,3,0,1,14,15,12,13,10,11,10,11,10,11,10,11]
; AVX2-FAST-NEXT: vpshufb %xmm5, %xmm4, %xmm4
; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm4, %ymm1, %ymm1
; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm8[0,1,2],ymm7[3],ymm8[4,5],ymm7[6],ymm8[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm6[0,1,2],ymm10[3],ymm6[4,5],ymm10[6],ymm6[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm4
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm1 = xmm4[0],xmm1[1],xmm4[2,3,4,5],xmm1[6],xmm4[7]
; AVX2-FAST-NEXT: vpshufb %xmm5, %xmm1, %xmm1
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm13[2],ymm6[3,4],ymm13[5],ymm6[6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm8[0,1],ymm7[2],ymm8[3,4],ymm7[5],ymm8[6,7]
; AVX2-FAST-NEXT: vpermd %ymm4, %ymm2, %ymm2
; AVX2-FAST-NEXT: vpshufb %ymm3, %ymm2, %ymm2
; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm1
; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm10[0,1],ymm12[2,3],ymm10[4,5],ymm12[6,7]
-; AVX2-FAST-NEXT: vmovdqa %ymm12, %ymm15
-; AVX2-FAST-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa %ymm10, %ymm14
-; AVX2-FAST-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm13[0,1],ymm14[2,3],ymm13[4,5],ymm14[6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm9[0],ymm11[1],ymm9[2,3],ymm11[4],ymm9[5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm11[0],ymm12[1],ymm11[2,3],ymm12[4],ymm11[5,6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3,4,5],xmm3[6],xmm2[7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm3 = <2,6,1,u,5,u,u,u>
@@ -5141,101 +5141,98 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vpshufb %xmm5, %xmm2, %xmm2
; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm2, %ymm1, %ymm1
; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0],ymm8[1],ymm7[2,3],ymm8[4],ymm7[5,6,7]
-; AVX2-FAST-NEXT: vmovdqa %ymm8, %ymm12
-; AVX2-FAST-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm10[0],ymm9[1],ymm10[2,3],ymm9[4],ymm10[5,6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2,3,4,5],xmm2[6],xmm1[7]
; AVX2-FAST-NEXT: vpshufb %xmm5, %xmm1, %xmm1
-; AVX2-FAST-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm13[2,3],ymm6[4,5],ymm13[6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm8[0,1],ymm7[2,3],ymm8[4,5],ymm7[6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpermd %ymm2, %ymm3, %ymm2
; AVX2-FAST-NEXT: vpshufb %ymm4, %ymm2, %ymm2
-; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm8
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm9[0],ymm11[1],ymm9[2,3,4],ymm11[5],ymm9[6,7]
-; AVX2-FAST-NEXT: vmovdqa %ymm9, %ymm10
+; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm9
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm11[0],ymm12[1],ymm11[2,3,4],ymm12[5],ymm11[6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4,5,6,7]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm14[0,1,2],ymm15[3],ymm14[4,5],ymm15[6],ymm14[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm13[0,1,2],ymm14[3],ymm13[4,5],ymm14[6],ymm13[7]
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm2 = ymm2[1,3,2,3]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm3 = [6,7,6,7,6,7,6,7,8,9,4,5,10,11,0,1,22,23,22,23,22,23,22,23,24,25,20,21,26,27,16,17]
; AVX2-FAST-NEXT: vpshufb %ymm3, %ymm2, %ymm2
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm4 = [6,7,4,5,2,3,0,1,14,15,14,15,14,15,14,15]
; AVX2-FAST-NEXT: vpshufb %xmm4, %xmm1, %xmm1
-; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm9
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0],ymm12[1],ymm7[2,3,4],ymm12[5],ymm7[6,7]
+; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm10[0],ymm6[1],ymm10[2,3,4],ymm6[5],ymm10[6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4,5,6,7]
; AVX2-FAST-NEXT: vpshufb %xmm4, %xmm1, %xmm1
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm6[0,1,2],ymm13[3],ymm6[4,5],ymm13[6],ymm6[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm8[0,1,2],ymm7[3],ymm8[4,5],ymm7[6],ymm8[7]
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm2 = ymm2[1,3,2,3]
; AVX2-FAST-NEXT: vpshufb %ymm3, %ymm2, %ymm2
-; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm6
-; AVX2-FAST-NEXT: vmovdqa 128(%rdi), %ymm5
-; AVX2-FAST-NEXT: vmovdqa 160(%rdi), %ymm1
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm1[0,1,2],ymm5[3],ymm1[4,5],ymm5[6],ymm1[7]
-; AVX2-FAST-NEXT: vmovdqa %ymm1, %ymm3
-; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm15 = [3,6,2,5,3,6,2,5]
-; AVX2-FAST-NEXT: # ymm15 = mem[0,1,0,1]
-; AVX2-FAST-NEXT: vpermd %ymm2, %ymm15, %ymm2
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm4 = [0,1,2,3,0,1,6,7,8,9,14,15,12,13,14,15,16,17,18,19,16,17,22,23,24,25,30,31,28,29,30,31]
-; AVX2-FAST-NEXT: vpshufb %ymm4, %ymm2, %ymm14
-; AVX2-FAST-NEXT: vmovdqa 192(%rdi), %ymm7
-; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm7[0,1,0,2]
+; AVX2-FAST-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm7
+; AVX2-FAST-NEXT: vmovdqa 128(%rdi), %ymm10
+; AVX2-FAST-NEXT: vmovdqa 160(%rdi), %ymm12
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm12[0,1,2],ymm10[3],ymm12[4,5],ymm10[6],ymm12[7]
+; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm4 = [3,6,2,5,3,6,2,5]
+; AVX2-FAST-NEXT: # ymm4 = mem[0,1,0,1]
+; AVX2-FAST-NEXT: vpermd %ymm2, %ymm4, %ymm2
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm5 = [0,1,2,3,0,1,6,7,8,9,14,15,12,13,14,15,16,17,18,19,16,17,22,23,24,25,30,31,28,29,30,31]
+; AVX2-FAST-NEXT: vpshufb %ymm5, %ymm2, %ymm14
+; AVX2-FAST-NEXT: vmovdqa 192(%rdi), %ymm8
+; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm8[0,1,0,2]
; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm2 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm0, %ymm13
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm13 = ymm14[0,1,2,3,4,5,6],ymm13[7]
-; AVX2-FAST-NEXT: vpblendd $31, (%rsp), %ymm13, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm1 = mem[0,1,2,3,4],ymm13[5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm1, (%rsp) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm15[0,1,2,3,4],ymm13[5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vmovdqa 352(%rdi), %ymm13
; AVX2-FAST-NEXT: vmovdqa 384(%rdi), %ymm14
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm12 = ymm14[0,1,2],ymm13[3],ymm14[4,5],ymm13[6],ymm14[7]
-; AVX2-FAST-NEXT: vpermd %ymm12, %ymm15, %ymm12
-; AVX2-FAST-NEXT: vpshufb %ymm4, %ymm12, %ymm4
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm11 = ymm14[0,1,2],ymm13[3],ymm14[4,5],ymm13[6],ymm14[7]
+; AVX2-FAST-NEXT: vpermd %ymm11, %ymm4, %ymm4
+; AVX2-FAST-NEXT: vpshufb %ymm5, %ymm4, %ymm5
; AVX2-FAST-NEXT: vmovdqa 416(%rdi), %ymm1
-; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm12 = ymm1[0,1,0,2]
-; AVX2-FAST-NEXT: vmovdqa %ymm1, %ymm15
-; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm12, %ymm11
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm11[7]
-; AVX2-FAST-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm1 = mem[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm3[2],ymm5[3,4,5],ymm3[6],ymm5[7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm11
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm11[4],xmm4[5],xmm11[6],xmm4[7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm11 = [0,1,0,1,14,15,12,13,10,11,8,9,12,13,14,15]
-; AVX2-FAST-NEXT: vpshufb %xmm11, %xmm4, %xmm4
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm11 = ymm1[0,1,0,2]
+; AVX2-FAST-NEXT: vmovdqa %ymm1, %ymm3
+; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm11, %ymm15
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm15[7]
+; AVX2-FAST-NEXT: vpblendd $31, (%rsp), %ymm5, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm1 = mem[0,1,2,3,4],ymm5[5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm1, (%rsp) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm10[0,1],ymm12[2],ymm10[3,4,5],ymm12[6],ymm10[7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm5, %xmm15
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm15[4],xmm5[5],xmm15[6],xmm5[7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm15 = [0,1,0,1,14,15,12,13,10,11,8,9,12,13,14,15]
+; AVX2-FAST-NEXT: vpshufb %xmm15, %xmm5, %xmm5
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm2 = [22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29]
; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2,3,4,5,6],ymm0[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1],ymm14[2],ymm13[3,4,5],ymm14[6],ymm13[7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm4
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm4[4],xmm0[5],xmm4[6],xmm0[7]
-; AVX2-FAST-NEXT: vpshufb %xmm11, %xmm0, %xmm0
-; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm12, %ymm2
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm5
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm5[4],xmm0[5],xmm5[6],xmm0[7]
+; AVX2-FAST-NEXT: vpshufb %xmm15, %xmm0, %xmm0
+; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm11, %ymm2
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm3[3],ymm5[4,5],ymm3[6],ymm5[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm12[3],ymm10[4,5],ymm12[6],ymm10[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5],xmm0[6],xmm2[7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,2,3,0,1,14,15,12,13,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm0, %xmm0
; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm4 = [2,5,2,5,2,5,2,5]
-; AVX2-FAST-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpermd %ymm7, %ymm4, %ymm11
+; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm5 = [2,5,2,5,2,5,2,5]
+; AVX2-FAST-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpermd %ymm8, %ymm5, %ymm11
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} ymm11 = ymm11[0,1,2,3,4,5,4,7,8,9,10,11,12,13,12,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm11[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
@@ -5247,206 +5244,207 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm11[0],xmm0[1],xmm11[2,3,4,5],xmm0[6],xmm11[7]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm0, %xmm0
; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpermd %ymm15, %ymm4, %ymm2
+; AVX2-FAST-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpermd %ymm3, %ymm5, %ymm2
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,4,7,8,9,10,11,12,13,12,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm8[0],ymm0[1,2,3,4,5,6,7],ymm8[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm9[0],ymm0[1,2,3,4,5,6,7],ymm9[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0],ymm5[1],ymm3[2,3],ymm5[4],ymm3[5,6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0],ymm10[1],ymm12[2,3],ymm10[4],ymm12[5,6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2,3,4,5],xmm2[6],xmm0[7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,2,3,0,1,14,15,12,13,12,13,14,15]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm0, %xmm0
; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm4 = ymm7[0,1,1,3]
-; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm8 = [18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25]
-; AVX2-FAST-NEXT: vpshufb %ymm8, %ymm4, %ymm11
+; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm5 = ymm8[0,1,1,3]
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm9 = [18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25]
+; AVX2-FAST-NEXT: vpshufb %ymm9, %ymm5, %ymm11
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm11[7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm9[0],ymm0[1,2,3,4,5,6,7],ymm9[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm4[0],ymm0[1,2,3,4,5,6,7],ymm4[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0],ymm13[1],ymm14[2,3],ymm13[4],ymm14[5,6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm9
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm9[1],xmm0[2,3,4,5],xmm9[6],xmm0[7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm8
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm8[1],xmm0[2,3,4,5],xmm8[6],xmm0[7]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm0, %xmm0
; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm2 = ymm15[0,1,1,3]
-; AVX2-FAST-NEXT: vpshufb %ymm8, %ymm2, %ymm8
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm6[0],ymm0[1,2,3,4,5,6,7],ymm6[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm8 = ymm3[0,1,1,3]
+; AVX2-FAST-NEXT: vpshufb %ymm9, %ymm8, %ymm2
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm7[0],ymm0[1,2,3,4,5,6,7],ymm7[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0],ymm5[1],ymm3[2,3,4],ymm5[5],ymm3[6,7]
-; AVX2-FAST-NEXT: vmovdqa %ymm3, %ymm15
-; AVX2-FAST-NEXT: vmovdqa %ymm5, %ymm11
-; AVX2-FAST-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm6
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm6[0],xmm0[1],xmm6[2],xmm0[3],xmm6[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
-; AVX2-FAST-NEXT: vpshufb %xmm6, %xmm0, %xmm0
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm7 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
-; AVX2-FAST-NEXT: vpshufb %ymm7, %ymm4, %ymm4
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm0[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-FAST-NEXT: vmovdqa %ymm10, %ymm1
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1],ymm11[2],ymm6[3,4],ymm11[5],ymm6[6,7]
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} xmm3 = [8,9,4,5,8,9,4,5,8,9,4,5,8,9,4,5]
+; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm0, %xmm2
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
+; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm12[0],ymm10[1],ymm12[2,3,4],ymm10[5],ymm12[6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm8 = ymm10[0,1],ymm3[2],ymm10[3,4],ymm3[5],ymm10[6,7]
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm9 = xmm8[8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm8, %xmm8
-; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm8 = xmm8[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm8 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm9 = ymm10[0,1],ymm0[2],ymm10[3,4,5],ymm0[6],ymm10[7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm5 = <0,3,7,2,6,u,u,u>
-; AVX2-FAST-NEXT: vpermd %ymm9, %ymm5, %ymm12
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm12 = ymm12[u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm8 = ymm8[0,1],ymm12[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm8[0],ymm4[1,2,3,4,5,6,7],ymm8[8],ymm4[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm8[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm14[0],ymm13[1],ymm14[2,3,4],ymm13[5],ymm14[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm8
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm8[0],xmm4[1],xmm8[2],xmm4[3],xmm8[4,5,6,7]
-; AVX2-FAST-NEXT: vpshufb %xmm6, %xmm4, %xmm4
-; AVX2-FAST-NEXT: vpshufb %ymm7, %ymm2, %ymm2
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm7[2],ymm5[3,4],ymm7[5],ymm5[6,7]
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm6 = xmm4[8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm4
-; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm6[0],xmm4[0],xmm6[1],xmm4[1],xmm6[2],xmm4[2],xmm6[3],xmm4[3]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm12[0,1],ymm9[2],ymm12[3,4,5],ymm9[6],ymm12[7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm8 = <0,3,7,2,6,u,u,u>
-; AVX2-FAST-NEXT: vpermd %ymm6, %ymm8, %ymm6
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm6 = ymm6[u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1],ymm6[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0],ymm2[1,2,3,4,5,6,7],ymm4[8],ymm2[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm2 = [0,3,3,3,0,3,7,7]
-; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm4 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm8 = ymm15[0,1],ymm11[2],ymm15[3,4],ymm11[5],ymm15[6,7]
-; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm11 = [2,5,1,4,2,5,1,4]
-; AVX2-FAST-NEXT: # ymm11 = mem[0,1,0,1]
-; AVX2-FAST-NEXT: vpermd %ymm8, %ymm11, %ymm8
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm8 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25,30,31,u,u,u,u,u,u]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm8[0,1,2,3,4],ymm4[5,6,7],ymm8[8,9,10,11,12],ymm4[13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm8 = ymm1[0,1],ymm3[2,3],ymm1[4,5],ymm3[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm8, %xmm6
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm3 = [8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm6, %xmm6
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm1 = [10,11,6,7,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm8, %xmm8
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm8[0],xmm6[0],xmm8[1],xmm6[1],xmm8[2],xmm6[2],xmm8[3],xmm6[3]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm8 = ymm10[0,1,2],ymm0[3],ymm10[4,5],ymm0[6],ymm10[7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm10 = <0,4,7,3,6,u,u,u>
-; AVX2-FAST-NEXT: vpermd %ymm8, %ymm10, %ymm8
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm8 = ymm8[u,u,u,u,u,u,u,u,2,3,4,5,10,11,12,13,18,19,18,19,18,19,18,19,18,19,20,21,26,27,28,29]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm6[0,1],ymm8[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm6[0],ymm4[1,2,3,4,5,6,7],ymm6[8],ymm4[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm7
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm2 = xmm7[0],xmm2[1],xmm7[2],xmm2[3],xmm7[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm7 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
+; AVX2-FAST-NEXT: vpshufb %xmm7, %xmm2, %xmm2
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,26,27]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm2[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm2 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm2 = mem[0,1],ymm1[2],mem[3,4,5],ymm1[6],mem[7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm9 = <0,3,7,2,6,u,u,u>
+; AVX2-FAST-NEXT: vpermd %ymm2, %ymm9, %ymm15
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm15 = ymm15[u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm15[2,3]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm5 = ymm15[0],ymm5[1,2,3,4,5,6,7],ymm15[8],ymm5[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm5[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT: vpermd %ymm4, %ymm2, %ymm2
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm14[0,1],ymm13[2],ymm14[3,4],ymm13[5],ymm14[6,7]
-; AVX2-FAST-NEXT: vpermd %ymm6, %ymm11, %ymm6
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm6 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25,30,31,u,u,u,u,u,u]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm6[0,1,2,3,4],ymm2[5,6,7],ymm6[8,9,10,11,12],ymm2[13,14,15]
-; AVX2-FAST-NEXT: vmovdqa %ymm7, %ymm11
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm5[0,1],ymm7[2,3],ymm5[4,5],ymm7[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm6, %xmm7
-; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm7, %xmm0
-; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm6, %xmm6
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm6[0],xmm0[0],xmm6[1],xmm0[1],xmm6[2],xmm0[2],xmm6[3],xmm0[3]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm12[0,1,2],ymm9[3],ymm12[4,5],ymm9[6],ymm12[7]
-; AVX2-FAST-NEXT: vpermd %ymm6, %ymm10, %ymm6
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm6 = ymm6[u,u,u,u,u,u,u,u,2,3,4,5,10,11,12,13,18,19,18,19,18,19,18,19,18,19,20,21,26,27,28,29]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm6[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0],ymm2[1,2,3,4,5,6,7],ymm0[8],ymm2[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm0[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FAST-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm0 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
+; AVX2-FAST-NEXT: vpblendd $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm0 = mem[0,1],ymm4[2],mem[3,4],ymm4[5],mem[6,7]
+; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm0, %xmm5
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
+; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm2[0,1],ymm3[2],ymm2[3,4,5],ymm3[6],ymm2[7]
+; AVX2-FAST-NEXT: vpermd %ymm5, %ymm9, %ymm5
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm9 = ymm14[0],ymm13[1],ymm14[2,3,4],ymm13[5],ymm14[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm9, %xmm15
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm9 = xmm15[0],xmm9[1],xmm15[2],xmm9[3],xmm15[4,5,6,7]
+; AVX2-FAST-NEXT: vpshufb %xmm7, %xmm9, %xmm7
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm8 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,26,27]
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm7, %ymm0, %ymm7
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6],ymm8[7]
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm5 = ymm5[0],ymm7[1,2,3,4,5,6,7],ymm5[8],ymm7[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm0 = [0,3,3,3,0,3,7,7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-NEXT: vpermd %ymm15, %ymm0, %ymm5
+; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm7 = [0,0,16,17,22,23,24,25,0,0,16,17,22,23,24,25,0,0,16,17,22,23,24,25,0,0,16,17,22,23,24,25]
+; AVX2-FAST-NEXT: vpshufb %ymm7, %ymm5, %ymm5
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm8 = ymm12[0,1],ymm10[2],ymm12[3,4],ymm10[5],ymm12[6,7]
+; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm10 = [2,5,1,4,2,5,1,4]
+; AVX2-FAST-NEXT: # ymm10 = mem[0,1,0,1]
+; AVX2-FAST-NEXT: vpermd %ymm8, %ymm10, %ymm8
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm8 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25,30,31,u,u,u,u,u,u]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm5 = ymm8[0,1,2,3,4],ymm5[5,6,7],ymm8[8,9,10,11,12],ymm5[13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm8 = ymm6[0,1],ymm11[2,3],ymm6[4,5],ymm11[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm8, %xmm9
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} xmm6 = [8,9,4,5,8,9,4,5,8,9,4,5,8,9,4,5]
+; AVX2-FAST-NEXT: vpshufb %xmm6, %xmm9, %xmm9
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm8 = xmm8[10,11,6,7,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm8 = xmm8[0],xmm9[0],xmm8[1],xmm9[1],xmm8[2],xmm9[2],xmm8[3],xmm9[3]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm9 = ymm11[0,1,2],ymm1[3],ymm11[4,5],ymm1[6],ymm11[7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm0 = <0,4,7,3,6,u,u,u>
+; AVX2-FAST-NEXT: vpermd %ymm9, %ymm0, %ymm9
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,u,u,2,3,4,5,10,11,12,13,18,19,18,19,18,19,18,19,18,19,20,21,26,27,28,29>
+; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm9, %ymm9
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm5 = ymm9[0],ymm5[1,2,3,4,5,6,7],ymm9[8],ymm5[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm8 = xmm8[0,1],xmm9[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm8[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm5 = [0,3,3,3,0,3,7,7]
+; AVX2-FAST-NEXT: vpermd %ymm9, %ymm5, %ymm5
+; AVX2-FAST-NEXT: vpshufb %ymm7, %ymm5, %ymm5
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm7 = ymm14[0,1],ymm13[2],ymm14[3,4],ymm13[5],ymm14[6,7]
+; AVX2-FAST-NEXT: vpermd %ymm7, %ymm10, %ymm7
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm7 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25,30,31,u,u,u,u,u,u]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm5 = ymm7[0,1,2,3,4],ymm5[5,6,7],ymm7[8,9,10,11,12],ymm5[13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm7 = ymm2[0,1,2],ymm3[3],ymm2[4,5],ymm3[6],ymm2[7]
+; AVX2-FAST-NEXT: vpermd %ymm7, %ymm0, %ymm0
+; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm0, %ymm0
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm7
+; AVX2-FAST-NEXT: vpshufb %xmm6, %xmm7, %xmm7
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm2 = xmm2[10,11,6,7,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm7[0],xmm2[1],xmm7[1],xmm2[2],xmm7[2],xmm2[3],xmm7[3]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm5 = ymm0[0],ymm5[1,2,3,4,5,6,7],ymm0[8],ymm5[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm0[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FAST-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm0 = ymm12[0,1],mem[2,3],ymm12[4,5],mem[6,7]
; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm1 = [0,4,7,0,0,4,7,0]
; AVX2-FAST-NEXT: # ymm1 = mem[0,1,0,1]
-; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm5 = [0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27]
-; AVX2-FAST-NEXT: vpshufb %ymm5, %ymm3, %ymm3
+; AVX2-FAST-NEXT: vpermd %ymm15, %ymm1, %ymm2
+; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm4 = [0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27]
+; AVX2-FAST-NEXT: vpshufb %ymm4, %ymm2, %ymm2
; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm6 = [2,6,1,5,2,6,1,5]
; AVX2-FAST-NEXT: # ymm6 = mem[0,1,0,1]
; AVX2-FAST-NEXT: vpermd %ymm0, %ymm6, %ymm0
; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm7 = [28,29,18,19,20,21,26,27,28,29,18,19,20,21,26,27,28,29,18,19,20,21,26,27,28,29,18,19,20,21,26,27]
; AVX2-FAST-NEXT: vpshufb %ymm7, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm3[5,6,7],ymm0[8,9,10,11,12],ymm3[13,14,15]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm3 = ymm3[0],mem[1],ymm3[2,3],mem[4],ymm3[5,6,7]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7],ymm0[8,9,10,11,12],ymm2[13,14,15]
+; AVX2-FAST-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm2 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm2 = mem[0],ymm11[1],mem[2,3],ymm11[4],mem[5,6,7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm8 = mem[0,1,2],ymm8[3],mem[4,5],ymm8[6],mem[7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm8, %xmm9
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm10 = [0,1,2,3,4,5,6,7,10,11,6,7,4,5,6,7]
-; AVX2-FAST-NEXT: vpshufb %xmm10, %xmm9, %xmm9
+; AVX2-FAST-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm8 = ymm8[0,1,2],mem[3],ymm8[4,5],mem[6],ymm8[7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm8, %xmm10
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} xmm11 = [10,11,6,7,10,11,6,7,10,11,6,7,10,11,6,7]
+; AVX2-FAST-NEXT: vpshufb %xmm11, %xmm10, %xmm10
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,6,4,6,7]
-; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm8 = xmm8[4],xmm9[4],xmm8[5],xmm9[5],xmm8[6],xmm9[6],xmm8[7],xmm9[7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm9 = <1,4,0,3,7,u,u,u>
-; AVX2-FAST-NEXT: vpermd %ymm3, %ymm9, %ymm3
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm15 = <u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31>
-; AVX2-FAST-NEXT: vpshufb %ymm15, %ymm3, %ymm3
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3,4,5,6,7],ymm3[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FAST-NEXT: vpermd %ymm4, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vpshufb %ymm5, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm14[0,1],ymm13[2,3],ymm14[4,5],ymm13[6,7]
-; AVX2-FAST-NEXT: vpermd %ymm3, %ymm6, %ymm3
-; AVX2-FAST-NEXT: vpshufb %ymm7, %ymm3, %ymm3
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm3[0,1,2,3,4],ymm1[5,6,7],ymm3[8,9,10,11,12],ymm1[13,14,15]
-; AVX2-FAST-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm3 = mem[0,1,2],ymm11[3],mem[4,5],ymm11[6],mem[7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm3, %xmm5
-; AVX2-FAST-NEXT: vpshufb %xmm10, %xmm5, %xmm5
+; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm8 = xmm8[4],xmm10[4],xmm8[5],xmm10[5],xmm8[6],xmm10[6],xmm8[7],xmm10[7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm10 = <1,4,0,3,7,u,u,u>
+; AVX2-FAST-NEXT: vpermd %ymm2, %ymm10, %ymm2
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm12 = <u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31>
+; AVX2-FAST-NEXT: vpshufb %ymm12, %ymm2, %ymm2
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm2 = xmm8[0,1],xmm2[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-NEXT: vpermd %ymm9, %ymm1, %ymm1
+; AVX2-FAST-NEXT: vpshufb %ymm4, %ymm1, %ymm1
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm14[0,1],ymm13[2,3],ymm14[4,5],ymm13[6,7]
+; AVX2-FAST-NEXT: vpermd %ymm2, %ymm6, %ymm2
+; AVX2-FAST-NEXT: vpshufb %ymm7, %ymm2, %ymm2
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7],ymm2[8,9,10,11,12],ymm1[13,14,15]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm2 = ymm2[0],mem[1],ymm2[2,3],mem[4],ymm2[5,6,7]
+; AVX2-FAST-NEXT: vpermd %ymm2, %ymm10, %ymm2
+; AVX2-FAST-NEXT: vpshufb %ymm12, %ymm2, %ymm2
+; AVX2-FAST-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5],mem[6],ymm3[7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm3, %xmm4
+; AVX2-FAST-NEXT: vpshufb %xmm11, %xmm4, %xmm4
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,6,4,6,7]
-; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm3[4],xmm5[4],xmm3[5],xmm5[5],xmm3[6],xmm5[6],xmm3[7],xmm5[7]
-; AVX2-FAST-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm5 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm5 = mem[0],ymm12[1],mem[2,3],ymm12[4],mem[5,6,7]
-; AVX2-FAST-NEXT: vpermd %ymm5, %ymm9, %ymm5
-; AVX2-FAST-NEXT: vpshufb %ymm15, %ymm5, %ymm5
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm5[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm3[0],ymm1[1,2,3,4,5,6,7],ymm3[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm3, 32(%rsi)
-; AVX2-FAST-NEXT: vmovups (%rsp), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm3, (%rsi)
-; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm3, 32(%rdx)
-; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm3, (%rdx)
-; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm3, 32(%rcx)
-; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm3, (%rcx)
-; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm3, 32(%r8)
-; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm3, (%r8)
-; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm3, 32(%r9)
-; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm3, (%r9)
+; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-NEXT: vmovups (%rsp), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm2, 32(%rsi)
+; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm2, (%rsi)
+; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm2, 32(%rdx)
+; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm2, (%rdx)
+; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm2, 32(%rcx)
+; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm2, (%rcx)
+; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm2, 32(%r8)
+; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm2, (%r8)
+; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm2, 32(%r9)
+; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovaps %ymm2, (%r9)
; AVX2-FAST-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-NEXT: vmovdqa %ymm2, 32(%rax)
+; AVX2-FAST-NEXT: vmovdqa %ymm5, 32(%rax)
; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-FAST-NEXT: vmovaps %ymm2, (%rax)
; AVX2-FAST-NEXT: movq {{[0-9]+}}(%rsp), %rax
@@ -5458,376 +5456,368 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
;
; AVX2-FAST-PERLANE-LABEL: load_i16_stride7_vf32:
; AVX2-FAST-PERLANE: # %bb.0:
-; AVX2-FAST-PERLANE-NEXT: subq $520, %rsp # imm = 0x208
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 288(%rdi), %ymm10
+; AVX2-FAST-PERLANE-NEXT: subq $488, %rsp # imm = 0x1E8
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 288(%rdi), %ymm9
; AVX2-FAST-PERLANE-NEXT: vmovdqa 320(%rdi), %ymm11
