[DAGCombiner] move splat-shuffle after binop with splat constant

[spatel]

Summary

This is a subset of a transform that we do in instcombine:

binop (splat X), SplatC --> splat (binop X, SplatC)

...and it seems to usually be working as intended (we seem to get more demanded-bits/elements optimizations), but I'm still going through these test diffs.

The end motivation is to reduce the number of patterns that we have to match when trying to scalarize with D60150.

This is intentionally limited to only work when the constant has no undef lanes, but we might be able to ease that as a follow-up. Otherwise, at least in the more general transform in instcombine, we have to deal with potential poison propagation and undef simplification.

[RKSimon]

My only concern is the width reduction on the BROADCAST - as these could regress AVX512 memory folding - but then again that isn't working great in many cases as it is (due to existing SimplifyDemanded calla, shuffle combining etc.) - @craig.topper what do you reckon?

[spatel]

Patch updated:

1. Rebase - D60150 allowed scalarization of x86 FP ops, so we see that happening in the motivating examples in 'test/CodeGen/X86/scalarize-fp.ll'.
2. Refined the conditions where we can use the existing shuffle mask vs. creating a new splat w/o undef lanes. For example:

%s = shufflevector <2 x i64> %x, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
%a = and <2 x i64> %s, <i64 42, i64 42>

To be safe, the new shuffle must actually splat element 1 of the source operand:

LCPI0_0:
.quad	42                      ## 0x2a
.quad	42                      ## 0x2a
pand	LCPI0_0(%rip), %xmm0
pshufd	$238, %xmm0, %xmm0      ## xmm0 = xmm0[2,3,2,3]

Rather than:

pand	LCPI0_0(%rip), %xmm0
pshufd	$78, %xmm0, %xmm0       ## xmm0 = xmm0[2,3,0,1]

[spatel]

Ping.

[spatel]

Ping * 2.

[RKSimon]

A couple of minors but this looks almost ready to me, the avx512 broadcast folds are a known issue

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

18844

Like in getKnownUndefForVectorBinop - add a TODO about not creating nodes on the fly.

llvm/test/CodeGen/ARM/reg_sequence.ll

274–290

Add the other checks for this test to trunk - to show the full codegen delta in this patch.

[spatel]

Patch updated:

1. Added TODO comment about adding a more effificient constant query.
2. Updated ARM test assertions to show more complete diff.

[RKSimon]

Cheers, LGTM

[spatel]

I haven't hand-written FileCheck lines in a long time, and I messed that up...another update coming up.

[spatel]

Patch updated:
Made ARM test checks more exact, so we can see there actually is an extra instruction here (vext.32). That seems like a regression, but I'm not sure if it's important and/or another known shortcoming for folding shuffles.

[efriedma]

llvm/test/CodeGen/ARM/reg_sequence.ll

280

I think the reason you're seeing a regression here is that ARM has a combined "splat-and-multiply" operation. If the splat isn't an operand to the multiply, you end up with an extra instruction. If DAGCombine is going to reorder the operations, we might need some ARM/AArch64-specific pattern-matching code for multiplies.

Granted, this isn't really a great testcase to demonstrate that issue; it's sort of degenerate. Instead, consider something like the following:

define <4 x i32> @x(<4 x i32> %a) {
entry:
  %splat = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> zeroinitializer
  %mul = mul <4 x i32> %splat, <i32 3, i32 3, i32 3, i32 3>
  ret <4 x i32> %mul
}

On AArch64, this currently compiles to something like "movi v1.4s, #3; mul v0.4s, v1.4s, v0.s[0]". If you reorder the operations, I think you end up with three instructions.

[RKSimon]

Please add support for arm/aarch64 splat-and-multiply instructions

[spatel]

Abandoning - see D79886 for a limited version of this.
I didn't locate this patch locally, so I wrote another version.

