[X86][AVX] Reduce unary target shuffles width if the upper elements aren't demanded.

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 1cbfd41..86825ce 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -36908,6 +36908,11 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
       return TLO.CombineTo(Op, insertSubVector(TLO.DAG.getUNDEF(VT), Src, 0,
                                                TLO.DAG, DL, ExtSizeInBits));
     }
+      // Target unary shuffles by immediate:
+    case X86ISD::PSHUFD:
+    case X86ISD::PSHUFLW:
+    case X86ISD::PSHUFHW:
+    case X86ISD::VPERMILPI:
       // Byte shifts by immediate.
     case X86ISD::VSHLDQ:
     case X86ISD::VSRLDQ:

diff --git a/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
index fee195a..295b527 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
+++ b/llvm/test/CodeGen/X86/avx512-intrinsics-fast-isel.ll
@@ -7742,7 +7742,7 @@ define i64 @test_mm512_reduce_max_epi64(<8 x i64> %__W) {
 ; X86-NEXT:    vpmaxsq %zmm0, %zmm1, %zmm0
 ; X86-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X86-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm0
-; X86-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X86-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X86-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm0
 ; X86-NEXT:    vmovd %xmm0, %eax
 ; X86-NEXT:    vpextrd $1, %xmm0, %edx
@@ -7755,7 +7755,7 @@ define i64 @test_mm512_reduce_max_epi64(<8 x i64> %__W) {
 ; X64-NEXT:    vpmaxsq %zmm0, %zmm1, %zmm0
 ; X64-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X64-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm0
-; X64-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X64-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vzeroupper
@@ -7781,7 +7781,7 @@ define i64 @test_mm512_reduce_max_epu64(<8 x i64> %__W) {
 ; X86-NEXT:    vpmaxuq %zmm0, %zmm1, %zmm0
 ; X86-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X86-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
-; X86-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X86-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X86-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
 ; X86-NEXT:    vmovd %xmm0, %eax
 ; X86-NEXT:    vpextrd $1, %xmm0, %edx
@@ -7794,7 +7794,7 @@ define i64 @test_mm512_reduce_max_epu64(<8 x i64> %__W) {
 ; X64-NEXT:    vpmaxuq %zmm0, %zmm1, %zmm0
 ; X64-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X64-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
-; X64-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X64-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vzeroupper
@@ -7867,7 +7867,7 @@ define i64 @test_mm512_reduce_min_epi64(<8 x i64> %__W) {
 ; X86-NEXT:    vpminsq %zmm0, %zmm1, %zmm0
 ; X86-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X86-NEXT:    vpminsq %zmm1, %zmm0, %zmm0
-; X86-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X86-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X86-NEXT:    vpminsq %zmm1, %zmm0, %zmm0
 ; X86-NEXT:    vmovd %xmm0, %eax
 ; X86-NEXT:    vpextrd $1, %xmm0, %edx
@@ -7880,7 +7880,7 @@ define i64 @test_mm512_reduce_min_epi64(<8 x i64> %__W) {
 ; X64-NEXT:    vpminsq %zmm0, %zmm1, %zmm0
 ; X64-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X64-NEXT:    vpminsq %zmm1, %zmm0, %zmm0
-; X64-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X64-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-NEXT:    vpminsq %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vzeroupper
@@ -7906,7 +7906,7 @@ define i64 @test_mm512_reduce_min_epu64(<8 x i64> %__W) {
 ; X86-NEXT:    vpminuq %zmm0, %zmm1, %zmm0
 ; X86-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X86-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
-; X86-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X86-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X86-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
 ; X86-NEXT:    vmovd %xmm0, %eax
 ; X86-NEXT:    vpextrd $1, %xmm0, %edx
@@ -7919,7 +7919,7 @@ define i64 @test_mm512_reduce_min_epu64(<8 x i64> %__W) {
 ; X64-NEXT:    vpminuq %zmm0, %zmm1, %zmm0
 ; X64-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X64-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
-; X64-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X64-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vzeroupper
@@ -7996,7 +7996,7 @@ define i64 @test_mm512_mask_reduce_max_epi64(i8 zeroext %__M, <8 x i64> %__W) {
 ; X86-NEXT:    vpmaxsq %zmm0, %zmm1, %zmm0
 ; X86-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X86-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm0
-; X86-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X86-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X86-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm0
 ; X86-NEXT:    vmovd %xmm0, %eax
 ; X86-NEXT:    vpextrd $1, %xmm0, %edx
@@ -8012,7 +8012,7 @@ define i64 @test_mm512_mask_reduce_max_epi64(i8 zeroext %__M, <8 x i64> %__W) {
 ; X64-NEXT:    vpmaxsq %zmm0, %zmm1, %zmm0
 ; X64-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X64-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm0
-; X64-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X64-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-NEXT:    vpmaxsq %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vzeroupper
@@ -8043,7 +8043,7 @@ define i64 @test_mm512_mask_reduce_max_epu64(i8 zeroext %__M, <8 x i64> %__W) {
 ; X86-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
 ; X86-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X86-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
-; X86-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X86-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X86-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
 ; X86-NEXT:    vmovd %xmm0, %eax
 ; X86-NEXT:    vpextrd $1, %xmm0, %edx
@@ -8058,7 +8058,7 @@ define i64 @test_mm512_mask_reduce_max_epu64(i8 zeroext %__M, <8 x i64> %__W) {
 ; X64-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X64-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
-; X64-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X64-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vzeroupper
@@ -8146,7 +8146,7 @@ define i64 @test_mm512_mask_reduce_min_epi64(i8 zeroext %__M, <8 x i64> %__W) {
 ; X86-NEXT:    vpminsq %zmm0, %zmm1, %zmm0
 ; X86-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X86-NEXT:    vpminsq %zmm1, %zmm0, %zmm0
-; X86-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X86-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X86-NEXT:    vpminsq %zmm1, %zmm0, %zmm0
 ; X86-NEXT:    vmovd %xmm0, %eax
 ; X86-NEXT:    vpextrd $1, %xmm0, %edx
@@ -8162,7 +8162,7 @@ define i64 @test_mm512_mask_reduce_min_epi64(i8 zeroext %__M, <8 x i64> %__W) {
 ; X64-NEXT:    vpminsq %zmm0, %zmm1, %zmm0
 ; X64-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X64-NEXT:    vpminsq %zmm1, %zmm0, %zmm0
-; X64-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X64-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-NEXT:    vpminsq %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vzeroupper
@@ -8194,7 +8194,7 @@ define i64 @test_mm512_mask_reduce_min_epu64(i8 zeroext %__M, <8 x i64> %__W) {
 ; X86-NEXT:    vpminuq %zmm0, %zmm1, %zmm0
 ; X86-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X86-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
-; X86-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X86-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X86-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
 ; X86-NEXT:    vmovd %xmm0, %eax
 ; X86-NEXT:    vpextrd $1, %xmm0, %edx
@@ -8210,7 +8210,7 @@ define i64 @test_mm512_mask_reduce_min_epu64(i8 zeroext %__M, <8 x i64> %__W) {
 ; X64-NEXT:    vpminuq %zmm0, %zmm1, %zmm0
 ; X64-NEXT:    vpermq {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5]
 ; X64-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
-; X64-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
+; X64-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vzeroupper