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 256(%rdi), %ymm5
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 256(%rdi), %ymm10
; AVX2-FAST-PERLANE-NEXT: vmovdqa 224(%rdi), %ymm12
; AVX2-FAST-PERLANE-NEXT: vmovdqa (%rdi), %ymm3
; AVX2-FAST-PERLANE-NEXT: vmovdqa 32(%rdi), %ymm4
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 64(%rdi), %ymm6
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 64(%rdi), %ymm5
; AVX2-FAST-PERLANE-NEXT: vmovdqa 96(%rdi), %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm6[1],ymm1[2,3,4],ymm6[5],ymm1[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm1, %ymm9
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm2 = [0,1,14,15,12,13,14,15,2,3,6,7,12,13,2,3,16,17,30,31,28,29,30,31,18,19,22,23,28,29,18,19]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm5[1],ymm1[2,3,4],ymm5[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm1, %ymm8
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm0 = [0,1,14,15,12,13,14,15,2,3,6,7,12,13,2,3,16,17,30,31,28,29,30,31,18,19,22,23,28,29,18,19]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm1, %ymm2
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1],ymm4[2],ymm3[3,4,5],ymm4[6],ymm3[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm4, %ymm7
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm3, %ymm8
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm4, %ymm6
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm3, %ymm7
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm3
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4],xmm1[5],xmm3[6],xmm1[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,14,15,12,13,10,11,8,9,8,9,8,9,8,9]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm1, %xmm4
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm4, %ymm0, %ymm15
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm5, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm12[0,1],ymm5[2],ymm12[3,4,5],ymm5[6],ymm12[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm5
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm5[4],xmm4[5],xmm5[6],xmm4[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm4, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0],ymm10[1],ymm11[2,3,4],ymm10[5],ymm11[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm4, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm3, %ymm2, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm11, %ymm13
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm2[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3],ymm2[4,5,6,7,8,9,10],ymm3[11],ymm2[12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm3 = [2,3,2,3,2,3,2,3,8,9,8,9,6,7,4,5,18,19,18,19,18,19,18,19,24,25,24,25,22,23,20,21]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm3, %ymm2, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm12[0,1,2],ymm0[3],ymm12[4,5],ymm0[6],ymm12[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm0, %ymm14
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm5
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2,3,4,5],xmm4[6],xmm5[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm5 = [2,3,0,1,14,15,12,13,10,11,10,11,10,11,10,11]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm4, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm4, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm4, %ymm2, %ymm15
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm10[2],ymm12[3,4,5],ymm10[6],ymm12[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm4[4],xmm2[5],xmm4[6],xmm2[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm2, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm11[0],ymm9[1],ymm11[2,3,4],ymm9[5],ymm11[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm3, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm6[2],ymm9[3,4],ymm6[5],ymm9[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm2[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0,1,2],ymm4[3],ymm2[4,5,6,7,8,9,10],ymm4[11],ymm2[12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm3, %ymm2, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0,1,2],ymm7[3],ymm8[4,5],ymm7[6],ymm8[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm11[0,1],ymm9[2],ymm11[3,4],ymm9[5],ymm11[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm0[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3],ymm0[4,5,6,7,8,9,10],ymm2[11],ymm0[12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm2 = [2,3,2,3,2,3,2,3,8,9,8,9,6,7,4,5,18,19,18,19,18,19,18,19,24,25,24,25,22,23,20,21]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm12[0,1,2],ymm10[3],ymm12[4,5],ymm10[6],ymm12[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2,3,4,5],xmm3[6],xmm4[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm3, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm4 = [2,3,0,1,14,15,12,13,10,11,10,11,10,11,10,11]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm3, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm6[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm9[0,1],ymm6[2,3],ymm9[4,5],ymm6[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm9, %ymm11
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4],ymm3[5,6,7,8,9,10,11],ymm2[12],ymm3[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm7[0],ymm8[1],ymm7[2,3],ymm8[4],ymm7[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm8, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1],ymm5[2],ymm8[3,4],ymm5[5],ymm8[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm0[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3],ymm0[4,5,6,7,8,9,10],ymm3[11],ymm0[12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm7[0,1,2],ymm6[3],ymm7[4,5],ymm6[6],ymm7[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0],xmm2[1],xmm3[2,3,4,5],xmm2[6],xmm3[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm2, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm0 = ymm5[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm8[0,1],ymm5[2,3],ymm8[4,5],ymm5[6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2,3,4,5],xmm4[6],xmm3[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm5 = [4,5,4,5,4,5,4,5,8,9,10,11,8,9,6,7,20,21,20,21,20,21,20,21,24,25,26,27,24,25,22,23]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm5, %ymm2, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm8 = [4,5,2,3,0,1,14,15,12,13,12,13,12,13,12,13]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm3, %ymm2, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm14[0],ymm12[1],ymm14[2,3],ymm12[4],ymm14[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4],ymm2[5,6,7,8,9,10,11],ymm0[12],ymm2[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm6[0],ymm7[1],ymm6[2,3],ymm7[4],ymm6[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3,4,5],xmm3[6],xmm2[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm2, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm13, (%rsp) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm13[0,1],ymm10[2,3],ymm13[4,5],ymm10[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm8 = ymm10[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm3 = [4,5,4,5,4,5,4,5,8,9,10,11,8,9,6,7,20,21,20,21,20,21,20,21,24,25,26,27,24,25,22,23]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm3, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm4 = [4,5,2,3,0,1,14,15,12,13,12,13,12,13,12,13]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm2, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm8[4],ymm3[5,6,7,8,9,10,11],ymm8[12],ymm3[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm5, %ymm3, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm2, %ymm3, %ymm9
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm7[0],ymm0[1],ymm7[2,3,4],ymm0[5],ymm7[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0],xmm2[1],xmm3[2],xmm2[3],xmm3[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm11[0,1,2],ymm6[3],ymm11[4,5],ymm6[6],ymm11[7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm3[1,3,2,3]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm5 = [6,7,6,7,6,7,6,7,8,9,4,5,10,11,0,1,22,23,22,23,22,23,22,23,24,25,20,21,26,27,16,17]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm5, %ymm3, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm8 = [6,7,4,5,2,3,0,1,14,15,14,15,14,15,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm2, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm2, %ymm3, %ymm11
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm14[0],ymm12[1],ymm14[2,3,4],ymm12[5],ymm14[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm12
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm12[0],xmm3[1],xmm12[2],xmm3[3],xmm12[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm8 = ymm13[0,1,2],ymm10[3],ymm13[4,5],ymm10[6],ymm13[7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm8 = ymm8[1,3,2,3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm5, %ymm8, %ymm5
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm3, %ymm5, %ymm10
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 128(%rdi), %ymm5
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 160(%rdi), %ymm6
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm6[0,1,2],ymm5[3],ymm6[4,5],ymm5[6],ymm6[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,2,3,4,5,6,7,8,9,10,11,6,7,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,4,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %ymm8
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm8[0,1,0,2]
-; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm7 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm7, %ymm3, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0],ymm12[1],ymm10[2,3],ymm12[4],ymm10[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2,3,4,5],xmm2[6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm11[0,1],ymm9[2,3],ymm11[4,5],ymm9[6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm11, %ymm13
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm11, (%rsp) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm9[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm9, %ymm14
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm4[4],ymm2[5,6,7,8,9,10,11],ymm4[12],ymm2[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm3, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm0, %ymm2, %ymm9
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0],ymm7[1],ymm6[2,3,4],ymm7[5],ymm6[6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm7, %ymm11
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2],xmm0[3],xmm2[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm8[0,1,2],ymm5[3],ymm8[4,5],ymm5[6],ymm8[7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm2[1,3,2,3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm4 = [6,7,6,7,6,7,6,7,8,9,4,5,10,11,0,1,22,23,22,23,22,23,22,23,24,25,20,21,26,27,16,17]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm3 = [6,7,4,5,2,3,0,1,14,15,14,15,14,15,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm0, %ymm2, %ymm8
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0],ymm12[1],ymm10[2,3,4],ymm12[5],ymm10[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2],xmm0[3],xmm2[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm13[0,1,2],ymm14[3],ymm13[4,5],ymm14[6],ymm13[7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm2[1,3,2,3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm1, %ymm0, %ymm2, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 128(%rdi), %ymm13
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 160(%rdi), %ymm14
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0,1,2],ymm13[3],ymm14[4,5],ymm13[6],ymm14[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm5 = [0,1,2,3,4,5,6,7,8,9,10,11,6,7,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,4,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %ymm4
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm4[0,1,0,2]
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm10 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm2, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1,2,3,4],ymm1[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 352(%rdi), %ymm13
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 384(%rdi), %ymm12
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0,1,2],ymm13[3],ymm12[4,5],ymm13[6],ymm12[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 352(%rdi), %ymm3
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 384(%rdi), %ymm1
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2],ymm3[3],ymm1[4,5],ymm3[6],ymm1[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm15
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm15, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm15, %xmm5
; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,4,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 416(%rdi), %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm15 = ymm4[0,1,0,2]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm7, %ymm15, %ymm14
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm14[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 416(%rdi), %ymm5
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm15 = ymm5[0,1,0,2]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm15, %ymm12
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm12[7]
; AVX2-FAST-PERLANE-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0,1,2,3,4],ymm0[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1],ymm12[2],ymm13[3,4,5],ymm12[6],ymm13[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm14
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm14[4],xmm0[5],xmm14[6],xmm0[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm14 = [0,1,0,1,14,15,12,13,10,11,8,9,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm14, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm1[2],ymm3[3,4,5],ymm1[6],ymm3[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm12
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm12[4],xmm0[5],xmm12[6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm12 = [0,1,0,1,14,15,12,13,10,11,8,9,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm12, %xmm0, %xmm0
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm1 = [22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm15, %ymm15
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm6 = [22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm6, %ymm15, %ymm15
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm15[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm15[0],ymm0[1,2,3,4,5,6,7],ymm15[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1],ymm6[2],ymm5[3,4,5],ymm6[6],ymm5[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1],ymm14[2],ymm13[3,4,5],ymm14[6],ymm13[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm15
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm15[4],xmm0[5],xmm15[6],xmm0[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm14, %xmm0, %xmm0
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm3, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm6[3],ymm5[4,5],ymm6[6],ymm5[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm5, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0],xmm0[1],xmm3[2,3,4,5],xmm0[6],xmm3[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,2,3,0,1,14,15,12,13,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm12, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm6, %ymm2, %ymm2
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm8[0,1,1,2]
-; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm14 = [16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm14, %ymm5, %ymm5
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1,2],ymm12[3],ymm13[4,5],ymm12[6],ymm13[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm5
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm5[0],xmm0[1],xmm5[2,3,4,5],xmm0[6],xmm5[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm0, %xmm0
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm4[0,1,1,2]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm14, %ymm3, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1,2],ymm14[3],ymm13[4,5],ymm14[6],ymm13[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5],xmm0[6],xmm2[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,2,3,0,1,14,15,12,13,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm6 = ymm4[0,1,1,2]
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm12 = [16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm12, %ymm6, %ymm6
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm6[0],ymm0[1,2,3,4,5,6,7],ymm6[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2],ymm1[3],ymm3[4,5],ymm1[6],ymm3[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm6
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm6[0],xmm0[1],xmm6[2,3,4,5],xmm0[6],xmm6[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm5[0,1,1,2]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm12, %ymm2, %ymm2
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm9[0],ymm0[1,2,3,4,5,6,7],ymm9[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0],ymm1[1],ymm6[2,3],ymm1[4],ymm6[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm6, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm3[1],xmm0[2,3,4,5],xmm3[6],xmm0[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,4,5,2,3,0,1,14,15,12,13,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0],ymm13[1],ymm14[2,3],ymm13[4],ymm14[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2,3,4,5],xmm2[6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,4,5,2,3,0,1,14,15,12,13,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm0, %xmm0
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm8[0,1,1,3]
-; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm8 = [18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm8, %ymm5, %ymm9
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm6 = [18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm6, %ymm4, %ymm9
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm9[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm11[0],ymm0[1,2,3,4,5,6,7],ymm11[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm11[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm8[0],ymm0[1,2,3,4,5,6,7],ymm8[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0],ymm13[1],ymm12[2,3],ymm13[4],ymm12[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm7
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm7[1],xmm0[2,3,4,5],xmm7[6],xmm0[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm3[1],ymm1[2,3],ymm3[4],ymm1[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm8
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm8[1],xmm0[2,3,4,5],xmm8[6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm0, %xmm0
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm4[0,1,1,3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm8, %ymm3, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm10[0],ymm0[1,2,3,4,5,6,7],ymm10[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm5[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm6, %ymm2, %ymm5
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm7[0],ymm0[1,2,3,4,5,6,7],ymm7[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu (%rsp), %ymm14 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm14[2],ymm15[3,4,5],ymm14[6],ymm15[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27>
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm0, %ymm6
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm6[0,1,2,3,4,5],ymm0[6],ymm6[7,8,9,10,11,12,13],ymm0[14],ymm6[15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm6 = ymm8[0,1],ymm11[2],ymm8[3,4],ymm11[5],ymm8[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm7 = xmm6[8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm6, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm6 = xmm6[3,1,2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm7[0],xmm6[0],xmm7[1],xmm6[1],xmm7[2],xmm6[2],xmm7[3],xmm6[3]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm6 = ymm12[0],ymm13[1],ymm12[2,3,4],ymm13[5],ymm12[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm6, %xmm7
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm6 = xmm7[0],xmm6[1],xmm7[2],xmm6[3],xmm7[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm7 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm7, %xmm6, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm6
-; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm10 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm3, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm0[0],ymm3[1,2,3,4,5,6,7],ymm0[8],ymm3[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1],ymm11[2],ymm9[3,4],ymm11[5],ymm9[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} xmm6 = [8,9,4,5,8,9,4,5,8,9,4,5,8,9,4,5]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm0, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm6, %xmm11
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm5 = mem[0,1],ymm5[2],mem[3,4,5],ymm5[6],mem[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27>
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm6, %ymm5, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm7[0,1,2,3,4,5],ymm5[6],ymm7[7,8,9,10,11,12,13],ymm5[14],ymm7[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm14[0],ymm13[1],ymm14[2,3,4],ymm13[5],ymm14[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm8
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm5 = xmm8[0],xmm5[1],xmm8[2],xmm5[3],xmm8[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm8 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm5, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm4, %ymm4
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm7[0],ymm4[1,2,3,4,5,6,7],ymm7[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm4[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0],ymm1[1],ymm2[2,3,4],ymm1[5],ymm2[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm2, %ymm9
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm1, %ymm6
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0],xmm0[1],xmm3[2],xmm0[3],xmm3[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm7, %xmm0, %xmm0
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm5, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm3 = mem[0,1],ymm1[2],mem[3,4,5],ymm1[6],mem[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm3, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm3[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6],ymm4[7,8,9,10,11,12,13],ymm3[14],ymm4[15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm7 = [8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm7, %xmm4, %xmm5
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3,4,5,6,7],ymm3[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm12[2],ymm15[3,4],ymm12[5],ymm15[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm11, %xmm0, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu (%rsp), %ymm11 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm7[0,1],ymm11[2],ymm7[3,4,5],ymm11[6],ymm7[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm6, %ymm4, %ymm5
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm4[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm5[0,1,2,3,4,5],ymm4[6],ymm5[7,8,9,10,11,12,13],ymm4[14],ymm5[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm4[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm1[0],ymm3[1],ymm1[2,3,4],ymm3[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm6
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm6[0],xmm4[1],xmm6[2],xmm4[3],xmm6[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm4, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6,7],ymm5[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1,2],ymm14[3],ymm15[4,5],ymm14[6],ymm15[7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm0[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3,4,5,6],ymm3[7,8],ymm0[9,10,11,12,13,14],ymm3[15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0,1],ymm11[2,3],ymm8[4,5],ymm11[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm7, %xmm4, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm3 = xmm3[10,11,6,7,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm12[0,1],ymm13[2],ymm12[3,4],ymm13[5],ymm12[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0,1],xmm3[2],xmm4[3],xmm3[4],xmm4[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1,2],ymm11[3],ymm7[4,5],ymm11[6],ymm7[7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm0[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm2[0],ymm0[1,2,3,4,5,6],ymm2[7,8],ymm0[9,10,11,12,13,14],ymm2[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm3[2],ymm1[3,4],ymm3[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1],xmm0[2],xmm2[3],xmm0[4],xmm2[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm5 = [8,9,8,9,6,7,4,5,2,3,2,3,2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm7
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 432(%rdi), %xmm4
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 416(%rdi), %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm10 = xmm3[0,1,2,3,4,5],xmm4[6],xmm3[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm14 = [0,1,2,3,4,5,6,7,0,1,0,1,14,15,12,13]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm14, %xmm10, %xmm10
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm10, %ymm0, %ymm10
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm10[5,6,7],ymm7[8,9,10,11,12],ymm10[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm7 = ymm0[0],ymm7[1,2,3,4,5,6,7],ymm0[8],ymm7[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1],ymm6[2],ymm9[3,4],ymm6[5],ymm9[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm9, %ymm11
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm7
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm7[0,1],xmm0[2],xmm7[3],xmm0[4],xmm7[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm0, %xmm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 208(%rdi), %xmm5
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %xmm15
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm7 = xmm15[0,1,2,3,4,5],xmm5[6],xmm15[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm14, %xmm7, %xmm7
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm6
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 432(%rdi), %xmm2
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 416(%rdi), %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm7 = xmm0[0,1,2,3,4,5],xmm2[6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm8 = [0,1,2,3,4,5,6,7,0,1,0,1,14,15,12,13]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm7, %xmm7
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm7, %ymm0, %ymm7
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm7[5,6,7],ymm0[8,9,10,11,12],ymm7[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm7 = ymm2[0,1],ymm1[2,3],ymm2[4,5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5,6,7],ymm6[8,9,10,11,12],ymm7[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm7 = ymm15[0,1],ymm12[2,3],ymm15[4,5],ymm12[6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm7, %xmm10
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm9 = xmm10[8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm6 = xmm7[10,11,6,7,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm6[0],xmm9[0],xmm6[1],xmm9[1],xmm6[2],xmm9[2],xmm6[3],xmm9[3]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm7 = ymm14[0,1,2],ymm10[3],ymm14[4,5],ymm10[6],ymm14[7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm9 = ymm7[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm7 = ymm9[0],ymm7[1,2,3,4,5,6],ymm9[7,8],ymm7[9,10,11,12,13,14],ymm9[15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm7 = ymm7[u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm6 = ymm6[0,1],ymm7[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm6[0],ymm0[1,2,3,4,5,6,7],ymm6[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm9 = ymm6[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $237, (%rsp), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0],ymm0[1],mem[2,3],ymm0[4],mem[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm6 = ymm0[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm6[1],ymm0[2,3,4,5,6,7,8],ymm6[9],ymm0[10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm7 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm7 = ymm8[0,1,2],mem[3],ymm8[4,5],mem[6],ymm8[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm7, %xmm8
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,2,3,4,5,6,7,10,11,6,7,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm8, %xmm8
-; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,6,4,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm7 = xmm7[4],xmm8[4],xmm7[5],xmm8[5],xmm7[6],xmm8[6],xmm7[7],xmm8[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31>
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm8, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm12[0,1],ymm13[2,3],ymm12[4,5],ymm13[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1,2],xmm1[3],xmm2[4],xmm1[5],xmm2[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm4[0],xmm3[1],xmm4[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm3 = [8,9,10,11,8,9,6,7,4,5,4,5,4,5,4,5]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm10 = xmm10[8,9,4,5,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm7 = xmm7[10,11,6,7,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm7 = xmm7[0],xmm10[0],xmm7[1],xmm10[1],xmm7[2],xmm10[2],xmm7[3],xmm10[3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29>
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm4, %ymm4
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm6 = ymm4[0],ymm6[1,2,3,4,5,6,7],ymm4[8],ymm6[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm4 = xmm7[0,1],xmm4[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm14[0,1],ymm13[2],ymm14[3,4],ymm13[5],ymm14[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm6
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm6[0,1],xmm4[2],xmm6[3],xmm4[4],xmm6[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm4, %xmm6
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 208(%rdi), %xmm15
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %xmm12
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm7 = xmm12[0,1,2,3,4,5],xmm15[6],xmm12[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm7, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm6
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm7, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm6 = ymm6[0,1,2,3,4],ymm7[5,6,7],ymm6[8,9,10,11,12],ymm7[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm7 = ymm5[0,1,2],ymm4[3],ymm5[4,5],ymm4[6],ymm5[7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm8 = ymm7[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm7 = ymm8[0],ymm7[1,2,3,4,5,6],ymm8[7,8],ymm7[9,10,11,12,13,14],ymm8[15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm7, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm8 = ymm9[0,1],ymm11[2,3],ymm9[4,5],ymm11[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm8, %xmm10
