[llvm] [X86] combineConcatVectorOps - convert ISD::VECTOR_SHUFFLE concatenation to use combineConcatVectorOps recursion (PR #130610)

Simon Pilgrim via llvm-commits llvm-commits at lists.llvm.org
Tue Mar 11 05:08:47 PDT 2025


https://github.com/RKSimon updated https://github.com/llvm/llvm-project/pull/130610

>From 1548a2db92715e9be1d8f4b05b47914e00292435 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev at redking.me.uk>
Date: Mon, 10 Mar 2025 09:08:52 +0000
Subject: [PATCH] [X86] combineConcatVectorOps - convert ISD::VECTOR_SHUFFLE
 concatenation to use combineConcatVectorOps recursion

Only concatenate ISD::VECTOR_SHUFFLE nodes if at least one operand is beneficial to concatenate
---
 llvm/lib/Target/X86/X86ISelLowering.cpp       |  38 +--
 llvm/test/CodeGen/X86/gfni-rotates.ll         |   6 +-
 llvm/test/CodeGen/X86/known-bits-vector.ll    |   8 +-
 llvm/test/CodeGen/X86/matrix-multiply.ll      |  88 ++++--
 llvm/test/CodeGen/X86/mulvi32.ll              |   8 +-
 llvm/test/CodeGen/X86/vector-fshr-rot-128.ll  |   6 +-
 .../vector-interleaved-store-i8-stride-3.ll   |  33 ++-
 .../test/CodeGen/X86/vector-shuffle-256-v4.ll |  19 +-
 llvm/test/CodeGen/X86/widen_bitcnt.ll         | 256 ++++++++----------
 .../CodeGen/X86/x86-interleaved-access.ll     |  56 ++--
 10 files changed, 267 insertions(+), 251 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 5e98ef70c578f..5e86b5ad28e6d 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -57971,24 +57971,28 @@ static SDValue combineConcatVectorOps(const SDLoc &DL, MVT VT,
       // TODO: Relax VBMI requirement for repeated shuffle ops - currently
       // limited to targets that should always have good cross lane shuffles.
       if (!IsSplat && NumOps == 2 && VT.is256BitVector() &&
-          (EltSizeInBits >= 32 || Subtarget.hasInt256()) &&
-          (IsConcatFree(VT, Ops, 0) || IsConcatFree(VT, Ops, 1) ||
-           (Ops[0].getOperand(0) == Ops[1].getOperand(0) &&
-            Ops[0].getOperand(1) == Ops[1].getOperand(1) &&
-            Subtarget.hasVBMI()))) {
-        int NumSubElts = Op0.getValueType().getVectorNumElements();
-        SmallVector<int> NewMask;
-        for (int M : cast<ShuffleVectorSDNode>(Ops[0])->getMask()) {
-          M = M >= NumSubElts ? M + NumSubElts : M;
-          NewMask.push_back(M);
-        }
-        for (int M : cast<ShuffleVectorSDNode>(Ops[1])->getMask()) {
-          if (0 <= M)
-            M = (M >= NumSubElts ? M + NumSubElts : M) + NumSubElts;
-          NewMask.push_back(M);
+          (EltSizeInBits >= 32 || Subtarget.hasInt256())) {
+        SDValue Concat0 = CombineSubOperand(VT, Ops, 0);
+        SDValue Concat1 = CombineSubOperand(VT, Ops, 1);
+        if (Concat0 || Concat1 ||
+            (Ops[0].getOperand(0) == Ops[1].getOperand(0) &&
+             Ops[0].getOperand(1) == Ops[1].getOperand(1) &&
+             Subtarget.hasVBMI())) {
+          int NumSubElts = Op0.getValueType().getVectorNumElements();
+          SmallVector<int> NewMask;
+          for (int M : cast<ShuffleVectorSDNode>(Ops[0])->getMask()) {
+            M = M >= NumSubElts ? M + NumSubElts : M;
+            NewMask.push_back(M);
+          }
+          for (int M : cast<ShuffleVectorSDNode>(Ops[1])->getMask()) {
+            if (0 <= M)
+              M = (M >= NumSubElts ? M + NumSubElts : M) + NumSubElts;
+            NewMask.push_back(M);
+          }
+          Concat0 = Concat0 ? Concat0 : ConcatSubOperand(VT, Ops, 0);
+          Concat1 = Concat1 ? Concat1 : ConcatSubOperand(VT, Ops, 1);
+          return DAG.getVectorShuffle(VT, DL, Concat0, Concat1, NewMask);
         }
-        return DAG.getVectorShuffle(VT, DL, ConcatSubOperand(VT, Ops, 0),
-                                    ConcatSubOperand(VT, Ops, 1), NewMask);
       }
       break;
     }
diff --git a/llvm/test/CodeGen/X86/gfni-rotates.ll b/llvm/test/CodeGen/X86/gfni-rotates.ll
index 5fd4dfa7cc262..967f26f70946a 100644
--- a/llvm/test/CodeGen/X86/gfni-rotates.ll
+++ b/llvm/test/CodeGen/X86/gfni-rotates.ll
@@ -255,9 +255,9 @@ define <16 x i8> @splatvar_rotr_v16i8(<16 x i8> %a, <16 x i8> %amt) nounwind {
 ;
 ; GFNIAVX512BW-LABEL: splatvar_rotr_v16i8:
 ; GFNIAVX512BW:       # %bb.0:
-; GFNIAVX512BW-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; GFNIAVX512BW-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; GFNIAVX512BW-NEXT:    vinserti128 $1, %xmm0, %ymm2, %ymm0
+; GFNIAVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; GFNIAVX512BW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; GFNIAVX512BW-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
 ; GFNIAVX512BW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
 ; GFNIAVX512BW-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; GFNIAVX512BW-NEXT:    vpmovwb %ymm0, %xmm0
diff --git a/llvm/test/CodeGen/X86/known-bits-vector.ll b/llvm/test/CodeGen/X86/known-bits-vector.ll
index dbf3d6635fb92..261908fafc06e 100644
--- a/llvm/test/CodeGen/X86/known-bits-vector.ll
+++ b/llvm/test/CodeGen/X86/known-bits-vector.ll
@@ -384,19 +384,19 @@ declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>)
 define <8 x float> @knownbits_mask_concat_uitofp(<4 x i32> %a0, <4 x i32> %a1) nounwind {
 ; X86-LABEL: knownbits_mask_concat_uitofp:
 ; X86:       # %bb.0:
-; X86-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2,0,2]
-; X86-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[1,3,1,3]
+; X86-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; X86-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; X86-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
+; X86-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,2,0,2,5,7,5,7]
 ; X86-NEXT:    vcvtdq2ps %ymm0, %ymm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: knownbits_mask_concat_uitofp:
 ; X64:       # %bb.0:
-; X64-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2,0,2]
-; X64-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[1,3,1,3]
+; X64-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; X64-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; X64-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; X64-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,2,0,2,5,7,5,7]
 ; X64-NEXT:    vcvtdq2ps %ymm0, %ymm0
 ; X64-NEXT:    retq
   %1 = and <4 x i32> %a0, <i32 131071, i32 -1, i32 131071, i32 -1>
diff --git a/llvm/test/CodeGen/X86/matrix-multiply.ll b/llvm/test/CodeGen/X86/matrix-multiply.ll
index 1ee03c5f1223f..d723ec849f328 100644
--- a/llvm/test/CodeGen/X86/matrix-multiply.ll
+++ b/llvm/test/CodeGen/X86/matrix-multiply.ll
@@ -974,35 +974,65 @@ define <16 x float> @test_mul4x4_f32(<16 x float> %a0, <16 x float> %a1) nounwin
 ; SSE-NEXT:    movaps %xmm5, %xmm2
 ; SSE-NEXT:    retq
 ;
-; AVX1OR2-LABEL: test_mul4x4_f32:
-; AVX1OR2:       # %bb.0: # %entry
-; AVX1OR2-NEXT:    vshufps {{.*#+}} ymm4 = ymm2[1,1,1,1,5,5,5,5]
-; AVX1OR2-NEXT:    vperm2f128 {{.*#+}} ymm5 = ymm0[2,3,2,3]
-; AVX1OR2-NEXT:    vmulps %ymm4, %ymm5, %ymm4
-; AVX1OR2-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm6
-; AVX1OR2-NEXT:    vshufps {{.*#+}} ymm0 = ymm2[0,0,0,0,4,4,4,4]
-; AVX1OR2-NEXT:    vmulps %ymm0, %ymm6, %ymm0
-; AVX1OR2-NEXT:    vaddps %ymm4, %ymm0, %ymm0
-; AVX1OR2-NEXT:    vinsertf128 $1, %xmm1, %ymm1, %ymm4
-; AVX1OR2-NEXT:    vshufps {{.*#+}} ymm7 = ymm2[2,2,2,2,6,6,6,6]
-; AVX1OR2-NEXT:    vmulps %ymm7, %ymm4, %ymm7
-; AVX1OR2-NEXT:    vaddps %ymm7, %ymm0, %ymm0
-; AVX1OR2-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[3,3,3,3,7,7,7,7]
-; AVX1OR2-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX1OR2-NEXT:    vmulps %ymm2, %ymm1, %ymm2
-; AVX1OR2-NEXT:    vaddps %ymm2, %ymm0, %ymm0
-; AVX1OR2-NEXT:    vshufps {{.*#+}} ymm2 = ymm3[1,1,1,1,5,5,5,5]
-; AVX1OR2-NEXT:    vmulps %ymm2, %ymm5, %ymm2
-; AVX1OR2-NEXT:    vshufps {{.*#+}} ymm5 = ymm3[0,0,0,0,4,4,4,4]
-; AVX1OR2-NEXT:    vmulps %ymm5, %ymm6, %ymm5
-; AVX1OR2-NEXT:    vaddps %ymm2, %ymm5, %ymm2
-; AVX1OR2-NEXT:    vshufps {{.*#+}} ymm5 = ymm3[2,2,2,2,6,6,6,6]
-; AVX1OR2-NEXT:    vmulps %ymm5, %ymm4, %ymm4
-; AVX1OR2-NEXT:    vaddps %ymm4, %ymm2, %ymm2
-; AVX1OR2-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[3,3,3,3,7,7,7,7]
-; AVX1OR2-NEXT:    vmulps %ymm3, %ymm1, %ymm1
-; AVX1OR2-NEXT:    vaddps %ymm1, %ymm2, %ymm1
-; AVX1OR2-NEXT:    retq
+; AVX1-LABEL: test_mul4x4_f32:
+; AVX1:       # %bb.0: # %entry
+; AVX1-NEXT:    vshufps {{.*#+}} ymm4 = ymm2[1,1,1,1,5,5,5,5]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm5 = ymm0[2,3,2,3]
+; AVX1-NEXT:    vmulps %ymm4, %ymm5, %ymm4
+; AVX1-NEXT:    vshufps {{.*#+}} ymm6 = ymm2[0,0,0,0,4,4,4,4]
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm7
+; AVX1-NEXT:    vmulps %ymm6, %ymm7, %ymm0
