[X86] Shrink zero extends of gather indices from type less than i32 to types larger...

Gather instructions can use i32 or i64 elements for indices. If
the index is zero extended from a type smaller than i32 to i64, we
can shrink the extend to just extend to i32.

llvm-svn: 373982

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 052300d..54ca372 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -42572,16 +42572,17 @@ static SDValue combineGatherScatter(SDNode *N, SelectionDAG &DAG,
   SDValue Base = GorS->getBasePtr();
   SDValue Scale = GorS->getScale();
 
-  // Shrink constant indices if they are larger than 32-bits.
-  // Only do this before legalize types since v2i64 could become v2i32.
-  // FIXME: We could check that the type is legal if we're after legalize types,
-  // but then we would need to construct test cases where that happens.
-  // FIXME: We could support more than just constant vectors, but we need to
-  // careful with costing. A truncate that can be optimized out would be fine.
-  // Otherwise we might only want to create a truncate if it avoids a split.
   if (DCI.isBeforeLegalize()) {
+    unsigned IndexWidth = Index.getScalarValueSizeInBits();
+
+    // Shrink constant indices if they are larger than 32-bits.
+    // Only do this before legalize types since v2i64 could become v2i32.
+    // FIXME: We could check that the type is legal if we're after legalize
+    // types, but then we would need to construct test cases where that happens.
+    // FIXME: We could support more than just constant vectors, but we need to
+    // careful with costing. A truncate that can be optimized out would be fine.
+    // Otherwise we might only want to create a truncate if it avoids a split.
     if (auto *BV = dyn_cast<BuildVectorSDNode>(Index)) {
-      unsigned IndexWidth = Index.getScalarValueSizeInBits();
       if (BV->isConstant() && IndexWidth > 32 &&
           DAG.ComputeNumSignBits(Index) > (IndexWidth - 32)) {
         unsigned NumElts = Index.getValueType().getVectorNumElements();
@@ -42604,16 +42605,18 @@ static SDValue combineGatherScatter(SDNode *N, SelectionDAG &DAG,
                                     Scatter->getIndexType());
       }
     }
-  }
 
-  if (DCI.isBeforeLegalizeOps()) {
-    // Remove any sign extends from 32 or smaller to larger than 32.
-    // Only do this before LegalizeOps in case we need the sign extend for
-    // legalization.
-    if (Index.getOpcode() == ISD::SIGN_EXTEND &&
-        Index.getScalarValueSizeInBits() > 32 &&
-        Index.getOperand(0).getScalarValueSizeInBits() <= 32) {
-      Index = Index.getOperand(0);
+    // Shrink any sign/zero extends from 32 or smaller to larger than 32 if
+    // there are sufficient sign bits. Only do this before legalize types to
+    // avoid creating illegal types in truncate.
+    if ((Index.getOpcode() == ISD::SIGN_EXTEND ||
+         Index.getOpcode() == ISD::ZERO_EXTEND) &&
+        IndexWidth > 32 &&
+        Index.getOperand(0).getScalarValueSizeInBits() <= 32 &&
+        DAG.ComputeNumSignBits(Index) > (IndexWidth - 32)) {
+      unsigned NumElts = Index.getValueType().getVectorNumElements();
+      EVT NewVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32, NumElts);
+      Index = DAG.getNode(ISD::TRUNCATE, DL, NewVT, Index);
       if (auto *Gather = dyn_cast<MaskedGatherSDNode>(GorS)) {
         SDValue Ops[] = { Chain, Gather->getPassThru(),
                           Mask, Base, Index, Scale } ;
@@ -42630,11 +42633,14 @@ static SDValue combineGatherScatter(SDNode *N, SelectionDAG &DAG,
                                   Ops, Scatter->getMemOperand(),
                                   Scatter->getIndexType());
     }
+  }
+
+  if (DCI.isBeforeLegalizeOps()) {
+    unsigned IndexWidth = Index.getScalarValueSizeInBits();
 