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

if (Shuf0 && Shuf1 && Shuf0->getMask().equals(Shuf1->getMask()) &&
         LHS.getOperand(1).isUndef() && RHS.getOperand(1).isUndef() &&
         (LHS.hasOneUse() || RHS.hasOneUse() || LHS == RHS)) {
       SDLoc DL(N);
       SDValue NewBinOp = DAG.getNode(Opcode, DL, VT, LHS.getOperand(0),
                                      RHS.getOperand(0), N->getFlags());
       SDValue UndefV = LHS.getOperand(1);
       return DAG.getVectorShuffle(VT, DL, NewBinOp, UndefV, Shuf0->getMask());
     }
+
+    SDLoc DL(N);
+    SDValue UndefV = DAG.getUNDEF(VT);
+    auto isMaskSafeForReorderedShuffle = [&](bool IsShufOp0) {
+      // If the shuffle mask has no undefined elements, the mask is safe.
+      ShuffleVectorSDNode *Shuf = IsShufOp0 ? Shuf0 : Shuf1;
+      if (all_of(Shuf->getMask(), [](int M) { return M != -1; }))
+        return true;
+
+      // If the binop would create undefined elements for the lanes where the
+      // shuffle does too, the mask is safe. If the mask has undefined elements,
+      // but the binop would not produce undefined elements, it is not safe (the
+      // new shuffle could produce undefined elements where the original code
+      // did not).
+      // TODO: We should have a query API that can answer whether some operation
+      //       on undef/constant values produces undef/constant without actually
+      //       creating constant nodes.
+      return IsShufOp0 ? DAG.getNode(Opcode, DL, VT, UndefV, RHS).isUndef()
+                       : DAG.getNode(Opcode, DL, VT, LHS, UndefV).isUndef();
+    };
+
+    // Move a splat after a binop to increase shuffle combining and demanded
+    // elements opportunities.
+    unsigned NumElts = VT.getVectorNumElements();
+    if (Shuf0 && Shuf0->isSplat() && Shuf0->hasOneUse() &&
+        (isConstOrConstSplat(RHS) || isConstOrConstSplatFP(RHS))) {
+      // bo (splat X), SplatC --> splat (bo X, SplatC)
+      SDValue NewBinOp = DAG.getNode(Opcode, DL, VT, LHS.getOperand(0), RHS,
+                                     N->getFlags());
+      // Re-use the existing shuffle mask if possible to preserve undef lanes.
+      if (isMaskSafeForReorderedShuffle(true))
+        return DAG.getVectorShuffle(VT, DL, NewBinOp, UndefV, Shuf0->getMask());
+
+      // We can't preserve undef lanes, so create a "full" splat.
+      SmallVector<int, 16> SplatMask(NumElts, Shuf0->getSplatIndex());
+      return DAG.getVectorShuffle(VT, DL, NewBinOp, UndefV, SplatMask);
+    }
+    if (Shuf1 && Shuf1->isSplat() && Shuf1->hasOneUse() &&
+        (isConstOrConstSplat(LHS) || isConstOrConstSplatFP(LHS))) {
+      // bo SplatC, (splat X) --> splat (bo SplatC, X)
+      SDValue NewBinOp = DAG.getNode(Opcode, DL, VT, LHS, RHS.getOperand(0),
+                                     N->getFlags());
+      // Re-use the existing shuffle mask if possible to preserve undef lanes.
+      if (isMaskSafeForReorderedShuffle(false))
+        return DAG.getVectorShuffle(VT, DL, NewBinOp, UndefV, Shuf1->getMask());
+
+      // We can't preserve undef lanes, so create a "full" splat.
+      SmallVector<int, 16> SplatMask(NumElts, Shuf1->getSplatIndex());
+      return DAG.getVectorShuffle(VT, DL, NewBinOp, UndefV, SplatMask);
+    }
   }
 