+; AVX1-NEXT:    vaddps %ymm4, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm4 = ymm2[2,2,2,2,6,6,6,6]
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm1, %ymm6
+; AVX1-NEXT:    vmulps %ymm4, %ymm6, %ymm4
+; AVX1-NEXT:    vaddps %ymm4, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[3,3,3,3,7,7,7,7]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX1-NEXT:    vmulps %ymm2, %ymm1, %ymm2
+; AVX1-NEXT:    vaddps %ymm2, %ymm0, %ymm0
+; AVX1-NEXT:    vshufps {{.*#+}} ymm2 = ymm3[1,1,1,1,5,5,5,5]
+; AVX1-NEXT:    vmulps %ymm2, %ymm5, %ymm2
+; AVX1-NEXT:    vshufps {{.*#+}} ymm4 = ymm3[0,0,0,0,4,4,4,4]
+; AVX1-NEXT:    vmulps %ymm4, %ymm7, %ymm4
+; AVX1-NEXT:    vaddps %ymm2, %ymm4, %ymm2
+; AVX1-NEXT:    vshufps {{.*#+}} ymm4 = ymm3[2,2,2,2,6,6,6,6]
+; AVX1-NEXT:    vmulps %ymm4, %ymm6, %ymm4
+; AVX1-NEXT:    vaddps %ymm4, %ymm2, %ymm2
+; AVX1-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[3,3,3,3,7,7,7,7]
+; AVX1-NEXT:    vmulps %ymm3, %ymm1, %ymm1
+; AVX1-NEXT:    vaddps %ymm1, %ymm2, %ymm1
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: test_mul4x4_f32:
+; AVX2:       # %bb.0: # %entry
+; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm2[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm5 = ymm0[2,3,2,3]
+; AVX2-NEXT:    vmulps %ymm4, %ymm5, %ymm4
+; AVX2-NEXT:    vshufps {{.*#+}} ymm6 = ymm2[0,0,0,0,4,4,4,4]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm7 = ymm0[0,1,0,1]
+; AVX2-NEXT:    vmulps %ymm6, %ymm7, %ymm0
+; AVX2-NEXT:    vaddps %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm2[2,2,2,2,6,6,6,6]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm6 = ymm1[0,1,0,1]
+; AVX2-NEXT:    vmulps %ymm4, %ymm6, %ymm4
+; AVX2-NEXT:    vaddps %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm2[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-NEXT:    vmulps %ymm2, %ymm1, %ymm2
+; AVX2-NEXT:    vaddps %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm2 = ymm3[1,1,1,1,5,5,5,5]
+; AVX2-NEXT:    vmulps %ymm2, %ymm5, %ymm2
+; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm3[0,0,0,0,4,4,4,4]
+; AVX2-NEXT:    vmulps %ymm4, %ymm7, %ymm4
+; AVX2-NEXT:    vaddps %ymm2, %ymm4, %ymm2
+; AVX2-NEXT:    vshufps {{.*#+}} ymm4 = ymm3[2,2,2,2,6,6,6,6]
+; AVX2-NEXT:    vmulps %ymm4, %ymm6, %ymm4
+; AVX2-NEXT:    vaddps %ymm4, %ymm2, %ymm2
+; AVX2-NEXT:    vshufps {{.*#+}} ymm3 = ymm3[3,3,3,3,7,7,7,7]
+; AVX2-NEXT:    vmulps %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vaddps %ymm1, %ymm2, %ymm1
+; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_mul4x4_f32:
 ; AVX512:       # %bb.0: # %entry
diff --git a/llvm/test/CodeGen/X86/mulvi32.ll b/llvm/test/CodeGen/X86/mulvi32.ll
index 46d5da54a7482..bbda4d68bb685 100644
--- a/llvm/test/CodeGen/X86/mulvi32.ll
+++ b/llvm/test/CodeGen/X86/mulvi32.ll
@@ -286,12 +286,8 @@ define <4 x i64> @_mul4xi32toi64c(<4 x i32>, <4 x i32>) {
 ;
 ; AVX2-LABEL: _mul4xi32toi64c:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm3, %ymm1
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm2, %ymm0
+; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
+; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; AVX2-NEXT:    vpmuludq %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
   %lower0 = shufflevector <4 x i32> %0, <4 x i32> undef, <4 x i32> <i32 0, i32 undef, i32 1, i32 undef>
diff --git a/llvm/test/CodeGen/X86/vector-fshr-rot-128.ll b/llvm/test/CodeGen/X86/vector-fshr-rot-128.ll
index f45405d885377..9ce682306f18b 100644
--- a/llvm/test/CodeGen/X86/vector-fshr-rot-128.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-rot-128.ll
@@ -1194,9 +1194,9 @@ define <16 x i8> @splatvar_funnnel_v16i8(<16 x i8> %x, <16 x i8> %amt) nounwind
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; AVX512VLBW-NEXT:    vinserti128 $1, %xmm0, %ymm2, %ymm0
+; AVX512VLBW-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; AVX512VLBW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
 ; AVX512VLBW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-3.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-3.ll
index ba1621c67f480..7f2210742e7f1 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-3.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-3.ll
@@ -671,15 +671,16 @@ define void @store_i8_stride3_vf16(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm3[5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4]