     // Make sure the index is either i32 or i64
-    unsigned ScalarSize = Index.getScalarValueSizeInBits();
-    if (ScalarSize != 32 && ScalarSize != 64) {
-      MVT EltVT = ScalarSize > 32 ? MVT::i64 : MVT::i32;
+    if (IndexWidth != 32 && IndexWidth != 64) {
+      MVT EltVT = IndexWidth > 32 ? MVT::i64 : MVT::i32;
       EVT IndexVT = EVT::getVectorVT(*DAG.getContext(), EltVT,
                                    Index.getValueType().getVectorNumElements());
       Index = DAG.getSExtOrTrunc(Index, DL, IndexVT);
@@ -42654,30 +42660,6 @@ static SDValue combineGatherScatter(SDNode *N, SelectionDAG &DAG,
                                   Ops, Scatter->getMemOperand(),
                                   Scatter->getIndexType());
     }
-
-    // Try to remove zero extends from 32->64 if we know the sign bit of
-    // the input is zero.
-    if (Index.getOpcode() == ISD::ZERO_EXTEND &&
-        Index.getScalarValueSizeInBits() == 64 &&
-        Index.getOperand(0).getScalarValueSizeInBits() == 32 &&
-        DAG.SignBitIsZero(Index.getOperand(0))) {
-      Index = Index.getOperand(0);
-      if (auto *Gather = dyn_cast<MaskedGatherSDNode>(GorS)) {
-        SDValue Ops[] = { Chain, Gather->getPassThru(),
-                          Mask, Base, Index, Scale } ;
-        return DAG.getMaskedGather(Gather->getVTList(),
-                                   Gather->getMemoryVT(), DL, Ops,
-                                   Gather->getMemOperand(),
-                                   Gather->getIndexType());
-      }
-      auto *Scatter = cast<MaskedScatterSDNode>(GorS);
-      SDValue Ops[] = { Chain, Scatter->getValue(),
-                        Mask, Base, Index, Scale };
-      return DAG.getMaskedScatter(Scatter->getVTList(),
-                                  Scatter->getMemoryVT(), DL,
-                                  Ops, Scatter->getMemOperand(),
-                                  Scatter->getIndexType());
-    }
   }
 
   // With vector masks we only demand the upper bit of the mask.

diff --git a/llvm/test/CodeGen/X86/masked_gather_scatter.ll b/llvm/test/CodeGen/X86/masked_gather_scatter.ll
index fe870e5..2e4edb9 100644 (file)
--- a/llvm/test/CodeGen/X86/masked_gather_scatter.ll
+++ b/llvm/test/CodeGen/X86/masked_gather_scatter.ll
@@ -2693,56 +2693,32 @@ declare <8 x float> @llvm.masked.gather.v8f32.v8p0f32(<8 x float*>, i32, <8 x i1
 define <16 x float> @zext_i8_index(float* %base, <16 x i8> %ind) {
 ; KNL_64-LABEL: zext_i8_index:
 ; KNL_64:       # %bb.0:
-; KNL_64-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; KNL_64-NEXT:    vpmovzxwq {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
-; KNL_64-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL_64-NEXT:    vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; KNL_64-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; KNL_64-NEXT:    kxnorw %k0, %k0, %k1
-; KNL_64-NEXT:    kxnorw %k0, %k0, %k2
-; KNL_64-NEXT:    vgatherqps (%rdi,%zmm0,4), %ymm2 {%k2}
-; KNL_64-NEXT:    vgatherqps (%rdi,%zmm1,4), %ymm0 {%k1}
-; KNL_64-NEXT:    vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
+; KNL_64-NEXT:    vgatherdps (%rdi,%zmm1,4), %zmm0 {%k1}
 ; KNL_64-NEXT:    retq
 ;
 ; KNL_32-LABEL: zext_i8_index:
 ; KNL_32:       # %bb.0:
 ; KNL_32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; KNL_32-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; KNL_32-NEXT:    vpmovzxwq {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
-; KNL_32-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL_32-NEXT:    vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; KNL_32-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; KNL_32-NEXT:    kxnorw %k0, %k0, %k1
-; KNL_32-NEXT:    kxnorw %k0, %k0, %k2
-; KNL_32-NEXT:    vgatherqps (%eax,%zmm0,4), %ymm2 {%k2}
-; KNL_32-NEXT:    vgatherqps (%eax,%zmm1,4), %ymm0 {%k1}
-; KNL_32-NEXT:    vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
+; KNL_32-NEXT:    vgatherdps (%eax,%zmm1,4), %zmm0 {%k1}
 ; KNL_32-NEXT:    retl
 ;
 ; SKX-LABEL: zext_i8_index:
 ; SKX:       # %bb.0:
-; SKX-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; SKX-NEXT:    vpmovzxwq {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
-; SKX-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; SKX-NEXT:    vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; SKX-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; SKX-NEXT:    kxnorw %k0, %k0, %k1
-; SKX-NEXT:    kxnorw %k0, %k0, %k2
-; SKX-NEXT:    vgatherqps (%rdi,%zmm0,4), %ymm2 {%k2}
-; SKX-NEXT:    vgatherqps (%rdi,%zmm1,4), %ymm0 {%k1}
-; SKX-NEXT:    vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
+; SKX-NEXT:    vgatherdps (%rdi,%zmm1,4), %zmm0 {%k1}
 ; SKX-NEXT:    retq
 ;
 ; SKX_32-LABEL: zext_i8_index:
 ; SKX_32:       # %bb.0:
 ; SKX_32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; SKX_32-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
-; SKX_32-NEXT:    vpmovzxwq {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
-; SKX_32-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; SKX_32-NEXT:    vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; SKX_32-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; SKX_32-NEXT:    kxnorw %k0, %k0, %k1
-; SKX_32-NEXT:    kxnorw %k0, %k0, %k2
-; SKX_32-NEXT:    vgatherqps (%eax,%zmm0,4), %ymm2 {%k2}
-; SKX_32-NEXT:    vgatherqps (%eax,%zmm1,4), %ymm0 {%k1}
-; SKX_32-NEXT:    vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
+; SKX_32-NEXT:    vgatherdps (%eax,%zmm1,4), %zmm0 {%k1}
 ; SKX_32-NEXT:    retl
 