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm9 = xmm10[8,9,4,5,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm8 = xmm8[10,11,6,7,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm8 = xmm8[0],xmm9[0],xmm8[1],xmm9[1],xmm8[2],xmm9[2],xmm8[3],xmm9[3]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm6 = ymm7[0],ymm6[1,2,3,4,5,6,7],ymm7[8],ymm6[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm7 = xmm8[0,1],xmm7[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm9 = ymm7[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $237, (%rsp), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm6 = mem[0],ymm6[1],mem[2,3],ymm6[4],mem[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm7 = ymm6[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm6 = ymm6[0],ymm7[1],ymm6[2,3,4,5,6,7,8],ymm7[9],ymm6[10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm3[2,3],ymm1[4,5],ymm3[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm3[0,1,2],xmm1[3],xmm3[4],xmm1[5],xmm3[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm2 = [8,9,10,11,8,9,6,7,4,5,4,5,4,5,4,5]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm1, %xmm1
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,2,3,4,5,6,7,0,1,2,3,0,1,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm2, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7],ymm1[8,9,10,11,12],ymm2[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm1 = ymm11[0,1],mem[2,3],ymm11[4,5],mem[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1,2],xmm1[3],xmm2[4],xmm1[5],xmm2[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm1, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm5[0],xmm15[1],xmm5[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm2, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,2,3,4,5,6,7,0,1,2,3,0,1,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7],ymm1[8,9,10,11,12],ymm0[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm1 = ymm1[0,1,2],mem[3],ymm1[4,5],mem[6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} xmm8 = [10,11,6,7,10,11,6,7,10,11,6,7,10,11,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm7, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,4,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm7[4],xmm1[5],xmm7[5],xmm1[6],xmm7[6],xmm1[7],xmm7[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm7 = <u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31>
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm7, %ymm6, %ymm6
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm6[0],ymm0[1,2,3,4,5,6,7],ymm6[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm6[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm14[0,1],ymm13[2,3],ymm14[4,5],ymm13[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm6
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm6[0,1,2],xmm1[3],xmm6[4],xmm1[5],xmm6[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm15[0],xmm12[1],xmm15[2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm2, %xmm2
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7],ymm1[8,9,10,11,12],ymm2[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm10[0],ymm14[1],ymm10[2,3],ymm14[4],ymm10[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0],ymm5[1],ymm4[2,3],ymm5[4],ymm4[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm2[2,3,0,1]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0],ymm3[1],ymm2[2,3,4,5,6,7,8],ymm3[9],ymm2[10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm8, %ymm2, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm3 = ymm3[0,1,2],mem[3],ymm3[4,5],mem[6],ymm3[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm7, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm3 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm3 = mem[0,1,2],ymm11[3],mem[4,5],ymm11[6],mem[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm4, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm4, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,6,4,6,7]
; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vmovaps %ymm2, 32(%rsi)
@@ -5856,7 +5846,7 @@ define void @load_i16_stride7_vf32(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-PERLANE-NEXT: movq {{[0-9]+}}(%rsp), %rax
; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm0, 32(%rax)
; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm1, (%rax)
-; AVX2-FAST-PERLANE-NEXT: addq $520, %rsp # imm = 0x208
+; AVX2-FAST-PERLANE-NEXT: addq $488, %rsp # imm = 0x1E8
; AVX2-FAST-PERLANE-NEXT: vzeroupper
; AVX2-FAST-PERLANE-NEXT: retq
;
@@ -10240,13 +10230,13 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
;
; AVX2-SLOW-LABEL: load_i16_stride7_vf64:
; AVX2-SLOW: # %bb.0:
-; AVX2-SLOW-NEXT: subq $1464, %rsp # imm = 0x5B8
-; AVX2-SLOW-NEXT: vmovdqa 256(%rdi), %ymm15
-; AVX2-SLOW-NEXT: vmovdqa 224(%rdi), %ymm9
-; AVX2-SLOW-NEXT: vmovdqa 512(%rdi), %ymm14
-; AVX2-SLOW-NEXT: vmovdqa 544(%rdi), %ymm5
-; AVX2-SLOW-NEXT: vmovdqa 480(%rdi), %ymm10
-; AVX2-SLOW-NEXT: vmovdqa 448(%rdi), %ymm12
+; AVX2-SLOW-NEXT: subq $1480, %rsp # imm = 0x5C8
+; AVX2-SLOW-NEXT: vmovdqa 256(%rdi), %ymm12
+; AVX2-SLOW-NEXT: vmovdqa 224(%rdi), %ymm8
+; AVX2-SLOW-NEXT: vmovdqa 512(%rdi), %ymm7
+; AVX2-SLOW-NEXT: vmovdqa 544(%rdi), %ymm11
+; AVX2-SLOW-NEXT: vmovdqa 480(%rdi), %ymm9
+; AVX2-SLOW-NEXT: vmovdqa 448(%rdi), %ymm14
; AVX2-SLOW-NEXT: vmovdqa (%rdi), %ymm2
; AVX2-SLOW-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vmovdqa 32(%rdi), %ymm4
@@ -10266,60 +10256,57 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpshufb %xmm2, %xmm0, %xmm4
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm6 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
; AVX2-SLOW-NEXT: vpblendvb %ymm6, %ymm4, %ymm3, %ymm0
-; AVX2-SLOW-NEXT: vmovdqa %ymm6, %ymm13
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm12[0,1],ymm10[2],ymm12[3,4,5],ymm10[6],ymm12[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm14[0,1],ymm9[2],ymm14[3,4,5],ymm9[6],ymm14[7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
; AVX2-SLOW-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0],ymm14[1],ymm5[2,3,4],ymm14[5],ymm5[6,7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm14, %ymm6
-; AVX2-SLOW-NEXT: vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa %ymm5, %ymm14
-; AVX2-SLOW-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0],ymm7[1],ymm11[2,3,4],ymm7[5],ymm11[6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-SLOW-NEXT: vpblendvb %ymm13, %ymm3, %ymm4, %ymm0
+; AVX2-SLOW-NEXT: vpblendvb %ymm6, %ymm3, %ymm4, %ymm0
+; AVX2-SLOW-NEXT: vmovdqa %ymm6, %ymm10
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm9[0,1],ymm15[2],ymm9[3,4,5],ymm15[6],ymm9[7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0,1],ymm12[2],ymm8[3,4,5],ymm12[6],ymm8[7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
; AVX2-SLOW-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-SLOW-NEXT: vmovdqa 288(%rdi), %ymm8
+; AVX2-SLOW-NEXT: vmovdqa 288(%rdi), %ymm6
; AVX2-SLOW-NEXT: vmovdqa 320(%rdi), %ymm5
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0],ymm8[1],ymm5[2,3,4],ymm8[5],ymm5[6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa %ymm5, %ymm7
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0],ymm6[1],ymm5[2,3,4],ymm6[5],ymm5[6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa %ymm5, %ymm0
; AVX2-SLOW-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-SLOW-NEXT: vpblendvb %ymm13, %ymm3, %ymm4, %ymm0
-; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa 704(%rdi), %ymm3
+; AVX2-SLOW-NEXT: vpblendvb %ymm10, %ymm3, %ymm4, %ymm3
; AVX2-SLOW-NEXT: vmovdqu %ymm3, (%rsp) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa 672(%rdi), %ymm0
-; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm0[0,1],ymm3[2],ymm0[3,4,5],ymm3[6],ymm0[7]
+; AVX2-SLOW-NEXT: vmovdqa 704(%rdi), %ymm3
+; AVX2-SLOW-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa 672(%rdi), %ymm5
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0,1],ymm3[2],ymm5[3,4,5],ymm3[6],ymm5[7]
+; AVX2-SLOW-NEXT: vmovdqa %ymm5, %ymm15
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
; AVX2-SLOW-NEXT: vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-SLOW-NEXT: vmovdqa 736(%rdi), %ymm0
-; AVX2-SLOW-NEXT: vmovdqa 768(%rdi), %ymm11
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm11[0],ymm0[1],ymm11[2,3,4],ymm0[5],ymm11[6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa 736(%rdi), %ymm10
+; AVX2-SLOW-NEXT: vmovdqa 768(%rdi), %ymm13
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm13[0],ymm10[1],ymm13[2,3,4],ymm10[5],ymm13[6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm3, %ymm1
-; AVX2-SLOW-NEXT: vmovdqa %ymm13, %ymm5
-; AVX2-SLOW-NEXT: vpblendvb %ymm13, %ymm2, %ymm1, %ymm1
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
+; AVX2-SLOW-NEXT: vpblendvb %ymm5, %ymm2, %ymm1, %ymm1
; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm14[0,1],ymm6[2],ymm14[3,4],ymm6[5],ymm14[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm11[0,1],ymm7[2],ymm11[3,4],ymm7[5],ymm11[6,7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm1[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3],ymm1[4,5,6,7,8,9,10],ymm2[11],ymm1[12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm12[0,1,2],ymm10[3],ymm12[4,5],ymm10[6],ymm12[7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm10, %ymm14
-; AVX2-SLOW-NEXT: vmovdqa %ymm12, %ymm13
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm14[0,1,2],ymm9[3],ymm14[4,5],ymm9[6],ymm14[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm2[1],xmm3[2,3,4,5],xmm2[6],xmm3[7]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm2 = [2,3,2,3,2,3,2,3,8,9,8,9,6,7,4,5,18,19,18,19,18,19,18,19,24,25,24,25,22,23,20,21]
@@ -10327,102 +10314,105 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm1 = [2,3,0,1,14,15,12,13,10,11,10,11,10,11,10,11]
; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm3, %xmm3
; AVX2-SLOW-NEXT: vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
-; AVX2-SLOW-NEXT: vmovdqa %ymm5, %ymm6
+; AVX2-SLOW-NEXT: vmovdqa %ymm5, %ymm7
; AVX2-SLOW-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm7[0,1],ymm8[2],ymm7[3,4],ymm8[5],ymm7[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm0[0,1],ymm6[2],ymm0[3,4],ymm6[5],ymm0[6,7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm3[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2],ymm4[3],ymm3[4,5,6,7,8,9,10],ymm4[11],ymm3[12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm9[0,1,2],ymm15[3],ymm9[4,5],ymm15[6],ymm9[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm8[0,1,2],ymm12[3],ymm8[4,5],ymm12[6],ymm8[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2,3,4,5],xmm4[6],xmm5[7]
; AVX2-SLOW-NEXT: vpshufb %ymm2, %ymm3, %ymm3
; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm4, %xmm4
-; AVX2-SLOW-NEXT: vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
-; AVX2-SLOW-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa %ymm11, %ymm12
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm11[0,1],ymm0[2],ymm11[3,4],ymm0[5],ymm11[6,7]
+; AVX2-SLOW-NEXT: vpblendvb %ymm7, %ymm4, %ymm3, %ymm0
+; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa %ymm13, %ymm0
+; AVX2-SLOW-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm13[0,1],ymm10[2],ymm13[3,4],ymm10[5],ymm13[6,7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm3[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2],ymm4[3],ymm3[4,5,6,7,8,9,10],ymm4[11],ymm3[12,13,14,15]
-; AVX2-SLOW-NEXT: vmovdqu (%rsp), %ymm15 # 32-byte Reload
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm10[0,1,2],ymm15[3],ymm10[4,5],ymm15[6],ymm10[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm15[0,1,2],ymm10[3],ymm15[4,5],ymm10[6],ymm15[7]
+; AVX2-SLOW-NEXT: vmovdqa %ymm15, %ymm11
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2,3,4,5],xmm4[6],xmm5[7]
; AVX2-SLOW-NEXT: vpshufb %ymm2, %ymm3, %ymm3
; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm4, %xmm4
-; AVX2-SLOW-NEXT: vmovdqa %ymm6, %ymm0
-; AVX2-SLOW-NEXT: vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
+; AVX2-SLOW-NEXT: vmovdqa %ymm7, %ymm5
+; AVX2-SLOW-NEXT: vpblendvb %ymm7, %ymm4, %ymm3, %ymm3
; AVX2-SLOW-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0,1],ymm6[2],ymm5[3,4],ymm6[5],ymm5[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm9[0,1],ymm6[2],ymm9[3,4],ymm6[5],ymm9[6,7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm3[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2],ymm4[3],ymm3[4,5,6,7,8,9,10],ymm4[11],ymm3[12,13,14,15]
; AVX2-SLOW-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm9[0,1,2],ymm7[3],ymm9[4,5],ymm7[6],ymm9[7]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm15[0,1,2],ymm7[3],ymm15[4,5],ymm7[6],ymm15[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2,3,4,5],xmm3[6],xmm4[7]
; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm3, %xmm1
-; AVX2-SLOW-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm1
+; AVX2-SLOW-NEXT: vpblendvb %ymm5, %ymm1, %ymm2, %ymm1
; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm6[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm6[2,3],ymm5[4,5],ymm6[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm6[2,3],ymm9[4,5],ymm6[6,7]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm1[4],ymm2[5,6,7,8,9,10,11],ymm1[12],ymm2[13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0],ymm9[1],ymm7[2,3],ymm9[4],ymm7[5,6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0],ymm15[1],ymm7[2,3],ymm15[4],ymm7[5,6,7]
+; AVX2-SLOW-NEXT: vmovdqa %ymm7, %ymm9
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm3
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm1[0],xmm3[1],xmm1[2,3,4,5],xmm3[6],xmm1[7]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm1 = [4,5,4,5,4,5,4,5,8,9,10,11,8,9,6,7,20,21,20,21,20,21,20,21,24,25,26,27,24,25,22,23]
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm2, %ymm4
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm2 = [4,5,2,3,0,1,14,15,12,13,12,13,12,13,12,13]
; AVX2-SLOW-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-SLOW-NEXT: vpblendvb %ymm0, %ymm3, %ymm4, %ymm3
+; AVX2-SLOW-NEXT: vmovdqa %ymm5, %ymm7
+; AVX2-SLOW-NEXT: vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
; AVX2-SLOW-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa %ymm14, %ymm13
; AVX2-SLOW-NEXT: vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm14[0],ymm13[1],ymm14[2,3],ymm13[4],ymm14[5,6,7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm13, %ymm8
-; AVX2-SLOW-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2,3,4,5],xmm4[6],xmm3[7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm11[2,3],ymm6[4,5],ymm11[6,7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm5 = ymm11[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm5[4],ymm4[5,6,7,8,9,10,11],ymm5[12],ymm4[13,14,15]
; AVX2-SLOW-NEXT: vpshufb %xmm2, %xmm3, %xmm3
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-SLOW-NEXT: vpblendvb %ymm0, %ymm3, %ymm4, %ymm3
+; AVX2-SLOW-NEXT: vpblendvb %ymm7, %ymm3, %ymm4, %ymm3
; AVX2-SLOW-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm15[0],ymm10[1],ymm15[2,3],ymm10[4],ymm15[5,6,7]
+; AVX2-SLOW-NEXT: vmovdqa %ymm10, %ymm8
+; AVX2-SLOW-NEXT: vmovdqa %ymm11, %ymm12
+; AVX2-SLOW-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm10[0],ymm11[1],ymm10[2,3],ymm11[4],ymm10[5,6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2,3,4,5],xmm4[6],xmm3[7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm12[0,1],ymm5[2,3],ymm12[4,5],ymm5[6,7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm12, %ymm15
-; AVX2-SLOW-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm0[0,1],ymm10[2,3],ymm0[4,5],ymm10[6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm5 = ymm10[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm5[4],ymm4[5,6,7,8,9,10,11],ymm5[12],ymm4[13,14,15]
; AVX2-SLOW-NEXT: vpshufb %xmm2, %xmm3, %xmm3
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-SLOW-NEXT: vmovdqa %ymm0, %ymm5
-; AVX2-SLOW-NEXT: vpblendvb %ymm0, %ymm3, %ymm4, %ymm0
+; AVX2-SLOW-NEXT: vpblendvb %ymm7, %ymm3, %ymm4, %ymm0
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm0[0],ymm10[1],ymm0[2,3],ymm10[4],ymm0[5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0],ymm10[1],ymm5[2,3],ymm10[4],ymm5[5,6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2,3,4,5],xmm4[6],xmm3[7]
; AVX2-SLOW-NEXT: vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm13[0,1],ymm12[2,3],ymm13[4,5],ymm12[6,7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm12[2,3,0,1]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm0[0,1],ymm11[2,3],ymm0[4,5],ymm11[6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm11[2,3,0,1]
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4],ymm3[5,6,7,8,9,10,11],ymm4[12],ymm3[13,14,15]
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm3, %ymm1
-; AVX2-SLOW-NEXT: vpblendvb %ymm5, %ymm2, %ymm1, %ymm1
+; AVX2-SLOW-NEXT: vpblendvb %ymm7, %ymm2, %ymm1, %ymm1
; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0],ymm9[1],ymm7[2,3,4],ymm9[5],ymm7[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm9[0],ymm15[1],ymm9[2,3,4],ymm15[5],ymm9[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
@@ -10433,64 +10423,62 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm3, %ymm3
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,7,7,7]
-; AVX2-SLOW-NEXT: vpblendvb %ymm5, %ymm2, %ymm3, %ymm2
+; AVX2-SLOW-NEXT: vpblendvb %ymm7, %ymm2, %ymm3, %ymm2
; AVX2-SLOW-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm14[0],ymm8[1],ymm14[2,3,4],ymm8[5],ymm14[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm14[0],ymm13[1],ymm14[2,3,4],ymm13[5],ymm14[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0],xmm2[1],xmm3[2],xmm2[3],xmm3[4,5,6,7]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,7,7,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0,1,2],ymm11[3],ymm6[4,5],ymm11[6],ymm6[7]
+; AVX2-SLOW-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm3 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm3 = ymm6[0,1,2],mem[3],ymm6[4,5],mem[6],ymm6[7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm3[1,3,2,3]
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm3, %ymm3
-; AVX2-SLOW-NEXT: vpblendvb %ymm5, %ymm2, %ymm3, %ymm2
+; AVX2-SLOW-NEXT: vpblendvb %ymm7, %ymm2, %ymm3, %ymm2
; AVX2-SLOW-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm0[0],ymm10[1],ymm0[2,3,4],ymm10[5],ymm0[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0],ymm10[1],ymm5[2,3,4],ymm10[5],ymm5[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0],xmm2[1],xmm3[2],xmm2[3],xmm3[4,5,6,7]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,7,7,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm13[0,1,2],ymm12[3],ymm13[4,5],ymm12[6],ymm13[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm0[0,1,2],ymm11[3],ymm0[4,5],ymm11[6],ymm0[7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm3[1,3,2,3]
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm3, %ymm3
-; AVX2-SLOW-NEXT: vpblendvb %ymm5, %ymm2, %ymm3, %ymm0
+; AVX2-SLOW-NEXT: vpblendvb %ymm7, %ymm2, %ymm3, %ymm0
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu (%rsp), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm2 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm2 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm8[0],ymm12[1],ymm8[2,3,4],ymm12[5],ymm8[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0],xmm2[1],xmm3[2],xmm2[3],xmm3[4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm3 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm3 = ymm15[0,1,2],mem[3],ymm15[4,5],mem[6],ymm15[7]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm3 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm3 = ymm0[0,1,2],mem[3],ymm0[4,5],mem[6],ymm0[7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm3[1,3,2,3]
; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm3, %ymm1
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,7,7,7]
-; AVX2-SLOW-NEXT: vpblendvb %ymm5, %ymm2, %ymm1, %ymm0
+; AVX2-SLOW-NEXT: vpblendvb %ymm7, %ymm2, %ymm1, %ymm0
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vmovdqa 128(%rdi), %ymm10
-; AVX2-SLOW-NEXT: vmovdqa 160(%rdi), %ymm12
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0,1,2],ymm10[3],ymm12[4,5],ymm10[6],ymm12[7]
+; AVX2-SLOW-NEXT: vmovdqa 160(%rdi), %ymm11
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm11[0,1,2],ymm10[3],ymm11[4,5],ymm10[6],ymm11[7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,6,4,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm0
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,7,7]
; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm1
-; AVX2-SLOW-NEXT: vmovdqa 192(%rdi), %ymm0
-; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm9 = ymm0[0,1,0,2]
+; AVX2-SLOW-NEXT: vmovdqa 192(%rdi), %ymm14
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm9 = ymm14[0,1,0,2]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm2 = ymm9[0,1,2,1,4,5,6,5]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
; AVX2-SLOW-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
; AVX2-SLOW-NEXT: # ymm0 = mem[0,1,2,3,4],ymm1[5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa 576(%rdi), %ymm6
+; AVX2-SLOW-NEXT: vmovdqa 576(%rdi), %ymm12
; AVX2-SLOW-NEXT: vmovdqa 608(%rdi), %ymm8
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm8[0,1,2],ymm6[3],ymm8[4,5],ymm6[6],ymm8[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm8[0,1,2],ymm12[3],ymm8[4,5],ymm12[6],ymm8[7]
; AVX2-SLOW-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm1[0,1,2,3,4,6,4,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm1
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
@@ -10506,24 +10494,23 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
; AVX2-SLOW-NEXT: # ymm0 = mem[0,1,2,3,4],ymm1[5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa 352(%rdi), %ymm0
-; AVX2-SLOW-NEXT: vmovdqa 384(%rdi), %ymm4
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm4[0,1,2],ymm0[3],ymm4[4,5],ymm0[6],ymm4[7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa %ymm0, %ymm11
-; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa 352(%rdi), %ymm4
+; AVX2-SLOW-NEXT: vmovdqa 384(%rdi), %ymm6
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm6[0,1,2],ymm4[3],ymm6[4,5],ymm4[6],ymm6[7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm1[0,1,2,3,4,6,4,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm1
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,7,7]
; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm1 = xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-SLOW-NEXT: vmovdqa 416(%rdi), %ymm15
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm15[0,1,0,2]
+; AVX2-SLOW-NEXT: vmovdqa 416(%rdi), %ymm0
+; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm0[0,1,0,2]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm7 = ymm2[0,1,2,1,4,5,6,5]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm7[7]
-; AVX2-SLOW-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: vpblendd $31, (%rsp), %ymm1, %ymm0 # 32-byte Folded Reload
; AVX2-SLOW-NEXT: # ymm0 = mem[0,1,2,3,4],ymm1[5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vmovdqa 800(%rdi), %ymm0
@@ -10536,15 +10523,15 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,7,7]
; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm1
-; AVX2-SLOW-NEXT: vmovdqa 864(%rdi), %ymm14
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm0 = ymm14[0,1,0,2]
+; AVX2-SLOW-NEXT: vmovdqa 864(%rdi), %ymm15
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm0 = ymm15[0,1,0,2]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm13 = ymm0[0,1,2,1,4,5,6,5]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm13 = ymm13[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm13[7]
; AVX2-SLOW-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
; AVX2-SLOW-NEXT: # ymm1 = mem[0,1,2,3,4],ymm1[5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm6[0,1],ymm8[2],ymm6[3,4,5],ymm8[6],ymm6[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm12[0,1],ymm8[2],ymm12[3,4,5],ymm8[6],ymm12[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm13
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm13[4],xmm1[5],xmm13[6],xmm1[7]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm13 = [0,1,0,1,14,15,12,13,10,11,8,9,12,13,14,15]
@@ -10557,7 +10544,8 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm3[0],ymm1[1,2,3,4,5,6,7],ymm3[8],ymm1[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm11[0,1],ymm4[2],ymm11[3,4,5],ymm4[6],ymm11[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm6[2],ymm4[3,4,5],ymm6[6],ymm4[7]
+; AVX2-SLOW-NEXT: vmovdqa %ymm4, %ymm8
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm3
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm3[4],xmm1[5],xmm3[6],xmm1[7]
; AVX2-SLOW-NEXT: vpshufb %xmm13, %xmm1, %xmm1
@@ -10569,7 +10557,6 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa %ymm5, %ymm11
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm5[2],ymm7[3,4,5],ymm5[6],ymm7[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4],xmm1[5],xmm2[6],xmm1[7]
@@ -10582,7 +10569,7 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm12[2],ymm10[3,4,5],ymm12[6],ymm10[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm11[2],ymm10[3,4,5],ymm11[6],ymm10[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm1[4],xmm0[5],xmm1[6],xmm0[7]
; AVX2-SLOW-NEXT: vpshufb %xmm13, %xmm0, %xmm0
@@ -10594,17 +10581,14 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm12[3],ymm10[4,5],ymm12[6],ymm10[7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm12, %ymm6
-; AVX2-SLOW-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm11[3],ymm10[4,5],ymm11[6],ymm10[7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3,4,5],xmm0[6],xmm1[7]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm1 = [0,1,2,3,0,1,14,15,12,13,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm8[0,1,1,2]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm14[0,1,1,2]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[0,1,0,3,4,5,4,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,4,7,8,9,10,11,12,13,12,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
@@ -10612,15 +10596,15 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa %ymm12, %ymm3
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm13[3],ymm4[4,5],ymm13[6],ymm4[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0,1,2],ymm4[3],ymm12[4,5],ymm4[6],ymm12[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5],xmm0[6],xmm2[7]
; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm12[0,1,1,2]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm6[0,1,1,2]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[0,1,0,3,4,5,4,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,4,7,8,9,10,11,12,13,12,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
@@ -10629,14 +10613,15 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1,2],ymm5[3],ymm7[4,5],ymm5[6],ymm7[7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa %ymm5, %ymm13
+; AVX2-SLOW-NEXT: vmovdqu %ymm5, (%rsp) # 32-byte Spill
; AVX2-SLOW-NEXT: vmovdqa %ymm7, %ymm5
; AVX2-SLOW-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5],xmm0[6],xmm2[7]
; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm14[0,1,1,2]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm15[0,1,1,2]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[0,1,0,3,4,5,4,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,4,7,8,9,10,11,12,13,12,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
@@ -10644,14 +10629,15 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1,2],ymm3[3],ymm9[4,5],ymm3[6],ymm9[7]
+; AVX2-SLOW-NEXT: vmovdqa %ymm8, %ymm12
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0,1,2],ymm8[3],ymm12[4,5],ymm8[6],ymm12[7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5],xmm0[6],xmm2[7]
; AVX2-SLOW-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm15[0,1,1,2]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm9[0,1,1,2]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,1,0,3,4,5,4,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,4,7,8,9,10,11,12,13,12,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
@@ -10659,169 +10645,152 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0],ymm10[1],ymm6[2,3],ymm10[4],ymm6[5,6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm11[0],ymm10[1],ymm11[2,3],ymm10[4],ymm11[5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5],xmm1[6],xmm0[7]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,6,7]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm8[0,1,1,3]
-; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,1,2,0,4,5,6,4]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm1 = ymm14[0,1,2,0,4,5,6,4]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,7,4,8,9,10,11,12,13,15,12]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0],ymm4[1],ymm13[2,3],ymm4[4],ymm13[5,6,7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm13, %ymm7
-; AVX2-SLOW-NEXT: vmovdqa %ymm4, %ymm8
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6,7]
+; AVX2-SLOW-NEXT: vmovdqa %ymm4, %ymm2
+; AVX2-SLOW-NEXT: vmovdqa %ymm3, %ymm7
+; AVX2-SLOW-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5],xmm1[6],xmm0[7]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,6,7]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm12[0,1,1,3]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm1 = ymm2[0,1,2,0,4,5,6,4]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm6[0,1,1,3]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm1 = ymm3[0,1,2,0,4,5,6,4]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,7,4,8,9,10,11,12,13,15,12]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0],ymm9[1],ymm3[2,3],ymm9[4],ymm3[5,6,7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm9, %ymm12
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0],ymm12[1],ymm8[2,3],ymm12[4],ymm8[5,6,7]
+; AVX2-SLOW-NEXT: vmovdqa %ymm12, %ymm11
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5],xmm1[6],xmm0[7]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,6,7]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm15[0,1,1,3]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm1 = ymm3[0,1,2,0,4,5,6,4]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm8 = ymm9[0,1,1,3]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm1 = ymm8[0,1,2,0,4,5,6,4]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,7,4,8,9,10,11,12,13,15,12]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm11[0],ymm5[1],ymm11[2,3],ymm5[4],ymm11[5,6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0],ymm5[1],ymm13[2,3],ymm5[4],ymm13[5,6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5],xmm1[6],xmm0[7]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,2,1,0,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,6,7]
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm14[0,1,1,3]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm4 = ymm1[0,1,2,0,4,5,6,4]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm1
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm0 = ymm15[0,1,1,3]
+; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm4 = ymm0[0,1,2,0,4,5,6,4]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,7,4,8,9,10,11,12,13,15,12]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm4[0],ymm0[1,2,3,4,5,6,7],ymm4[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm14[0,1],ymm10[2],ymm14[3,4,5],ymm10[6],ymm14[7]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27>
-; AVX2-SLOW-NEXT: vpshufb %ymm0, %ymm4, %ymm5
-; AVX2-SLOW-NEXT: vmovdqa %ymm0, %ymm9