   // The following pattern is likely to emerge with vector reduction ops. Moving
   // the binary operation ahead of insertion may allow using a narrower vector
   // instruction that has better performance than the wide version of the op:
   // VBinOp (ins undef, X, Z), (ins undef, Y, Z) --> ins VecC, (VBinOp X, Y), Z
   if (LHS.getOpcode() == ISD::INSERT_SUBVECTOR && LHS.getOperand(0).isUndef() &&
       RHS.getOpcode() == ISD::INSERT_SUBVECTOR && RHS.getOperand(0).isUndef() &&

llvm/test/CodeGen/ARM/reg_sequence.ll

 
 ; <label>:10                                      ; preds = %6
   ret float 9.990000e+02
 }
 
 ; PR7162
 define arm_aapcs_vfpcc i32 @t10(float %x) nounwind {
 ; CHECK-LABEL: t10:
 ; CHECK:       vdup.32  [[Q0:q[0-9]+]], d0[0]
 ; CHECK:       vmov.i32 [[Q9:q[0-9]+]], #0x3f000000
 ; CHECK:       vmul.f32 [[Q8:q[0-9]+]], [[Q0]], [[Q0]]
 ; CHECK-NEXT:  vadd.f32 [[Q8]], [[Q8]], [[Q8]]
 ; CHECK-NEXT:  vadd.f32 [[Q1:q[0-9]+]], [[Q8]], [[Q8]]
-; CHECK-NEXT:  vmul.f32 [[Q8]], [[Q9]], d1[0]
+; CHECK-NEXT:  vext.32  [[Q8]], [[Q8]], [[Q8]], #2
+; CHECK-NEXT:  vmul.f32  [[Q8]], [[Q8]], [[Q9]]
 ; CHECK-NEXT:  vmul.f32 [[Q8]], [[Q8]], [[Q8]]
 ; CHECK-NEXT:  vadd.f32 [[Q8]], [[Q8]], [[Q8]]
 ; CHECK-NEXT:  vmul.f32 [[Q8]], [[Q8]], [[Q8]]
-; CHECK-NEXT:  vst1.32  {d17[1]}, [r0:32]
+; CHECK-NEXT:  vst1.32  {d16[0]}, [r0:32]
 ; CHECK-NEXT:  mov  r0, #0
 ; CHECK-NEXT:  cmp  r0, #0
 ; CHECK-NEXT:  movne  r0, #0
 ; CHECK-NEXT:  bxne lr
 ; CHECK-NEXT:  trap
 entry:
   %0 = shufflevector <4 x float> zeroinitializer, <4 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]
   %1 = insertelement <4 x float> %0, float %x, i32 1 ; <<4 x float>> [#uses=1]
   %2 = insertelement <4 x float> %1, float %x, i32 2 ; <<4 x float>> [#uses=1]
   %3 = insertelement <4 x float> %2, float %x, i32 3 ; <<4 x float>> [#uses=1]
   %tmp54.i = bitcast <4 x float> %3 to <2 x double> ; <<2 x double>> [#uses=1]
   %4 = extractelement <2 x double> %tmp54.i, i32 1 ; <double> [#uses=1]
   %5 = bitcast double %4 to <2 x float>           ; <<2 x float>> [#uses=1]
   %6 = shufflevector <2 x float> %5, <2 x float> undef, <4 x i32> zeroinitializer ; <<4 x float>> [#uses=1]

llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll

 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
 ; X86-NEXT:    vpbroadcastd (%eax), %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x58,0x08]
 ; X86-NEXT:    vpmuldq %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x28,0xc1]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_mask_mul_epi32_rmb_128:
 ; X64:       # %bb.0:
-; X64-NEXT:    vpmuldq (%rdi){1to2}, %xmm0, %xmm0 # encoding: [0x62,0xf2,0xfd,0x18,0x28,0x07]
+; X64-NEXT:    vpbroadcastd (%rdi), %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x58,0x0f]
+; X64-NEXT:    vpmuldq %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x28,0xc1]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %q = load i64, i64* %ptr_b
   %vecinit.i = insertelement < 2 x i64> undef, i64 %q, i32 0
   %b64 = shufflevector < 2 x i64> %vecinit.i, < 2 x i64> undef, <2 x i32> zeroinitializer
   %b = bitcast < 2 x i64> %b64 to < 4 x i32>
   %res = call < 2 x i64> @llvm.x86.avx512.mask.pmul.dq.128(< 4 x i32> %a, < 4 x i32> %b, < 2 x i64> zeroinitializer, i8 -1)
   ret < 2 x i64> %res
 }
 
 define < 2 x i64> @test_mask_mul_epi32_rmbk_128(< 4 x i32> %a, i64* %ptr_b, < 2 x i64> %passThru, i8 %mask) {
 ; X86-LABEL: test_mask_mul_epi32_rmbk_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
 ; X86-NEXT:    vpbroadcastd (%eax), %xmm2 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x58,0x10]
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x08]
 ; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
 ; X86-NEXT:    vpmuldq %xmm2, %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x28,0xca]
 ; X86-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_mask_mul_epi32_rmbk_128:
 ; X64:       # %bb.0:
+; X64-NEXT:    vpbroadcastd (%rdi), %xmm2 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x58,0x17]
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vpmuldq (%rdi){1to2}, %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x19,0x28,0x0f]
+; X64-NEXT:    vpmuldq %xmm2, %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x28,0xca]
 ; X64-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %q = load i64, i64* %ptr_b
   %vecinit.i = insertelement < 2 x i64> undef, i64 %q, i32 0
   %b64 = shufflevector < 2 x i64> %vecinit.i, < 2 x i64> undef, <2 x i32> zeroinitializer
   %b = bitcast < 2 x i64> %b64 to < 4 x i32>
   %res = call < 2 x i64> @llvm.x86.avx512.mask.pmul.dq.128(< 4 x i32> %a, < 4 x i32> %b, < 2 x i64> %passThru, i8 %mask)
   ret < 2 x i64> %res
 }
 
 define < 2 x i64> @test_mask_mul_epi32_rmbkz_128(< 4 x i32> %a, i64* %ptr_b, i8 %mask) {
 ; X86-LABEL: test_mask_mul_epi32_rmbkz_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
 ; X86-NEXT:    vpbroadcastd (%eax), %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x58,0x08]
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x08]
 ; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
 ; X86-NEXT:    vpmuldq %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0x89,0x28,0xc1]
 ; X86-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-LABEL: test_mask_mul_epi32_rmbkz_128:
 ; X64:       # %bb.0:
+; X64-NEXT:    vpbroadcastd (%rdi), %xmm1 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x58,0x0f]
 ; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
-; X64-NEXT:    vpmuldq (%rdi){1to2}, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0x99,0x28,0x07]
+; X64-NEXT:    vpmuldq %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0x89,0x28,0xc1]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %q = load i64, i64* %ptr_b
   %vecinit.i = insertelement < 2 x i64> undef, i64 %q, i32 0
   %b64 = shufflevector < 2 x i64> %vecinit.i, < 2 x i64> undef, < 2 x i32> zeroinitializer
   %b = bitcast < 2 x i64> %b64 to < 4 x i32>
   %res = call < 2 x i64> @llvm.x86.avx512.mask.pmul.dq.128(< 4 x i32> %a, < 4 x i32> %b, < 2 x i64> zeroinitializer, i8 %mask)
   ret < 2 x i64> %res
 }

llvm/test/CodeGen/X86/horizontal-reduce-umax.ll

 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v8i16:
 ; X86-SSE42:       ## %bb.0:

 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v8i16:
 ; X64-SSE42:       ## %bb.0:

 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i16:
 ; X86-SSE42:       ## %bb.0:

 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i16:
 ; X64-SSE42:       ## %bb.0:

 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i16:
 ; X86-SSE42:       ## %bb.0:

 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i16:
 ; X64-SSE42:       ## %bb.0:

llvm/test/CodeGen/X86/horizontal-reduce-umin.ll

 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v8i16:
 ; X86-SSE42:       ## %bb.0:

 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v8i16:
 ; X64-SSE42:       ## %bb.0:

 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i16:
 ; X86-SSE42:       ## %bb.0:

 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i16:
 ; X64-SSE42:       ## %bb.0:

 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i16:
 ; X86-SSE42:       ## %bb.0:

 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i16:
 ; X64-SSE42:       ## %bb.0:

llvm/test/CodeGen/X86/scalarize-fp.ll

; AVX-NEXT:    retq
   %splaty = shufflevector <2 x i64> %vy, <2 x i64> undef, <2 x i32> <i32 0, i32 0>
   %r = add <2 x i64> %splatx, %splaty
   ret <2 x i64> %r
 }
 
 define <2 x double> @fadd_splat_const_op1_v2f64(<2 x double> %vx) {
 ; SSE-LABEL: fadd_splat_const_op1_v2f64:
 ; SSE:       # %bb.0:
+; SSE-NEXT:    addsd {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0,0]
-; SSE-NEXT:    addpd {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: fadd_splat_const_op1_v2f64:
 ; AVX:       # %bb.0:
+; AVX-NEXT:    vaddsd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
-; AVX-NEXT:    vaddpd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %splatx = shufflevector <2 x double> %vx, <2 x double> undef, <2 x i32> zeroinitializer
   %r = fadd <2 x double> %splatx, <double 42.0, double 42.0>
   ret <2 x double> %r
 }
 
 define <4 x double> @fsub_const_op0_splat_v4f64(double %x) {
 ; SSE-LABEL: fsub_const_op0_splat_v4f64:
; AVX-NEXT:    retq
   %splaty = shufflevector <4 x double> %vy, <4 x double> undef, <4 x i32> zeroinitializer
   %r = fsub <4 x double> %splatx, %splaty
   ret <4 x double> %r
 }
 
 define <4 x float> @fmul_splat_const_op1_v4f32(<4 x float> %vx, <4 x float> %vy) {
 ; SSE-LABEL: fmul_splat_const_op1_v4f32:
 ; SSE:       # %bb.0:
+; SSE-NEXT:    mulss {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    mulps {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: fmul_splat_const_op1_v4f32:
 ; AVX:       # %bb.0:
+; AVX-NEXT:    vmulss {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; AVX-NEXT:    vmulps {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %splatx = shufflevector <4 x float> %vx, <4 x float> undef, <4 x i32> zeroinitializer
   %r = fmul fast <4 x float> %splatx, <float 17.0, float 17.0, float 17.0, float 17.0>
   ret <4 x float> %r
 }
 
 define <8 x float> @fdiv_splat_const_op0_v8f32(<8 x float> %vy) {
 ; SSE-LABEL: fdiv_splat_const_op0_v8f32:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    rcpps %xmm0, %xmm2
-; SSE-NEXT:    mulps %xmm2, %xmm0
-; SSE-NEXT:    movaps {{.*#+}} xmm1 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0]
-; SSE-NEXT:    subps %xmm0, %xmm1
-; SSE-NEXT:    mulps %xmm2, %xmm1
-; SSE-NEXT:    addps %xmm2, %xmm1
-; SSE-NEXT:    mulps {{.*}}(%rip), %xmm1
+; SSE-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; SSE-NEXT:    divss %xmm0, %xmm1
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE-NEXT:    movaps %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: fdiv_splat_const_op0_v8f32:
 ; AVX:       # %bb.0:
+; AVX-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; AVX-NEXT:    vdivss %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
 ; AVX-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
-; AVX-NEXT:    vrcpps %ymm0, %ymm1
-; AVX-NEXT:    vmulps %ymm1, %ymm0, %ymm0
-; AVX-NEXT:    vmovaps {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
-; AVX-NEXT:    vsubps %ymm0, %ymm2, %ymm0
-; AVX-NEXT:    vmulps %ymm0, %ymm1, %ymm0
-; AVX-NEXT:    vaddps %ymm0, %ymm1, %ymm0
-; AVX-NEXT:    vmulps {{.*}}(%rip), %ymm0, %ymm0
 ; AVX-NEXT:    retq
   %splatx = shufflevector <8 x float> <float 4.5, float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0>, <8 x float> undef, <8 x i32> zeroinitializer
   %splaty = shufflevector <8 x float> %vy, <8 x float> undef, <8 x i32> zeroinitializer
   %r = fdiv fast <8 x float> %splatx, %splaty
   ret <8 x float> %r
 }
 