diff --git a/llvm/test/CodeGen/X86/vector-reduce-mul.ll b/llvm/test/CodeGen/X86/vector-reduce-mul.ll
index 09d1472..e6f9bb5 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-reduce-mul.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-mul.ll
@@ -1969,8 +1969,7 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX2-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm3, %xmm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
@@ -2052,8 +2051,7 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512DQ-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vpand %xmm3, %xmm0, %xmm1
-; AVX512DQ-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,2,3,6,7,6,7]
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
@@ -2079,8 +2077,7 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512DQVL-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQVL-NEXT:    vpand %xmm3, %xmm0, %xmm1
-; AVX512DQVL-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,2,3,6,7,6,7]
+; AVX512DQVL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
@@ -2238,8 +2235,7 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX2-NEXT:    vpmullw %xmm0, %xmm3, %xmm0
 ; AVX2-NEXT:    vpmullw %xmm0, %xmm2, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm2 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX2-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm2 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
@@ -2274,8 +2270,7 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BW-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpand %xmm3, %xmm0, %xmm1
-; AVX512BW-NEXT:    vpshufd {{.*#+}} zmm1 = zmm1[2,3,2,3,6,7,6,7,10,11,10,11,14,15,14,15]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
@@ -2311,8 +2306,7 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; AVX512BWVL-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX512BWVL-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} zmm1 = zmm1[2,3,2,3,6,7,6,7,10,11,10,11,14,15,14,15]
+; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
@@ -2608,8 +2602,7 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,2,3,6,7,6,7]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
@@ -2647,8 +2640,7 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BW-NEXT:    vpmullw %xmm0, %xmm1, %xmm0
 ; AVX512BW-NEXT:    vpmullw %xmm0, %xmm2, %xmm0
-; AVX512BW-NEXT:    vpand %xmm3, %xmm0, %xmm1
-; AVX512BW-NEXT:    vpshufd {{.*#+}} zmm1 = zmm1[2,3,2,3,6,7,6,7,10,11,10,11,14,15,14,15]
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
@@ -2687,8 +2679,7 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; AVX512BWVL-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX512BWVL-NEXT:    vpmullw %xmm0, %xmm1, %xmm0
 ; AVX512BWVL-NEXT:    vpmullw %xmm0, %xmm2, %xmm0
-; AVX512BWVL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} zmm1 = zmm1[2,3,2,3,6,7,6,7,10,11,10,11,14,15,14,15]
+; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0

diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll
index 50a250b..6ffbe09 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll
@@ -145,8 +145,7 @@ define <8 x float> @combine_vpermilvar_vperm2f128_zero_8f32(<8 x float> %a0) {
 define <4 x double> @combine_vperm2f128_vpermilvar_as_vpblendpd(<4 x double> %a0) {
 ; CHECK-LABEL: combine_vperm2f128_vpermilvar_as_vpblendpd:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
-; CHECK-NEXT:    vmovapd %xmm0, %xmm0
+; CHECK-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
 ; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
 ; CHECK-NEXT:    ret{{[l|q]}}
   %1 = tail call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %a0, <4 x i64> <i64 2, i64 0, i64 2, i64 0>)