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm4[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm5[0,1,2,3,4,5],ymm4[6],ymm5[7,8,9,10,11,12,13],ymm4[14],ymm5[15]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm1 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm1 = mem[0,1],ymm12[2],mem[3,4],ymm12[5],mem[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm4
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm5 = ymm11[0,1],ymm13[2],ymm11[3,4],ymm13[5],ymm11[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm6 = xmm6[3,1,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm5 = ymm7[0],ymm8[1],ymm7[2,3,4],ymm8[5],ymm7[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm5 = xmm6[0],xmm5[1],xmm6[2],xmm5[3],xmm6[4,5,6,7]
+; AVX2-SLOW-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm5 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm5 = mem[0,1],ymm13[2],mem[3,4,5],ymm13[6],mem[7]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27>
+; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm5, %ymm6
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm5 = ymm6[0,1,2,3,4,5],ymm5[6],ymm6[7,8,9,10,11,12,13],ymm5[14],ymm6[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm5 = ymm2[0],ymm7[1],ymm2[2,3,4],ymm7[5],ymm2[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm9
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm5 = xmm9[0],xmm5[1],xmm9[2],xmm5[3],xmm9[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm0 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm5, %xmm5
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[0,1,2,1,4,5,6,5]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0],ymm2[1,2,3,4,5,6,7],ymm4[8],ymm2[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm2 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm2 = mem[0,1],ymm6[2],mem[3,4,5],ymm6[6],mem[7]
-; AVX2-SLOW-NEXT: vpshufb %ymm9, %ymm2, %ymm4
-; AVX2-SLOW-NEXT: vmovdqa %ymm9, %ymm7
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm2[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5],ymm2[6],ymm4[7,8,9,10,11,12,13],ymm2[14],ymm4[15]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm15[0,1],ymm9[2],ymm15[3,4],ymm9[5],ymm15[6,7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm3 = ymm3[0,1,2,1,4,5,6,5]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm6[0],ymm3[1,2,3,4,5,6,7],ymm6[8],ymm3[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm3 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm3 = mem[0],ymm10[1],mem[2,3,4],ymm10[5],mem[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm3, %xmm3
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm4 = ymm14[0,1,2,1,4,5,6,5]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm4 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm4 = ymm2[0,1],mem[2],ymm2[3,4],mem[5],ymm2[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm5 = xmm5[3,1,2,3,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[2,1,2,3]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm4 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm4 = mem[0],ymm12[1],mem[2,3,4],ymm12[5],mem[6,7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm12, %ymm8
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm5
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2],xmm4[3],xmm5[4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm4, %xmm4
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm3 = ymm3[0,1,2,1,4,5,6,5]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm2[0],ymm3[1,2,3,4,5,6,7],ymm2[8],ymm3[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm2 = ymm2[0,1],mem[2],ymm2[3,4,5],mem[6],ymm2[7]
-; AVX2-SLOW-NEXT: vpshufb %ymm7, %ymm2, %ymm3
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm2[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7,8,9,10,11,12,13],ymm2[14],ymm3[15]
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $219, (%rsp), %ymm5, %ymm3 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm3 = mem[0,1],ymm5[2],mem[3,4],ymm5[5],mem[6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm5 = ymm2[0,1],ymm5[2],ymm2[3,4,5],ymm5[6],ymm2[7]
+; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm5, %ymm6
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm6[0],ymm3[1,2,3,4,5,6,7],ymm6[8],ymm3[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm5 = ymm6[0,1,2,3,4,5],ymm5[6],ymm6[7,8,9,10,11,12,13],ymm5[14],ymm6[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm14[0,1],ymm9[2],ymm14[3,4],ymm9[5],ymm14[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[2,1,2,3]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm3 = mem[0],ymm3[1],mem[2,3,4],ymm3[5],mem[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm3, %xmm3
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,1,2,1,4,5,6,5]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5,6],ymm1[7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm1 = mem[0],ymm1[1],mem[2,3,4],ymm1[5],mem[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm1, %xmm1
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-SLOW-NEXT: vpshufd $100, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm2 = mem[0,1,2,1,4,5,6,5]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm12[0,1],ymm7[2],ymm12[3,4,5],ymm7[6],ymm12[7]
-; AVX2-SLOW-NEXT: vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm2[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6],ymm0[7,8,9,10,11,12,13],ymm2[14],ymm0[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm7[0,1],ymm15[2],ymm7[3,4,5],ymm15[6],ymm7[7]
+; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm4, %ymm5
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm4[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm5[0,1,2,3,4,5],ymm4[6],ymm5[7,8,9,10,11,12,13],ymm4[14],ymm5[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm3 = xmm3[0,1],xmm4[2,3]
+; AVX2-SLOW-NEXT: vmovdqa %ymm11, %ymm10
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0],ymm10[1],ymm11[2,3,4],ymm10[5],ymm11[6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm6
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm6[0],xmm4[1],xmm6[2],xmm4[3],xmm6[4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm4, %xmm4
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} ymm2 = ymm8[0,1,2,1,4,5,6,5]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6,7],ymm5[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-SLOW-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
; AVX2-SLOW-NEXT: # ymm2 = ymm2[0,1],mem[2],ymm2[3,4],mem[5],ymm2[6,7]
@@ -10830,308 +10799,324 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0,1,2],ymm10[3],ymm14[4,5],ymm10[6],ymm14[7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7,8],ymm0[9,10,11,12,13,14],ymm1[15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm11[0,1],ymm13[2,3],ymm11[4,5],ymm13[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[0,2,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[1,3,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29>
-; AVX2-SLOW-NEXT: vpshufb %ymm11, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm1 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2],xmm2[3],xmm1[4],xmm2[5,6,7]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm10 = [8,9,8,9,6,7,4,5,2,3,2,3,2,3,2,3]
-; AVX2-SLOW-NEXT: vpshufb %xmm10, %xmm1, %xmm1
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-SLOW-NEXT: vmovdqa 656(%rdi), %xmm3
-; AVX2-SLOW-NEXT: vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa 640(%rdi), %xmm2
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm2[0,1,2,3,4,5],xmm3[6],xmm2[7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[0,1,0,3]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,7,6]
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm4[5,6,7],ymm1[8,9,10,11,12],ymm4[13,14,15]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0,1],ymm6[2],ymm8[3,4,5],ymm6[6],ymm8[7]
+; AVX2-SLOW-NEXT: vpshufb %ymm1, %ymm3, %ymm1
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm3[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm1[0,1,2,3,4,5],ymm3[6],ymm1[7,8,9,10,11,12,13],ymm3[14],ymm1[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2,3]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $221, (%rsp), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm3 = mem[0],ymm3[1],mem[2,3,4],ymm3[5],mem[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm4
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm3, %xmm3
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-SLOW-NEXT: vpshufd $100, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm0 = mem[0,1,2,1,4,5,6,5]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6,5,8,9,10,11,12,13,14,13]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm0 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm0 = mem[0,1,2],ymm6[3],mem[4,5],ymm6[6],mem[7]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2],ymm13[3],ymm3[4,5],ymm13[6],ymm3[7]
; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7,8],ymm0[9,10,11,12,13,14],ymm1[15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm15[0,1],ymm9[2,3],ymm15[4,5],ymm9[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm4
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7,8],ymm0[9,10,11,12,13,14],ymm1[15]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm0 = ymm0[0,1],mem[2],ymm0[3,4],mem[5],ymm0[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1],xmm0[2],xmm2[3],xmm0[4],xmm2[5,6,7]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm13 = [8,9,8,9,6,7,4,5,2,3,2,3,2,3,2,3]
+; AVX2-SLOW-NEXT: vpshufb %xmm13, %xmm0, %xmm0
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-SLOW-NEXT: vmovdqa 656(%rdi), %xmm2
+; AVX2-SLOW-NEXT: vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-SLOW-NEXT: vmovdqa 640(%rdi), %xmm4
+; AVX2-SLOW-NEXT: vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm4[0,1,2,3,4,5],xmm2[6],xmm4[7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,3]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,7,6]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0,1,2,3,4],ymm2[5,6,7],ymm0[8,9,10,11,12],ymm2[13,14,15]
+; AVX2-SLOW-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm0 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm0 = mem[0,1],ymm12[2,3],mem[4,5],ymm12[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm4
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[2,1,2,3]
; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[1,3,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
-; AVX2-SLOW-NEXT: vpshufb %ymm11, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm1 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm1 = mem[0,1],ymm8[2],mem[3,4],ymm8[5],mem[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm4
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm4[0,1],xmm1[2],xmm4[3],xmm1[4],xmm4[5,6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm10, %xmm1, %xmm1
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-SLOW-NEXT: vmovdqa 432(%rdi), %xmm3
-; AVX2-SLOW-NEXT: vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa 416(%rdi), %xmm4
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[1,3,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29>
+; AVX2-SLOW-NEXT: vpshufb %ymm0, %ymm1, %ymm1
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0],ymm2[1,2,3,4,5,6,7],ymm1[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm1 = xmm4[0,1],xmm1[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0,1,2],ymm15[3],ymm7[4,5],ymm15[6],ymm7[7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm1[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6],ymm2[7,8],ymm1[9,10,11,12,13,14],ymm2[15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm4
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm4[0,1],xmm2[2],xmm4[3],xmm2[4],xmm4[5,6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm13, %xmm2, %xmm2
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-SLOW-NEXT: vmovdqa 432(%rdi), %xmm4
; AVX2-SLOW-NEXT: vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,5],xmm3[6],xmm4[7]
+; AVX2-SLOW-NEXT: vmovdqa 416(%rdi), %xmm5
+; AVX2-SLOW-NEXT: vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0,1,2,3,4,5],xmm4[6],xmm5[7]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[0,1,0,3]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,7,6]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm4[5,6,7],ymm1[8,9,10,11,12],ymm4[13,14,15]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1,2],ymm14[3],ymm13[4,5],ymm14[6],ymm13[7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7,8],ymm0[9,10,11,12,13,14],ymm1[15]
-; AVX2-SLOW-NEXT: vmovdqu (%rsp), %ymm9 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm9[0,1],ymm5[2,3],ymm9[4,5],ymm5[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm4
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[0,2,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[1,3,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
-; AVX2-SLOW-NEXT: vpshufb %ymm11, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm4[5,6,7],ymm2[8,9,10,11,12],ymm4[13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm14[0,1],ymm9[2,3],ymm14[4,5],ymm9[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm5
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm5 = xmm5[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[1,3,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3]
+; AVX2-SLOW-NEXT: vpshufb %ymm0, %ymm1, %ymm1
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0],ymm2[1,2,3,4,5,6,7],ymm1[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm1 = xmm4[0,1],xmm1[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm8[0,1,2],ymm6[3],ymm8[4,5],ymm6[6],ymm8[7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm2 = ymm1[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm2[0],ymm1[1,2,3,4,5,6],ymm2[7,8],ymm1[9,10,11,12,13,14],ymm2[15]
+; AVX2-SLOW-NEXT: vmovdqu (%rsp), %ymm7 # 32-byte Reload
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1],ymm3[2],ymm8[3,4],ymm3[5],ymm8[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2],xmm1[3],xmm0[4],xmm1[5,6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm10, %xmm0, %xmm0
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm5
-; AVX2-SLOW-NEXT: vmovdqa 880(%rdi), %xmm1
-; AVX2-SLOW-NEXT: vmovdqa 864(%rdi), %xmm0
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm6 = xmm0[0,1,2,3,4,5],xmm1[6],xmm0[7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0,1],ymm8[2],ymm7[3,4],ymm8[5],ymm7[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm1, %xmm2
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2],xmm2[3],xmm1[4],xmm2[5,6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm13, %xmm1, %xmm1
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm5
+; AVX2-SLOW-NEXT: vmovdqa 880(%rdi), %xmm2
+; AVX2-SLOW-NEXT: vmovdqa 864(%rdi), %xmm1
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm6 = xmm1[0,1,2,3,4,5],xmm2[6],xmm1[7]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm6 = xmm6[0,1,0,3]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,4,7,6]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm6
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm6 = ymm9[0,1],ymm15[2,3],ymm9[4,5],ymm15[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm6, %xmm14
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm14 = xmm14[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm14 = xmm14[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm6 = xmm6[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm6 = xmm6[1,3,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm6[0],xmm14[0],xmm6[1],xmm14[1],xmm6[2],xmm14[2],xmm6[3],xmm14[3]
+; AVX2-SLOW-NEXT: vpshufb %ymm0, %ymm4, %ymm4
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm5 = ymm4[0],ymm5[1,2,3,4,5,6,7],ymm4[8],ymm5[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm5[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm4 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm4 = mem[0,1],ymm15[2],mem[3,4],ymm15[5],mem[6,7]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0,1],ymm12[2],ymm11[3,4],ymm12[5],ymm11[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0,1],xmm4[2],xmm5[3],xmm4[4],xmm5[5,6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm10, %xmm4, %xmm4
+; AVX2-SLOW-NEXT: vpshufb %xmm13, %xmm4, %xmm4
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-SLOW-NEXT: vmovdqa 208(%rdi), %xmm6
-; AVX2-SLOW-NEXT: vmovdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-SLOW-NEXT: vmovdqa 192(%rdi), %xmm5
-; AVX2-SLOW-NEXT: vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5],xmm6[6],xmm5[7]
+; AVX2-SLOW-NEXT: vmovdqa 208(%rdi), %xmm10
+; AVX2-SLOW-NEXT: vmovdqa 192(%rdi), %xmm14
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm5 = xmm14[0,1,2,3,4,5],xmm10[6],xmm14[7]
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[0,1,0,3]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,4,7,6]
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7],ymm4[8,9,10,11,12],ymm5[13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm5 = ymm12[0,1,2],ymm7[3],ymm12[4,5],ymm7[6],ymm12[7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm6 = ymm5[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm5 = ymm6[0],ymm5[1,2,3,4,5,6],ymm6[7,8],ymm5[9,10,11,12,13,14],ymm6[15]
-; AVX2-SLOW-NEXT: vpshufb %ymm11, %ymm5, %ymm5
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm6 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm6 = mem[0,1],ymm10[2,3],mem[4,5],ymm10[6,7]
-; AVX2-SLOW-NEXT: vmovdqa %ymm10, %ymm12
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm6, %xmm11
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm11 = xmm11[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm11 = xmm11[0,2,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm6 = xmm6[2,1,2,3]
-; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm6 = xmm6[1,3,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm6[0],xmm11[0],xmm6[1],xmm11[1],xmm6[2],xmm11[2],xmm6[3],xmm11[3]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm5 = ymm6[0,1],ymm5[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm5[0],ymm4[1,2,3,4,5,6,7],ymm5[8],ymm4[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm10 = ymm5[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm4 = ymm4[0],mem[1],ymm4[2,3],mem[4],ymm4[5,6,7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm5 = ymm4[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0],ymm5[1],ymm4[2,3,4,5,6,7,8],ymm5[9],ymm4[10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
; AVX2-SLOW-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
; AVX2-SLOW-NEXT: # ymm5 = mem[0,1,2],ymm5[3],mem[4,5],ymm5[6],mem[7]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm6 = xmm5[0,1,2,3,6,4,6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm5
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[0,1,2,1]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,7,6,7]
-; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm5 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31>
-; AVX2-SLOW-NEXT: vpshufb %ymm11, %ymm4, %ymm4
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm5 = ymm5[0,1],mem[2,3],ymm5[4,5],mem[6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm6 = ymm5[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm5 = ymm6[0],ymm5[1,2,3,4,5,6],ymm6[7,8],ymm5[9,10,11,12,13,14],ymm6[15]
+; AVX2-SLOW-NEXT: vpshufb %ymm0, %ymm5, %ymm0
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm5 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm5 = mem[0,1],ymm13[2,3],mem[4,5],ymm13[6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm5 = xmm6[0,1,2],xmm5[3],xmm6[4],xmm5[5],xmm6[6,7]
-; AVX2-SLOW-NEXT: vpblendw $253, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm2 # 16-byte Folded Reload
-; AVX2-SLOW-NEXT: # xmm2 = mem[0],xmm2[1],mem[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm7 = [8,9,10,11,8,9,6,7,4,5,4,5,4,5,4,5]
-; AVX2-SLOW-NEXT: vpshufb %xmm7, %xmm5, %xmm5
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm6 = xmm6[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm6 = xmm6[0,2,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[2,1,2,3]
+; AVX2-SLOW-NEXT: vpshuflw {{.*#+}} xmm5 = xmm5[1,3,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpunpcklwd {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm0[0],ymm4[1,2,3,4,5,6,7],ymm0[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm0 = xmm5[0,1],xmm0[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm0 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm0 = mem[0],ymm3[1],mem[2,3],ymm3[4],mem[5,6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm0[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm0[0],ymm4[1],ymm0[2,3,4,5,6,7,8],ymm4[9],ymm0[10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm0 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm0, %xmm5
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm5 = xmm5[0,1,2],xmm0[3],xmm5[4],xmm0[5],xmm5[6,7]
+; AVX2-SLOW-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX2-SLOW-NEXT: vpblendw $253, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm3 # 16-byte Folded Reload
+; AVX2-SLOW-NEXT: # xmm3 = mem[0],xmm0[1],mem[2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} xmm0 = [8,9,10,11,8,9,6,7,4,5,4,5,4,5,4,5]
+; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm5, %xmm5
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,3]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,7]
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm5[0,1,2,3,4],ymm2[5,6,7],ymm5[8,9,10,11,12],ymm2[13,14,15]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0],ymm2[1,2,3,4,5,6,7],ymm4[8],ymm2[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm14[0],ymm13[1],ymm14[2,3],ymm13[4],ymm14[5,6,7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm5 = ymm4[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0],ymm5[1],ymm4[2,3,4,5,6,7,8],ymm5[9],ymm4[10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm5 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm5 = ymm9[0,1,2],mem[3],ymm9[4,5],mem[6],ymm9[7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm3 = xmm3[0,1,0,3]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,4,7]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm5[0,1,2,3,4],ymm3[5,6,7],ymm5[8,9,10,11,12],ymm3[13,14,15]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm5 = ymm5[0,1,2],mem[3],ymm5[4,5],mem[6],ymm5[7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm6 = xmm5[0,1,2,3,6,4,6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm5
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[0,1,2,1]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,7,6,7]
-; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm5 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
-; AVX2-SLOW-NEXT: vpshufb %ymm11, %ymm4, %ymm4
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm5 = ymm8[0,1],ymm3[2,3],ymm8[4,5],ymm3[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm5 = xmm6[0,1,2],xmm5[3],xmm6[4],xmm5[5],xmm6[6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm7, %xmm5, %xmm1
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,3]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,4,7]
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4],ymm0[5,6,7],ymm1[8,9,10,11,12],ymm0[13,14,15]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm4[0],ymm0[1,2,3,4,5,6,7],ymm4[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm4[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm0 = ymm0[0],mem[1],ymm0[2,3],mem[4],ymm0[5,6,7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm4 = ymm0[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm4[1],ymm0[2,3,4,5,6,7,8],ymm4[9],ymm0[10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm6 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
+; AVX2-SLOW-NEXT: vmovdqa {{.*#+}} ymm5 = <u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31>
+; AVX2-SLOW-NEXT: vpshufb %ymm5, %ymm4, %ymm4
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm4[0],ymm3[1,2,3,4,5,6,7],ymm4[8],ymm3[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm4 = xmm6[0,1],xmm4[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm3[4,5,6,7]
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm4 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm4 = mem[0,1,2],ymm3[3],mem[4,5],ymm3[6],mem[7]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm5 = xmm4[0,1,2,3,6,4,6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm4
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[0,1,2,1]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,5,7,6,7]
-; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
-; AVX2-SLOW-NEXT: vpshufb %ymm11, %ymm0, %ymm0
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm3 = mem[0],ymm3[1],mem[2,3],ymm3[4],mem[5,6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm6 = ymm3[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0],ymm6[1],ymm3[2,3,4,5,6,7,8],ymm6[9],ymm3[10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm6 = ymm7[0,1],ymm8[2,3],ymm7[4,5],ymm8[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm6, %xmm7
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm6 = xmm7[0,1,2],xmm6[3],xmm7[4],xmm6[5],xmm7[6,7]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm6, %xmm2
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,3]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,4,7]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7],ymm2[8,9,10,11,12],ymm1[13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm9[0,1,2],ymm15[3],ymm9[4,5],ymm15[6],ymm9[7]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm6 = xmm2[0,1,2,3,6,4,6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm2
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,1,2,1]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,7,6,7]
+; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm2 = xmm6[4],xmm2[4],xmm6[5],xmm2[5],xmm6[6],xmm2[6],xmm6[7],xmm2[7]
+; AVX2-SLOW-NEXT: vpshufb %ymm5, %ymm3, %ymm3
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm1 = ymm3[0],ymm1[1,2,3,4,5,6,7],ymm3[8],ymm1[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm3[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm2 = mem[0],ymm2[1],mem[2,3],ymm2[4],mem[5,6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm3 = ymm2[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0],ymm3[1],ymm2[2,3,4,5,6,7,8],ymm3[9],ymm2[10,11,12,13,14,15]
; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm4 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm4 = mem[0,1],ymm3[2,3],mem[4,5],ymm3[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm5
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[3],xmm5[4],xmm4[5],xmm5[6,7]
-; AVX2-SLOW-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; AVX2-SLOW-NEXT: vpblendw $253, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm5 # 16-byte Folded Reload
-; AVX2-SLOW-NEXT: # xmm5 = mem[0],xmm3[1],mem[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm7, %xmm4, %xmm4
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[0,1,0,3]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,4,7]
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm5[5,6,7],ymm4[8,9,10,11,12],ymm5[13,14,15]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm0[0],ymm4[1,2,3,4,5,6,7],ymm0[8],ymm4[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm4 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm4 = mem[0,1],ymm15[2,3],mem[4,5],ymm15[6,7]
-; AVX2-SLOW-NEXT: vextracti128 $1, %ymm4, %xmm5
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0,1,2],xmm4[3],xmm5[4],xmm4[5],xmm5[6,7]
-; AVX2-SLOW-NEXT: vpshufb %xmm7, %xmm4, %xmm3
-; AVX2-SLOW-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; AVX2-SLOW-NEXT: vpblendw $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
-; AVX2-SLOW-NEXT: # xmm4 = xmm4[0],mem[1],xmm4[2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm3 = ymm3[0,1],mem[2,3],ymm3[4,5],mem[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm3, %xmm6
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm3 = xmm6[0,1,2],xmm3[3],xmm6[4],xmm3[5],xmm6[6,7]
+; AVX2-SLOW-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; AVX2-SLOW-NEXT: vpblendw $253, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6, %xmm6 # 16-byte Folded Reload
+; AVX2-SLOW-NEXT: # xmm6 = mem[0],xmm6[1],mem[2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm3, %xmm3
; AVX2-SLOW-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm4 = xmm4[0,1,0,3]
-; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,5,4,7]
-; AVX2-SLOW-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7],ymm3[8,9,10,11,12],ymm4[13,14,15]
-; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm4 = mem[0],ymm4[1],mem[2,3],ymm4[4],mem[5,6,7]
-; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm5 = ymm4[2,3,0,1]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0],ymm5[1],ymm4[2,3,4,5,6,7,8],ymm5[9],ymm4[10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpshufb %ymm11, %ymm4, %ymm4
-; AVX2-SLOW-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm5 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT: # ymm5 = mem[0,1,2],ymm12[3],mem[4,5],ymm12[6],mem[7]
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm6 = xmm6[0,1,0,3]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,5,4,7]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm6
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm6[5,6,7],ymm3[8,9,10,11,12],ymm6[13,14,15]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm6 = mem[0,1,2],ymm6[3],mem[4,5],ymm6[6],mem[7]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm7 = xmm6[0,1,2,3,6,4,6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm6, %xmm6
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm6 = xmm6[0,1,2,1]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,5,7,6,7]
+; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm6 = xmm7[4],xmm6[4],xmm7[5],xmm6[5],xmm7[6],xmm6[6],xmm7[7],xmm6[7]
+; AVX2-SLOW-NEXT: vpshufb %ymm5, %ymm2, %ymm2
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm2[0],ymm3[1,2,3,4,5,6,7],ymm2[8],ymm3[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm2 = xmm6[0,1],xmm2[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm2 = ymm11[0,1],ymm12[2,3],ymm11[4,5],ymm12[6,7]
+; AVX2-SLOW-NEXT: vextracti128 $1, %ymm2, %xmm6
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm6[0,1,2],xmm2[3],xmm6[4],xmm2[5],xmm6[6,7]
+; AVX2-SLOW-NEXT: vpshufb %xmm0, %xmm2, %xmm0
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} xmm2 = xmm10[0],xmm14[1],xmm10[2,3,4,5,6,7]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,3]
+; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,7]
+; AVX2-SLOW-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7],ymm0[8,9,10,11,12],ymm2[13,14,15]
+; AVX2-SLOW-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm2 = mem[0],ymm2[1],mem[2,3],ymm2[4],mem[5,6,7]
+; AVX2-SLOW-NEXT: vpermq {{.*#+}} ymm6 = ymm2[2,3,0,1]
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0],ymm6[1],ymm2[2,3,4,5,6,7,8],ymm6[9],ymm2[10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpshufb %ymm5, %ymm2, %ymm2
+; AVX2-SLOW-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm5 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT: # ymm5 = mem[0,1,2],ymm13[3],mem[4,5],ymm13[6],mem[7]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm6 = xmm5[0,1,2,3,6,4,6,7]
; AVX2-SLOW-NEXT: vextracti128 $1, %ymm5, %xmm5
; AVX2-SLOW-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[0,1,2,1]
