X86 ISel: Basic support for variable-index vector permutations

[zvi]

Summary

Try to lower a BUILD_VECTOR composed of extract-extract chains that can be
reasoned to be a permutation of a vector by indices in a non-constant vector.

We saw this pattern created by ISPC, which resolts to creating it due to the
requirement that shufflevector's mask operand be a *constant* vector.
I didn't check this but we could possibly use this pattern for lowering the X86 permute
C-instrinsics instead of llvm.x86 instrinsics.

This change can be followed by more improvements:

1. Handle vectors with undef elements.
2. Utilize pshufb and zero-mask-blending to support more effiecient construction of vectors with constant-0 elements.
3. Use smaller-element vectors of same width, and "interpolate" the indices, when no native operation available.

[zvi]

ping

[RKSimon]

Why no float/double support?

[zvi]

Add support for floating-point types.

[RKSimon]

lib/Target/X86/X86ISelLowering.cpp

7719 ↗

(On Diff #120887)

efficient

7726 ↗

(On Diff #120887)

(style) Full stops at the end of comments.

7777 ↗

(On Diff #120887)

Is it worth putting all this code above the switch statement? As soon as you want to start adding more cases you're going to have to know more details of the source vectors/indices to determine whether the subtarget supports them.

7788 ↗

(On Diff #120887)

You can probably pull the '0th idx' code above into the for loop quite easily, it'd also make future support for UNDEF/ZERO elements easier.

[zvi]

Follow Simon's suggestion to pull the oth index scan into the loop.
Fix typo + style defects.

[RKSimon]

LGTM with one minor comment

lib/Target/X86/X86ISelLowering.cpp

7802 ↗

(On Diff #121398)

Probably clearer to just do something like:

SDValue ExtrIndices = ExtractedIndex.getOperand(0);
if (V && V != ExtrIndices)
  return SDValue();
V = ExtrIndices;

[zvi]

Last minute changes in the spirit of Simon's suggestion

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

if (VT == MVT::v4i32 || VT == MVT::v16i32 ||
       return Op;
 
     return getOnesVector(VT, DAG, DL);
   }
 
   return SDValue();
 }
 
+// Tries to lower a BUILD_VECTOR composed of extract-extract chains that can be
+// reasoned to be a permutation of a vector by indices in a non-constant vector.
+// (build_vector (extract_elt V, (extract_elt I, 0)),
+//               (extract_elt V, (extract_elt I, 1)),
+//                    ...
+// ->
+// (vpermv I, V)
+//
+// TODO: Handle undefs
+// TODO: Utilize pshufb and zero mask blending to support more efficient
+// construction of vectors with constant-0 elements.
+// TODO: Use smaller-element vectors of same width, and "interpolate" the indices,
+// when no native operation available.
+static SDValue
+LowerBUILD_VECTORAsVariablePermute(SDValue V, SelectionDAG &DAG,
+                                   const X86Subtarget &Subtarget) {
+  // Look for VPERMV and PSHUFB opportunities.
+  MVT VT = V.getSimpleValueType();
+  switch (VT.SimpleTy) {
+  default:
+    return SDValue();
+  case MVT::v16i8:
+    if (!Subtarget.hasSSE3())
+      return SDValue();
+    break;
+  case MVT::v8f32:
+  case MVT::v8i32:
+    if (!Subtarget.hasAVX2())
+      return SDValue();
+    break;
+  case MVT::v4i64:
+  case MVT::v4f64:
+    if (!Subtarget.hasVLX())
+      return SDValue();
+    break;
+  case MVT::v16f32:
+  case MVT::v8f64:
+  case MVT::v16i32:
+  case MVT::v8i64:
+    if (!Subtarget.hasAVX512())
+      return SDValue();
+    break;
+  case MVT::v32i16:
+    if (!Subtarget.hasBWI())
+      return SDValue();
+    break;
+  case MVT::v8i16:
+  case MVT::v16i16:
+    if (!Subtarget.hasVLX() || !Subtarget.hasBWI())
+      return SDValue();
+    break;
+  case MVT::v64i8:
+    if (!Subtarget.hasVBMI())
+      return SDValue();
+    break;
+  case MVT::v32i8:
+    if (!Subtarget.hasVLX() || !Subtarget.hasVBMI())
+      return SDValue();
+    break;
+  }
+  SDValue SrcVec, IndicesVec;
+  // Check for a match of the permute source vector and permute index elements.
+  // This is done by checking that the i-th build_vector operand is of the form:
+  // (extract_elt SrcVec, (extract_elt IndicesVec, i)).
+  for (unsigned Idx = 0, E = V.getNumOperands(); Idx != E; ++Idx) {
+    SDValue Op = V.getOperand(Idx);
+    if (Op.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
+      return SDValue();
+
+    // If this is the first extract encountered in V, set the source vector,
+    // otherwise verify the extract is from the previously defined source
+    // vector.
+    if (!SrcVec)
+      SrcVec = Op.getOperand(0);
+    else if (SrcVec != Op.getOperand(0))
+      return SDValue();
+    SDValue ExtractedIndex = Op->getOperand(1);
+    // Peek through extends.
+    if (ExtractedIndex.getOpcode() == ISD::ZERO_EXTEND ||
+        ExtractedIndex.getOpcode() == ISD::SIGN_EXTEND)
+      ExtractedIndex = ExtractedIndex.getOperand(0);
+    if (ExtractedIndex.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
+      return SDValue();
+
+    // If this is the first extract from the index vector candidate, set the
+    // indices vector, otherwise verify the extract is from the previously
+    // defined indices vector.
+    if (!IndicesVec)
+      IndicesVec = ExtractedIndex.getOperand(0);
+    else if (IndicesVec != ExtractedIndex.getOperand(0))
+      return SDValue();
+
+    auto *PermIdx = dyn_cast<ConstantSDNode>(ExtractedIndex.getOperand(1));
+    if (!PermIdx || PermIdx->getZExtValue() != Idx)
+      return SDValue();
+  }
+  MVT IndicesVT = VT;
+  if (VT.isFloatingPoint())
+    IndicesVT = MVT::getVectorVT(MVT::getIntegerVT(VT.getScalarSizeInBits()),
+                                 VT.getVectorNumElements());
+  IndicesVec = DAG.getZExtOrTrunc(IndicesVec, SDLoc(IndicesVec), IndicesVT);
+  return DAG.getNode(VT == MVT::v16i8 ? X86ISD::PSHUFB : X86ISD::VPERMV,
+                     SDLoc(V), VT, IndicesVec, SrcVec);
+}
+
 SDValue
 X86TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const {
   SDLoc dl(Op);
 
   MVT VT = Op.getSimpleValueType();
   MVT ExtVT = VT.getVectorElementType();
   unsigned NumElems = Op.getNumOperands();
 
if (Values.size() == 1) {
     return SDValue();
   }
 
   // A vector full of immediates; various special cases are already
   // handled, so this is best done with a single constant-pool load.
   if (IsAllConstants)
     return SDValue();
 
+  if (SDValue V = LowerBUILD_VECTORAsVariablePermute(Op, DAG, Subtarget))
+      return V;
+
   // See if we can use a vector load to get all of the elements.
   if (VT.is128BitVector() || VT.is256BitVector() || VT.is512BitVector()) {
     SmallVector<SDValue, 64> Ops(Op->op_begin(), Op->op_begin() + NumElems);
     if (SDValue LD =
             EltsFromConsecutiveLoads(VT, Ops, dl, DAG, Subtarget, false))
       return LD;
   }
 