 ; AVX2-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm3[0,1,2,3,4]
 ; AVX2-NEXT:    vpalignr {{.*#+}} xmm1 = xmm4[5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4]
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5]
-; AVX2-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[5,6,7,8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4]
-; AVX2-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
 ; AVX2-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm4[0,1,2,3,4]
+; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5,0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5]
+; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
 ; AVX2-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vmovdqa %xmm0, 16(%rcx)
-; AVX2-NEXT:    vmovdqa %xmm1, (%rcx)
+; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-NEXT:    vmovdqa %xmm2, 32(%rcx)
+; AVX2-NEXT:    vmovdqa %ymm0, (%rcx)
+; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX2-FP-LABEL: store_i8_stride3_vf16:
@@ -693,15 +694,16 @@ define void @store_i8_stride3_vf16(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX2-FP-NEXT:    vpalignr {{.*#+}} xmm0 = xmm3[5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4]
 ; AVX2-FP-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm3[0,1,2,3,4]
 ; AVX2-FP-NEXT:    vpalignr {{.*#+}} xmm1 = xmm4[5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4]
-; AVX2-FP-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5]
-; AVX2-FP-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX2-FP-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[5,6,7,8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4]
-; AVX2-FP-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
 ; AVX2-FP-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm4[0,1,2,3,4]
+; AVX2-FP-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5,0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5]
+; AVX2-FP-NEXT:    # ymm3 = mem[0,1,0,1]
 ; AVX2-FP-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX2-FP-NEXT:    vmovdqa %xmm0, 16(%rcx)
-; AVX2-FP-NEXT:    vmovdqa %xmm1, (%rcx)
+; AVX2-FP-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-FP-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-FP-NEXT:    vmovdqa %xmm2, 32(%rcx)
+; AVX2-FP-NEXT:    vmovdqa %ymm0, (%rcx)
+; AVX2-FP-NEXT:    vzeroupper
 ; AVX2-FP-NEXT:    retq
 ;
 ; AVX2-FCP-LABEL: store_i8_stride3_vf16:
@@ -715,15 +717,16 @@ define void @store_i8_stride3_vf16(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vec
 ; AVX2-FCP-NEXT:    vpalignr {{.*#+}} xmm0 = xmm3[5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4]
 ; AVX2-FCP-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm3[0,1,2,3,4]
 ; AVX2-FCP-NEXT:    vpalignr {{.*#+}} xmm1 = xmm4[5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4]
-; AVX2-FCP-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5]
-; AVX2-FCP-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX2-FCP-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[5,6,7,8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4]
-; AVX2-FCP-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
 ; AVX2-FCP-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm4[0,1,2,3,4]
+; AVX2-FCP-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5,0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5]
+; AVX2-FCP-NEXT:    # ymm3 = mem[0,1,0,1]
 ; AVX2-FCP-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX2-FCP-NEXT:    vmovdqa %xmm0, 16(%rcx)
-; AVX2-FCP-NEXT:    vmovdqa %xmm1, (%rcx)
+; AVX2-FCP-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-FCP-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
 ; AVX2-FCP-NEXT:    vmovdqa %xmm2, 32(%rcx)
+; AVX2-FCP-NEXT:    vmovdqa %ymm0, (%rcx)
+; AVX2-FCP-NEXT:    vzeroupper
 ; AVX2-FCP-NEXT:    retq
 ;
 ; AVX512-LABEL: store_i8_stride3_vf16:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll
index d6208aca3b2b7..fb8618be17f06 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll
@@ -1682,12 +1682,19 @@ define <4 x i64> @shuffle_v4i64_0044_v2i64(<2 x i64> %a, <2 x i64> %b) {
 }
 
 define <4 x i64> @shuffle_v4i64_1032_v2i64(<2 x i64> %a, <2 x i64> %b) {
-; AVX1OR2-LABEL: shuffle_v4i64_1032_v2i64:
-; AVX1OR2:       # %bb.0:
-; AVX1OR2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; AVX1OR2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1OR2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
-; AVX1OR2-NEXT:    retq
+; AVX1-LABEL: shuffle_v4i64_1032_v2i64:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vshufpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuffle_v4i64_1032_v2i64:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
+; AVX2-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v4i64_1032_v2i64:
 ; AVX512VL-SLOW:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/widen_bitcnt.ll b/llvm/test/CodeGen/X86/widen_bitcnt.ll
index cca9d4aa2a9f0..56001468898e4 100644
--- a/llvm/test/CodeGen/X86/widen_bitcnt.ll
+++ b/llvm/test/CodeGen/X86/widen_bitcnt.ll
@@ -241,77 +241,59 @@ define <8 x i32> @widen_ctpop_v2i32_v8i32(<2 x i32> %a0, <2 x i32> %a1, <2 x i32
 ;
 ; AVX2-LABEL: widen_ctpop_v2i32_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX2-NEXT:    vpbroadcastb {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm4, %xmm6, %xmm4
-; AVX2-NEXT:    vpand %xmm5, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm6, %xmm0
-; AVX2-NEXT:    vpsrlw $4, %xmm1, %xmm7
-; AVX2-NEXT:    vpand %xmm5, %xmm7, %xmm7
-; AVX2-NEXT:    vpshufb %xmm7, %xmm6, %xmm7
-; AVX2-NEXT:    vpand %xmm5, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm1, %xmm6, %xmm1
-; AVX2-NEXT:    vpsrlw $4, %xmm2, %xmm8
-; AVX2-NEXT:    vpand %xmm5, %xmm8, %xmm8
-; AVX2-NEXT:    vpshufb %xmm8, %xmm6, %xmm8
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm8[0]
-; AVX2-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX2-NEXT:    vpshufb %xmm2, %xmm6, %xmm2
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
-; AVX2-NEXT:    vpaddb %xmm0, %xmm4, %xmm0
-; AVX2-NEXT:    vpsrlw $4, %xmm3, %xmm2
-; AVX2-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX2-NEXT:    vpshufb %xmm2, %xmm6, %xmm2
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm7[0],xmm2[0]
-; AVX2-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX2-NEXT:    vpshufb %xmm3, %xmm6, %xmm3
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
-; AVX2-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
-; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-NEXT:    vpsadbw %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX2-NEXT:    vpsadbw %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
+; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpbroadcastb {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpand %ymm3, %ymm1, %ymm4
+; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT:    # ymm5 = mem[0,1,0,1]
+; AVX2-NEXT:    vpshufb %ymm4, %ymm5, %ymm4
+; AVX2-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX2-NEXT:    vpand %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpshufb %ymm1, %ymm5, %ymm1
+; AVX2-NEXT:    vpaddb %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm1[0],ymm4[0],ymm1[1],ymm4[1],ymm1[4],ymm4[4],ymm1[5],ymm4[5]
+; AVX2-NEXT:    vpsadbw %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm3, %ymm0, %ymm2
+; AVX2-NEXT:    vpshufb %ymm2, %ymm5, %ymm2
+; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufb %ymm0, %ymm5, %ymm0
+; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm4[0],ymm0[1],ymm4[1],ymm0[4],ymm4[4],ymm0[5],ymm4[5]
+; AVX2-NEXT:    vpsadbw %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VL-LABEL: widen_ctpop_v2i32_v8i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX512VL-NEXT:    vpbroadcastb {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %xmm5, %xmm4, %xmm4
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX512VL-NEXT:    vpshufb %xmm4, %xmm6, %xmm4
-; AVX512VL-NEXT:    vpand %xmm5, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpshufb %xmm0, %xmm6, %xmm0
-; AVX512VL-NEXT:    vpsrlw $4, %xmm1, %xmm7
-; AVX512VL-NEXT:    vpand %xmm5, %xmm7, %xmm7
-; AVX512VL-NEXT:    vpshufb %xmm7, %xmm6, %xmm7
-; AVX512VL-NEXT:    vpand %xmm5, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpshufb %xmm1, %xmm6, %xmm1
-; AVX512VL-NEXT:    vpsrlw $4, %xmm2, %xmm8
-; AVX512VL-NEXT:    vpand %xmm5, %xmm8, %xmm8
-; AVX512VL-NEXT:    vpshufb %xmm8, %xmm6, %xmm8
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm8[0]
-; AVX512VL-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm6, %xmm2
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
-; AVX512VL-NEXT:    vpaddb %xmm0, %xmm4, %xmm0