; AVX2-SLOW-NEXT: vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,7,6,7]
; AVX2-SLOW-NEXT: vpunpckhwd {{.*#+}} xmm5 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0,1],ymm4[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm3 = ymm4[0],ymm3[1,2,3,4,5,6,7],ymm4[8],ymm3[9,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm3 = ymm4[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 96(%rsi)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 32(%rsi)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 64(%rsi)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, (%rsi)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 96(%rdx)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 32(%rdx)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 64(%rdx)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, (%rdx)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 32(%rcx)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 96(%rcx)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 64(%rcx)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, (%rcx)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 96(%r8)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 32(%r8)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 64(%r8)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, (%r8)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 96(%r9)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 32(%r9)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, (%r9)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 64(%r9)
+; AVX2-SLOW-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} xmm2 = xmm5[0,1],xmm2[2,3]
+; AVX2-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 96(%rsi)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 32(%rsi)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 64(%rsi)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, (%rsi)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 96(%rdx)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 32(%rdx)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 64(%rdx)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, (%rdx)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 32(%rcx)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 96(%rcx)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 64(%rcx)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, (%rcx)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 96(%r8)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 32(%r8)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 64(%r8)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, (%r8)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 96(%r9)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 32(%r9)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, (%r9)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 64(%r9)
; AVX2-SLOW-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 96(%rax)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 32(%rax)
-; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT: vmovaps %ymm4, 64(%rax)
-; AVX2-SLOW-NEXT: vmovdqa %ymm10, (%rax)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 96(%rax)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 32(%rax)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, 64(%rax)
+; AVX2-SLOW-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT: vmovaps %ymm2, (%rax)
; AVX2-SLOW-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; AVX2-SLOW-NEXT: vmovdqa %ymm0, 32(%rax)
-; AVX2-SLOW-NEXT: vmovdqa %ymm3, (%rax)
+; AVX2-SLOW-NEXT: vmovdqa %ymm3, 32(%rax)
+; AVX2-SLOW-NEXT: vmovdqa %ymm0, (%rax)
; AVX2-SLOW-NEXT: vmovdqa %ymm1, 96(%rax)
-; AVX2-SLOW-NEXT: vmovdqa %ymm2, 64(%rax)
-; AVX2-SLOW-NEXT: addq $1464, %rsp # imm = 0x5B8
+; AVX2-SLOW-NEXT: vmovdqa %ymm4, 64(%rax)
+; AVX2-SLOW-NEXT: addq $1480, %rsp # imm = 0x5C8
; AVX2-SLOW-NEXT: vzeroupper
; AVX2-SLOW-NEXT: retq
;
@@ -11140,223 +11125,228 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: subq $1544, %rsp # imm = 0x608
; AVX2-FAST-NEXT: vmovdqa 256(%rdi), %ymm5
; AVX2-FAST-NEXT: vmovdqa 224(%rdi), %ymm6
-; AVX2-FAST-NEXT: vmovdqa 512(%rdi), %ymm9
-; AVX2-FAST-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa 544(%rdi), %ymm7
-; AVX2-FAST-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa 480(%rdi), %ymm11
-; AVX2-FAST-NEXT: vmovdqa 448(%rdi), %ymm12
-; AVX2-FAST-NEXT: vmovdqa (%rdi), %ymm10
-; AVX2-FAST-NEXT: vmovdqa 32(%rdi), %ymm13
-; AVX2-FAST-NEXT: vmovdqa 64(%rdi), %ymm0
+; AVX2-FAST-NEXT: vmovdqa 512(%rdi), %ymm12
+; AVX2-FAST-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqa 544(%rdi), %ymm13
+; AVX2-FAST-NEXT: vmovdqa 480(%rdi), %ymm10
+; AVX2-FAST-NEXT: vmovdqa 448(%rdi), %ymm11
+; AVX2-FAST-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqa (%rdi), %ymm2
+; AVX2-FAST-NEXT: vmovdqa 32(%rdi), %ymm4
+; AVX2-FAST-NEXT: vmovdqa 64(%rdi), %ymm1
+; AVX2-FAST-NEXT: vmovdqa 96(%rdi), %ymm0
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa 96(%rdi), %ymm1
-; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3,4],ymm1[5],ymm0[6,7]
+; AVX2-FAST-NEXT: vmovdqa %ymm1, %ymm9
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = [0,1,14,15,12,13,14,15,2,3,6,7,12,13,2,3,16,17,30,31,28,29,30,31,18,19,22,23,28,29,18,19]
; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm0, %ymm3
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm13[2],ymm10[3,4,5],ymm13[6],ymm10[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm4[2],ymm2[3,4,5],ymm4[6],ymm2[7]
+; AVX2-FAST-NEXT: vmovdqa %ymm4, %ymm7
+; AVX2-FAST-NEXT: vmovdqa %ymm2, %ymm8
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4],xmm0[5],xmm2[6],xmm0[7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,14,15,12,13,10,11,8,9,8,9,8,9,8,9]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm0, %xmm4
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm15 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-FAST-NEXT: vpblendvb %ymm15, %ymm4, %ymm3, %ymm0
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm14 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
+; AVX2-FAST-NEXT: vpblendvb %ymm14, %ymm4, %ymm3, %ymm0
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm12[0,1],ymm11[2],ymm12[3,4,5],ymm11[6],ymm12[7]
-; AVX2-FAST-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm11[0,1],ymm10[2],ymm11[3,4,5],ymm10[6],ymm11[7]
+; AVX2-FAST-NEXT: vmovdqa %ymm10, %ymm15
; AVX2-FAST-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm7[0],ymm9[1],ymm7[2,3,4],ymm9[5],ymm7[6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm13[0],ymm12[1],ymm13[2,3,4],ymm12[5],ymm13[6,7]
+; AVX2-FAST-NEXT: vmovdqa %ymm13, %ymm10
+; AVX2-FAST-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-FAST-NEXT: vpblendvb %ymm15, %ymm3, %ymm4, %ymm0
+; AVX2-FAST-NEXT: vmovdqa %ymm14, %ymm13
+; AVX2-FAST-NEXT: vpblendvb %ymm14, %ymm3, %ymm4, %ymm0
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0,1],ymm5[2],ymm6[3,4,5],ymm5[6],ymm6[7]
-; AVX2-FAST-NEXT: vmovdqa %ymm6, %ymm9
+; AVX2-FAST-NEXT: vmovdqa %ymm6, %ymm12
; AVX2-FAST-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa %ymm5, %ymm7
-; AVX2-FAST-NEXT: vmovdqu %ymm5, (%rsp) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqa %ymm5, %ymm6
+; AVX2-FAST-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-NEXT: vmovdqa 288(%rdi), %ymm6
-; AVX2-FAST-NEXT: vmovdqa 320(%rdi), %ymm14
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm14[0],ymm6[1],ymm14[2,3,4],ymm6[5],ymm14[6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqa 288(%rdi), %ymm11
+; AVX2-FAST-NEXT: vmovdqa 320(%rdi), %ymm0
+; AVX2-FAST-NEXT: vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm0[0],ymm11[1],ymm0[2,3,4],ymm11[5],ymm0[6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-FAST-NEXT: vpblendvb %ymm15, %ymm3, %ymm4, %ymm0
+; AVX2-FAST-NEXT: vpblendvb %ymm14, %ymm3, %ymm4, %ymm0
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa 704(%rdi), %ymm3
-; AVX2-FAST-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa 672(%rdi), %ymm0
+; AVX2-FAST-NEXT: vmovdqa 704(%rdi), %ymm0
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm0[0,1],ymm3[2],ymm0[3,4,5],ymm3[6],ymm0[7]
+; AVX2-FAST-NEXT: vmovdqa 672(%rdi), %ymm3
+; AVX2-FAST-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm0[2],ymm3[3,4,5],ymm0[6],ymm3[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-FAST-NEXT: vmovdqa 736(%rdi), %ymm0
-; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa 768(%rdi), %ymm3
+; AVX2-FAST-NEXT: vmovdqa 736(%rdi), %ymm3
; AVX2-FAST-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0],ymm0[1],ymm3[2,3,4],ymm0[5],ymm3[6,7]
+; AVX2-FAST-NEXT: vmovdqa 768(%rdi), %ymm0
+; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm0[0],ymm3[1],ymm0[2,3,4],ymm3[5],ymm0[6,7]
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm3, %ymm1
-; AVX2-FAST-NEXT: vpblendvb %ymm15, %ymm2, %ymm1, %ymm0
+; AVX2-FAST-NEXT: vpblendvb %ymm14, %ymm2, %ymm1, %ymm0
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa %ymm13, %ymm0
-; AVX2-FAST-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm10[0,1,2],ymm13[3],ymm10[4,5],ymm13[6],ymm10[7]
+; AVX2-FAST-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm8[0,1,2],ymm7[3],ymm8[4,5],ymm7[6],ymm8[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm2[0],xmm1[1],xmm2[2,3,4,5],xmm1[6],xmm2[7]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm13[0,1],ymm8[2],ymm13[3,4],ymm8[5],ymm13[6,7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm0[0,1],ymm9[2],ymm0[3,4],ymm9[5],ymm0[6,7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = <2,5,1,u,4,u,u,u>
; AVX2-FAST-NEXT: vpermd %ymm2, %ymm1, %ymm3
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm2 = [2,3,2,3,2,3,2,3,8,9,0,1,6,7,8,9,18,19,18,19,18,19,18,19,24,25,16,17,22,23,24,25]
; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm3, %ymm5
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm3 = [2,3,0,1,14,15,12,13,10,11,10,11,10,11,10,11]
; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm4, %xmm4
-; AVX2-FAST-NEXT: vpblendvb %ymm15, %ymm4, %ymm5, %ymm4
+; AVX2-FAST-NEXT: vpblendvb %ymm14, %ymm4, %ymm5, %ymm4
; AVX2-FAST-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm12[0,1,2],ymm11[3],ymm12[4,5],ymm11[6],ymm12[7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm14[0,1,2],ymm15[3],ymm14[4,5],ymm15[6],ymm14[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2,3,4,5],xmm4[6],xmm5[7]
; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm4, %xmm4
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm11[0,1],ymm12[2],ymm11[3,4],ymm12[5],ymm11[6,7]
+; AVX2-FAST-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm5 = ymm10[0,1],mem[2],ymm10[3,4],mem[5],ymm10[6,7]
; AVX2-FAST-NEXT: vpermd %ymm5, %ymm1, %ymm5
; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm5, %ymm5
-; AVX2-FAST-NEXT: vpblendvb %ymm15, %ymm4, %ymm5, %ymm4
+; AVX2-FAST-NEXT: vpblendvb %ymm13, %ymm4, %ymm5, %ymm4
; AVX2-FAST-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm9[0,1,2],ymm7[3],ymm9[4,5],ymm7[6],ymm9[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm12[0,1,2],ymm6[3],ymm12[4,5],ymm6[6],ymm12[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2,3,4,5],xmm4[6],xmm5[7]
; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm4, %xmm4
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm14[0,1],ymm6[2],ymm14[3,4],ymm6[5],ymm14[6,7]
+; AVX2-FAST-NEXT: vpblendd $219, (%rsp), %ymm11, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm5 = mem[0,1],ymm11[2],mem[3,4],ymm11[5],mem[6,7]
; AVX2-FAST-NEXT: vpermd %ymm5, %ymm1, %ymm5
; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm5, %ymm5
-; AVX2-FAST-NEXT: vpblendvb %ymm15, %ymm4, %ymm5, %ymm4
-; AVX2-FAST-NEXT: vmovdqa %ymm15, %ymm14
+; AVX2-FAST-NEXT: vpblendvb %ymm13, %ymm4, %ymm5, %ymm4
; AVX2-FAST-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm7[0,1,2],ymm15[3],ymm7[4,5],ymm15[6],ymm7[7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1,2],ymm12[3],ymm6[4,5],ymm12[6],ymm6[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2,3,4,5],xmm4[6],xmm5[7]
; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm4, %xmm3
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm9[2],ymm6[3,4],ymm9[5],ymm6[6,7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7]
; AVX2-FAST-NEXT: vpermd %ymm4, %ymm1, %ymm1
; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vpblendvb %ymm14, %ymm3, %ymm1, %ymm1
+; AVX2-FAST-NEXT: vpblendvb %ymm13, %ymm3, %ymm1, %ymm1
; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0],ymm10[1],ymm0[2,3],ymm10[4],ymm0[5,6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0],ymm8[1],ymm7[2,3],ymm8[4],ymm7[5,6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm1[0],xmm2[1],xmm1[2,3,4,5],xmm2[6],xmm1[7]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm13[0,1],ymm8[2,3],ymm13[4,5],ymm8[6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm0[0,1],ymm9[2,3],ymm0[4,5],ymm9[6,7]
+; AVX2-FAST-NEXT: vmovdqa %ymm9, %ymm0
+; AVX2-FAST-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = <2,6,1,u,5,u,u,u>
; AVX2-FAST-NEXT: vpermd %ymm2, %ymm1, %ymm3
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,2,3,8,9,2,3,4,5,10,11,16,17,18,19,20,21,18,19,24,25,18,19,20,21,26,27]
; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm3, %ymm5
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm3 = [4,5,2,3,0,1,14,15,12,13,12,13,12,13,12,13]
; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm4, %xmm4
-; AVX2-FAST-NEXT: vpblendvb %ymm14, %ymm4, %ymm5, %ymm0
-; AVX2-FAST-NEXT: vmovdqa %ymm14, %ymm8
-; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm14[0],ymm13[1],ymm14[2,3],ymm13[4],ymm14[5,6,7]
+; AVX2-FAST-NEXT: vmovdqa %ymm13, %ymm8
+; AVX2-FAST-NEXT: vpblendvb %ymm13, %ymm4, %ymm5, %ymm4
+; AVX2-FAST-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm15[0],ymm14[1],ymm15[2,3],ymm14[4],ymm15[5,6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0],xmm5[1],xmm4[2,3,4,5],xmm5[6],xmm4[7]
; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm4, %xmm4
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm11[0,1],ymm12[2,3],ymm11[4,5],ymm12[6,7]
-; AVX2-FAST-NEXT: vmovdqa %ymm11, %ymm0
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm13[0,1],ymm7[2,3],ymm13[4,5],ymm7[6,7]
; AVX2-FAST-NEXT: vpermd %ymm5, %ymm1, %ymm5
; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm5, %ymm5
; AVX2-FAST-NEXT: vpblendvb %ymm8, %ymm4, %ymm5, %ymm4
-; AVX2-FAST-NEXT: vmovdqa %ymm8, %ymm10
; AVX2-FAST-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa %ymm15, %ymm8
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm15[0],ymm7[1],ymm15[2,3],ymm7[4],ymm15[5,6,7]
+; AVX2-FAST-NEXT: vmovdqa %ymm12, %ymm15
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm12[0],ymm6[1],ymm12[2,3],ymm6[4],ymm12[5,6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0],xmm5[1],xmm4[2,3,4,5],xmm5[6],xmm4[7]
; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm4, %xmm4
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm6[0,1],ymm9[2,3],ymm6[4,5],ymm9[6,7]
-; AVX2-FAST-NEXT: vmovdqa %ymm9, %ymm15
+; AVX2-FAST-NEXT: vmovdqa %ymm11, %ymm9
+; AVX2-FAST-NEXT: vmovdqa %ymm10, %ymm14
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm11[0,1],ymm10[2,3],ymm11[4,5],ymm10[6,7]
; AVX2-FAST-NEXT: vpermd %ymm5, %ymm1, %ymm5
; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm5, %ymm5
-; AVX2-FAST-NEXT: vpblendvb %ymm10, %ymm4, %ymm5, %ymm4
+; AVX2-FAST-NEXT: vpblendvb %ymm8, %ymm4, %ymm5, %ymm4
; AVX2-FAST-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu (%rsp), %ymm11 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0],ymm10[1],ymm11[2,3],ymm10[4],ymm11[5,6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm10[0],ymm6[1],ymm10[2,3],ymm6[4],ymm10[5,6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0],xmm5[1],xmm4[2,3,4,5],xmm5[6],xmm4[7]
; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm4, %xmm3
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm9[2,3],ymm6[4,5],ymm9[6,7]
+; AVX2-FAST-NEXT: vmovdqu (%rsp), %ymm11 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0,1],ymm12[2,3],ymm11[4,5],ymm12[6,7]
; AVX2-FAST-NEXT: vpermd %ymm4, %ymm1, %ymm1
; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-FAST-NEXT: vpblendvb %ymm5, %ymm3, %ymm1, %ymm1
+; AVX2-FAST-NEXT: vpblendvb %ymm8, %ymm3, %ymm1, %ymm1
; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm1 = ymm1[0],mem[1],ymm1[2,3,4],mem[5],ymm1[6,7]
+; AVX2-FAST-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm1 = mem[0],ymm1[1],mem[2,3,4],ymm1[5],mem[6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm3 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm1 = mem[0,1,2],ymm1[3],mem[4,5],ymm1[6],mem[7]
+; AVX2-FAST-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm1 = mem[0,1,2],ymm0[3],mem[4,5],ymm0[6],mem[7]
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm2 = ymm1[1,3,2,3]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = [6,7,6,7,6,7,6,7,8,9,4,5,10,11,0,1,22,23,22,23,22,23,22,23,24,25,20,21,26,27,16,17]
; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm2, %ymm4
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm2 = [6,7,4,5,2,3,0,1,14,15,14,15,14,15,14,15]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-NEXT: vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
-; AVX2-FAST-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm14[0],ymm13[1],ymm14[2,3,4],ymm13[5],ymm14[6,7]
+; AVX2-FAST-NEXT: vpblendvb %ymm8, %ymm3, %ymm4, %ymm0
+; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm3 = mem[0],ymm3[1],mem[2,3,4],ymm3[5],mem[6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm0[0,1,2],ymm12[3],ymm0[4,5],ymm12[6],ymm0[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm13[0,1,2],ymm7[3],ymm13[4,5],ymm7[6],ymm13[7]
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm4 = ymm4[1,3,2,3]
; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-FAST-NEXT: vpblendvb %ymm5, %ymm3, %ymm4, %ymm0
-; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm11[0],ymm10[1],ymm11[2,3,4],ymm10[5],ymm11[6,7]
+; AVX2-FAST-NEXT: vpblendvb %ymm8, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm10[0],ymm6[1],ymm10[2,3,4],ymm6[5],ymm10[6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1,2],ymm9[3],ymm6[4,5],ymm9[6],ymm6[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0,1,2],ymm12[3],ymm11[4,5],ymm12[6],ymm11[7]
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm4 = ymm4[1,3,2,3]
; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-FAST-NEXT: vpblendvb %ymm5, %ymm3, %ymm4, %ymm0
-; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0],ymm7[1],ymm8[2,3,4],ymm7[5],ymm8[6,7]
+; AVX2-FAST-NEXT: vpblendvb %ymm8, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm3 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm3 = ymm15[0],mem[1],ymm15[2,3,4],mem[5],ymm15[6,7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-FAST-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm3 = mem[0,1,2],ymm15[3],mem[4,5],ymm15[6],mem[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm9[0,1,2],ymm14[3],ymm9[4,5],ymm14[6],ymm9[7]
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm3 = ymm3[1,3,2,3]
; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm3, %ymm1
-; AVX2-FAST-NEXT: vpblendvb %ymm5, %ymm2, %ymm1, %ymm0
+; AVX2-FAST-NEXT: vpblendvb %ymm8, %ymm2, %ymm1, %ymm0
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa 128(%rdi), %ymm8
-; AVX2-FAST-NEXT: vmovdqa 160(%rdi), %ymm11
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm11[0,1,2],ymm8[3],ymm11[4,5],ymm8[6],ymm11[7]
+; AVX2-FAST-NEXT: vmovdqa 128(%rdi), %ymm10
+; AVX2-FAST-NEXT: vmovdqa 160(%rdi), %ymm2
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm10[3],ymm2[4,5],ymm10[6],ymm2[7]
+; AVX2-FAST-NEXT: vmovdqa %ymm2, %ymm11
; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm1 = [3,6,2,5,3,6,2,5]
; AVX2-FAST-NEXT: # ymm1 = mem[0,1,0,1]
; AVX2-FAST-NEXT: vpermd %ymm0, %ymm1, %ymm0
@@ -11365,21 +11355,21 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vmovdqa 192(%rdi), %ymm2
; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm5 = ymm2[0,1,0,2]
-; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm13 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
-; AVX2-FAST-NEXT: vpshufb %ymm13, %ymm5, %ymm4
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm12 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
+; AVX2-FAST-NEXT: vpshufb %ymm12, %ymm5, %ymm4
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm4[7]
; AVX2-FAST-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FAST-NEXT: # ymm0 = mem[0,1,2,3,4],ymm0[5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa 576(%rdi), %ymm15
-; AVX2-FAST-NEXT: vmovdqa 608(%rdi), %ymm14
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0,1,2],ymm15[3],ymm14[4,5],ymm15[6],ymm14[7]
+; AVX2-FAST-NEXT: vmovdqa 576(%rdi), %ymm14
+; AVX2-FAST-NEXT: vmovdqa 608(%rdi), %ymm15
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1,2],ymm14[3],ymm15[4,5],ymm14[6],ymm15[7]
; AVX2-FAST-NEXT: vpermd %ymm0, %ymm1, %ymm0
; AVX2-FAST-NEXT: vpshufb %ymm6, %ymm0, %ymm0
; AVX2-FAST-NEXT: vmovdqa 640(%rdi), %ymm2
; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm3 = ymm2[0,1,0,2]
-; AVX2-FAST-NEXT: vpshufb %ymm13, %ymm3, %ymm7
+; AVX2-FAST-NEXT: vpshufb %ymm12, %ymm3, %ymm7
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm7[7]
; AVX2-FAST-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FAST-NEXT: # ymm0 = mem[0,1,2,3,4],ymm0[5,6,7]
@@ -11387,32 +11377,33 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vmovdqa 352(%rdi), %ymm2
; AVX2-FAST-NEXT: vmovdqa 384(%rdi), %ymm4
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0,1,2],ymm2[3],ymm4[4,5],ymm2[6],ymm4[7]
-; AVX2-FAST-NEXT: vmovdqa %ymm4, %ymm10
-; AVX2-FAST-NEXT: vmovdqa %ymm2, %ymm9
+; AVX2-FAST-NEXT: vmovdqa %ymm4, %ymm9
+; AVX2-FAST-NEXT: vmovdqa %ymm2, %ymm8
; AVX2-FAST-NEXT: vpermd %ymm0, %ymm1, %ymm0
; AVX2-FAST-NEXT: vpshufb %ymm6, %ymm0, %ymm0
; AVX2-FAST-NEXT: vmovdqa 416(%rdi), %ymm2
; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,1,0,2]
-; AVX2-FAST-NEXT: vpshufb %ymm13, %ymm2, %ymm7
+; AVX2-FAST-NEXT: vpshufb %ymm12, %ymm2, %ymm7
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm7[7]
; AVX2-FAST-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FAST-NEXT: # ymm0 = mem[0,1,2,3,4],ymm0[5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa 800(%rdi), %ymm12
-; AVX2-FAST-NEXT: vmovdqa 832(%rdi), %ymm7
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1,2],ymm12[3],ymm7[4,5],ymm12[6],ymm7[7]
+; AVX2-FAST-NEXT: vmovdqa 800(%rdi), %ymm4
+; AVX2-FAST-NEXT: vmovdqa 832(%rdi), %ymm13
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1,2],ymm4[3],ymm13[4,5],ymm4[6],ymm13[7]
+; AVX2-FAST-NEXT: vmovdqa %ymm4, %ymm7
; AVX2-FAST-NEXT: vpermd %ymm0, %ymm1, %ymm0
; AVX2-FAST-NEXT: vpshufb %ymm6, %ymm0, %ymm0
; AVX2-FAST-NEXT: vmovdqa 864(%rdi), %ymm1
; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,1,0,2]
-; AVX2-FAST-NEXT: vpshufb %ymm13, %ymm1, %ymm6
+; AVX2-FAST-NEXT: vpshufb %ymm12, %ymm1, %ymm6
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm6[7]
; AVX2-FAST-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
; AVX2-FAST-NEXT: # ymm0 = mem[0,1,2,3,4],ymm0[5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1],ymm11[2],ymm8[3,4,5],ymm11[6],ymm8[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm11[2],ymm10[3,4,5],ymm11[6],ymm10[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm6
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm6[4],xmm0[5],xmm6[6],xmm0[7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,0,1,14,15,12,13,10,11,8,9,12,13,14,15]
@@ -11425,9 +11416,7 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm5[0],ymm0[1,2,3,4,5,6,7],ymm5[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa %ymm14, %ymm13
-; AVX2-FAST-NEXT: vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1],ymm14[2],ymm15[3,4,5],ymm14[6],ymm15[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0,1],ymm15[2],ymm14[3,4,5],ymm15[6],ymm14[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm5
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm5[4],xmm0[5],xmm5[6],xmm0[7]
; AVX2-FAST-NEXT: vpshufb %xmm6, %xmm0, %xmm0
@@ -11438,7 +11427,7 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3,4,5,6,7],ymm3[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1],ymm10[2],ymm9[3,4,5],ymm10[6],ymm9[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1],ymm9[2],ymm8[3,4,5],ymm9[6],ymm8[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm3
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4],xmm0[5],xmm3[6],xmm0[7]
; AVX2-FAST-NEXT: vpshufb %xmm6, %xmm0, %xmm0
@@ -11449,7 +11438,8 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm7[2],ymm12[3,4,5],ymm7[6],ymm12[7]
+; AVX2-FAST-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1],ymm13[2],ymm7[3,4,5],ymm13[6],ymm7[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4],xmm0[5],xmm2[6],xmm0[7]
; AVX2-FAST-NEXT: vpshufb %xmm6, %xmm0, %xmm0
@@ -11460,61 +11450,60 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1,2],ymm11[3],ymm8[4,5],ymm11[6],ymm8[7]
-; AVX2-FAST-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa %ymm8, %ymm4
-; AVX2-FAST-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm11[3],ymm10[4,5],ymm11[6],ymm10[7]
+; AVX2-FAST-NEXT: vmovdqa %ymm10, %ymm4
+; AVX2-FAST-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3,4,5],xmm0[6],xmm1[7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm1 = [0,1,2,3,0,1,14,15,12,13,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm2 = [2,5,2,5,2,5,2,5]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-NEXT: vpermd %ymm14, %ymm2, %ymm3
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FAST-NEXT: vpermd %ymm12, %ymm2, %ymm3
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,4,7,8,9,10,11,12,13,12,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm3[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3,4,5,6,7],ymm3[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0,1,2],ymm13[3],ymm15[4,5],ymm13[6],ymm15[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm14[0,1,2],ymm15[3],ymm14[4,5],ymm15[6],ymm14[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm3
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0],xmm0[1],xmm3[2,3,4,5],xmm0[6],xmm3[7]
; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT: vpermd %ymm5, %ymm2, %ymm3
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT: vpermd %ymm6, %ymm2, %ymm3
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,4,7,8,9,10,11,12,13,12,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm3[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3,4,5,6,7],ymm3[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0,1,2],ymm7[3],ymm12[4,5],ymm7[6],ymm12[7]
-; AVX2-FAST-NEXT: vmovdqa %ymm12, %ymm13
-; AVX2-FAST-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa %ymm7, %ymm8
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1,2],ymm13[3],ymm7[4,5],ymm13[6],ymm7[7]
+; AVX2-FAST-NEXT: vmovdqa %ymm7, %ymm10
; AVX2-FAST-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm3
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0],xmm0[1],xmm3[2,3,4,5],xmm0[6],xmm3[7]
; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-NEXT: vpermd %ymm12, %ymm2, %ymm3
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FAST-NEXT: vpermd %ymm13, %ymm2, %ymm3
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,4,7,8,9,10,11,12,13,12,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm3[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm3[0],ymm0[1,2,3,4,5,6,7],ymm3[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1,2],ymm10[3],ymm9[4,5],ymm10[6],ymm9[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1,2],ymm9[3],ymm8[4,5],ymm9[6],ymm8[7]
+; AVX2-FAST-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm3
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0],xmm0[1],xmm3[2,3,4,5],xmm0[6],xmm3[7]
; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-NEXT: vpermd %ymm6, %ymm2, %ymm1
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FAST-NEXT: vpermd %ymm7, %ymm2, %ymm1
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,4,7,8,9,10,11,12,13,12,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
@@ -11526,356 +11515,354 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5],xmm1[6],xmm0[7]
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm1 = [0,1,4,5,2,3,0,1,14,15,12,13,12,13,14,15]
; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm0, %xmm0
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm3 = ymm14[0,1,1,3]
-; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm2 = [18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25]
-; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm3, %ymm4
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm2
+; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm12[0,1,1,3]
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm3 = [18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25]
+; AVX2-FAST-NEXT: vpshufb %ymm3, %ymm0, %ymm4
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm4[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm4[0],ymm0[1,2,3,4,5,6,7],ymm4[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0],ymm15[1],ymm7[2,3],ymm15[4],ymm7[5,6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm4
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm4[1],xmm0[2,3,4,5],xmm4[6],xmm0[7]
-; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm0, %xmm0
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm4 = ymm5[0,1,1,3]
-; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm4, %ymm5
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm5[7]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm5[0],ymm0[1,2,3,4,5,6,7],ymm5[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0],ymm9[1],ymm10[2,3],ymm9[4],ymm10[5,6,7]
-; AVX2-FAST-NEXT: vmovdqa %ymm10, %ymm14
-; AVX2-FAST-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa %ymm9, %ymm11
-; AVX2-FAST-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm5
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm5[1],xmm0[2,3,4,5],xmm5[6],xmm0[7]
-; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm0, %xmm0
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm5
-; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm6[0,1,1,3]
-; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm0, %ymm6
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm6[7]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm5 = ymm6[0],ymm5[1,2,3,4,5,6,7],ymm6[8],ymm5[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm6[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm8[0],ymm13[1],ymm8[2,3],ymm13[4],ymm8[5,6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm5 = xmm5[0],xmm6[1],xmm5[2,3,4,5],xmm6[6],xmm5[7]
-; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm5, %xmm1
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0],ymm2[1,2,3,4,5,6,7],ymm4[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm15[0],ymm14[1],ymm15[2,3],ymm14[4],ymm15[5,6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm4
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm4[1],xmm2[2,3,4,5],xmm4[6],xmm2[7]
+; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm2, %xmm2
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm4 = ymm6[0,1,1,3]
+; AVX2-FAST-NEXT: vpshufb %ymm3, %ymm4, %ymm5
