[X86][AVX] More aggressively simplify BROADCAST source operand

Aim to use scalar source or lowest 128-bit vector directly.

We're still missing some VZMOVL_LOAD combines.

llvm-svn: 352994

6 files changed, 31 insertions, 48 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index bd132cd7340..9ea3c64697e 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -32004,12 +32004,13 @@ static SDValue combineTargetShuffle(SDValue N, SelectionDAG &DAG,
 
   switch (Opcode) {
   case X86ISD::VBROADCAST: {
-    // If broadcasting from another shuffle, attempt to simplify it.
-    // TODO - we really need a general SimplifyDemandedVectorElts mechanism.
     SDValue Src = N.getOperand(0);
     SDValue BC = peekThroughBitcasts(Src);
     EVT SrcVT = Src.getValueType();
     EVT BCVT = BC.getValueType();
+
+    // If broadcasting from another shuffle, attempt to simplify it.
+    // TODO - we really need a general SimplifyDemandedVectorElts mechanism.
     if (isTargetShuffle(BC.getOpcode()) &&
         VT.getScalarSizeInBits() % BCVT.getScalarSizeInBits() == 0) {
       unsigned Scale = VT.getScalarSizeInBits() / BCVT.getScalarSizeInBits();
@@ -32023,6 +32024,7 @@ static SDValue combineTargetShuffle(SDValue N, SelectionDAG &DAG,
         return DAG.getNode(X86ISD::VBROADCAST, DL, VT,
                            DAG.getBitcast(SrcVT, Res));
     }
+
     // broadcast(bitcast(src)) -> bitcast(broadcast(src))
     // 32-bit targets have to bitcast i64 to f64, so better to bitcast upward.
     if (Src.getOpcode() == ISD::BITCAST &&
@@ -32031,6 +32033,16 @@ static SDValue combineTargetShuffle(SDValue N, SelectionDAG &DAG,
                                    VT.getVectorNumElements());
       return DAG.getBitcast(VT, DAG.getNode(X86ISD::VBROADCAST, DL, NewVT, BC));
     }
+
+    // Reduce broadcast source vector to lowest 128-bits.
+    if (SrcVT.getSizeInBits() > 128)
+      return DAG.getNode(X86ISD::VBROADCAST, DL, VT,
+                         extract128BitVector(Src, 0, DAG, DL));
+
+    // broadcast(scalar_to_vector(x)) -> broadcast(x).
+    if (Src.getOpcode() == ISD::SCALAR_TO_VECTOR)
+      return DAG.getNode(X86ISD::VBROADCAST, DL, VT, Src.getOperand(0));
+
     return SDValue();
   }
   case X86ISD::PSHUFD:
diff --git a/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-vector-int.ll b/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-vector-int.ll
index e33399e97cd..9c9df48c03c 100644
--- a/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-vector-int.ll
+++ b/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-vector-int.ll
@@ -536,8 +536,7 @@ define <8 x i32> @test_masked_z_2xi32_to_8xi32_mem_mask3(<2 x i32>* %vp, <8 x i3
 define <16 x i32> @test_2xi32_to_16xi32_mem(<2 x i32>* %vp) {
 ; CHECK-LABEL: test_2xi32_to_16xi32_mem:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0
+; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0
 ; CHECK-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %res = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
@@ -546,10 +545,8 @@ define <16 x i32> @test_2xi32_to_16xi32_mem(<2 x i32>* %vp) {
 define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask0(<2 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
 ; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask0:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
-; CHECK-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; CHECK-NEXT:    vptestnmd %zmm1, %zmm1, %k1
-; CHECK-NEXT:    vmovdqa32 %zmm2, %zmm0 {%k1}
+; CHECK-NEXT:    vbroadcasti32x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
 ; CHECK-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
@@ -561,10 +558,8 @@ define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask0(<2 x i32>* %vp, <16 x i
 define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask0(<2 x i32>* %vp, <16 x i32> %mask) {
 ; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask0:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
-; CHECK-NEXT:    vpbroadcastq %xmm1, %zmm1
 ; CHECK-NEXT:    vptestnmd %zmm0, %zmm0, %k1
-; CHECK-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
+; CHECK-NEXT:    vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
 ; CHECK-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
@@ -575,10 +570,8 @@ define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask0(<2 x i32>* %vp, <16 x
 define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask1(<2 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
 ; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask1:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
-; CHECK-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; CHECK-NEXT:    vptestnmd %zmm1, %zmm1, %k1
-; CHECK-NEXT:    vmovdqa32 %zmm2, %zmm0 {%k1}
+; CHECK-NEXT:    vbroadcasti32x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
 ; CHECK-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
@@ -590,10 +583,8 @@ define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask1(<2 x i32>* %vp, <16 x i
 define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask1(<2 x i32>* %vp, <16 x i32> %mask) {
 ; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask1:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
-; CHECK-NEXT:    vpbroadcastq %xmm1, %zmm1
 ; CHECK-NEXT:    vptestnmd %zmm0, %zmm0, %k1
-; CHECK-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
+; CHECK-NEXT:    vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
 ; CHECK-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
@@ -604,10 +595,8 @@ define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask1(<2 x i32>* %vp, <16 x
 define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask2(<2 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
 ; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask2:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
-; CHECK-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; CHECK-NEXT:    vptestnmd %zmm1, %zmm1, %k1
-; CHECK-NEXT:    vmovdqa32 %zmm2, %zmm0 {%k1}
+; CHECK-NEXT:    vbroadcasti32x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
 ; CHECK-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
@@ -619,10 +608,8 @@ define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask2(<2 x i32>* %vp, <16 x i
 define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask2(<2 x i32>* %vp, <16 x i32> %mask) {
 ; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask2:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
-; CHECK-NEXT:    vpbroadcastq %xmm1, %zmm1
 ; CHECK-NEXT:    vptestnmd %zmm0, %zmm0, %k1
-; CHECK-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
+; CHECK-NEXT:    vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
 ; CHECK-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
@@ -633,10 +620,8 @@ define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask2(<2 x i32>* %vp, <16 x
 define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask3(<2 x i32>* %vp, <16 x i32> %default, <16 x i32> %mask) {
 ; CHECK-LABEL: test_masked_2xi32_to_16xi32_mem_mask3:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
-; CHECK-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; CHECK-NEXT:    vptestnmd %zmm1, %zmm1, %k1
-; CHECK-NEXT:    vmovdqa32 %zmm2, %zmm0 {%k1}
+; CHECK-NEXT:    vbroadcasti32x2 {{.*#+}} zmm0 {%k1} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
 ; CHECK-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
@@ -648,10 +633,8 @@ define <16 x i32> @test_masked_2xi32_to_16xi32_mem_mask3(<2 x i32>* %vp, <16 x i
 define <16 x i32> @test_masked_z_2xi32_to_16xi32_mem_mask3(<2 x i32>* %vp, <16 x i32> %mask) {
 ; CHECK-LABEL: test_masked_z_2xi32_to_16xi32_mem_mask3:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