-; AVX512VL-NEXT:    vpsrlw $4, %xmm3, %xmm2
-; AVX512VL-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm6, %xmm2
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm7[0],xmm2[0]
-; AVX512VL-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpshufb %xmm3, %xmm6, %xmm3
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
-; AVX512VL-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
-; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpsadbw %ymm2, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX512VL-NEXT:    vpsadbw %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
+; AVX512VL-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; AVX512VL-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpbroadcastb {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512VL-NEXT:    vpand %ymm3, %ymm1, %ymm4
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX512VL-NEXT:    # ymm5 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpshufb %ymm4, %ymm5, %ymm4
+; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpand %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpshufb %ymm1, %ymm5, %ymm1
+; AVX512VL-NEXT:    vpaddb %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512VL-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm1[0],ymm4[0],ymm1[1],ymm4[1],ymm1[4],ymm4[4],ymm1[5],ymm4[5]
+; AVX512VL-NEXT:    vpsadbw %ymm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpand %ymm3, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpshufb %ymm2, %ymm5, %ymm2
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpand %ymm3, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm5, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm4[0],ymm0[1],ymm4[1],ymm0[4],ymm4[4],ymm0[5],ymm4[5]
+; AVX512VL-NEXT:    vpsadbw %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
@@ -1297,48 +1279,35 @@ define <8 x i32> @widen_cttz_v2i32_v8i32(<2 x i32> %a0, <2 x i32> %a1, <2 x i32>
 ;
 ; AVX2-LABEL: widen_cttz_v2i32_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
-; AVX2-NEXT:    vpaddd %xmm4, %xmm0, %xmm5
-; AVX2-NEXT:    vpandn %xmm5, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm5
-; AVX2-NEXT:    vpbroadcastb {{.*#+}} xmm6 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm6, %xmm5, %xmm5
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm7 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm5, %xmm7, %xmm5
-; AVX2-NEXT:    vpand %xmm6, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm7, %xmm0
-; AVX2-NEXT:    vpaddd %xmm4, %xmm1, %xmm8
-; AVX2-NEXT:    vpandn %xmm8, %xmm1, %xmm1
-; AVX2-NEXT:    vpsrlw $4, %xmm1, %xmm8
-; AVX2-NEXT:    vpand %xmm6, %xmm8, %xmm8
-; AVX2-NEXT:    vpshufb %xmm8, %xmm7, %xmm8
-; AVX2-NEXT:    vpand %xmm6, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm1, %xmm7, %xmm1
-; AVX2-NEXT:    vpaddd %xmm4, %xmm2, %xmm9
-; AVX2-NEXT:    vpandn %xmm9, %xmm2, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm2, %xmm9
-; AVX2-NEXT:    vpand %xmm6, %xmm9, %xmm9
-; AVX2-NEXT:    vpshufb %xmm9, %xmm7, %xmm9
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm5 = xmm5[0],xmm9[0]
-; AVX2-NEXT:    vpand %xmm6, %xmm2, %xmm2
-; AVX2-NEXT:    vpshufb %xmm2, %xmm7, %xmm2
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
-; AVX2-NEXT:    vpaddb %xmm0, %xmm5, %xmm0
-; AVX2-NEXT:    vpaddd %xmm4, %xmm3, %xmm2
-; AVX2-NEXT:    vpandn %xmm2, %xmm3, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm2, %xmm3
-; AVX2-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; AVX2-NEXT:    vpshufb %xmm3, %xmm7, %xmm3
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm3 = xmm8[0],xmm3[0]
-; AVX2-NEXT:    vpand %xmm6, %xmm2, %xmm2
-; AVX2-NEXT:    vpshufb %xmm2, %xmm7, %xmm2
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; AVX2-NEXT:    vpaddb %xmm1, %xmm3, %xmm1
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
-; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-NEXT:    vpsadbw %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX2-NEXT:    vpsadbw %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
+; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpaddd %ymm3, %ymm1, %ymm4
+; AVX2-NEXT:    vpandn %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpbroadcastb {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpand %ymm4, %ymm1, %ymm5
+; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT:    # ymm6 = mem[0,1,0,1]