llvm/trunk/test/CodeGen/X86/var-permute-128.ll

 ; SSSE3-NEXT:    movd %eax, %xmm3
 ; SSSE3-NEXT:    movzwl -24(%rsp,%r8,2), %eax
 ; SSSE3-NEXT:    movd %eax, %xmm0
 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
 ; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm2[0]
 ; SSSE3-NEXT:    retq
 ;
-; AVX-LABEL: var_shuffle_v8i16:
-; AVX:       # BB#0:
-; AVX-NEXT:    vmovd %xmm1, %eax
-; AVX-NEXT:    vpextrw $1, %xmm1, %r10d
-; AVX-NEXT:    vpextrw $2, %xmm1, %ecx
-; AVX-NEXT:    vpextrw $3, %xmm1, %edx
-; AVX-NEXT:    vpextrw $4, %xmm1, %esi
-; AVX-NEXT:    vpextrw $5, %xmm1, %edi
-; AVX-NEXT:    vpextrw $6, %xmm1, %r8d
-; AVX-NEXT:    vpextrw $7, %xmm1, %r9d
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $7, %eax
-; AVX-NEXT:    andl $7, %r10d
-; AVX-NEXT:    andl $7, %ecx
-; AVX-NEXT:    andl $7, %edx
-; AVX-NEXT:    andl $7, %esi
-; AVX-NEXT:    andl $7, %edi
-; AVX-NEXT:    andl $7, %r8d
-; AVX-NEXT:    andl $7, %r9d
-; AVX-NEXT:    movzwl -24(%rsp,%rax,2), %eax
-; AVX-NEXT:    vmovd %eax, %xmm0
-; AVX-NEXT:    vpinsrw $1, -24(%rsp,%r10,2), %xmm0, %xmm0
-; AVX-NEXT:    vpinsrw $2, -24(%rsp,%rcx,2), %xmm0, %xmm0
-; AVX-NEXT:    vpinsrw $3, -24(%rsp,%rdx,2), %xmm0, %xmm0
-; AVX-NEXT:    vpinsrw $4, -24(%rsp,%rsi,2), %xmm0, %xmm0
-; AVX-NEXT:    vpinsrw $5, -24(%rsp,%rdi,2), %xmm0, %xmm0
-; AVX-NEXT:    vpinsrw $6, -24(%rsp,%r8,2), %xmm0, %xmm0
-; AVX-NEXT:    vpinsrw $7, -24(%rsp,%r9,2), %xmm0, %xmm0
-; AVX-NEXT:    retq
+; AVXNOVLBW-LABEL: var_shuffle_v8i16:
+; AVXNOVLBW:       # BB#0:
+; AVXNOVLBW-NEXT:    vmovd %xmm1, %eax
+; AVXNOVLBW-NEXT:    vpextrw $1, %xmm1, %r10d
+; AVXNOVLBW-NEXT:    vpextrw $2, %xmm1, %ecx
+; AVXNOVLBW-NEXT:    vpextrw $3, %xmm1, %edx
+; AVXNOVLBW-NEXT:    vpextrw $4, %xmm1, %esi
+; AVXNOVLBW-NEXT:    vpextrw $5, %xmm1, %edi
+; AVXNOVLBW-NEXT:    vpextrw $6, %xmm1, %r8d
+; AVXNOVLBW-NEXT:    vpextrw $7, %xmm1, %r9d
+; AVXNOVLBW-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
+; AVXNOVLBW-NEXT:    andl $7, %eax
+; AVXNOVLBW-NEXT:    andl $7, %r10d
+; AVXNOVLBW-NEXT:    andl $7, %ecx
+; AVXNOVLBW-NEXT:    andl $7, %edx
+; AVXNOVLBW-NEXT:    andl $7, %esi
+; AVXNOVLBW-NEXT:    andl $7, %edi
+; AVXNOVLBW-NEXT:    andl $7, %r8d
+; AVXNOVLBW-NEXT:    andl $7, %r9d
+; AVXNOVLBW-NEXT:    movzwl -24(%rsp,%rax,2), %eax
+; AVXNOVLBW-NEXT:    vmovd %eax, %xmm0
+; AVXNOVLBW-NEXT:    vpinsrw $1, -24(%rsp,%r10,2), %xmm0, %xmm0
+; AVXNOVLBW-NEXT:    vpinsrw $2, -24(%rsp,%rcx,2), %xmm0, %xmm0
+; AVXNOVLBW-NEXT:    vpinsrw $3, -24(%rsp,%rdx,2), %xmm0, %xmm0
+; AVXNOVLBW-NEXT:    vpinsrw $4, -24(%rsp,%rsi,2), %xmm0, %xmm0
+; AVXNOVLBW-NEXT:    vpinsrw $5, -24(%rsp,%rdi,2), %xmm0, %xmm0
+; AVXNOVLBW-NEXT:    vpinsrw $6, -24(%rsp,%r8,2), %xmm0, %xmm0
+; AVXNOVLBW-NEXT:    vpinsrw $7, -24(%rsp,%r9,2), %xmm0, %xmm0
+; AVXNOVLBW-NEXT:    retq
+;
+; AVX512VLBW-LABEL: var_shuffle_v8i16:
+; AVX512VLBW:       # BB#0:
+; AVX512VLBW-NEXT:    vpermw %xmm0, %xmm1, %xmm0
+; AVX512VLBW-NEXT:    retq
   %index0 = extractelement <8 x i16> %indices, i32 0
   %index1 = extractelement <8 x i16> %indices, i32 1
   %index2 = extractelement <8 x i16> %indices, i32 2
   %index3 = extractelement <8 x i16> %indices, i32 3
   %index4 = extractelement <8 x i16> %indices, i32 4
   %index5 = extractelement <8 x i16> %indices, i32 5
   %index6 = extractelement <8 x i16> %indices, i32 6
   %index7 = extractelement <8 x i16> %indices, i32 7
; AVX512VLBW-NEXT:    retq
   %ret6 = insertelement <8 x i16> %ret5, i16 %v6, i32 6
   %ret7 = insertelement <8 x i16> %ret6, i16 %v7, i32 7
   ret <8 x i16> %ret7
 }
 
 define <16 x i8> @var_shuffle_v16i8(<16 x i8> %v, <16 x i8> %indices) nounwind {
 ; SSSE3-LABEL: var_shuffle_v16i8:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    movaps %xmm1, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    leaq -{{[0-9]+}}(%rsp), %rax
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm8
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm15
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm9
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm3
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm10
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm7
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm11
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm6
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm12
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm5
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm13
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm4
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm14
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm1
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %ecx
-; SSSE3-NEXT:    movd %ecx, %xmm2
-; SSSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSSE3-NEXT:    andl $15, %ecx
-; SSSE3-NEXT:    movzbl (%rcx,%rax), %eax
-; SSSE3-NEXT:    movd %eax, %xmm0
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0],xmm8[0],xmm15[1],xmm8[1],xmm15[2],xmm8[2],xmm15[3],xmm8[3],xmm15[4],xmm8[4],xmm15[5],xmm8[5],xmm15[6],xmm8[6],xmm15[7],xmm8[7]
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm9[0],xmm3[1],xmm9[1],xmm3[2],xmm9[2],xmm3[3],xmm9[3],xmm3[4],xmm9[4],xmm3[5],xmm9[5],xmm3[6],xmm9[6],xmm3[7],xmm9[7]
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm15[0],xmm3[1],xmm15[1],xmm3[2],xmm15[2],xmm3[3],xmm15[3]
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm10[0],xmm7[1],xmm10[1],xmm7[2],xmm10[2],xmm7[3],xmm10[3],xmm7[4],xmm10[4],xmm7[5],xmm10[5],xmm7[6],xmm10[6],xmm7[7],xmm10[7]
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm11[0],xmm6[1],xmm11[1],xmm6[2],xmm11[2],xmm6[3],xmm11[3],xmm6[4],xmm11[4],xmm6[5],xmm11[5],xmm6[6],xmm11[6],xmm6[7],xmm11[7]
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1],xmm6[2],xmm7[2],xmm6[3],xmm7[3]
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm6 = xmm6[0],xmm3[0],xmm6[1],xmm3[1]
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm12[0],xmm5[1],xmm12[1],xmm5[2],xmm12[2],xmm5[3],xmm12[3],xmm5[4],xmm12[4],xmm5[5],xmm12[5],xmm5[6],xmm12[6],xmm5[7],xmm12[7]
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm13[0],xmm4[1],xmm13[1],xmm4[2],xmm13[2],xmm4[3],xmm13[3],xmm4[4],xmm13[4],xmm4[5],xmm13[5],xmm4[6],xmm13[6],xmm4[7],xmm13[7]
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3]
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm14[0],xmm1[1],xmm14[1],xmm1[2],xmm14[2],xmm1[3],xmm14[3],xmm1[4],xmm14[4],xmm1[5],xmm14[5],xmm1[6],xmm14[6],xmm1[7],xmm14[7]
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
-; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm6[0]
+; SSSE3-NEXT:    pshufb %xmm0, %xmm1
+; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; AVX-LABEL: var_shuffle_v16i8:
 ; AVX:       # BB#0:
-; AVX-NEXT:    vpextrb $0, %xmm1, %eax
-; AVX-NEXT:    vmovaps %xmm0, -{{[0-9]+}}(%rsp)
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    leaq -{{[0-9]+}}(%rsp), %rcx
-; AVX-NEXT:    movzbl (%rax,%rcx), %eax
-; AVX-NEXT:    vmovd %eax, %xmm0
-; AVX-NEXT:    vpextrb $1, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $1, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $2, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $2, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $3, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $3, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $4, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $4, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $5, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $5, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $6, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $6, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $7, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $7, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $8, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $8, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $9, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $9, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $10, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $10, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $11, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $11, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $12, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $13, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $13, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $14, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $14, (%rax,%rcx), %xmm0, %xmm0
-; AVX-NEXT:    vpextrb $15, %xmm1, %eax
-; AVX-NEXT:    andl $15, %eax
-; AVX-NEXT:    vpinsrb $15, (%rax,%rcx), %xmm0, %xmm0
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
   %index0 = extractelement <16 x i8> %indices, i32 0
   %index1 = extractelement <16 x i8> %indices, i32 1
   %index2 = extractelement <16 x i8> %indices, i32 2
   %index3 = extractelement <16 x i8> %indices, i32 3
   %index4 = extractelement <16 x i8> %indices, i32 4
   %index5 = extractelement <16 x i8> %indices, i32 5
   %index6 = extractelement <16 x i8> %indices, i32 6