-; CHECK-NEXT:    vpbroadcastq %xmm1, %zmm1
 ; CHECK-NEXT:    vptestnmd %zmm0, %zmm0, %k1
-; CHECK-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
+; CHECK-NEXT:    vbroadcasti32x2 {{.*#+}} zmm0 {%k1} {z} = mem[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
 ; CHECK-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %shuf = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
diff --git a/llvm/test/CodeGen/X86/subvector-broadcast.ll b/llvm/test/CodeGen/X86/subvector-broadcast.ll
index 0005bcc26c6..926f00099a4 100644
--- a/llvm/test/CodeGen/X86/subvector-broadcast.ll
+++ b/llvm/test/CodeGen/X86/subvector-broadcast.ll
@@ -1598,8 +1598,7 @@ define <16 x i32> @test_2xi32_to_16xi32_mem(<2 x i32>* %vp) {
 ; X32-AVX512-LABEL: test_2xi32_to_16xi32_mem:
 ; X32-AVX512:       # %bb.0:
 ; X32-AVX512-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-AVX512-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; X32-AVX512-NEXT:    vbroadcastsd %xmm0, %zmm0
+; X32-AVX512-NEXT:    vbroadcastsd (%eax), %zmm0
 ; X32-AVX512-NEXT:    retl
 ;
 ; X64-AVX-LABEL: test_2xi32_to_16xi32_mem:
@@ -1610,8 +1609,7 @@ define <16 x i32> @test_2xi32_to_16xi32_mem(<2 x i32>* %vp) {
 ;
 ; X64-AVX512-LABEL: test_2xi32_to_16xi32_mem:
 ; X64-AVX512:       # %bb.0:
-; X64-AVX512-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; X64-AVX512-NEXT:    vbroadcastsd %xmm0, %zmm0
+; X64-AVX512-NEXT:    vbroadcastsd (%rdi), %zmm0
 ; X64-AVX512-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %res = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
index 40ba3c1a9ac..ff9a6210ca4 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
@@ -141,7 +141,6 @@ define <16 x i8> @combine_pshufb_as_vpbroadcastb128(<16 x i8> %a) {
 define <32 x i8> @combine_pshufb_as_vpbroadcastb256(<2 x i64> %a) {
 ; CHECK-LABEL: combine_pshufb_as_vpbroadcastb256:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; CHECK-NEXT:    vpbroadcastb %xmm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %1 = shufflevector <2 x i64> %a, <2 x i64> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
@@ -165,7 +164,6 @@ define <16 x i8> @combine_pshufb_as_vpbroadcastw128(<16 x i8> %a) {
 define <32 x i8> @combine_pshufb_as_vpbroadcastw256(<2 x i64> %a) {
 ; CHECK-LABEL: combine_pshufb_as_vpbroadcastw256:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; CHECK-NEXT:    vpbroadcastw %xmm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %1 = shufflevector <2 x i64> %a, <2 x i64> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
@@ -197,14 +195,12 @@ define <16 x i8> @combine_pshufb_as_vpbroadcastd128(<16 x i8> %a) {
 define <8 x i32> @combine_permd_as_vpbroadcastd256(<4 x i32> %a) {
 ; X86-LABEL: combine_permd_as_vpbroadcastd256:
 ; X86:       # %bb.0:
-; X86-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; X86-NEXT:    vpbroadcastd %xmm0, %ymm0
 ; X86-NEXT:    vpaddd {{\.LCPI.*}}, %ymm0, %ymm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: combine_permd_as_vpbroadcastd256:
 ; X64:       # %bb.0:
-; X64-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; X64-NEXT:    vpbroadcastd %xmm0, %ymm0
 ; X64-NEXT:    vpaddd {{.*}}(%rip), %ymm0, %ymm0
 ; X64-NEXT:    retq
@@ -226,14 +222,12 @@ define <16 x i8> @combine_pshufb_as_vpbroadcastq128(<16 x i8> %a) {
 define <8 x i32> @combine_permd_as_vpbroadcastq256(<4 x i32> %a) {
 ; X86-LABEL: combine_permd_as_vpbroadcastq256:
 ; X86:       # %bb.0:
-; X86-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; X86-NEXT:    vpbroadcastq %xmm0, %ymm0
 ; X86-NEXT:    vpaddd {{\.LCPI.*}}, %ymm0, %ymm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: combine_permd_as_vpbroadcastq256:
 ; X64:       # %bb.0:
-; X64-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; X64-NEXT:    vpbroadcastq %xmm0, %ymm0
 ; X64-NEXT:    vpaddd {{.*}}(%rip), %ymm0, %ymm0
 ; X64-NEXT:    retq