+; AVX2-NEXT:    vpshufb %ymm5, %ymm6, %ymm5
+; AVX2-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX2-NEXT:    vpand %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpshufb %ymm1, %ymm6, %ymm1
+; AVX2-NEXT:    vpaddb %ymm5, %ymm1, %ymm1
+; AVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm1[0],ymm5[0],ymm1[1],ymm5[1],ymm1[4],ymm5[4],ymm1[5],ymm5[5]
+; AVX2-NEXT:    vpsadbw %ymm5, %ymm1, %ymm1
+; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddd %ymm3, %ymm0, %ymm2
+; AVX2-NEXT:    vpandn %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm4, %ymm0, %ymm2
+; AVX2-NEXT:    vpshufb %ymm2, %ymm6, %ymm2
+; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufb %ymm0, %ymm6, %ymm0
+; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm5[0],ymm0[1],ymm5[1],ymm0[4],ymm5[4],ymm0[5],ymm5[5]
+; AVX2-NEXT:    vpsadbw %ymm5, %ymm0, %ymm0
 ; AVX2-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
@@ -1640,48 +1609,35 @@ define <8 x i32> @widen_cttz_undef_v2i32_v8i32(<2 x i32> %a0, <2 x i32> %a1, <2
 ;
 ; AVX2-LABEL: widen_cttz_undef_v2i32_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
-; AVX2-NEXT:    vpaddd %xmm4, %xmm0, %xmm5
-; AVX2-NEXT:    vpandn %xmm5, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $4, %xmm0, %xmm5
-; AVX2-NEXT:    vpbroadcastb {{.*#+}} xmm6 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %xmm6, %xmm5, %xmm5
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm7 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; AVX2-NEXT:    vpshufb %xmm5, %xmm7, %xmm5
-; AVX2-NEXT:    vpand %xmm6, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm0, %xmm7, %xmm0
-; AVX2-NEXT:    vpaddd %xmm4, %xmm1, %xmm8
-; AVX2-NEXT:    vpandn %xmm8, %xmm1, %xmm1
-; AVX2-NEXT:    vpsrlw $4, %xmm1, %xmm8
-; AVX2-NEXT:    vpand %xmm6, %xmm8, %xmm8
-; AVX2-NEXT:    vpshufb %xmm8, %xmm7, %xmm8
-; AVX2-NEXT:    vpand %xmm6, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm1, %xmm7, %xmm1
-; AVX2-NEXT:    vpaddd %xmm4, %xmm2, %xmm9
-; AVX2-NEXT:    vpandn %xmm9, %xmm2, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm2, %xmm9
-; AVX2-NEXT:    vpand %xmm6, %xmm9, %xmm9
-; AVX2-NEXT:    vpshufb %xmm9, %xmm7, %xmm9
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm5 = xmm5[0],xmm9[0]
-; AVX2-NEXT:    vpand %xmm6, %xmm2, %xmm2
-; AVX2-NEXT:    vpshufb %xmm2, %xmm7, %xmm2
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
-; AVX2-NEXT:    vpaddb %xmm0, %xmm5, %xmm0
-; AVX2-NEXT:    vpaddd %xmm4, %xmm3, %xmm2
-; AVX2-NEXT:    vpandn %xmm2, %xmm3, %xmm2
-; AVX2-NEXT:    vpsrlw $4, %xmm2, %xmm3
-; AVX2-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; AVX2-NEXT:    vpshufb %xmm3, %xmm7, %xmm3
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm3 = xmm8[0],xmm3[0]
-; AVX2-NEXT:    vpand %xmm6, %xmm2, %xmm2
-; AVX2-NEXT:    vpshufb %xmm2, %xmm7, %xmm2
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; AVX2-NEXT:    vpaddb %xmm1, %xmm3, %xmm1
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
-; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-NEXT:    vpsadbw %ymm2, %ymm1, %ymm1
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
-; AVX2-NEXT:    vpsadbw %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
+; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
+; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm3, %ymm3
+; AVX2-NEXT:    vpaddd %ymm3, %ymm1, %ymm4
+; AVX2-NEXT:    vpandn %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpbroadcastb {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX2-NEXT:    vpand %ymm4, %ymm1, %ymm5
+; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; AVX2-NEXT:    # ymm6 = mem[0,1,0,1]
+; AVX2-NEXT:    vpshufb %ymm5, %ymm6, %ymm5
+; AVX2-NEXT:    vpsrlw $4, %ymm1, %ymm1
+; AVX2-NEXT:    vpand %ymm4, %ymm1, %ymm1
+; AVX2-NEXT:    vpshufb %ymm1, %ymm6, %ymm1
+; AVX2-NEXT:    vpaddb %ymm5, %ymm1, %ymm1
+; AVX2-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm1 = ymm1[0],ymm5[0],ymm1[1],ymm5[1],ymm1[4],ymm5[4],ymm1[5],ymm5[5]
+; AVX2-NEXT:    vpsadbw %ymm5, %ymm1, %ymm1
+; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpaddd %ymm3, %ymm0, %ymm2
+; AVX2-NEXT:    vpandn %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm4, %ymm0, %ymm2
+; AVX2-NEXT:    vpshufb %ymm2, %ymm6, %ymm2
+; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; AVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufb %ymm0, %ymm6, %ymm0
+; AVX2-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm5[0],ymm0[1],ymm5[1],ymm0[4],ymm5[4],ymm0[5],ymm5[5]
+; AVX2-NEXT:    vpsadbw %ymm5, %ymm0, %ymm0
 ; AVX2-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
diff --git a/llvm/test/CodeGen/X86/x86-interleaved-access.ll b/llvm/test/CodeGen/X86/x86-interleaved-access.ll
index 3e76bffb77a66..edadcdadad42e 100644
--- a/llvm/test/CodeGen/X86/x86-interleaved-access.ll
+++ b/llvm/test/CodeGen/X86/x86-interleaved-access.ll
@@ -915,24 +915,44 @@ ret void
 }
 
 define void @interleaved_store_vf16_i8_stride3(<16 x i8> %a, <16 x i8> %b, <16 x i8> %c, ptr %p) nounwind {
-; AVX1OR2-LABEL: interleaved_store_vf16_i8_stride3:
-; AVX1OR2:       # %bb.0:
-; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15,0,1,2,3,4,5]
-; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm3 = xmm1[11,12,13,14,15,0,1,2,3,4,5,6,7,8,9,10]
-; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm4 = xmm0[5,6,7,8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4]
-; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm3[5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4]
-; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm3[0,1,2,3,4]
-; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm1 = xmm4[5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4]
-; AVX1OR2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5]
-; AVX1OR2-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
-; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[5,6,7,8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4]
-; AVX1OR2-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
-; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm4[0,1,2,3,4]
-; AVX1OR2-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX1OR2-NEXT:    vmovdqu %xmm0, 16(%rdi)
-; AVX1OR2-NEXT:    vmovdqu %xmm1, (%rdi)
-; AVX1OR2-NEXT:    vmovdqu %xmm2, 32(%rdi)
-; AVX1OR2-NEXT:    retq
+; AVX1-LABEL: interleaved_store_vf16_i8_stride3:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15,0,1,2,3,4,5]
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm3 = xmm1[11,12,13,14,15,0,1,2,3,4,5,6,7,8,9,10]
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm4 = xmm0[5,6,7,8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4]
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm0 = xmm3[5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4]
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm3[0,1,2,3,4]
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm1 = xmm4[5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5]
+; AVX1-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[5,6,7,8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4]
+; AVX1-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm4[0,1,2,3,4]
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqu %xmm0, 16(%rdi)
+; AVX1-NEXT:    vmovdqu %xmm1, (%rdi)
+; AVX1-NEXT:    vmovdqu %xmm2, 32(%rdi)
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: interleaved_store_vf16_i8_stride3:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15,0,1,2,3,4,5]
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm3 = xmm1[11,12,13,14,15,0,1,2,3,4,5,6,7,8,9,10]
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm4 = xmm0[5,6,7,8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4]
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm3[5,6,7,8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4]
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm3[0,1,2,3,4]
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm1 = xmm4[5,6,7,8,9,10,11,12,13,14,15],xmm1[0,1,2,3,4]
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[5,6,7,8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4]
+; AVX2-NEXT:    vpalignr {{.*#+}} xmm2 = xmm2[5,6,7,8,9,10,11,12,13,14,15],xmm4[0,1,2,3,4]
+; AVX2-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5,0,11,6,1,12,7,2,13,8,3,14,9,4,15,10,5]
+; AVX2-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX2-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vmovdqu %xmm2, 32(%rdi)
+; AVX2-NEXT:    vmovdqu %ymm0, (%rdi)
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: interleaved_store_vf16_i8_stride3:
 ; AVX512:       # %bb.0:



More information about the llvm-commits mailing list