llvm/trunk/test/CodeGen/X86/var-permute-256.ll

 ; AVX1-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
 ; AVX1-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
 ; AVX1-NEXT:    vmovlhps {{.*#+}} xmm1 = xmm2[0],xmm1[0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    movq %rbp, %rsp
 ; AVX1-NEXT:    popq %rbp
 ; AVX1-NEXT:    retq
 ;
-; INT256-LABEL: var_shuffle_v4i64:
-; INT256:       # BB#0:
-; INT256-NEXT:    pushq %rbp
-; INT256-NEXT:    movq %rsp, %rbp
-; INT256-NEXT:    andq $-32, %rsp
-; INT256-NEXT:    subq $64, %rsp
-; INT256-NEXT:    vmovq %xmm1, %rax
-; INT256-NEXT:    andl $3, %eax
-; INT256-NEXT:    vpextrq $1, %xmm1, %rcx
-; INT256-NEXT:    andl $3, %ecx
-; INT256-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; INT256-NEXT:    vmovq %xmm1, %rdx
-; INT256-NEXT:    andl $3, %edx
-; INT256-NEXT:    vpextrq $1, %xmm1, %rsi
-; INT256-NEXT:    andl $3, %esi
-; INT256-NEXT:    vmovaps %ymm0, (%rsp)
-; INT256-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; INT256-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
-; INT256-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; INT256-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
-; INT256-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
-; INT256-NEXT:    vmovlhps {{.*#+}} xmm1 = xmm2[0],xmm1[0]
-; INT256-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
-; INT256-NEXT:    movq %rbp, %rsp
-; INT256-NEXT:    popq %rbp
-; INT256-NEXT:    retq
+; AVX2-LABEL: var_shuffle_v4i64:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    pushq %rbp
+; AVX2-NEXT:    movq %rsp, %rbp
+; AVX2-NEXT:    andq $-32, %rsp
+; AVX2-NEXT:    subq $64, %rsp
+; AVX2-NEXT:    vmovq %xmm1, %rax
+; AVX2-NEXT:    andl $3, %eax
+; AVX2-NEXT:    vpextrq $1, %xmm1, %rcx
+; AVX2-NEXT:    andl $3, %ecx
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm1
+; AVX2-NEXT:    vmovq %xmm1, %rdx
+; AVX2-NEXT:    andl $3, %edx
+; AVX2-NEXT:    vpextrq $1, %xmm1, %rsi
+; AVX2-NEXT:    andl $3, %esi
+; AVX2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX2-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX2-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX2-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
+; AVX2-NEXT:    vmovlhps {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX2-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    movq %rbp, %rsp
+; AVX2-NEXT:    popq %rbp
+; AVX2-NEXT:    retq
+;
+; AVX512F-LABEL: var_shuffle_v4i64:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    pushq %rbp
+; AVX512F-NEXT:    movq %rsp, %rbp
+; AVX512F-NEXT:    andq $-32, %rsp
+; AVX512F-NEXT:    subq $64, %rsp
+; AVX512F-NEXT:    vmovq %xmm1, %rax
+; AVX512F-NEXT:    andl $3, %eax
+; AVX512F-NEXT:    vpextrq $1, %xmm1, %rcx
+; AVX512F-NEXT:    andl $3, %ecx
+; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm1
+; AVX512F-NEXT:    vmovq %xmm1, %rdx
+; AVX512F-NEXT:    andl $3, %edx
+; AVX512F-NEXT:    vpextrq $1, %xmm1, %rsi
+; AVX512F-NEXT:    andl $3, %esi
+; AVX512F-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX512F-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX512F-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX512F-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX512F-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX512F-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
+; AVX512F-NEXT:    vmovlhps {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX512F-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX512F-NEXT:    movq %rbp, %rsp
+; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_shuffle_v4i64:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    vpermpd %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    retq
+;
+; AVX512VLBW-LABEL: var_shuffle_v4i64:
+; AVX512VLBW:       # BB#0:
+; AVX512VLBW-NEXT:    vpermpd %ymm0, %ymm1, %ymm0
+; AVX512VLBW-NEXT:    retq
   %index0 = extractelement <4 x i64> %indices, i32 0
   %index1 = extractelement <4 x i64> %indices, i32 1
   %index2 = extractelement <4 x i64> %indices, i32 2
   %index3 = extractelement <4 x i64> %indices, i32 3
   %v0 = extractelement <4 x i64> %v, i64 %index0
   %v1 = extractelement <4 x i64> %v, i64 %index1
   %v2 = extractelement <4 x i64> %v, i64 %index2
   %v3 = extractelement <4 x i64> %v, i64 %index3

 ; AVX1-NEXT:    vpinsrd $3, (%rcx,%rax), %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    movq %rbp, %rsp
 ; AVX1-NEXT:    popq %rbp
 ; AVX1-NEXT:    retq
 ;
 ; INT256-LABEL: var_shuffle_v8i32:
 ; INT256:       # BB#0:
-; INT256-NEXT:    pushq %rbp
-; INT256-NEXT:    movq %rsp, %rbp
-; INT256-NEXT:    andq $-32, %rsp
-; INT256-NEXT:    subq $64, %rsp
-; INT256-NEXT:    vpextrq $1, %xmm1, %r8
-; INT256-NEXT:    movq %r8, %rcx
-; INT256-NEXT:    shrq $30, %rcx
-; INT256-NEXT:    vmovq %xmm1, %r9
-; INT256-NEXT:    movq %r9, %rsi
-; INT256-NEXT:    shrq $30, %rsi
-; INT256-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; INT256-NEXT:    vpextrq $1, %xmm1, %r10
-; INT256-NEXT:    movq %r10, %rdi
-; INT256-NEXT:    shrq $30, %rdi
-; INT256-NEXT:    vmovq %xmm1, %rax
-; INT256-NEXT:    movq %rax, %rdx
-; INT256-NEXT:    shrq $30, %rdx
-; INT256-NEXT:    vmovaps %ymm0, (%rsp)
-; INT256-NEXT:    andl $7, %r9d
-; INT256-NEXT:    andl $28, %esi
-; INT256-NEXT:    andl $7, %r8d
-; INT256-NEXT:    andl $28, %ecx
-; INT256-NEXT:    andl $7, %eax
-; INT256-NEXT:    andl $28, %edx
-; INT256-NEXT:    andl $7, %r10d
-; INT256-NEXT:    andl $28, %edi
-; INT256-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; INT256-NEXT:    movq %rsp, %rax
-; INT256-NEXT:    vpinsrd $1, (%rdx,%rax), %xmm0, %xmm0
-; INT256-NEXT:    vpinsrd $2, (%rsp,%r10,4), %xmm0, %xmm0
-; INT256-NEXT:    vpinsrd $3, (%rdi,%rax), %xmm0, %xmm0
-; INT256-NEXT:    vmovd {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; INT256-NEXT:    vpinsrd $1, (%rsi,%rax), %xmm1, %xmm1
-; INT256-NEXT:    vpinsrd $2, (%rsp,%r8,4), %xmm1, %xmm1
-; INT256-NEXT:    vpinsrd $3, (%rcx,%rax), %xmm1, %xmm1
-; INT256-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; INT256-NEXT:    movq %rbp, %rsp
-; INT256-NEXT:    popq %rbp
+; INT256-NEXT:    vpermps %ymm0, %ymm1, %ymm0
 ; INT256-NEXT:    retq
   %index0 = extractelement <8 x i32> %indices, i32 0
   %index1 = extractelement <8 x i32> %indices, i32 1
   %index2 = extractelement <8 x i32> %indices, i32 2
   %index3 = extractelement <8 x i32> %indices, i32 3
   %index4 = extractelement <8 x i32> %indices, i32 4
   %index5 = extractelement <8 x i32> %indices, i32 5
   %index6 = extractelement <8 x i32> %indices, i32 6