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6,7],ymm5[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm9[0],ymm8[1],ymm9[2,3],ymm8[4],ymm9[5,6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm5
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm5[1],xmm2[2,3,4,5],xmm5[6],xmm2[7]
+; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm2, %xmm2
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm8 = ymm7[0,1,1,3]
+; AVX2-FAST-NEXT: vpshufb %ymm3, %ymm8, %ymm5
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6,7],ymm5[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm9[0],ymm10[1],ymm9[2,3],ymm10[4],ymm9[5,6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm5
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm5[1],xmm2[2,3,4,5],xmm5[6],xmm2[7]
+; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm2, %xmm1
; AVX2-FAST-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm13 = ymm12[0,1,1,3]
-; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm13, %ymm2
+; AVX2-FAST-NEXT: vpermq {{.*#+}} ymm2 = ymm13[0,1,1,3]
+; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpshufb %ymm3, %ymm2, %ymm2
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0],ymm15[1],ymm7[2,3,4],ymm15[5],ymm7[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
-; AVX2-FAST-NEXT: vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-FAST-NEXT: vmovdqa %xmm2, %xmm8
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm6 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
-; AVX2-FAST-NEXT: vpshufb %ymm6, %ymm4, %ymm2
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3,4,5,6],ymm2[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm1 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm5 = [8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpshufb %xmm5, %xmm1, %xmm4
-; AVX2-FAST-NEXT: vmovdqa %xmm5, %xmm12
+; AVX2-FAST-NEXT: vpblendd $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm1 = mem[0,1],ymm1[2],mem[3,4],ymm1[5],mem[6,7]
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} xmm3 = [8,9,4,5,8,9,4,5,8,9,4,5,8,9,4,5]
+; AVX2-FAST-NEXT: vpshufb %xmm3, %xmm1, %xmm2
+; AVX2-FAST-NEXT: vmovdqa %xmm3, %xmm12
; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm1
; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1],xmm4[2],xmm1[2],xmm4[3],xmm1[3]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm5 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm15[0],ymm14[1],ymm15[2,3,4],ymm14[5],ymm15[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm2
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm13 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
+; AVX2-FAST-NEXT: vpshufb %xmm13, %xmm1, %xmm1
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} ymm10 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
+; AVX2-FAST-NEXT: vpshufb %ymm10, %ymm4, %ymm4
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm1[0,1,2,3,4,5,6],ymm4[7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm5 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm5 = ymm1[0,1],mem[2],ymm1[3,4,5],mem[6],ymm1[7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = <0,3,7,2,6,u,u,u>
-; AVX2-FAST-NEXT: vpermd %ymm5, %ymm1, %ymm5
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1],ymm5[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0],ymm2[1,2,3,4,5,6,7],ymm4[8],ymm2[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm10[0],ymm5[1],ymm10[2,3,4],ymm5[5],ymm10[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm4
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm2 = xmm4[0],xmm2[1],xmm4[2],xmm2[3],xmm4[4,5,6,7]
-; AVX2-FAST-NEXT: vpshufb %xmm8, %xmm2, %xmm2
-; AVX2-FAST-NEXT: vmovdqa %xmm8, %xmm15
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FAST-NEXT: vpshufb %ymm6, %ymm3, %ymm3
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm6 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm6 = mem[0,1],ymm1[2],mem[3,4,5],ymm1[6],mem[7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm2 = <0,3,7,2,6,u,u,u>
+; AVX2-FAST-NEXT: vpermd %ymm6, %ymm2, %ymm6
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm3 = <u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31>
+; AVX2-FAST-NEXT: vpshufb %ymm3, %ymm6, %ymm6
+; AVX2-FAST-NEXT: vmovdqa %ymm3, %ymm1
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm5 = xmm5[0,1],xmm6[2,3]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm6[0],ymm4[1,2,3,4,5,6,7],ymm6[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0,1],ymm7[2],ymm6[3,4],ymm7[5],ymm6[6,7]
-; AVX2-FAST-NEXT: vpshufb %xmm12, %xmm3, %xmm4
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm3, %xmm3
-; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm3 = xmm3[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm9[0,1],ymm8[2],ymm9[3,4,5],ymm8[6],ymm9[7]
-; AVX2-FAST-NEXT: vpermd %ymm4, %ymm1, %ymm4
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm3[0],ymm2[1,2,3,4,5,6,7],ymm3[8],ymm2[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm14[0],ymm11[1],ymm14[2,3,4],ymm11[5],ymm14[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm3
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0],xmm2[1],xmm3[2],xmm2[3],xmm3[4,5,6,7]
-; AVX2-FAST-NEXT: vpshufb %xmm15, %xmm2, %xmm2
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,26,27]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm7[0,1],ymm15[2],ymm7[3,4],ymm15[5],ymm7[6,7]
+; AVX2-FAST-NEXT: vpshufb %xmm12, %xmm4, %xmm5
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm4
+; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
+; AVX2-FAST-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $219, (%rsp), %ymm14, %ymm2 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm2 = mem[0,1],ymm14[2],mem[3,4],ymm14[5],mem[6,7]
-; AVX2-FAST-NEXT: vpshufb %xmm12, %xmm2, %xmm3
-; AVX2-FAST-NEXT: vmovdqa %xmm12, %xmm15
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm2
-; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm11[0],ymm14[1],ymm11[2,3,4],ymm14[5],ymm11[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm5, %xmm6
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm5 = xmm6[0],xmm5[1],xmm6[2],xmm5[3],xmm6[4,5,6,7]
+; AVX2-FAST-NEXT: vpshufb %xmm13, %xmm5, %xmm5
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
+; AVX2-FAST-NEXT: vpshufb %ymm10, %ymm0, %ymm0
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm3[0,1],ymm6[2],ymm3[3,4,5],ymm6[6],ymm3[7]
+; AVX2-FAST-NEXT: vpermd %ymm5, %ymm2, %ymm5
+; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm5, %ymm5
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm5[0],ymm0[1,2,3,4,5,6,7],ymm5[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm11[0,1],ymm12[2],ymm11[3,4,5],ymm12[6],ymm11[7]
-; AVX2-FAST-NEXT: vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31>
-; AVX2-FAST-NEXT: vpshufb %ymm4, %ymm3, %ymm3
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0,1],ymm0[2],ymm12[3,4],ymm0[5],ymm12[6,7]
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} xmm1 = [8,9,4,5,8,9,4,5,8,9,4,5,8,9,4,5]
+; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm0, %xmm4
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
+; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm4 = ymm10[0],mem[1],ymm10[2,3,4],mem[5],ymm10[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2],xmm4[3],xmm5[4,5,6,7]
+; AVX2-FAST-NEXT: vpshufb %xmm13, %xmm4, %xmm4
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm5 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,26,27]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FAST-NEXT: vmovdqu (%rsp), %ymm8 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm5 = mem[0,1],ymm8[2],mem[3,4,5],ymm8[6],mem[7]
+; AVX2-FAST-NEXT: vpermd %ymm5, %ymm2, %ymm5
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm5[0],ymm4[1,2,3,4,5,6,7],ymm5[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm4[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm0 = ymm0[0],mem[1],ymm0[2,3,4],mem[5],ymm0[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm2
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2],xmm0[3],xmm2[4,5,6,7]
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm2 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,26,27]
-; AVX2-FAST-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm2 = ymm2[0,1],mem[2],ymm2[3,4],mem[5],ymm2[6,7]
-; AVX2-FAST-NEXT: vpshufb %xmm15, %xmm2, %xmm3
-; AVX2-FAST-NEXT: vmovdqa %xmm15, %xmm13
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm2
-; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm3 = mem[0,1],ymm3[2],mem[3,4,5],ymm3[6],mem[7]
-; AVX2-FAST-NEXT: vpermd %ymm3, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vpshufb %ymm4, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm0 = ymm0[0,1],mem[2],ymm0[3,4],mem[5],ymm0[6,7]
+; AVX2-FAST-NEXT: vpshufb %xmm1, %xmm0, %xmm4
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm0
+; AVX2-FAST-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm4 = ymm4[0,1],mem[2],ymm4[3,4,5],mem[6],ymm4[7]
+; AVX2-FAST-NEXT: vpermd %ymm4, %ymm2, %ymm1
+; AVX2-FAST-NEXT: vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm4 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2],xmm4[3],xmm5[4,5,6,7]
+; AVX2-FAST-NEXT: vpshufb %xmm13, %xmm4, %xmm2
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,26,27]
+; AVX2-FAST-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,6],ymm4[7]
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0],ymm2[1,2,3,4,5,6,7],ymm1[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = [0,3,3,3,0,3,7,7]
-; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm2 = [0,0,16,17,22,23,24,25,0,0,16,17,22,23,24,25,0,0,16,17,22,23,24,25,0,0,16,17,22,23,24,25]
-; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm0, %ymm4
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1],ymm5[2],ymm10[3,4],ymm5[5],ymm10[6,7]
-; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [2,5,1,4,2,5,1,4]
-; AVX2-FAST-NEXT: # ymm3 = mem[0,1,0,1]
-; AVX2-FAST-NEXT: vpermd %ymm0, %ymm3, %ymm5
-; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm0 = [30,31,16,17,22,23,24,25,30,31,16,17,22,23,24,25,30,31,16,17,22,23,24,25,30,31,16,17,22,23,24,25]
-; AVX2-FAST-NEXT: vpshufb %ymm0, %ymm5, %ymm5
-; AVX2-FAST-NEXT: vmovdqa %ymm0, %ymm15
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm4[5,6,7],ymm5[8,9,10,11,12],ymm4[13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm6[0,1],ymm7[2,3],ymm6[4,5],ymm7[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-FAST-NEXT: vpshufb %xmm13, %xmm6, %xmm6
-; AVX2-FAST-NEXT: vmovdqa %xmm13, %xmm7
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm5 = xmm5[10,11,6,7,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm10 = ymm9[0,1,2],ymm8[3],ymm9[4,5],ymm8[6],ymm9[7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm9 = <0,4,7,3,6,u,u,u>
-; AVX2-FAST-NEXT: vpermd %ymm10, %ymm9, %ymm13
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm0 = [0,3,3,3,0,3,7,7]
+; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm2 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm1 = [0,0,16,17,22,23,24,25,0,0,16,17,22,23,24,25,0,0,16,17,22,23,24,25,0,0,16,17,22,23,24,25]
+; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm2, %ymm2
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0,1],ymm14[2],ymm11[3,4],ymm14[5],ymm11[6,7]
+; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm9 = [2,5,1,4,2,5,1,4]
+; AVX2-FAST-NEXT: # ymm9 = mem[0,1,0,1]
+; AVX2-FAST-NEXT: vpermd %ymm4, %ymm9, %ymm4
+; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm13 = [30,31,16,17,22,23,24,25,30,31,16,17,22,23,24,25,30,31,16,17,22,23,24,25,30,31,16,17,22,23,24,25]
+; AVX2-FAST-NEXT: vpshufb %ymm13, %ymm4, %ymm4
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0,1,2,3,4],ymm2[5,6,7],ymm4[8,9,10,11,12],ymm2[13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm7[0,1],ymm15[2,3],ymm7[4,5],ymm15[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} xmm7 = [8,9,4,5,8,9,4,5,8,9,4,5,8,9,4,5]
+; AVX2-FAST-NEXT: vpshufb %xmm7, %xmm5, %xmm5
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} xmm0 = [10,11,6,7,10,11,6,7,10,11,6,7,10,11,6,7]
+; AVX2-FAST-NEXT: vpshufb %xmm0, %xmm4, %xmm4
+; AVX2-FAST-NEXT: vmovdqa %xmm0, %xmm11
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm3[0,1,2],ymm6[3],ymm3[4,5],ymm6[6],ymm3[7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm3 = <0,4,7,3,6,u,u,u>
+; AVX2-FAST-NEXT: vpermd %ymm5, %ymm3, %ymm5
; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,2,3,4,5,10,11,12,13,18,19,18,19,18,19,18,19,18,19,20,21,26,27,28,29>
-; AVX2-FAST-NEXT: vpshufb %ymm0, %ymm13, %ymm13
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm6[0,1],ymm13[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm6[0],ymm4[1,2,3,4,5,6,7],ymm6[8],ymm4[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-NEXT: vpermd %ymm8, %ymm1, %ymm4
-; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm4, %ymm4
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm6 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm6 = ymm5[0,1],mem[2],ymm5[3,4],mem[5],ymm5[6,7]
-; AVX2-FAST-NEXT: vpermd %ymm6, %ymm3, %ymm6
-; AVX2-FAST-NEXT: vpshufb %ymm15, %ymm6, %ymm6
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm6[0,1,2,3,4],ymm4[5,6,7],ymm6[8,9,10,11,12],ymm4[13,14,15]
+; AVX2-FAST-NEXT: vpshufb %ymm0, %ymm5, %ymm5
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6,7],ymm5[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm6 = [0,3,3,3,0,3,7,7]
+; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm2 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm2, %ymm2
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm4 = mem[0,1],ymm4[2],mem[3,4],ymm4[5],mem[6,7]
+; AVX2-FAST-NEXT: vpermd %ymm4, %ymm9, %ymm4
+; AVX2-FAST-NEXT: vpshufb %ymm13, %ymm4, %ymm4
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0,1,2,3,4],ymm2[5,6,7],ymm4[8,9,10,11,12],ymm2[13,14,15]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm6 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm6 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm6, %xmm13
-; AVX2-FAST-NEXT: vpshufb %xmm7, %xmm13, %xmm13
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm6 = xmm6[10,11,6,7,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm6[0],xmm13[0],xmm6[1],xmm13[1],xmm6[2],xmm13[2],xmm6[3],xmm13[3]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm13 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm13 = ymm10[0,1,2],mem[3],ymm10[4,5],mem[6],ymm10[7]
-; AVX2-FAST-NEXT: vpermd %ymm13, %ymm9, %ymm13
-; AVX2-FAST-NEXT: vpshufb %ymm0, %ymm13, %ymm13
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm6[0,1],ymm13[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm6[0],ymm4[1,2,3,4,5,6,7],ymm6[8],ymm4[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm4 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm4, %ymm4
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm6 = mem[0,1],ymm6[2],mem[3,4],ymm6[5],mem[6,7]
-; AVX2-FAST-NEXT: vpermd %ymm6, %ymm3, %ymm6
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm6 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25,30,31,u,u,u,u,u,u]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm6[0,1,2,3,4],ymm4[5,6,7],ymm6[8,9,10,11,12],ymm4[13,14,15]
-; AVX2-FAST-NEXT: vpblendd $51, (%rsp), %ymm14, %ymm6 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm6 = mem[0,1],ymm14[2,3],mem[4,5],ymm14[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm6, %xmm13
-; AVX2-FAST-NEXT: vpshufb %xmm7, %xmm13, %xmm13
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm7 = [10,11,6,7,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpshufb %xmm7, %xmm6, %xmm6
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm6[0],xmm13[0],xmm6[1],xmm13[1],xmm6[2],xmm13[2],xmm6[3],xmm13[3]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm13 = ymm11[0,1,2],ymm12[3],ymm11[4,5],ymm12[6],ymm11[7]
-; AVX2-FAST-NEXT: vpermd %ymm13, %ymm9, %ymm13
-; AVX2-FAST-NEXT: vpshufb %ymm0, %ymm13, %ymm13
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm6 = ymm6[0,1],ymm13[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm4 = ymm6[0],ymm4[1,2,3,4,5,6,7],ymm6[8],ymm4[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-NEXT: vpermd %ymm7, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vpshufb %ymm2, %ymm1, %ymm1
+; AVX2-FAST-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm4 = mem[0,1],ymm15[2,3],mem[4,5],ymm15[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
+; AVX2-FAST-NEXT: vpshufb %xmm7, %xmm5, %xmm5
+; AVX2-FAST-NEXT: vpshufb %xmm11, %xmm4, %xmm4
+; AVX2-FAST-NEXT: vmovdqa %xmm11, %xmm7
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm11[0,1,2],ymm14[3],ymm11[4,5],ymm14[6],ymm11[7]
+; AVX2-FAST-NEXT: vpermd %ymm5, %ymm3, %ymm5
+; AVX2-FAST-NEXT: vpshufb %ymm0, %ymm5, %ymm5
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6,7],ymm5[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm2 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm2, %ymm2
+; AVX2-FAST-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm4 = ymm10[0,1],mem[2],ymm10[3,4],mem[5],ymm10[6,7]
+; AVX2-FAST-NEXT: vpermd %ymm4, %ymm9, %ymm4
+; AVX2-FAST-NEXT: vpshufb %ymm13, %ymm4, %ymm4
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0,1,2,3,4],ymm2[5,6,7],ymm4[8,9,10,11,12],ymm2[13,14,15]
+; AVX2-FAST-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm4 = ymm12[0,1],mem[2,3],ymm12[4,5],mem[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm4, %xmm5
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm5 = xmm5[8,9,4,5,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT: vpshufb %xmm7, %xmm4, %xmm4
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3]
+; AVX2-FAST-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm5 = mem[0,1,2],ymm8[3],mem[4,5],ymm8[6],mem[7]
+; AVX2-FAST-NEXT: vpermd %ymm5, %ymm3, %ymm5
+; AVX2-FAST-NEXT: vpshufb %ymm0, %ymm5, %ymm5
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6,7],ymm5[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FAST-NEXT: vpermd %ymm8, %ymm6, %ymm2
+; AVX2-FAST-NEXT: vpshufb %ymm1, %ymm2, %ymm1
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm4[2],ymm6[3,4],ymm4[5],ymm6[6,7]
-; AVX2-FAST-NEXT: vpermd %ymm2, %ymm3, %ymm0
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25,30,31,u,u,u,u,u,u]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7],ymm0[8,9,10,11,12],ymm1[13,14,15]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm7[2],ymm4[3,4],ymm7[5],ymm4[6,7]
+; AVX2-FAST-NEXT: vpermd %ymm2, %ymm9, %ymm2
+; AVX2-FAST-NEXT: vpshufb %ymm13, %ymm2, %ymm2
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0,1,2,3,4],ymm1[5,6,7],ymm2[8,9,10,11,12],ymm1[13,14,15]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm12[0,1],ymm14[2,3],ymm12[4,5],ymm14[6,7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm2 = xmm2[8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[10,11,6,7,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm12[0,1,2],ymm6[3],ymm12[4,5],ymm6[6],ymm12[7]
+; AVX2-FAST-NEXT: vpermd %ymm2, %ymm3, %ymm2
+; AVX2-FAST-NEXT: vpshufb %ymm0, %ymm2, %ymm0
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm13[0,1,2],ymm11[3],ymm13[4,5],ymm11[6],ymm13[7]
-; AVX2-FAST-NEXT: vpermd %ymm2, %ymm9, %ymm2
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,2,3,4,5,10,11,12,13,18,19,18,19,18,19,18,19,18,19,20,21,26,27,28,29]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm13[0,1],ymm10[2,3],ymm13[4,5],ymm10[6,7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm2, %xmm3
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm3 = xmm3[8,9,4,5,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} xmm2 = xmm2[10,11,6,7,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm1 = ymm5[0,1],mem[2,3],ymm5[4,5],mem[6,7]
-; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm0 = [0,4,7,0,0,4,7,0]
-; AVX2-FAST-NEXT: # ymm0 = mem[0,1,0,1]
-; AVX2-FAST-NEXT: vpermd %ymm8, %ymm0, %ymm2
-; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm8 = [0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27]
-; AVX2-FAST-NEXT: vpshufb %ymm8, %ymm2, %ymm2
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm0 = mem[0,1],ymm0[2,3],mem[4,5],ymm0[6,7]
+; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm1 = [0,4,7,0,0,4,7,0]
+; AVX2-FAST-NEXT: # ymm1 = mem[0,1,0,1]
+; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm2 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,26,27]
; AVX2-FAST-NEXT: vbroadcasti128 {{.*#+}} ymm9 = [2,6,1,5,2,6,1,5]
; AVX2-FAST-NEXT: # ymm9 = mem[0,1,0,1]
-; AVX2-FAST-NEXT: vpermd %ymm1, %ymm9, %ymm1
-; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,26,27,28,29,u,u,u,u,u,u]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm1[0,1,2,3,4],ymm2[5,6,7],ymm1[8,9,10,11,12],ymm2[13,14,15]
-; AVX2-FAST-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm1 = ymm15[0,1,2],mem[3],ymm15[4,5],mem[6],ymm15[7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm1, %xmm3
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} xmm15 = [0,1,2,3,4,5,6,7,10,11,6,7,4,5,6,7]
-; AVX2-FAST-NEXT: vpshufb %xmm15, %xmm3, %xmm3
-; AVX2-FAST-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,4,6,7]
-; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
-; AVX2-FAST-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm5 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm5 = mem[0],ymm10[1],mem[2,3],ymm10[4],mem[5,6,7]
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm1 = <1,4,0,3,7,u,u,u>
-; AVX2-FAST-NEXT: vpermd %ymm5, %ymm1, %ymm5
-; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31>
-; AVX2-FAST-NEXT: vpshufb %ymm10, %ymm5, %ymm5
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm5[2,3,4,5,6,7]
+; AVX2-FAST-NEXT: vpermd %ymm0, %ymm9, %ymm0
+; AVX2-FAST-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,26,27,28,29,u,u,u,u,u,u]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0,1,2,3,4],ymm2[5,6,7],ymm0[8,9,10,11,12],ymm2[13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm14[0],ymm11[1],ymm14[2,3],ymm11[4],ymm14[5,6,7]
+; AVX2-FAST-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm0 = mem[0,1,2],ymm15[3],mem[4,5],ymm15[6],mem[7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm0, %xmm5
+; AVX2-FAST-NEXT: vpbroadcastd {{.*#+}} xmm11 = [10,11,6,7,10,11,6,7,10,11,6,7,10,11,6,7]
+; AVX2-FAST-NEXT: vpshufb %xmm11, %xmm5, %xmm5
+; AVX2-FAST-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,4,6,7]
+; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm5 = xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm0 = <1,4,0,3,7,u,u,u>
+; AVX2-FAST-NEXT: vpermd %ymm3, %ymm0, %ymm3
+; AVX2-FAST-NEXT: vmovdqa {{.*#+}} ymm14 = <u,u,u,u,u,u,u,u,0,1,6,7,8,9,14,15,16,17,22,23,20,21,22,23,16,17,22,23,24,25,30,31>
+; AVX2-FAST-NEXT: vpshufb %ymm14, %ymm3, %ymm3
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm3[0],ymm2[1,2,3,4,5,6,7],ymm3[8],ymm2[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FAST-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm6[0,1],ymm4[2,3],ymm6[4,5],ymm4[6,7]
-; AVX2-FAST-NEXT: vpermd %ymm7, %ymm0, %ymm3
-; AVX2-FAST-NEXT: vpshufb %ymm8, %ymm3, %ymm3
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm15 = ymm3[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm7[2,3],ymm4[4,5],ymm7[6,7]
+; AVX2-FAST-NEXT: vpermd %ymm8, %ymm1, %ymm3
+; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm7 = [0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27,0,0,18,19,20,21,26,27]
+; AVX2-FAST-NEXT: vpshufb %ymm7, %ymm3, %ymm3
; AVX2-FAST-NEXT: vpermd %ymm2, %ymm9, %ymm2
-; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm4 = [28,29,18,19,20,21,26,27,28,29,18,19,20,21,26,27,28,29,18,19,20,21,26,27,28,29,18,19,20,21,26,27]
-; AVX2-FAST-NEXT: vpshufb %ymm4, %ymm2, %ymm2
+; AVX2-FAST-NEXT: vpbroadcastq {{.*#+}} ymm8 = [28,29,18,19,20,21,26,27,28,29,18,19,20,21,26,27,28,29,18,19,20,21,26,27,28,29,18,19,20,21,26,27]
+; AVX2-FAST-NEXT: vpshufb %ymm8, %ymm2, %ymm2
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7],ymm2[8,9,10,11,12],ymm3[13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm12[0,1,2],ymm14[3],ymm12[4,5],ymm14[6],ymm12[7]
-; AVX2-FAST-NEXT: vextracti128 $1, %ymm3, %xmm5
-; AVX2-FAST-NEXT: vpshufb %xmm15, %xmm5, %xmm5
-; AVX2-FAST-NEXT: vpshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,6,4,6,7]
-; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm3[4],xmm5[4],xmm3[5],xmm5[5],xmm3[6],xmm5[6],xmm3[7],xmm5[7]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm11[0],ymm13[1],ymm11[2,3],ymm13[4],ymm11[5,6,7]
-; AVX2-FAST-NEXT: vpermd %ymm5, %ymm1, %ymm5
-; AVX2-FAST-NEXT: vpshufb %ymm10, %ymm5, %ymm5
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm5[2,3,4,5,6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0],ymm12[1],ymm6[2,3],ymm12[4],ymm6[5,6,7]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm13[0,1,2],ymm10[3],ymm13[4,5],ymm10[6],ymm13[7]
+; AVX2-FAST-NEXT: vextracti128 $1, %ymm5, %xmm6
+; AVX2-FAST-NEXT: vpshufb %xmm11, %xmm6, %xmm6
+; AVX2-FAST-NEXT: vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,6,4,6,7]
+; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm5 = xmm5[4],xmm6[4],xmm5[5],xmm6[5],xmm5[6],xmm6[6],xmm5[7],xmm6[7]
+; AVX2-FAST-NEXT: vpermd %ymm3, %ymm0, %ymm3
+; AVX2-FAST-NEXT: vpshufb %ymm14, %ymm3, %ymm3
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm2 = ymm3[0],ymm2[1,2,3,4,5,6,7],ymm3[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm3 = xmm5[0,1],xmm3[2,3]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm3 = mem[0,1],ymm3[2,3],mem[4,5],ymm3[6,7]
-; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm5 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: vpshufb %ymm8, %ymm5, %ymm5
+; AVX2-FAST-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm3 = ymm3[0,1],mem[2,3],ymm3[4,5],mem[6,7]
+; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: vpshufb %ymm7, %ymm5, %ymm5
+; AVX2-FAST-NEXT: vmovdqa %ymm7, %ymm10
; AVX2-FAST-NEXT: vpermd %ymm3, %ymm9, %ymm3
-; AVX2-FAST-NEXT: vpshufb %ymm4, %ymm3, %ymm3
-; AVX2-FAST-NEXT: vmovdqa %ymm4, %ymm11
+; AVX2-FAST-NEXT: vpshufb %ymm8, %ymm3, %ymm3
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm5[5,6,7],ymm3[8,9,10,11,12],ymm5[13,14,15]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm5 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm5 = mem[0],ymm4[1],mem[2,3],ymm4[4],mem[5,6,7]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm6 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm6 = mem[0,1,2],ymm4[3],mem[4,5],ymm4[6],mem[7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm5 = mem[0],ymm5[1],mem[2,3],ymm5[4],mem[5,6,7]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm6 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm6 = ymm6[0,1,2],mem[3],ymm6[4,5],mem[6],ymm6[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm6, %xmm7
-; AVX2-FAST-NEXT: vpshufb %xmm15, %xmm7, %xmm7
+; AVX2-FAST-NEXT: vpshufb %xmm11, %xmm7, %xmm7
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,6,4,6,7]
; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm6 = xmm6[4],xmm7[4],xmm6[5],xmm7[5],xmm6[6],xmm7[6],xmm6[7],xmm7[7]
-; AVX2-FAST-NEXT: vpermd %ymm5, %ymm1, %ymm5
-; AVX2-FAST-NEXT: vpshufb %ymm10, %ymm5, %ymm5
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm5 = ymm6[0,1],ymm5[2,3,4,5,6,7]
+; AVX2-FAST-NEXT: vpermd %ymm5, %ymm0, %ymm5
+; AVX2-FAST-NEXT: vpshufb %ymm14, %ymm5, %ymm5
; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm3 = ymm5[0],ymm3[1,2,3,4,5,6,7],ymm5[8],ymm3[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3]
; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: vpshufb %ymm8, %ymm0, %ymm0
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm5 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm5 = mem[0,1],ymm4[2,3],mem[4,5],ymm4[6,7]
+; AVX2-FAST-NEXT: vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: vpshufb %ymm10, %ymm1, %ymm1
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm5 = ymm5[0,1],mem[2,3],ymm5[4,5],mem[6,7]
; AVX2-FAST-NEXT: vpermd %ymm5, %ymm9, %ymm5
-; AVX2-FAST-NEXT: vpshufb %ymm11, %ymm5, %ymm5
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm5[0,1,2,3,4],ymm0[5,6,7],ymm5[8,9,10,11,12],ymm0[13,14,15]
+; AVX2-FAST-NEXT: vpshufb %ymm8, %ymm5, %ymm5
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm5[0,1,2,3,4],ymm1[5,6,7],ymm5[8,9,10,11,12],ymm1[13,14,15]
+; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-NEXT: vpblendd $237, (%rsp), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm5 = mem[0],ymm5[1],mem[2,3],ymm5[4],mem[5,6,7]
+; AVX2-FAST-NEXT: vpermd %ymm5, %ymm0, %ymm0
+; AVX2-FAST-NEXT: vpshufb %ymm14, %ymm0, %ymm0
; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $183, (%rsp), %ymm4, %ymm5 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm5 = mem[0,1,2],ymm4[3],mem[4,5],ymm4[6],mem[7]
+; AVX2-FAST-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-NEXT: # ymm5 = ymm4[0,1,2],mem[3],ymm4[4,5],mem[6],ymm4[7]
; AVX2-FAST-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-FAST-NEXT: vpshufb %xmm15, %xmm6, %xmm4
+; AVX2-FAST-NEXT: vpshufb %xmm11, %xmm6, %xmm4
; AVX2-FAST-NEXT: vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,6,4,6,7]
; AVX2-FAST-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
-; AVX2-FAST-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
-; AVX2-FAST-NEXT: # ymm5 = ymm5[0],mem[1],ymm5[2,3],mem[4],ymm5[5,6,7]
-; AVX2-FAST-NEXT: vpermd %ymm5, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vpshufb %ymm10, %ymm1, %ymm1
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm1 = ymm4[0,1],ymm1[2,3,4,5,6,7]
-; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} xmm0 = xmm4[0,1],xmm0[2,3]
+; AVX2-FAST-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FAST-NEXT: vmovaps %ymm1, 96(%rsi)
; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
@@ -11929,21 +11916,20 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-NEXT: vmovdqa %ymm0, 32(%rax)
; AVX2-FAST-NEXT: vmovdqa %ymm3, (%rax)
; AVX2-FAST-NEXT: vmovdqa %ymm2, 96(%rax)
-; AVX2-FAST-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-NEXT: vmovaps %ymm0, 64(%rax)
+; AVX2-FAST-NEXT: vmovdqa %ymm15, 64(%rax)
; AVX2-FAST-NEXT: addq $1544, %rsp # imm = 0x608
; AVX2-FAST-NEXT: vzeroupper
; AVX2-FAST-NEXT: retq
;
; AVX2-FAST-PERLANE-LABEL: load_i16_stride7_vf64:
; AVX2-FAST-PERLANE: # %bb.0:
-; AVX2-FAST-PERLANE-NEXT: subq $1528, %rsp # imm = 0x5F8
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 256(%rdi), %ymm15
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 224(%rdi), %ymm9
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 512(%rdi), %ymm7
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 544(%rdi), %ymm5
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 480(%rdi), %ymm11
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 448(%rdi), %ymm10
+; AVX2-FAST-PERLANE-NEXT: subq $1512, %rsp # imm = 0x5E8
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 256(%rdi), %ymm14
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 224(%rdi), %ymm15
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 512(%rdi), %ymm6
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 544(%rdi), %ymm10
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 480(%rdi), %ymm5
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 448(%rdi), %ymm12
; AVX2-FAST-PERLANE-NEXT: vmovdqa (%rdi), %ymm2
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vmovdqa 32(%rdi), %ymm4