   %zext_ind = zext <16 x i8> %ind to <16 x i64>
@@ -2756,32 +2732,36 @@ define <16 x float> @zext_i8_index(float* %base, <16 x i8> %ind) {
 define <8 x float> @zext_v8i8_index(float* %base, <8 x i8> %ind) {
 ; KNL_64-LABEL: zext_v8i8_index:
 ; KNL_64:       # %bb.0:
-; KNL_64-NEXT:    vpmovzxbq {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero,xmm0[2],zero,zero,zero,zero,zero,zero,zero,xmm0[3],zero,zero,zero,zero,zero,zero,zero,xmm0[4],zero,zero,zero,zero,zero,zero,zero,xmm0[5],zero,zero,zero,zero,zero,zero,zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,xmm0[7],zero,zero,zero,zero,zero,zero,zero
-; KNL_64-NEXT:    kxnorw %k0, %k0, %k1
-; KNL_64-NEXT:    vgatherqps (%rdi,%zmm1,4), %ymm0 {%k1}
+; KNL_64-NEXT:    vpmovzxbd {{.*#+}} ymm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; KNL_64-NEXT:    movw $255, %ax
+; KNL_64-NEXT:    kmovw %eax, %k1
+; KNL_64-NEXT:    vgatherdps (%rdi,%zmm1,4), %zmm0 {%k1}
+; KNL_64-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; KNL_64-NEXT:    retq
 ;
 ; KNL_32-LABEL: zext_v8i8_index:
 ; KNL_32:       # %bb.0:
 ; KNL_32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; KNL_32-NEXT:    vpmovzxbq {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero,xmm0[2],zero,zero,zero,zero,zero,zero,zero,xmm0[3],zero,zero,zero,zero,zero,zero,zero,xmm0[4],zero,zero,zero,zero,zero,zero,zero,xmm0[5],zero,zero,zero,zero,zero,zero,zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,xmm0[7],zero,zero,zero,zero,zero,zero,zero
-; KNL_32-NEXT:    kxnorw %k0, %k0, %k1
-; KNL_32-NEXT:    vgatherqps (%eax,%zmm1,4), %ymm0 {%k1}
+; KNL_32-NEXT:    vpmovzxbd {{.*#+}} ymm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; KNL_32-NEXT:    movw $255, %cx
+; KNL_32-NEXT:    kmovw %ecx, %k1
+; KNL_32-NEXT:    vgatherdps (%eax,%zmm1,4), %zmm0 {%k1}
+; KNL_32-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; KNL_32-NEXT:    retl
 ;
 ; SKX-LABEL: zext_v8i8_index:
 ; SKX:       # %bb.0:
-; SKX-NEXT:    vpmovzxbq {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero,xmm0[2],zero,zero,zero,zero,zero,zero,zero,xmm0[3],zero,zero,zero,zero,zero,zero,zero,xmm0[4],zero,zero,zero,zero,zero,zero,zero,xmm0[5],zero,zero,zero,zero,zero,zero,zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,xmm0[7],zero,zero,zero,zero,zero,zero,zero
+; SKX-NEXT:    vpmovzxbd {{.*#+}} ymm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
 ; SKX-NEXT:    kxnorw %k0, %k0, %k1
-; SKX-NEXT:    vgatherqps (%rdi,%zmm1,4), %ymm0 {%k1}
+; SKX-NEXT:    vgatherdps (%rdi,%ymm1,4), %ymm0 {%k1}
 ; SKX-NEXT:    retq
 ;
 ; SKX_32-LABEL: zext_v8i8_index:
 ; SKX_32:       # %bb.0:
 ; SKX_32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; SKX_32-NEXT:    vpmovzxbq {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero,xmm0[2],zero,zero,zero,zero,zero,zero,zero,xmm0[3],zero,zero,zero,zero,zero,zero,zero,xmm0[4],zero,zero,zero,zero,zero,zero,zero,xmm0[5],zero,zero,zero,zero,zero,zero,zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,xmm0[7],zero,zero,zero,zero,zero,zero,zero
+; SKX_32-NEXT:    vpmovzxbd {{.*#+}} ymm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
 ; SKX_32-NEXT:    kxnorw %k0, %k0, %k1
-; SKX_32-NEXT:    vgatherqps (%eax,%zmm1,4), %ymm0 {%k1}
+; SKX_32-NEXT:    vgatherdps (%eax,%ymm1,4), %ymm0 {%k1}
 ; SKX_32-NEXT:    retl
 
   %zext_ind = zext <8 x i8> %ind to <8 x i64>