 ; AVX1-NEXT:    vpextrw $7, %xmm1, %eax
 ; AVX1-NEXT:    andl $15, %eax
 ; AVX1-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm2, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    movq %rbp, %rsp
 ; AVX1-NEXT:    popq %rbp
 ; AVX1-NEXT:    retq
 ;
-; INT256-LABEL: var_shuffle_v16i16:
-; INT256:       # BB#0:
-; INT256-NEXT:    pushq %rbp
-; INT256-NEXT:    movq %rsp, %rbp
-; INT256-NEXT:    andq $-32, %rsp
-; INT256-NEXT:    subq $64, %rsp
-; INT256-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; INT256-NEXT:    vmovd %xmm2, %eax
-; INT256-NEXT:    vmovaps %ymm0, (%rsp)
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    movzwl (%rsp,%rax,2), %eax
-; INT256-NEXT:    vmovd %eax, %xmm0
-; INT256-NEXT:    vpextrw $1, %xmm2, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm0, %xmm0
-; INT256-NEXT:    vpextrw $2, %xmm2, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm0, %xmm0
-; INT256-NEXT:    vpextrw $3, %xmm2, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm0, %xmm0
-; INT256-NEXT:    vpextrw $4, %xmm2, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm0, %xmm0
-; INT256-NEXT:    vpextrw $5, %xmm2, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm0, %xmm0
-; INT256-NEXT:    vpextrw $6, %xmm2, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm0, %xmm0
-; INT256-NEXT:    vpextrw $7, %xmm2, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm0, %xmm0
-; INT256-NEXT:    vmovd %xmm1, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    movzwl (%rsp,%rax,2), %eax
-; INT256-NEXT:    vmovd %eax, %xmm2
-; INT256-NEXT:    vpextrw $1, %xmm1, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm2, %xmm2
-; INT256-NEXT:    vpextrw $2, %xmm1, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm2, %xmm2
-; INT256-NEXT:    vpextrw $3, %xmm1, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm2, %xmm2
-; INT256-NEXT:    vpextrw $4, %xmm1, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm2, %xmm2
-; INT256-NEXT:    vpextrw $5, %xmm1, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm2, %xmm2
-; INT256-NEXT:    vpextrw $6, %xmm1, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm2, %xmm2
-; INT256-NEXT:    vpextrw $7, %xmm1, %eax
-; INT256-NEXT:    andl $15, %eax
-; INT256-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm2, %xmm1
-; INT256-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; INT256-NEXT:    movq %rbp, %rsp
-; INT256-NEXT:    popq %rbp
-; INT256-NEXT:    retq
+; AVX2-LABEL: var_shuffle_v16i16:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    pushq %rbp
+; AVX2-NEXT:    movq %rsp, %rbp
+; AVX2-NEXT:    andq $-32, %rsp
+; AVX2-NEXT:    subq $64, %rsp
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT:    vmovd %xmm2, %eax
+; AVX2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    movzwl (%rsp,%rax,2), %eax
+; AVX2-NEXT:    vmovd %eax, %xmm0
+; AVX2-NEXT:    vpextrw $1, %xmm2, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpextrw $2, %xmm2, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpextrw $3, %xmm2, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpextrw $4, %xmm2, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpextrw $5, %xmm2, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpextrw $6, %xmm2, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vpextrw $7, %xmm2, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX2-NEXT:    vmovd %xmm1, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    movzwl (%rsp,%rax,2), %eax
+; AVX2-NEXT:    vmovd %eax, %xmm2
+; AVX2-NEXT:    vpextrw $1, %xmm1, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrw $2, %xmm1, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrw $3, %xmm1, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrw $4, %xmm1, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrw $5, %xmm1, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrw $6, %xmm1, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrw $7, %xmm1, %eax
+; AVX2-NEXT:    andl $15, %eax
+; AVX2-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm2, %xmm1
+; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    movq %rbp, %rsp
+; AVX2-NEXT:    popq %rbp
+; AVX2-NEXT:    retq
+;
+; AVX512F-LABEL: var_shuffle_v16i16:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    pushq %rbp
+; AVX512F-NEXT:    movq %rsp, %rbp
+; AVX512F-NEXT:    andq $-32, %rsp
+; AVX512F-NEXT:    subq $64, %rsp
+; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX512F-NEXT:    vmovd %xmm2, %eax
+; AVX512F-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    movzwl (%rsp,%rax,2), %eax
+; AVX512F-NEXT:    vmovd %eax, %xmm0
+; AVX512F-NEXT:    vpextrw $1, %xmm2, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrw $2, %xmm2, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrw $3, %xmm2, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrw $4, %xmm2, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrw $5, %xmm2, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrw $6, %xmm2, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrw $7, %xmm2, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512F-NEXT:    vmovd %xmm1, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    movzwl (%rsp,%rax,2), %eax
+; AVX512F-NEXT:    vmovd %eax, %xmm2
+; AVX512F-NEXT:    vpextrw $1, %xmm1, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrw $2, %xmm1, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrw $3, %xmm1, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrw $4, %xmm1, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrw $5, %xmm1, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrw $6, %xmm1, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrw $7, %xmm1, %eax
+; AVX512F-NEXT:    andl $15, %eax
+; AVX512F-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm2, %xmm1
+; AVX512F-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512F-NEXT:    movq %rbp, %rsp
+; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_shuffle_v16i16:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    pushq %rbp
+; AVX512VL-NEXT:    movq %rsp, %rbp
+; AVX512VL-NEXT:    andq $-32, %rsp
+; AVX512VL-NEXT:    subq $64, %rsp
+; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX512VL-NEXT:    vmovd %xmm2, %eax
+; AVX512VL-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    movzwl (%rsp,%rax,2), %eax
+; AVX512VL-NEXT:    vmovd %eax, %xmm0
+; AVX512VL-NEXT:    vpextrw $1, %xmm2, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrw $2, %xmm2, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrw $3, %xmm2, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrw $4, %xmm2, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrw $5, %xmm2, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrw $6, %xmm2, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrw $7, %xmm2, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm0, %xmm0
+; AVX512VL-NEXT:    vmovd %xmm1, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    movzwl (%rsp,%rax,2), %eax
+; AVX512VL-NEXT:    vmovd %eax, %xmm2
+; AVX512VL-NEXT:    vpextrw $1, %xmm1, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrw $2, %xmm1, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrw $3, %xmm1, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrw $4, %xmm1, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrw $5, %xmm1, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrw $6, %xmm1, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrw $7, %xmm1, %eax
+; AVX512VL-NEXT:    andl $15, %eax
+; AVX512VL-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm2, %xmm1
+; AVX512VL-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    movq %rbp, %rsp
+; AVX512VL-NEXT:    popq %rbp
+; AVX512VL-NEXT:    retq
+;
+; AVX512VLBW-LABEL: var_shuffle_v16i16:
+; AVX512VLBW:       # BB#0:
+; AVX512VLBW-NEXT:    vpermw %ymm0, %ymm1, %ymm0
+; AVX512VLBW-NEXT:    retq
   %index0 = extractelement <16 x i16> %indices, i32 0
   %index1 = extractelement <16 x i16> %indices, i32 1
   %index2 = extractelement <16 x i16> %indices, i32 2
   %index3 = extractelement <16 x i16> %indices, i32 3
   %index4 = extractelement <16 x i16> %indices, i32 4
   %index5 = extractelement <16 x i16> %indices, i32 5
   %index6 = extractelement <16 x i16> %indices, i32 6
   %index7 = extractelement <16 x i16> %indices, i32 7