@@ -11961,44 +11947,44 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4],xmm0[5],xmm2[6],xmm0[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,14,15,12,13,10,11,8,9,8,9,8,9,8,9]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm0, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm14 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm4, %ymm3, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm11 = [255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm4, %ymm3, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm11, %ymm9
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm10[0,1],ymm11[2],ymm10[3,4,5],ymm11[6],ymm10[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm12[0,1],ymm5[2],ymm12[3,4,5],ymm5[6],ymm12[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm5, %ymm11
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm5[0],ymm7[1],ymm5[2,3,4],ymm7[5],ymm5[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm10[0],ymm6[1],ymm10[2,3,4],ymm6[5],ymm10[6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm3, %ymm4, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm9, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm9[0,1],ymm15[2],ymm9[3,4,5],ymm15[6],ymm9[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm15, %ymm12
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm9, %ymm3, %ymm4, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm15[0,1],ymm14[2],ymm15[3,4,5],ymm14[6],ymm15[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 288(%rdi), %ymm8
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 320(%rdi), %ymm13
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm13[0],ymm8[1],ymm13[2,3,4],ymm8[5],ymm13[6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 288(%rdi), %ymm7
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 320(%rdi), %ymm8
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm8[0],ymm7[1],ymm8[2,3,4],ymm7[5],ymm8[6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm13, %ymm6
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm3, %ymm4, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm9, %ymm5
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm9, %ymm3, %ymm4, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vmovdqa 704(%rdi), %ymm3
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 672(%rdi), %ymm15
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm15[0,1],ymm3[2],ymm15[3,4,5],ymm3[6],ymm15[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 672(%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm0[0,1],ymm3[2],ymm0[3,4,5],ymm3[6],ymm0[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm4[4],xmm3[5],xmm4[6],xmm3[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm2
@@ -12008,41 +11994,43 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm3, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm2, %ymm1, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm5[0,1],ymm7[2],ymm5[3,4],ymm7[5],ymm5[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm5, %ymm2, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm10[0,1],ymm6[2],ymm10[3,4],ymm6[5],ymm10[6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm1[2,3,0,1]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3],ymm1[4,5,6,7,8,9,10],ymm2[11],ymm1[12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm10[0,1,2],ymm11[3],ymm10[4,5],ymm11[6],ymm10[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm12[0,1,2],ymm11[3],ymm12[4,5],ymm11[6],ymm12[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm3
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm2[1],xmm3[2,3,4,5],xmm2[6],xmm3[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm2 = [2,3,2,3,2,3,2,3,8,9,8,9,6,7,4,5,18,19,18,19,18,19,18,19,24,25,24,25,22,23,20,21]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm1, %ymm4
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm1 = [2,3,0,1,14,15,12,13,10,11,10,11,10,11,10,11]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm3, %ymm4, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0,1],ymm8[2],ymm6[3,4],ymm8[5],ymm6[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm5, %ymm3, %ymm4, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm5, %ymm11
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0,1],ymm7[2],ymm8[3,4],ymm7[5],ymm8[6,7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm3[2,3,0,1]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2],ymm4[3],ymm3[4,5,6,7,8,9,10],ymm4[11],ymm3[12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm0[0,1,2],ymm12[3],ymm0[4,5],ymm12[6],ymm0[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm15[0,1,2],ymm14[3],ymm15[4,5],ymm14[6],ymm15[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2,3,4,5],xmm4[6],xmm5[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm3, %ymm3
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm4, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm4, %ymm3, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm4, %ymm3, %ymm0
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm9[0,1],ymm13[2],ymm9[3,4],ymm13[5],ymm9[6,7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm3[2,3,0,1]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2],ymm4[3],ymm3[4,5,6,7,8,9,10],ymm4[11],ymm3[12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm15[0,1,2],ymm11[3],ymm15[4,5],ymm11[6],ymm15[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm15, %ymm10
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm12[0,1,2],ymm14[3],ymm12[4,5],ymm14[6],ymm12[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm5
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2,3,4,5],xmm4[6],xmm5[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm3, %ymm3
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm4, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm4, %ymm3, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm4, %ymm3, %ymm0
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vmovdqu (%rsp), %ymm9 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
@@ -12051,241 +12039,240 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2],ymm4[3],ymm3[4,5,6,7,8,9,10],ymm4[11],ymm3[12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm3, %ymm2
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0,1,2],ymm7[3],ymm5[4,5],ymm7[6],ymm5[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0,1,2],ymm7[3],ymm8[4,5],ymm7[6],ymm8[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2,3,4,5],xmm3[6],xmm4[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm3, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm1, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm1, %ymm2, %ymm0
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm1 = ymm6[2,3,0,1]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm9[0,1],ymm6[2,3],ymm9[4,5],ymm6[6,7]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm1[4],ymm2[5,6,7,8,9,10,11],ymm1[12],ymm2[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0],ymm5[1],ymm7[2,3],ymm5[4],ymm7[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm7, %ymm9
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0],ymm8[1],ymm7[2,3],ymm8[4],ymm7[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm8, %ymm9
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm7, %ymm0
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm3
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm1[0],xmm3[1],xmm1[2,3,4,5],xmm3[6],xmm1[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm1 = [4,5,4,5,4,5,4,5,8,9,10,11,8,9,6,7,20,21,20,21,20,21,20,21,24,25,26,27,24,25,22,23]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm2, %ymm4
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm2 = [4,5,2,3,0,1,14,15,12,13,12,13,12,13,12,13]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm14, %ymm6
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm3, %ymm4, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm0[0],ymm14[1],ymm0[2,3],ymm14[4],ymm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0],ymm15[1],ymm6[2,3],ymm15[4],ymm6[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2,3,4,5],xmm4[6],xmm3[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm8[0,1],ymm5[2,3],ymm8[4,5],ymm5[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm10[0,1],ymm5[2,3],ymm10[4,5],ymm5[6,7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm5[4],ymm4[5,6,7,8,9,10,11],ymm5[12],ymm4[13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm3
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm6, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm3, %ymm4, %ymm3
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm11, %ymm15
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm10, %ymm11
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm15[0],ymm10[1],ymm15[2,3],ymm10[4],ymm15[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm12, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm14, %ymm8
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm14[0],ymm12[1],ymm14[2,3],ymm12[4],ymm14[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2,3,4,5],xmm4[6],xmm3[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm13, %ymm12
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm7[0,1],ymm13[2,3],ymm7[4,5],ymm13[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm13[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm13[0,1],ymm12[2,3],ymm13[4,5],ymm12[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm12[2,3,0,1]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm5[4],ymm4[5,6,7,8,9,10,11],ymm5[12],ymm4[13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm3
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm6, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm11, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm11, %ymm14
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0],ymm10[1],ymm5[2,3],ymm10[4],ymm5[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2,3,4,5],xmm4[6],xmm3[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm2
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm13[0,1],ymm12[2,3],ymm13[4,5],ymm12[6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm11[0,1],ymm12[2,3],ymm11[4,5],ymm12[6,7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm12[2,3,0,1]
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2,3],ymm4[4],ymm3[5,6,7,8,9,10,11],ymm4[12],ymm3[13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm3, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm6, %ymm2, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm2, %ymm1, %ymm1
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd $34, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm1 = ymm9[0],mem[1],ymm9[2,3,4],mem[5],ymm9[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0],ymm9[1],ymm0[2,3,4],ymm9[5],ymm0[6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm2
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $183, (%rsp), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm1 = mem[0,1,2],ymm1[3],mem[4,5],ymm1[6],mem[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $183, (%rsp), %ymm0, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm1 = mem[0,1,2],ymm0[3],mem[4,5],ymm0[6],mem[7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm1[1,3,2,3]
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm1 = [6,7,6,7,6,7,6,7,8,9,4,5,10,11,0,1,22,23,22,23,22,23,22,23,24,25,20,21,26,27,16,17]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm2, %ymm4
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm2 = [6,7,4,5,2,3,0,1,14,15,14,15,14,15,14,15]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm6, %ymm3, %ymm4, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm0[0],ymm14[1],ymm0[2,3,4],ymm14[5],ymm0[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm3, %ymm4, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm6[0],ymm15[1],ymm6[2,3,4],ymm15[5],ymm6[6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm4 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm4 = ymm8[0,1,2],mem[3],ymm8[4,5],mem[6],ymm8[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm4 = mem[0,1,2],ymm0[3],mem[4,5],ymm0[6],mem[7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm4[1,3,2,3]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm6, %ymm3, %ymm4, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm5[0],ymm10[1],ymm5[2,3,4],ymm10[5],ymm5[6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm13[0,1,2],ymm12[3],ymm13[4,5],ymm12[6],ymm13[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0,1,2],ymm12[3],ymm11[4,5],ymm12[6],ymm11[7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm4[1,3,2,3]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm4, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm6, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm3, %ymm4, %ymm3
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm15[0],ymm11[1],ymm15[2,3,4],ymm11[5],ymm15[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0],ymm7[1],ymm8[2,3,4],ymm7[5],ymm8[6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm3 = ymm7[0,1,2],mem[3],ymm7[4,5],mem[6],ymm7[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm3 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm3 = ymm13[0,1,2],mem[3],ymm13[4,5],mem[6],ymm13[7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm3 = ymm3[1,3,2,3]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm3, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm6, %ymm2, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendvb %ymm14, %ymm2, %ymm1, %ymm0
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 128(%rdi), %ymm9
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 160(%rdi), %ymm13
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0,1,2],ymm9[3],ymm13[4,5],ymm9[6],ymm13[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 128(%rdi), %ymm8
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 160(%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm8[3],ymm0[4,5],ymm8[6],ymm0[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm4 = [0,1,2,3,4,5,6,7,8,9,10,11,6,7,6,7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm1, %xmm1
; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,4,7]
; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %ymm1
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm12 = ymm1[0,1,0,2]
-; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm15 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm15, %ymm12, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm12 = ymm0[0,1,0,2]
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm10 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm12, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0,1,2,3,4],ymm1[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 576(%rdi), %ymm6
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 576(%rdi), %ymm11
; AVX2-FAST-PERLANE-NEXT: vmovdqa 608(%rdi), %ymm7
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm7[0,1,2],ymm6[3],ymm7[4,5],ymm6[6],ymm7[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0,1,2],ymm11[3],ymm7[4,5],ymm11[6],ymm7[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm1, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,4,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 640(%rdi), %ymm1
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm1[0,1,0,2]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm15, %ymm2, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm3[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm3, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,4,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 640(%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm0[0,1,0,2]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm2, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0,1,2,3,4],ymm1[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 352(%rdi), %ymm10
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 384(%rdi), %ymm5
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm10[3],ymm5[4,5],ymm10[6],ymm5[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 352(%rdi), %ymm5
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 384(%rdi), %ymm6
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm6[0,1,2],ymm5[3],ymm6[4,5],ymm5[6],ymm6[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm3
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,4,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 416(%rdi), %ymm1
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,1,0,2]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm15, %ymm1, %ymm8
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm8[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0,1,2,3,4],ymm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,4,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 416(%rdi), %ymm0
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,0,2]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm0, %ymm9
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm9[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm1 = mem[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vmovdqa 800(%rdi), %ymm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 832(%rdi), %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3],ymm0[4,5],ymm3[6],ymm0[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm3, %ymm8
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm11
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm11, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,4,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 864(%rdi), %ymm14
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm11 = ymm14[0,1,0,2]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm15, %ymm11, %ymm15
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm15[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm6[0,1],ymm7[2],ymm6[3,4,5],ymm7[6],ymm6[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm15
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm15[4],xmm4[5],xmm15[6],xmm4[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm15 = [0,1,0,1,14,15,12,13,10,11,8,9,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm15, %xmm4, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 832(%rdi), %ymm9
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm9[0,1,2],ymm3[3],ymm9[4,5],ymm3[6],ymm9[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm15
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm15, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,4,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 864(%rdi), %ymm14
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm15 = ymm14[0,1,0,2]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm10, %ymm15, %ymm13
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd $31, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm1 = mem[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0,1],ymm7[2],ymm11[3,4,5],ymm7[6],ymm11[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm13
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm13[4],xmm4[5],xmm13[6],xmm4[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm13 = [0,1,0,1,14,15,12,13,10,11,8,9,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm4, %xmm4
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm0 = [22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm1 = [22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29,22,23,28,29]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm2, %ymm2
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0],ymm2[1,2,3,4,5,6,7],ymm4[8],ymm2[9,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm10[0,1],ymm5[2],ymm10[3,4,5],ymm5[6],ymm10[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm6[2],ymm5[3,4,5],ymm6[6],ymm5[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm4
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm4[4],xmm2[5],xmm4[6],xmm2[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm15, %xmm2, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm2, %xmm2
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm1, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5,6],ymm1[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm8[0,1],ymm3[2],ymm8[3,4,5],ymm3[6],ymm8[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4],xmm1[5],xmm2[6],xmm1[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm15, %xmm1, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm11, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3,4,5,6],ymm0[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm9[2],ymm3[3,4,5],ymm9[6],ymm3[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm9, %ymm7
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm9[0,1],ymm13[2],ymm9[3,4,5],ymm13[6],ymm9[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4],xmm1[5],xmm2[6],xmm1[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm15, %xmm1, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm12, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4],xmm0[5],xmm2[6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm15, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1],ymm6[2],ymm8[3,4,5],ymm6[6],ymm8[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4],xmm0[5],xmm2[6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm12, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1,2],ymm13[3],ymm9[4,5],ymm13[6],ymm9[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm13, %ymm9
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1,2],ymm6[3],ymm8[4,5],ymm6[6],ymm8[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3,4,5],xmm0[6],xmm1[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm1 = [0,1,2,3,0,1,14,15,12,13,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm5[0,1,1,2]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm10[0,1,1,2]
; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm4 = [16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31,16,17,30,31]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm2, %ymm2
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
@@ -12293,24 +12280,25 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm6, %ymm12
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm11, %ymm5
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1,2],ymm15[3],ymm6[4,5],ymm15[6],ymm6[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm11[0,1,2],ymm15[3],ymm11[4,5],ymm15[6],ymm11[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5],xmm0[6],xmm2[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm6[0,1,1,2]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm3[0,1,1,2]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm2, %ymm2
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm3, %ymm11
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm8[0,1,2],ymm3[3],ymm8[4,5],ymm3[6],ymm8[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm7, %ymm13
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0,1,2],ymm7[3],ymm9[4,5],ymm7[6],ymm9[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5],xmm0[6],xmm2[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm0, %xmm0
@@ -12322,14 +12310,14 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm10[0,1,2],ymm13[3],ymm10[4,5],ymm13[6],ymm10[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm11[0,1,2],ymm12[3],ymm11[4,5],ymm12[6],ymm11[7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5],xmm0[6],xmm2[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm0, %xmm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm1 = ymm3[0,1,1,2]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm1 = ymm7[0,1,1,2]
; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm1, %ymm1
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7]
@@ -12337,401 +12325,407 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm9[0],ymm7[1],ymm9[2,3],ymm7[4],ymm9[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0],ymm8[1],ymm6[2,3],ymm8[4],ymm6[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm1
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3,4,5],xmm1[6],xmm0[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm1 = [0,1,4,5,2,3,0,1,14,15,12,13,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm5[0,1,1,3]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm2 = [18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm4, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm4[0],ymm0[1,2,3,4,5,6,7],ymm4[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm10[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm4 = [18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25,18,19,24,25]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0],ymm12[1],ymm15[2,3],ymm12[4],ymm15[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm12, %ymm9
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm4[1],xmm0[2,3,4,5],xmm4[6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm15[0],ymm5[1],ymm15[2,3],ymm5[4],ymm15[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm15, %ymm8
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm5, %ymm10
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2,3,4,5],xmm2[6],xmm0[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm0, %xmm0
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm6[0,1,1,3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm4, %ymm6
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm3[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm5, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0],ymm11[1],ymm12[2,3],ymm11[4],ymm12[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2,3,4,5],xmm2[6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm7[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm2, %ymm6
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm6[7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm6[0],ymm0[1,2,3,4,5,6,7],ymm6[8],ymm0[9,10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0],ymm10[1],ymm13[2,3],ymm10[4],ymm13[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm13[0],ymm9[1],ymm13[2,3],ymm9[4],ymm13[5,6,7]
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm6
; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm6[1],xmm0[2,3,4,5],xmm6[6],xmm0[7]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm0, %xmm0
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm13 = ymm3[0,1,1,3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm13, %ymm6
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm6[7]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm0 = ymm14[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm4, %ymm0, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm3[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm3[0],ymm1[1,2,3,4,5,6,7],ymm3[8],ymm1[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm1 = ymm1[0,1],mem[2],ymm1[3,4],mem[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} xmm4 = [8,9,4,5,8,9,4,5,8,9,4,5,8,9,4,5]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm1, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm4, %xmm9
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm12[0,1],ymm15[2],ymm12[3,4,5],ymm15[6],ymm12[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27>
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm4, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm4[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm7[0,1,2,3,4,5],ymm4[6],ymm7[7,8,9,10,11,12,13],ymm4[14],ymm7[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm4 = xmm3[0,1],xmm4[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm8[0],ymm10[1],ymm8[2,3,4],ymm10[5],ymm8[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm13
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm13 = xmm13[0],xmm3[1],xmm13[2],xmm3[3],xmm13[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm13, %xmm13
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm13, %ymm0, %ymm13
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm6 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm6, %ymm5, %ymm5
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm13[0,1,2,3,4,5,6],ymm5[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm7[0],ymm5[1,2,3,4,5,6,7],ymm7[8],ymm5[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm4 = ymm4[0,1],mem[2],ymm4[3,4],mem[5],ymm4[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm9, %xmm4, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm5[0],ymm0[1,2,3,4,5,6,7],ymm5[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm11[0],ymm8[1],ymm11[2,3],ymm8[4],ymm11[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm6[1],xmm0[2,3,4,5],xmm6[6],xmm0[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm0, %xmm0
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm1 = ymm14[0,1,1,3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm1, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd $68, (%rsp), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm5 = ymm5[0,1],mem[2],ymm5[3,4,5],mem[6],ymm5[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm5, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm7[0,1,2,3,4,5],ymm5[6],ymm7[7,8,9,10,11,12,13],ymm5[14],ymm7[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm5 = mem[0],ymm5[1],mem[2,3,4],ymm5[5],mem[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm13
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm5 = xmm13[0],xmm5[1],xmm13[2],xmm5[3],xmm13[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm5, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm8[2],ymm5[3,4,5],ymm8[6],ymm5[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27>
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm2, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm2[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7,8,9,10,11,12,13],ymm2[14],ymm3[15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm12[0,1],ymm7[2],ymm12[3,4],ymm7[5],ymm12[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm6 = xmm3[8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm3 = xmm3[3,1,2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm6[0],xmm3[0],xmm6[1],xmm3[1],xmm6[2],xmm3[2],xmm6[3],xmm3[3]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm15[0],ymm9[1],ymm15[2,3,4],ymm9[5],ymm15[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm6[0],xmm3[1],xmm6[2],xmm3[3],xmm6[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm6 = [0,1,6,7,4,5,2,3,0,1,14,15,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} ymm14 = [20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27,20,21,26,27]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm14, %ymm4, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm2[0],ymm3[1,2,3,4,5,6,7],ymm2[8],ymm3[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm6, %ymm8, %ymm13
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,6],ymm13[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm7[0],ymm5[1,2,3,4,5,6,7],ymm7[8],ymm5[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm4[0,1],ymm14[2],ymm4[3,4],ymm14[5],ymm4[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm9, %xmm4, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm9, %xmm8
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm10[0,1],ymm6[2],ymm10[3,4,5],ymm6[6],ymm10[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm5, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm7[0,1,2,3,4,5],ymm5[6],ymm7[7,8,9,10,11,12,13],ymm5[14],ymm7[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm5[2,3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm11, %ymm9