@@ -257,7 +251,6 @@ define <4 x float> @combine_pshufb_as_vpbroadcastss128(<4 x float> %a) {
 define <8 x float> @combine_permps_as_vpbroadcastss256(<4 x float> %a) {
 ; CHECK-LABEL: combine_permps_as_vpbroadcastss256:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; CHECK-NEXT:    vbroadcastss %xmm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %1 = shufflevector <4 x float> %a, <4 x float> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
@@ -268,7 +261,6 @@ define <8 x float> @combine_permps_as_vpbroadcastss256(<4 x float> %a) {
 define <4 x double> @combine_permps_as_vpbroadcastsd256(<2 x double> %a) {
 ; CHECK-LABEL: combine_permps_as_vpbroadcastsd256:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %1 = shufflevector <2 x double> %a, <2 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
@@ -313,7 +305,6 @@ define <4 x float> @combine_vpbroadcast_pshufb_as_vpbroadcastss128(<4 x float> %
 define <8 x float> @combine_vpbroadcast_permd_as_vpbroadcastss256(<4 x float> %a) {
 ; CHECK-LABEL: combine_vpbroadcast_permd_as_vpbroadcastss256:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; CHECK-NEXT:    vbroadcastss %xmm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %1 = shufflevector <4 x float> %a, <4 x float> undef, <8 x i32> zeroinitializer
@@ -324,7 +315,6 @@ define <8 x float> @combine_vpbroadcast_permd_as_vpbroadcastss256(<4 x float> %a
 define <4 x double> @combine_vpbroadcast_permd_as_vpbroadcastsd256(<2 x double> %a) {
 ; CHECK-LABEL: combine_vpbroadcast_permd_as_vpbroadcastsd256:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
 ; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %1 = shufflevector <2 x double> %a, <2 x double> undef, <4 x i32> zeroinitializer
@@ -665,7 +655,8 @@ define <16 x i8> @combine_pshufb_insertion_as_broadcast_v2i64(i64 %a0) {
 define <8 x i32> @combine_permd_insertion_as_broadcast_v4i64(i64 %a0) {
 ; X86-LABEL: combine_permd_insertion_as_broadcast_v4i64:
 ; X86:       # %bb.0:
-; X86-NEXT:    vbroadcastsd {{[0-9]+}}(%esp), %ymm0
+; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X86-NEXT:    vbroadcastsd %xmm0, %ymm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: combine_permd_insertion_as_broadcast_v4i64:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll
index 729863cada8..6a295ba8cc5 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll
@@ -975,13 +975,13 @@ define <16 x float> @combine_vpermi2var_vpermvar_16f32_as_vperm2_zero(<16 x floa
 define <8 x i64> @combine_vpermvar_insertion_as_broadcast_v8i64(i64 %a0) {
 ; X86-LABEL: combine_vpermvar_insertion_as_broadcast_v8i64:
 ; X86:       # %bb.0:
-; X86-NEXT:    vbroadcastsd {{[0-9]+}}(%esp), %zmm0
+; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X86-NEXT:    vbroadcastsd %xmm0, %zmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: combine_vpermvar_insertion_as_broadcast_v8i64:
 ; X64:       # %bb.0:
-; X64-NEXT:    vmovq %rdi, %xmm0
-; X64-NEXT:    vpbroadcastq %xmm0, %zmm0
+; X64-NEXT:    vpbroadcastq %rdi, %zmm0
 ; X64-NEXT:    retq
   %1 = insertelement <8 x i64> undef, i64 %a0, i32 0
   %2 = call <8 x i64> @llvm.x86.avx512.permvar.di.512(<8 x i64> %1, <8 x i64> zeroinitializer)
diff --git a/llvm/test/CodeGen/X86/widened-broadcast.ll b/llvm/test/CodeGen/X86/widened-broadcast.ll
index dcc84a4d82c..ae4f901b157 100644
--- a/llvm/test/CodeGen/X86/widened-broadcast.ll
+++ b/llvm/test/CodeGen/X86/widened-broadcast.ll
@@ -630,8 +630,7 @@ define <16 x i32> @load_splat_16i32_2i32_0101(<2 x i32>* %vp) {
 ;
 ; AVX512-LABEL: load_splat_16i32_2i32_0101:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; AVX512-NEXT:    vbroadcastsd %xmm0, %zmm0
+; AVX512-NEXT:    vbroadcastsd (%rdi), %zmm0
 ; AVX512-NEXT:    retq
   %vec = load <2 x i32>, <2 x i32>* %vp
   %res = shufflevector <2 x i32> %vec, <2 x i32> undef, <16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>