 ; AVX1-NEXT:    andl $31, %eax
 ; AVX1-NEXT:    movzbl (%rax,%rcx), %eax
 ; AVX1-NEXT:    vpinsrb $15, %eax, %xmm2, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    movq %rbp, %rsp
 ; AVX1-NEXT:    popq %rbp
 ; AVX1-NEXT:    retq
 ;
-; INT256-LABEL: var_shuffle_v32i8:
-; INT256:       # BB#0:
-; INT256-NEXT:    pushq %rbp
-; INT256-NEXT:    movq %rsp, %rbp
-; INT256-NEXT:    andq $-32, %rsp
-; INT256-NEXT:    subq $64, %rsp
-; INT256-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; INT256-NEXT:    vpextrb $0, %xmm2, %eax
-; INT256-NEXT:    vmovaps %ymm0, (%rsp)
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movq %rsp, %rcx
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vmovd %eax, %xmm0
-; INT256-NEXT:    vpextrb $1, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $2, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $3, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $4, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $5, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $6, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $7, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $8, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $8, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $9, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $9, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $10, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $10, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $11, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $11, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $12, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $12, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $13, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $13, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $14, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $15, %xmm2, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
-; INT256-NEXT:    vpextrb $0, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vmovd %eax, %xmm2
-; INT256-NEXT:    vpextrb $1, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $1, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $2, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $2, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $3, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $3, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $4, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $4, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $5, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $5, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $6, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $6, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $7, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $7, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $8, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $8, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $9, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $9, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $10, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $10, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $11, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $11, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $12, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $12, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $13, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $13, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $14, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    vpinsrb $14, (%rax,%rcx), %xmm2, %xmm2
-; INT256-NEXT:    vpextrb $15, %xmm1, %eax
-; INT256-NEXT:    andl $31, %eax
-; INT256-NEXT:    movzbl (%rax,%rcx), %eax
-; INT256-NEXT:    vpinsrb $15, %eax, %xmm2, %xmm1
-; INT256-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
-; INT256-NEXT:    movq %rbp, %rsp
-; INT256-NEXT:    popq %rbp
-; INT256-NEXT:    retq
+; AVX2-LABEL: var_shuffle_v32i8:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    pushq %rbp
+; AVX2-NEXT:    movq %rsp, %rbp
+; AVX2-NEXT:    andq $-32, %rsp
+; AVX2-NEXT:    subq $64, %rsp
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movq %rsp, %rcx
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vmovd %eax, %xmm0
+; AVX2-NEXT:    vpextrb $1, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $2, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $3, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $4, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $5, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $6, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $7, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $8, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $8, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $9, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $9, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $10, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $10, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $11, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $11, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $12, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $12, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $13, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $13, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $14, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $15, %xmm2, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
+; AVX2-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vmovd %eax, %xmm2
+; AVX2-NEXT:    vpextrb $1, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $1, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $2, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $2, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $3, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $3, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $4, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $4, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $5, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $5, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $6, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $6, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $7, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $7, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $8, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $8, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $9, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $9, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $10, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $10, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $11, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $11, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $12, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $12, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $13, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $13, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $14, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    vpinsrb $14, (%rax,%rcx), %xmm2, %xmm2
+; AVX2-NEXT:    vpextrb $15, %xmm1, %eax
+; AVX2-NEXT:    andl $31, %eax
+; AVX2-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX2-NEXT:    vpinsrb $15, %eax, %xmm2, %xmm1
+; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    movq %rbp, %rsp
+; AVX2-NEXT:    popq %rbp
+; AVX2-NEXT:    retq
+;
+; AVX512F-LABEL: var_shuffle_v32i8:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    pushq %rbp
+; AVX512F-NEXT:    movq %rsp, %rbp
+; AVX512F-NEXT:    andq $-32, %rsp
+; AVX512F-NEXT:    subq $64, %rsp
+; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX512F-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512F-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movq %rsp, %rcx
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vmovd %eax, %xmm0
+; AVX512F-NEXT:    vpextrb $1, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $2, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $3, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $4, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $5, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $6, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $7, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $8, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $8, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $9, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $9, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $10, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $10, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $11, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $11, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $12, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $12, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $13, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $13, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $14, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $15, %xmm2, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
+; AVX512F-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vmovd %eax, %xmm2
+; AVX512F-NEXT:    vpextrb $1, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $1, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $2, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $2, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $3, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $3, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $4, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $4, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $5, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $5, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $6, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $6, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $7, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $7, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $8, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $8, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $9, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $9, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $10, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $10, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $11, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $11, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $12, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $12, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $13, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $13, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $14, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    vpinsrb $14, (%rax,%rcx), %xmm2, %xmm2
+; AVX512F-NEXT:    vpextrb $15, %xmm1, %eax
+; AVX512F-NEXT:    andl $31, %eax
+; AVX512F-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512F-NEXT:    vpinsrb $15, %eax, %xmm2, %xmm1
+; AVX512F-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512F-NEXT:    movq %rbp, %rsp
+; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_shuffle_v32i8:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    pushq %rbp
+; AVX512VL-NEXT:    movq %rsp, %rbp
+; AVX512VL-NEXT:    andq $-32, %rsp
+; AVX512VL-NEXT:    subq $64, %rsp
+; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX512VL-NEXT:    vpextrb $0, %xmm2, %eax
+; AVX512VL-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movq %rsp, %rcx
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vmovd %eax, %xmm0
+; AVX512VL-NEXT:    vpextrb $1, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $1, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $2, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $2, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $3, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $4, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $4, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $5, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $5, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $6, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $6, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $7, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $7, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $8, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $8, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $9, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $9, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $10, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $10, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $11, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $11, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $12, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $12, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $13, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $13, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $14, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $14, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $15, %xmm2, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $15, %eax, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $0, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vmovd %eax, %xmm2
+; AVX512VL-NEXT:    vpextrb $1, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $1, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $2, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $2, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $3, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $3, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $4, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $4, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $5, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $5, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $6, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $6, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $7, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $7, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $8, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $8, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $9, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $9, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $10, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $10, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $11, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $11, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $12, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $12, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $13, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $13, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $14, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    vpinsrb $14, (%rax,%rcx), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpextrb $15, %xmm1, %eax
+; AVX512VL-NEXT:    andl $31, %eax
+; AVX512VL-NEXT:    movzbl (%rax,%rcx), %eax
+; AVX512VL-NEXT:    vpinsrb $15, %eax, %xmm2, %xmm1
+; AVX512VL-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    movq %rbp, %rsp
+; AVX512VL-NEXT:    popq %rbp
+; AVX512VL-NEXT:    retq
+;
+; VBMI-LABEL: var_shuffle_v32i8:
+; VBMI:       # BB#0:
+; VBMI-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; VBMI-NEXT:    retq
   %index0 = extractelement <32 x i8> %indices, i32 0
   %index1 = extractelement <32 x i8> %indices, i32 1
   %index2 = extractelement <32 x i8> %indices, i32 2
   %index3 = extractelement <32 x i8> %indices, i32 3
   %index4 = extractelement <32 x i8> %indices, i32 4
   %index5 = extractelement <32 x i8> %indices, i32 5
   %index6 = extractelement <32 x i8> %indices, i32 6
   %index7 = extractelement <32 x i8> %indices, i32 7

 ; AVX1-NEXT:    vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
 ; AVX1-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
 ; AVX1-NEXT:    vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    movq %rbp, %rsp
 ; AVX1-NEXT:    popq %rbp
 ; AVX1-NEXT:    retq
 ;
-; INT256-LABEL: var_shuffle_v4f64:
-; INT256:       # BB#0:
-; INT256-NEXT:    pushq %rbp
-; INT256-NEXT:    movq %rsp, %rbp
-; INT256-NEXT:    andq $-32, %rsp
-; INT256-NEXT:    subq $64, %rsp
-; INT256-NEXT:    vmovq %xmm1, %rax
-; INT256-NEXT:    andl $3, %eax
-; INT256-NEXT:    vpextrq $1, %xmm1, %rcx
-; INT256-NEXT:    andl $3, %ecx
-; INT256-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; INT256-NEXT:    vmovq %xmm1, %rdx
-; INT256-NEXT:    andl $3, %edx
-; INT256-NEXT:    vpextrq $1, %xmm1, %rsi
-; INT256-NEXT:    andl $3, %esi
-; INT256-NEXT:    vmovaps %ymm0, (%rsp)
-; INT256-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; INT256-NEXT:    vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
-; INT256-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
-; INT256-NEXT:    vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0]
-; INT256-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
-; INT256-NEXT:    movq %rbp, %rsp
-; INT256-NEXT:    popq %rbp
-; INT256-NEXT:    retq
+; AVX2-LABEL: var_shuffle_v4f64:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    pushq %rbp
+; AVX2-NEXT:    movq %rsp, %rbp
+; AVX2-NEXT:    andq $-32, %rsp
+; AVX2-NEXT:    subq $64, %rsp
+; AVX2-NEXT:    vmovq %xmm1, %rax
+; AVX2-NEXT:    andl $3, %eax
+; AVX2-NEXT:    vpextrq $1, %xmm1, %rcx
+; AVX2-NEXT:    andl $3, %ecx
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm1
+; AVX2-NEXT:    vmovq %xmm1, %rdx
+; AVX2-NEXT:    andl $3, %edx
+; AVX2-NEXT:    vpextrq $1, %xmm1, %rsi
+; AVX2-NEXT:    andl $3, %esi
+; AVX2-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX2-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX2-NEXT:    vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
+; AVX2-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX2-NEXT:    vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0]
+; AVX2-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    movq %rbp, %rsp
+; AVX2-NEXT:    popq %rbp
+; AVX2-NEXT:    retq
+;
+; AVX512F-LABEL: var_shuffle_v4f64:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    pushq %rbp
+; AVX512F-NEXT:    movq %rsp, %rbp
+; AVX512F-NEXT:    andq $-32, %rsp
+; AVX512F-NEXT:    subq $64, %rsp
+; AVX512F-NEXT:    vmovq %xmm1, %rax
+; AVX512F-NEXT:    andl $3, %eax
+; AVX512F-NEXT:    vpextrq $1, %xmm1, %rcx
+; AVX512F-NEXT:    andl $3, %ecx
+; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm1
+; AVX512F-NEXT:    vmovq %xmm1, %rdx
+; AVX512F-NEXT:    andl $3, %edx
+; AVX512F-NEXT:    vpextrq $1, %xmm1, %rsi
+; AVX512F-NEXT:    andl $3, %esi
+; AVX512F-NEXT:    vmovaps %ymm0, (%rsp)
+; AVX512F-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; AVX512F-NEXT:    vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
+; AVX512F-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX512F-NEXT:    vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0]
+; AVX512F-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX512F-NEXT:    movq %rbp, %rsp
+; AVX512F-NEXT:    popq %rbp
+; AVX512F-NEXT:    retq
+;
+; AVX512VL-LABEL: var_shuffle_v4f64:
+; AVX512VL:       # BB#0:
+; AVX512VL-NEXT:    vpermpd %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    retq
+;
+; AVX512VLBW-LABEL: var_shuffle_v4f64:
+; AVX512VLBW:       # BB#0:
+; AVX512VLBW-NEXT:    vpermpd %ymm0, %ymm1, %ymm0
+; AVX512VLBW-NEXT:    retq
   %index0 = extractelement <4 x i64> %indices, i32 0
   %index1 = extractelement <4 x i64> %indices, i32 1
   %index2 = extractelement <4 x i64> %indices, i32 2
   %index3 = extractelement <4 x i64> %indices, i32 3
   %v0 = extractelement <4 x double> %v, i64 %index0
   %v1 = extractelement <4 x double> %v, i64 %index1
   %v2 = extractelement <4 x double> %v, i64 %index2
   %v3 = extractelement <4 x double> %v, i64 %index3