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm10[0,1],ymm11[2],ymm10[3,4,5],ymm11[6],ymm10[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm2, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm2[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7,8,9,10,11,12,13],ymm2[14],ymm3[15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm15[0,1],ymm9[2],ymm15[3,4],ymm9[5],ymm15[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm0 = [8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm0, %xmm3, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm3 = xmm3[3,1,2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm3 = mem[0],ymm3[1],mem[2,3,4],ymm3[5],mem[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm14, %ymm13, %ymm4
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,6],ymm4[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm2[0],ymm3[1,2,3,4,5,6,7],ymm2[8],ymm3[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm3[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm11[0],ymm9[1],ymm11[2,3,4],ymm9[5],ymm11[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm13
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm5 = xmm13[0],xmm5[1],xmm13[2],xmm5[3],xmm13[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm5, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,26,27]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0,1,2,3,4,5,6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm7[0],ymm2[1,2,3,4,5,6,7],ymm7[8],ymm2[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm2 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm2 = ymm13[0,1],mem[2],ymm13[3,4,5],mem[6],ymm13[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm3 = ymm2[u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm2[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5],ymm2[6],ymm3[7,8,9,10,11,12,13],ymm2[14],ymm3[15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm3 = mem[0,1],ymm3[2],mem[3,4],ymm3[5],mem[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm0, %xmm3, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm3 = xmm3[3,1,2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm3 = mem[0],ymm3[1],mem[2,3,4],ymm3[5],mem[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm3, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0],xmm3[1],xmm4[2],xmm3[3],xmm4[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm14, %ymm1, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5,6],ymm1[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm1 = mem[0],ymm1[1],mem[2,3,4],ymm1[5],mem[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm1, %xmm1
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm14, %ymm2, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu (%rsp), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $187, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm2 = mem[0,1],ymm2[2],mem[3,4,5],ymm2[6],mem[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,0,1,14,15,u,u,10,11,24,25,24,25,24,25,24,25,16,17,30,31,u,u,26,27]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm2[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6],ymm0[7,8,9,10,11,12,13],ymm2[14],ymm0[15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm14[0,1],ymm4[2],ymm14[3,4],ymm4[5],ymm14[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm6 = [8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm2, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm2 = ymm2[0,1],mem[2],ymm2[3,4],mem[5],ymm2[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm8, %xmm2, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm8, %xmm7
; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm2
; AVX2-FAST-PERLANE-NEXT: vpshuflw {{.*#+}} xmm2 = xmm2[3,1,2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm2 = xmm4[0],xmm2[0],xmm4[1],xmm2[1],xmm4[2],xmm2[2],xmm4[3],xmm2[3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm8[0,1],ymm13[2],ymm8[3,4,5],ymm13[6],ymm8[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm1, %ymm4, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm4[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm1[0,1,2,3,4,5],ymm4[6],ymm1[7,8,9,10,11,12,13],ymm4[14],ymm1[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm4[2,3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $221, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm4 = mem[0],ymm4[1],mem[2,3,4],ymm4[5],mem[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0],xmm4[1],xmm5[2],xmm4[3],xmm5[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,26,27]
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3,4,5,6],ymm0[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm5[0,1,2],ymm8[3],ymm5[4,5],ymm8[6],ymm5[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm12[0,1,2],ymm15[3],ymm12[4,5],ymm15[6],ymm12[7]
; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7,8],ymm0[9,10,11,12,13,14],ymm1[15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm12[0,1],ymm7[2,3],ymm12[4,5],ymm7[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm2, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm6, %xmm7
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[10,11,6,7,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm2 = <u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29>
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm2, %ymm8
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0,1],ymm12[2],mem[3,4],ymm12[5],mem[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2],xmm1[3],xmm0[4],xmm1[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm0 = [8,9,8,9,6,7,4,5,2,3,2,3,2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm0, %xmm1, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm5
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 656(%rdi), %xmm1
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7,8],ymm0[9,10,11,12,13,14],ymm1[15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $36, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm0 = ymm0[0,1],mem[2],ymm0[3,4],mem[5],ymm0[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0,1],xmm0[2],xmm2[3],xmm0[4],xmm2[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm15 = [8,9,8,9,6,7,4,5,2,3,2,3,2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm15, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 656(%rdi), %xmm0
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX2-FAST-PERLANE-NEXT: vmovdqa 640(%rdi), %xmm3
; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm6 = xmm3[0,1,2,3,4,5],xmm1[6],xmm3[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm1 = [0,1,2,3,4,5,6,7,0,1,0,1,14,15,12,13]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm6, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm2[0],ymm5[1,2,3,4,5,6,7],ymm2[8],ymm5[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm10[0,1,2],ymm11[3],ymm10[4,5],ymm11[6],ymm10[7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm2[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6],ymm5[7,8],ymm2[9,10,11,12,13,14],ymm5[15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm15[0,1],ymm9[2,3],ymm15[4,5],ymm9[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm7, %xmm6, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm7, %xmm15
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm11 = [10,11,6,7,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm11, %xmm5, %xmm5
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm8, %ymm2, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm5[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $219, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm5 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm5 = mem[0,1],ymm3[2],mem[3,4],ymm3[5],mem[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm5 = xmm6[0,1],xmm5[2],xmm6[3],xmm5[4],xmm6[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm0, %xmm5, %xmm5
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 432(%rdi), %xmm3
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5],xmm0[6],xmm3[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm0 = [0,1,2,3,4,5,6,7,0,1,0,1,14,15,12,13]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm0, %xmm3, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm0, %xmm12
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm2[0,1,2,3,4],ymm3[5,6,7],ymm2[8,9,10,11,12],ymm3[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm2 = ymm2[0,1],mem[2,3],ymm2[4,5],mem[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm7, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm7, %xmm3, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} xmm3 = [10,11,6,7,10,11,6,7,10,11,6,7,10,11,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm2, %xmm2
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm7 = xmm2[0],xmm7[0],xmm2[1],xmm7[1],xmm2[2],xmm7[2],xmm2[3],xmm7[3]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm2 = <u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29>
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm1[0],ymm4[1,2,3,4,5,6,7],ymm1[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm1 = xmm7[0,1],xmm1[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm10[0,1,2],ymm6[3],ymm10[4,5],ymm6[6],ymm10[7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm1[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm4[0],ymm1[1,2,3,4,5,6],ymm4[7,8],ymm1[9,10,11,12,13,14],ymm4[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm11[0,1],ymm9[2],ymm11[3,4],ymm9[5],ymm11[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm7[0,1],xmm4[2],xmm7[3],xmm4[4],xmm7[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm15, %xmm4, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 432(%rdi), %xmm5
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX2-FAST-PERLANE-NEXT: vmovdqa 416(%rdi), %xmm6
; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,5],xmm3[6],xmm6[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm6, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm2[0],ymm5[1,2,3,4,5,6,7],ymm2[8],ymm5[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm7 = xmm6[0,1,2,3,4,5],xmm5[6],xmm6[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm12, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm12, %xmm7, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm7, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm7[5,6,7],ymm4[8,9,10,11,12],ymm7[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm7 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm7 = mem[0,1],ymm14[2,3],mem[4,5],ymm14[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm7, %xmm12
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm0, %xmm6
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm0, %xmm12, %xmm12
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm7, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm7 = xmm7[0],xmm12[0],xmm7[1],xmm12[1],xmm7[2],xmm12[2],xmm7[3],xmm12[3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm1[0],ymm4[1,2,3,4,5,6,7],ymm1[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm1 = xmm7[0,1],xmm1[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm8[0,1,2],ymm13[3],ymm8[4,5],ymm13[6],ymm8[7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm1[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm7 = ymm4[0],ymm1[1,2,3,4,5,6],ymm4[7,8],ymm1[9,10,11,12,13,14],ymm4[15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm13[0,1,2],ymm9[3],ymm13[4,5],ymm9[6],ymm13[7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm2[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm5[0],ymm2[1,2,3,4,5,6],ymm5[7,8],ymm2[9,10,11,12,13,14],ymm5[15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm9[0,1],ymm8[2],ymm9[3,4],ymm8[5],ymm9[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm4[0,1],xmm1[2],xmm4[3],xmm1[4],xmm4[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm15, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm12
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 880(%rdi), %xmm4
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 864(%rdi), %xmm1
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm13 = xmm1[0,1,2,3,4,5],xmm4[6],xmm1[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm13, %xmm13
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm5, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm13, %ymm0, %ymm13
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm12 = ymm12[0,1,2,3,4],ymm13[5,6,7],ymm12[8,9,10,11,12],ymm13[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm7[0,1],ymm10[2,3],ymm7[4,5],ymm10[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm15, %xmm6, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm11, %xmm5, %xmm5
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm8, %ymm2, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm5[0,1],ymm2[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm8[2],ymm3[3,4],ymm8[5],ymm3[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm6[0,1],xmm2[2],xmm6[3],xmm2[4],xmm6[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm0, %xmm2, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm13
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 880(%rdi), %xmm2
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 864(%rdi), %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm15 = xmm6[0,1,2,3,4,5],xmm2[6],xmm6[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm15, %xmm15
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm15, %ymm0, %ymm15
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm13 = ymm13[0,1,2,3,4],ymm15[5,6,7],ymm13[8,9,10,11,12],ymm15[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm13 = ymm5[0],ymm13[1,2,3,4,5,6,7],ymm5[8],ymm13[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm13[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm13 = ymm14[0,1],ymm10[2,3],ymm14[4,5],ymm10[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm13, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm5, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm13, %xmm13
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm5 = xmm13[0],xmm5[0],xmm13[1],xmm5[1],xmm13[2],xmm5[2],xmm13[3],xmm5[3]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm7, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm12 = ymm7[0],ymm12[1,2,3,4,5,6,7],ymm7[8],ymm12[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm5 = xmm5[0,1],xmm7[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm12[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm11[0,1],ymm15[2],ymm11[3,4],ymm15[5],ymm11[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm13
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm5 = xmm13[0,1],xmm5[2],xmm13[3],xmm5[4],xmm13[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm0, %xmm5, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 208(%rdi), %xmm5
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %xmm0
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm5 = xmm0[0,1,2,3,4,5],xmm5[6],xmm0[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm1, %xmm5, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7],ymm0[8,9,10,11,12],ymm1[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm14[0,1],ymm4[2,3],ymm14[4,5],ymm4[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm5
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm5 = xmm5[8,9,4,5,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[10,11,6,7,4,5,6,7,8,9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm5[0],xmm1[1],xmm5[1],xmm1[2],xmm5[2],xmm1[3],xmm5[3]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu (%rsp), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm4 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm4 = mem[0,1,2],ymm2[3],mem[4,5],ymm2[6],mem[7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm5 = ymm4[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm5[0],ymm4[1,2,3,4,5,6],ymm5[7,8],ymm4[9,10,11,12,13,14],ymm5[15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,u,u,u,u,2,3,0,1,14,15,12,13,26,27,26,27,26,27,26,27,18,19,16,17,30,31,28,29]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm4[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7],ymm1[8],ymm0[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm12[0,1],ymm11[2],ymm12[3,4],ymm11[5],ymm12[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm5 = xmm7[0,1],xmm5[2],xmm7[3],xmm5[4],xmm7[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm15, %xmm5, %xmm5
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 208(%rdi), %xmm7
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vmovdqa 192(%rdi), %xmm13
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm7 = xmm13[0,1,2,3,4,5],xmm7[6],xmm13[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm0, %xmm7, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm5[0,1,2,3,4],ymm0[5,6,7],ymm5[8,9,10,11,12],ymm0[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu (%rsp), %ymm5 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm5 = mem[0,1,2],ymm5[3],mem[4,5],ymm5[6],mem[7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm7 = ymm5[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm5 = ymm7[0],ymm5[1,2,3,4,5,6],ymm7[7,8],ymm5[9,10,11,12,13,14],ymm7[15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm5, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm5 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm5 = ymm15[0,1],mem[2,3],ymm15[4,5],mem[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm5, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm6, %xmm7, %xmm6
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm5, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vpunpcklwd {{.*#+}} xmm3 = xmm3[0],xmm6[0],xmm3[1],xmm6[1],xmm3[2],xmm6[2],xmm3[3],xmm6[3]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1,2,3,4,5,6,7],ymm2[8],ymm0[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm0 = ymm0[0],mem[1],ymm0[2,3],mem[4],ymm0[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3,4,5,6,7,8],ymm1[9],ymm0[10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm1 = mem[0,1,2],ymm1[3],mem[4,5],ymm1[6],mem[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpshufb {{.*#+}} xmm4 = xmm4[0,1,2,3,4,5,6,7,10,11,6,7,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,4,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm14 = <u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31>
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm14, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm1 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm1 = mem[0,1],ymm12[2,3],mem[4,5],ymm12[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm4
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm4[0,1,2],xmm1[3],xmm4[4],xmm1[5],xmm4[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendw $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2, %xmm5 # 16-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # xmm5 = xmm2[0],mem[1],xmm2[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm4 = [8,9,10,11,8,9,6,7,4,5,4,5,4,5,4,5]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm1, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0],ymm0[1],mem[2,3],ymm0[4],mem[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm0[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0],ymm2[1],ymm0[2,3,4,5,6,7,8],ymm2[9],ymm0[10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm0 = ymm0[0,1],mem[2,3],ymm0[4,5],mem[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0,1,2],xmm0[3],xmm3[4],xmm0[5],xmm3[6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendw $2, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm5 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # xmm5 = xmm3[0],mem[1],xmm3[2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm3 = [8,9,10,11,8,9,6,7,4,5,4,5,4,5,4,5]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm13 = [0,1,2,3,4,5,6,7,0,1,2,3,0,1,14,15]
; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm5, %xmm5
; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm5, %ymm0, %ymm5
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm5[5,6,7],ymm1[8,9,10,11,12],ymm5[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm5 = ymm0[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm0 = ymm9[0],mem[1],ymm9[2,3],mem[4],ymm9[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3,4,5,6,7,8],ymm1[9],ymm0[10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm7[0,1,2],ymm10[3],ymm7[4,5],ymm10[6],ymm7[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm7
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} xmm9 = [0,1,2,3,4,5,6,7,10,11,6,7,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm9, %xmm7, %xmm7
-; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,4,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm7[4],xmm1[5],xmm7[5],xmm1[6],xmm7[6],xmm1[7],xmm7[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm14, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1],ymm8[2,3],ymm3[4,5],ymm8[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm1, %xmm7
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm7[0,1,2],xmm1[3],xmm7[4],xmm1[5],xmm7[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw $253, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6, %xmm2 # 16-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # xmm2 = mem[0],xmm6[1],mem[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm1, %xmm1
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm2, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7],ymm1[8,9,10,11,12],ymm2[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0],ymm1[1,2,3,4,5,6,7],ymm0[8],ymm1[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm6 = ymm0[0,1,2,3,4],ymm5[5,6,7],ymm0[8,9,10,11,12],ymm5[13,14,15]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm0 = ymm0[0],mem[1],ymm0[2,3],mem[4],ymm0[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm2 = ymm0[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2,3,4,5,6,7,8],ymm2[9],ymm0[10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm2 = mem[0,1,2],ymm2[3],mem[4,5],ymm2[6],mem[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm9, %xmm6, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,6,4,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm2 = xmm2[4],xmm6[4],xmm2[5],xmm6[5],xmm2[6],xmm6[6],xmm2[7],xmm6[7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm14, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd $183, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm0 = mem[0,1,2],ymm0[3],mem[4,5],ymm0[6],mem[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm0, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpbroadcastd {{.*#+}} xmm5 = [10,11,6,7,10,11,6,7,10,11,6,7,10,11,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm7, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,4,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm7 = xmm0[4],xmm7[4],xmm0[5],xmm7[5],xmm0[6],xmm7[6],xmm0[7],xmm7[7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,4,5,2,3,0,1,14,15,28,29,28,29,28,29,28,29,20,21,18,19,16,17,30,31>
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm6 = ymm2[0],ymm6[1,2,3,4,5,6,7],ymm2[8],ymm6[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm2 = xmm7[0,1],xmm2[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm6 = ymm2[0,1,2,3],ymm6[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $51, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm2 = mem[0,1],ymm2[2,3],mem[4,5],ymm2[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm6[0,1,2],xmm2[3],xmm6[4],xmm2[5],xmm6[6,7]
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendw $253, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm6 # 16-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # xmm6 = mem[0],xmm3[1],mem[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm2, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm6, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm6, %ymm0, %ymm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm6[5,6,7],ymm2[8,9,10,11,12],ymm6[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm0[0],ymm2[1,2,3,4,5,6,7],ymm0[8],ymm2[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm11[0,1],ymm15[2,3],ymm11[4,5],ymm15[6,7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm2, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm2 = xmm6[0,1,2],xmm2[3],xmm6[4],xmm2[5],xmm6[6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm4, %xmm2, %xmm2
-; AVX2-FAST-PERLANE-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendw $253, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm4 # 16-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # xmm4 = mem[0],xmm3[1],mem[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm4, %xmm3
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm2
-; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0,1,2,3,4],ymm3[5,6,7],ymm2[8,9,10,11,12],ymm3[13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vmovdqu (%rsp), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm3 = ymm3[0],mem[1],ymm3[2,3],mem[4],ymm3[5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm3[2,3,0,1]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0],ymm4[1],ymm3[2,3,4,5,6,7,8],ymm4[9],ymm3[10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm14, %ymm3, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm2 = mem[0],ymm2[1],mem[2,3],ymm2[4],mem[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm7 = ymm2[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm2[0],ymm7[1],ymm2[2,3,4,5,6,7,8],ymm7[9],ymm2[10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm7 = ymm9[0,1],ymm8[2,3],ymm9[4,5],ymm8[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm7, %xmm8
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm7 = xmm8[0,1,2],xmm7[3],xmm8[4],xmm7[5],xmm8[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm1 = xmm4[0],xmm1[1],xmm4[2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm7, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm4[0,1,2,3,4],ymm1[5,6,7],ymm4[8,9,10,11,12],ymm1[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm14[0,1,2],ymm10[3],ymm14[4,5],ymm10[6],ymm14[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm7, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,6,4,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm4[4],xmm7[4],xmm4[5],xmm7[5],xmm4[6],xmm7[6],xmm4[7],xmm7[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1,2,3,4,5,6,7],ymm2[8],ymm1[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm2 = xmm4[0,1],xmm2[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $237, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm1 = mem[0],ymm1[1],mem[2,3],ymm1[4],mem[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm4 = ymm1[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm1 = ymm1[0],ymm4[1],ymm1[2,3,4,5,6,7,8],ymm4[9],ymm1[10,11,12,13,14,15]
; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT: # ymm4 = ymm4[0,1,2],mem[3],ymm4[4,5],mem[6],ymm4[7]
-; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm6
-; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm9, %xmm6, %xmm6
+; AVX2-FAST-PERLANE-NEXT: vpblendd $204, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm4 = ymm4[0,1],mem[2,3],ymm4[4,5],mem[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm7[0,1,2],xmm4[3],xmm7[4],xmm4[5],xmm7[6,7]
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendw $253, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7, %xmm7 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # xmm7 = mem[0],xmm7[1],mem[2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm4, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm7, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm7, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0,1,2,3,4],ymm7[5,6,7],ymm4[8,9,10,11,12],ymm7[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm7 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm7 = ymm7[0,1,2],mem[3],ymm7[4,5],mem[6],ymm7[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm7, %xmm8
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm8, %xmm8
+; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,6,4,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm7 = xmm7[4],xmm8[4],xmm7[5],xmm8[5],xmm7[6],xmm8[6],xmm7[7],xmm8[7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm1[0],ymm4[1,2,3,4,5,6,7],ymm1[8],ymm4[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm1 = xmm7[0,1],xmm1[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm4 = ymm12[0,1],ymm11[2,3],ymm12[4,5],ymm11[6,7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} xmm4 = xmm7[0,1,2],xmm4[3],xmm7[4],xmm4[5],xmm7[6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm3, %xmm4, %xmm3
+; AVX2-FAST-PERLANE-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendw $253, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # xmm4 = mem[0],xmm4[1],mem[2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm13, %xmm4, %xmm4
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-FAST-PERLANE-NEXT: vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm4[5,6,7],ymm3[8,9,10,11,12],ymm4[13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vmovdqu (%rsp), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT: vpblendd $18, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm4 = ymm4[0],mem[1],ymm4[2,3],mem[4],ymm4[5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm7 = ymm4[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0],ymm7[1],ymm4[2,3,4,5,6,7,8],ymm7[9],ymm4[10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpshufb %ymm0, %ymm4, %ymm0
+; AVX2-FAST-PERLANE-NEXT: vpblendd $72, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT: # ymm4 = ymm15[0,1,2],mem[3],ymm15[4,5],mem[6],ymm15[7]
+; AVX2-FAST-PERLANE-NEXT: vextracti128 $1, %ymm4, %xmm7
+; AVX2-FAST-PERLANE-NEXT: vpshufb %xmm5, %xmm7, %xmm5
; AVX2-FAST-PERLANE-NEXT: vpshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,6,4,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm4[4],xmm6[4],xmm4[5],xmm6[5],xmm4[6],xmm6[6],xmm4[7],xmm6[7]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm2 = ymm3[0],ymm2[1,2,3,4,5,6,7],ymm3[8],ymm2[9,10,11,12,13,14,15]
-; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm4[4],xmm5[4],xmm4[5],xmm5[5],xmm4[6],xmm5[6],xmm4[7],xmm5[7]
+; AVX2-FAST-PERLANE-NEXT: vpblendw {{.*#+}} ymm3 = ymm0[0],ymm3[1,2,3,4,5,6,7],ymm0[8],ymm3[9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} xmm0 = xmm4[0,1],xmm0[2,3]
+; AVX2-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm3[4,5,6,7]
; AVX2-FAST-PERLANE-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vmovaps %ymm3, 96(%rsi)
; AVX2-FAST-PERLANE-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
@@ -12782,11 +12776,11 @@ define void @load_i16_stride7_vf64(ptr %in.vec, ptr %out.vec0, ptr %out.vec1, pt
; AVX2-FAST-PERLANE-NEXT: vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
; AVX2-FAST-PERLANE-NEXT: vmovaps %ymm3, (%rax)
; AVX2-FAST-PERLANE-NEXT: movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm0, 32(%rax)
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm2, (%rax)
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm1, 96(%rax)
-; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm5, 64(%rax)
-; AVX2-FAST-PERLANE-NEXT: addq $1528, %rsp # imm = 0x5F8
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm1, 32(%rax)
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm0, (%rax)
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm2, 96(%rax)
+; AVX2-FAST-PERLANE-NEXT: vmovdqa %ymm6, 64(%rax)
+; AVX2-FAST-PERLANE-NEXT: addq $1512, %rsp # imm = 0x5E8
; AVX2-FAST-PERLANE-NEXT: vzeroupper
; AVX2-FAST-PERLANE-NEXT: retq
;
More information about the llvm-commits
mailing list