 ; AVX1-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],mem[0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    movq %rbp, %rsp
 ; AVX1-NEXT:    popq %rbp
 ; AVX1-NEXT:    retq
 ;
 ; INT256-LABEL: var_shuffle_v8f32:
 ; INT256:       # BB#0:
-; INT256-NEXT:    pushq %rbp
-; INT256-NEXT:    movq %rsp, %rbp
-; INT256-NEXT:    andq $-32, %rsp
-; INT256-NEXT:    subq $64, %rsp
-; INT256-NEXT:    vpextrq $1, %xmm1, %r8
-; INT256-NEXT:    movq %r8, %rcx
-; INT256-NEXT:    shrq $30, %rcx
-; INT256-NEXT:    vmovq %xmm1, %r9
-; INT256-NEXT:    movq %r9, %rdx
-; INT256-NEXT:    shrq $30, %rdx
-; INT256-NEXT:    vextracti128 $1, %ymm1, %xmm1
-; INT256-NEXT:    vpextrq $1, %xmm1, %r10
-; INT256-NEXT:    movq %r10, %rdi
-; INT256-NEXT:    shrq $30, %rdi
-; INT256-NEXT:    vmovq %xmm1, %rax
-; INT256-NEXT:    movq %rax, %rsi
-; INT256-NEXT:    shrq $30, %rsi
-; INT256-NEXT:    vmovaps %ymm0, (%rsp)
-; INT256-NEXT:    andl $7, %r9d
-; INT256-NEXT:    andl $28, %edx
-; INT256-NEXT:    andl $7, %r8d
-; INT256-NEXT:    andl $28, %ecx
-; INT256-NEXT:    andl $7, %eax
-; INT256-NEXT:    andl $28, %esi
-; INT256-NEXT:    andl $7, %r10d
-; INT256-NEXT:    andl $28, %edi
-; INT256-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; INT256-NEXT:    movq %rsp, %rax
-; INT256-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3]
-; INT256-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
-; INT256-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0]
-; INT256-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; INT256-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[2,3]
-; INT256-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1],mem[0],xmm1[3]
-; INT256-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],mem[0]
-; INT256-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
-; INT256-NEXT:    movq %rbp, %rsp
-; INT256-NEXT:    popq %rbp
+; INT256-NEXT:    vpermps %ymm0, %ymm1, %ymm0
 ; INT256-NEXT:    retq
   %index0 = extractelement <8 x i32> %indices, i32 0
   %index1 = extractelement <8 x i32> %indices, i32 1
   %index2 = extractelement <8 x i32> %indices, i32 2
   %index3 = extractelement <8 x i32> %indices, i32 3
   %index4 = extractelement <8 x i32> %indices, i32 4
   %index5 = extractelement <8 x i32> %indices, i32 5
   %index6 = extractelement <8 x i32> %indices, i32 6

llvm/trunk/test/CodeGen/X86/var-permute-512.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,NOBW,NOVBMI,AVX512F
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefixes=AVX512,NOVBMI,AVX512BW
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vbmi | FileCheck %s --check-prefixes=AVX512,AVX512BW,VBMI
 
 define <8 x i64> @var_shuffle_v8i64(<8 x i64> %v, <8 x i64> %indices) nounwind {
 ; AVX512-LABEL: var_shuffle_v8i64:
 ; AVX512:       # BB#0:
-; AVX512-NEXT:    pushq %rbp
-; AVX512-NEXT:    movq %rsp, %rbp
-; AVX512-NEXT:    andq $-64, %rsp
-; AVX512-NEXT:    subq $128, %rsp
-; AVX512-NEXT:    vmovq %xmm1, %r8
-; AVX512-NEXT:    andl $7, %r8d
-; AVX512-NEXT:    vpextrq $1, %xmm1, %r9
-; AVX512-NEXT:    andl $7, %r9d
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512-NEXT:    vmovq %xmm2, %r10
-; AVX512-NEXT:    andl $7, %r10d
-; AVX512-NEXT:    vpextrq $1, %xmm2, %rsi
-; AVX512-NEXT:    andl $7, %esi
-; AVX512-NEXT:    vextracti32x4 $2, %zmm1, %xmm2
-; AVX512-NEXT:    vmovq %xmm2, %rdi
-; AVX512-NEXT:    andl $7, %edi
-; AVX512-NEXT:    vpextrq $1, %xmm2, %rax
-; AVX512-NEXT:    andl $7, %eax
-; AVX512-NEXT:    vextracti32x4 $3, %zmm1, %xmm1
-; AVX512-NEXT:    vmovq %xmm1, %rcx
-; AVX512-NEXT:    andl $7, %ecx
-; AVX512-NEXT:    vpextrq $1, %xmm1, %rdx
-; AVX512-NEXT:    andl $7, %edx
-; AVX512-NEXT:    vmovaps %zmm0, (%rsp)
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
-; AVX512-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
-; AVX512-NEXT:    vmovlhps {{.*#+}} xmm1 = xmm2[0],xmm1[0]
-; AVX512-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
-; AVX512-NEXT:    vmovlhps {{.*#+}} xmm1 = xmm2[0],xmm1[0]
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm3 = mem[0],zero
-; AVX512-NEXT:    vmovlhps {{.*#+}} xmm2 = xmm3[0],xmm2[0]
-; AVX512-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; AVX512-NEXT:    vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512-NEXT:    movq %rbp, %rsp
-; AVX512-NEXT:    popq %rbp
+; AVX512-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
 ; AVX512-NEXT:    retq
   %index0 = extractelement <8 x i64> %indices, i32 0
   %index1 = extractelement <8 x i64> %indices, i32 1
   %index2 = extractelement <8 x i64> %indices, i32 2
   %index3 = extractelement <8 x i64> %indices, i32 3
   %index4 = extractelement <8 x i64> %indices, i32 4
   %index5 = extractelement <8 x i64> %indices, i32 5
   %index6 = extractelement <8 x i64> %indices, i32 6
; AVX512-NEXT:    retq
   %ret6 = insertelement <8 x i64> %ret5, i64 %v6, i32 6
   %ret7 = insertelement <8 x i64> %ret6, i64 %v7, i32 7
   ret <8 x i64> %ret7
 }
 
 define <16 x i32> @var_shuffle_v16i32(<16 x i32> %v, <16 x i32> %indices) nounwind {
 ; AVX512-LABEL: var_shuffle_v16i32:
 ; AVX512:       # BB#0:
-; AVX512-NEXT:    pushq %rbp
-; AVX512-NEXT:    movq %rsp, %rbp
-; AVX512-NEXT:    andq $-64, %rsp
-; AVX512-NEXT:    subq $128, %rsp
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512-NEXT:    vextracti32x4 $2, %zmm1, %xmm3
-; AVX512-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
-; AVX512-NEXT:    vpextrq $1, %xmm4, %rax
-; AVX512-NEXT:    vmovq %xmm4, %rdx
-; AVX512-NEXT:    movq %rdx, %rcx
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    vmovaps %zmm0, (%rsp)
-; AVX512-NEXT:    andl $15, %edx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; AVX512-NEXT:    movq %rsp, %rdx
-; AVX512-NEXT:    vpinsrd $1, (%rcx,%rdx), %xmm0, %xmm0
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vpinsrd $2, (%rsp,%rax,4), %xmm0, %xmm0
-; AVX512-NEXT:    vmovq %xmm3, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vpinsrd $3, (%rcx,%rdx), %xmm0, %xmm0
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vmovd {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX512-NEXT:    vpextrq $1, %xmm3, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vpinsrd $1, (%rcx,%rdx), %xmm4, %xmm3
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vpinsrd $2, (%rsp,%rax,4), %xmm3, %xmm3
-; AVX512-NEXT:    vmovq %xmm2, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vpinsrd $3, (%rcx,%rdx), %xmm3, %xmm3
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vmovd {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX512-NEXT:    vpextrq $1, %xmm2, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vpinsrd $1, (%rcx,%rdx), %xmm4, %xmm2
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vpinsrd $2, (%rsp,%rax,4), %xmm2, %xmm2
-; AVX512-NEXT:    vmovq %xmm1, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vpinsrd $3, (%rcx,%rdx), %xmm2, %xmm2
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vmovd {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX512-NEXT:    vpextrq $1, %xmm1, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vpinsrd $1, (%rcx,%rdx), %xmm4, %xmm1
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vpinsrd $2, (%rsp,%rax,4), %xmm1, %xmm1
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vpinsrd $3, (%rcx,%rdx), %xmm1, %xmm1
-; AVX512-NEXT:    vinserti128 $1, %xmm0, %ymm3, %ymm0
-; AVX512-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512-NEXT:    movq %rbp, %rsp
-; AVX512-NEXT:    popq %rbp
+; AVX512-NEXT:    vpermps %zmm0, %zmm1, %zmm0
 ; AVX512-NEXT:    retq
   %index0 = extractelement <16 x i32> %indices, i32 0
   %index1 = extractelement <16 x i32> %indices, i32 1
   %index2 = extractelement <16 x i32> %indices, i32 2
   %index3 = extractelement <16 x i32> %indices, i32 3
   %index4 = extractelement <16 x i32> %indices, i32 4
   %index5 = extractelement <16 x i32> %indices, i32 5
   %index6 = extractelement <16 x i32> %indices, i32 6

 ; NOBW-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
 ; NOBW-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm1
 ; NOBW-NEXT:    movq %rbp, %rsp
 ; NOBW-NEXT:    popq %rbp
 ; NOBW-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_shuffle_v32i16:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    pushq %rbp
-; AVX512BW-NEXT:    movq %rsp, %rbp
-; AVX512BW-NEXT:    andq $-64, %rsp
-; AVX512BW-NEXT:    subq $128, %rsp
-; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm1, %xmm3
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
-; AVX512BW-NEXT:    vmovd %xmm4, %eax
-; AVX512BW-NEXT:    vmovaps %zmm0, (%rsp)
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vmovd %eax, %xmm0
-; AVX512BW-NEXT:    vpextrw $1, %xmm4, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $1, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrw $2, %xmm4, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $2, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrw $3, %xmm4, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $3, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrw $4, %xmm4, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrw $5, %xmm4, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $5, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrw $6, %xmm4, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $6, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrw $7, %xmm4, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $7, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovd %xmm3, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vmovd %eax, %xmm4
-; AVX512BW-NEXT:    vpextrw $1, %xmm3, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $1, %eax, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $2, %xmm3, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $2, %eax, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $3, %xmm3, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $3, %eax, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $4, %xmm3, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $4, %eax, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $5, %xmm3, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $5, %eax, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $6, %xmm3, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $6, %eax, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $7, %xmm3, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $7, %eax, %xmm4, %xmm3
-; AVX512BW-NEXT:    vmovd %xmm2, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vmovd %eax, %xmm4
-; AVX512BW-NEXT:    vpextrw $1, %xmm2, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $2, %xmm2, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $3, %xmm2, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $4, %xmm2, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $5, %xmm2, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $5, %eax, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $6, %xmm2, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $6, %eax, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $7, %xmm2, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vpinsrw $7, %eax, %xmm4, %xmm2
-; AVX512BW-NEXT:    vmovd %xmm1, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    movzwl (%rsp,%rax,2), %eax
-; AVX512BW-NEXT:    vmovd %eax, %xmm4
-; AVX512BW-NEXT:    vpextrw $1, %xmm1, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $1, (%rsp,%rax,2), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $2, %xmm1, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $2, (%rsp,%rax,2), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $3, %xmm1, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $3, (%rsp,%rax,2), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $4, %xmm1, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $4, (%rsp,%rax,2), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $5, %xmm1, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $5, (%rsp,%rax,2), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $6, %xmm1, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $6, (%rsp,%rax,2), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrw $7, %xmm1, %eax
-; AVX512BW-NEXT:    andl $31, %eax
-; AVX512BW-NEXT:    vpinsrw $7, (%rsp,%rax,2), %xmm4, %xmm1
-; AVX512BW-NEXT:    vinserti128 $1, %xmm0, %ymm3, %ymm0
-; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512BW-NEXT:    movq %rbp, %rsp
-; AVX512BW-NEXT:    popq %rbp
+; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
   %index0 = extractelement <32 x i16> %indices, i32 0
   %index1 = extractelement <32 x i16> %indices, i32 1
   %index2 = extractelement <32 x i16> %indices, i32 2
   %index3 = extractelement <32 x i16> %indices, i32 3
   %index4 = extractelement <32 x i16> %indices, i32 4
   %index5 = extractelement <32 x i16> %indices, i32 5
   %index6 = extractelement <32 x i16> %indices, i32 6

 ; NOBW-NEXT:    leaq {{[0-9]+}}(%rsp), %rcx
 ; NOBW-NEXT:    vpinsrb $15, (%rax,%rcx), %xmm4, %xmm3
 ; NOBW-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
 ; NOBW-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm1
 ; NOBW-NEXT:    movq %rbp, %rsp
 ; NOBW-NEXT:    popq %rbp
 ; NOBW-NEXT:    retq
 ;
-; AVX512BW-LABEL: var_shuffle_v64i8:
-; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    pushq %rbp
-; AVX512BW-NEXT:    movq %rsp, %rbp
-; AVX512BW-NEXT:    andq $-64, %rsp
-; AVX512BW-NEXT:    subq $128, %rsp
-; AVX512BW-NEXT:    vpextrb $0, %xmm1, %ecx
-; AVX512BW-NEXT:    vpextrb $1, %xmm1, %eax
-; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm1, %xmm3
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
-; AVX512BW-NEXT:    vpextrb $0, %xmm4, %edx
-; AVX512BW-NEXT:    vmovaps %zmm0, (%rsp)
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movq %rsp, %rsi
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vmovd %edx, %xmm0
-; AVX512BW-NEXT:    vpextrb $1, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $2, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $2, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $3, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $3, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $4, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $4, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $5, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $5, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $6, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $6, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $7, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $7, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $8, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $8, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $9, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $9, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $10, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $10, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $11, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $11, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $12, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $12, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $13, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $13, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $14, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $14, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $15, %xmm4, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $15, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $0, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vmovd %edx, %xmm4
-; AVX512BW-NEXT:    vpextrb $1, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $1, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $2, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $2, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $3, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $3, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $4, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $4, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $5, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $5, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $6, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $6, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $7, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $7, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $8, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $8, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $9, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $9, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $10, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $10, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $11, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $11, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $12, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $12, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $13, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $13, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $14, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $14, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $15, %xmm3, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $15, %edx, %xmm4, %xmm3
-; AVX512BW-NEXT:    vpextrb $0, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vmovd %edx, %xmm4
-; AVX512BW-NEXT:    vpextrb $1, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $1, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $2, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $2, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $3, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $3, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $4, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $4, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $5, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $5, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $6, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $6, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $7, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $7, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $8, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $8, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $9, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $9, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $10, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $10, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $11, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $11, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $12, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $12, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $13, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $13, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $14, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $14, %edx, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $15, %xmm2, %edx
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    vpinsrb $15, %edx, %xmm4, %xmm2
-; AVX512BW-NEXT:    vpextrb $2, %xmm1, %edx
-; AVX512BW-NEXT:    andl $63, %ecx
-; AVX512BW-NEXT:    movzbl (%rcx,%rsi), %ecx
-; AVX512BW-NEXT:    vmovd %ecx, %xmm4
-; AVX512BW-NEXT:    vpextrb $3, %xmm1, %ecx
-; AVX512BW-NEXT:    andl $63, %eax
-; AVX512BW-NEXT:    vpinsrb $1, (%rax,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $4, %xmm1, %eax
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    vpinsrb $2, (%rdx,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $5, %xmm1, %edx
-; AVX512BW-NEXT:    andl $63, %ecx
-; AVX512BW-NEXT:    vpinsrb $3, (%rcx,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $6, %xmm1, %ecx
-; AVX512BW-NEXT:    andl $63, %eax
-; AVX512BW-NEXT:    vpinsrb $4, (%rax,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $7, %xmm1, %eax
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    vpinsrb $5, (%rdx,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $8, %xmm1, %edx
-; AVX512BW-NEXT:    andl $63, %ecx
-; AVX512BW-NEXT:    vpinsrb $6, (%rcx,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $9, %xmm1, %ecx
-; AVX512BW-NEXT:    andl $63, %eax
-; AVX512BW-NEXT:    vpinsrb $7, (%rax,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $10, %xmm1, %eax
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    vpinsrb $8, (%rdx,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $11, %xmm1, %edx
-; AVX512BW-NEXT:    andl $63, %ecx
-; AVX512BW-NEXT:    vpinsrb $9, (%rcx,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $12, %xmm1, %ecx
-; AVX512BW-NEXT:    andl $63, %eax
-; AVX512BW-NEXT:    vpinsrb $10, (%rax,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $13, %xmm1, %eax
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    vpinsrb $11, (%rdx,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $14, %xmm1, %edx
-; AVX512BW-NEXT:    andl $63, %ecx
-; AVX512BW-NEXT:    vpinsrb $12, (%rcx,%rsi), %xmm4, %xmm4
-; AVX512BW-NEXT:    vpextrb $15, %xmm1, %ecx
-; AVX512BW-NEXT:    andl $63, %eax
-; AVX512BW-NEXT:    andl $63, %edx
-; AVX512BW-NEXT:    andl $63, %ecx
-; AVX512BW-NEXT:    movzbl (%rcx,%rsi), %ecx
-; AVX512BW-NEXT:    movzbl (%rdx,%rsi), %edx
-; AVX512BW-NEXT:    movzbl (%rax,%rsi), %eax
-; AVX512BW-NEXT:    vpinsrb $13, %eax, %xmm4, %xmm1
-; AVX512BW-NEXT:    vpinsrb $14, %edx, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpinsrb $15, %ecx, %xmm1, %xmm1
-; AVX512BW-NEXT:    vinserti128 $1, %xmm0, %ymm3, %ymm0
-; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm1, %ymm1
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512BW-NEXT:    movq %rbp, %rsp
-; AVX512BW-NEXT:    popq %rbp
-; AVX512BW-NEXT:    retq
+; VBMI-LABEL: var_shuffle_v64i8:
+; VBMI:       # BB#0:
+; VBMI-NEXT:    vpermb %zmm0, %zmm1, %zmm0
+; VBMI-NEXT:    retq
   %index0 = extractelement <64 x i8> %indices, i32 0
   %index1 = extractelement <64 x i8> %indices, i32 1
   %index2 = extractelement <64 x i8> %indices, i32 2
   %index3 = extractelement <64 x i8> %indices, i32 3
   %index4 = extractelement <64 x i8> %indices, i32 4
   %index5 = extractelement <64 x i8> %indices, i32 5
   %index6 = extractelement <64 x i8> %indices, i32 6
   %index7 = extractelement <64 x i8> %indices, i32 7
; VBMI-NEXT:    retq
   %ret62 = insertelement <64 x i8> %ret61, i8 %v62, i32 62
   %ret63 = insertelement <64 x i8> %ret62, i8 %v63, i32 63
   ret <64 x i8> %ret63
 }
 
 define <8 x double> @var_shuffle_v8f64(<8 x double> %v, <8 x i64> %indices) nounwind {
 ; AVX512-LABEL: var_shuffle_v8f64:
 ; AVX512:       # BB#0:
-; AVX512-NEXT:    pushq %rbp
-; AVX512-NEXT:    movq %rsp, %rbp
-; AVX512-NEXT:    andq $-64, %rsp
-; AVX512-NEXT:    subq $128, %rsp
-; AVX512-NEXT:    vmovq %xmm1, %r8
-; AVX512-NEXT:    andl $7, %r8d
-; AVX512-NEXT:    vpextrq $1, %xmm1, %r9
-; AVX512-NEXT:    andl $7, %r9d
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512-NEXT:    vmovq %xmm2, %r10
-; AVX512-NEXT:    andl $7, %r10d
-; AVX512-NEXT:    vpextrq $1, %xmm2, %rsi
-; AVX512-NEXT:    andl $7, %esi
-; AVX512-NEXT:    vextracti32x4 $2, %zmm1, %xmm2
-; AVX512-NEXT:    vmovq %xmm2, %rdi
-; AVX512-NEXT:    andl $7, %edi
-; AVX512-NEXT:    vpextrq $1, %xmm2, %rax
-; AVX512-NEXT:    andl $7, %eax
-; AVX512-NEXT:    vextracti32x4 $3, %zmm1, %xmm1
-; AVX512-NEXT:    vmovq %xmm1, %rcx
-; AVX512-NEXT:    andl $7, %ecx
-; AVX512-NEXT:    vpextrq $1, %xmm1, %rdx
-; AVX512-NEXT:    andl $7, %edx
-; AVX512-NEXT:    vmovaps %zmm0, (%rsp)
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; AVX512-NEXT:    vmovhpd {{.*#+}} xmm0 = xmm0[0],mem[0]
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
-; AVX512-NEXT:    vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0]
-; AVX512-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
-; AVX512-NEXT:    vmovhpd {{.*#+}} xmm1 = xmm1[0],mem[0]
-; AVX512-NEXT:    vmovsd {{.*#+}} xmm2 = mem[0],zero
-; AVX512-NEXT:    vmovhpd {{.*#+}} xmm2 = xmm2[0],mem[0]
-; AVX512-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
-; AVX512-NEXT:    vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512-NEXT:    movq %rbp, %rsp
-; AVX512-NEXT:    popq %rbp
+; AVX512-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
 ; AVX512-NEXT:    retq
   %index0 = extractelement <8 x i64> %indices, i32 0
   %index1 = extractelement <8 x i64> %indices, i32 1
   %index2 = extractelement <8 x i64> %indices, i32 2
   %index3 = extractelement <8 x i64> %indices, i32 3
   %index4 = extractelement <8 x i64> %indices, i32 4
   %index5 = extractelement <8 x i64> %indices, i32 5
   %index6 = extractelement <8 x i64> %indices, i32 6
; AVX512-NEXT:    retq
   %ret6 = insertelement <8 x double> %ret5, double %v6, i32 6
   %ret7 = insertelement <8 x double> %ret6, double %v7, i32 7
   ret <8 x double> %ret7
 }
 
 define <16 x float> @var_shuffle_v16f32(<16 x float> %v, <16 x i32> %indices) nounwind {
 ; AVX512-LABEL: var_shuffle_v16f32:
 ; AVX512:       # BB#0:
-; AVX512-NEXT:    pushq %rbp
-; AVX512-NEXT:    movq %rsp, %rbp
-; AVX512-NEXT:    andq $-64, %rsp
-; AVX512-NEXT:    subq $128, %rsp
-; AVX512-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512-NEXT:    vextracti32x4 $2, %zmm1, %xmm3
-; AVX512-NEXT:    vextracti32x4 $3, %zmm1, %xmm4
-; AVX512-NEXT:    vpextrq $1, %xmm4, %rax
-; AVX512-NEXT:    vmovq %xmm4, %rdx
-; AVX512-NEXT:    movq %rdx, %rcx
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    vmovaps %zmm0, (%rsp)
-; AVX512-NEXT:    andl $15, %edx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; AVX512-NEXT:    movq %rsp, %rdx
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3]
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
-; AVX512-NEXT:    vmovq %xmm3, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],mem[0]
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vmovss {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX512-NEXT:    vpextrq $1, %xmm3, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm3 = xmm4[0],mem[0],xmm4[2,3]
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm3 = xmm3[0,1],mem[0],xmm3[3]
-; AVX512-NEXT:    vmovq %xmm2, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm3 = xmm3[0,1,2],mem[0]
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vmovss {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX512-NEXT:    vpextrq $1, %xmm2, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm2 = xmm4[0],mem[0],xmm4[2,3]
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1],mem[0],xmm2[3]
-; AVX512-NEXT:    vmovq %xmm1, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm2 = xmm2[0,1,2],mem[0]
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vmovss {{.*#+}} xmm4 = mem[0],zero,zero,zero
-; AVX512-NEXT:    vpextrq $1, %xmm1, %rax
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm1 = xmm4[0],mem[0],xmm4[2,3]
-; AVX512-NEXT:    movq %rax, %rcx
-; AVX512-NEXT:    andl $15, %eax
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1],mem[0],xmm1[3]
-; AVX512-NEXT:    shrq $30, %rcx
-; AVX512-NEXT:    andl $60, %ecx
-; AVX512-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1,2],mem[0]
-; AVX512-NEXT:    vinsertf128 $1, %xmm0, %ymm3, %ymm0
-; AVX512-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX512-NEXT:    vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512-NEXT:    movq %rbp, %rsp
-; AVX512-NEXT:    popq %rbp
+; AVX512-NEXT:    vpermps %zmm0, %zmm1, %zmm0
 ; AVX512-NEXT:    retq
   %index0 = extractelement <16 x i32> %indices, i32 0
   %index1 = extractelement <16 x i32> %indices, i32 1
   %index2 = extractelement <16 x i32> %indices, i32 2
   %index3 = extractelement <16 x i32> %indices, i32 3
   %index4 = extractelement <16 x i32> %indices, i32 4
   %index5 = extractelement <16 x i32> %indices, i32 5
   %index6 = extractelement <16 x i32> %indices, i32 6