 define <8 x float> @fdiv_const_op1_splat_v8f32(<8 x float> %vx) {
 ; SSE-LABEL: fdiv_const_op1_splat_v8f32:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
 ; SSE-NEXT:    xorps %xmm1, %xmm1
-; SSE-NEXT:    rcpps %xmm1, %xmm1
-; SSE-NEXT:    addps %xmm1, %xmm1
-; SSE-NEXT:    mulps %xmm0, %xmm1
-; SSE-NEXT:    movaps %xmm1, %xmm0
+; SSE-NEXT:    divss %xmm1, %xmm0
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; SSE-NEXT:    movaps %xmm0, %xmm1
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: fdiv_const_op1_splat_v8f32:
 ; AVX:       # %bb.0:
+; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
+; AVX-NEXT:    vdivss %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
 ; AVX-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
-; AVX-NEXT:    vxorps %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vrcpps %ymm1, %ymm1
-; AVX-NEXT:    vaddps %ymm1, %ymm1, %ymm1
-; AVX-NEXT:    vmulps %ymm1, %ymm0, %ymm0
 ; AVX-NEXT:    retq
   %splatx = shufflevector <8 x float> %vx, <8 x float> undef, <8 x i32> zeroinitializer
   %splaty = shufflevector <8 x float> <float 0.0, float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0>, <8 x float> undef, <8 x i32> zeroinitializer
   %r = fdiv fast <8 x float> %splatx, %splaty
   ret <8 x float> %r
 }
 
 define <2 x double> @splat0_fadd_v2f64(<2 x double> %vx, <2 x double> %vy) {

llvm/test/CodeGen/X86/vector-fshl-128.ll

 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <2 x i64> @splatvar_funnnel_v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v2i64:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    psllq %xmm2, %xmm3
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [64,64]
 ; SSE2-NEXT:    psubq %xmm2, %xmm4
 ; SSE2-NEXT:    psrlq %xmm4, %xmm1
 ; SSE2-NEXT:    por %xmm3, %xmm1
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    pcmpeqd %xmm2, %xmm3
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
 ; SSE2-NEXT:    pand %xmm3, %xmm2
 ; SSE2-NEXT:    pand %xmm2, %xmm0
 ; SSE2-NEXT:    pandn %xmm1, %xmm2
 ; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v2i64:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[0,1,0,1]
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psllq %xmm4, %xmm2
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    psllq %xmm2, %xmm4
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [64,64]
-; SSE41-NEXT:    psubq %xmm4, %xmm0
+; SSE41-NEXT:    psubq %xmm2, %xmm0
 ; SSE41-NEXT:    psrlq %xmm0, %xmm1
-; SSE41-NEXT:    por %xmm1, %xmm2
+; SSE41-NEXT:    por %xmm1, %xmm4
 ; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqq %xmm4, %xmm0
-; SSE41-NEXT:    blendvpd %xmm0, %xmm3, %xmm2
-; SSE41-NEXT:    movapd %xmm2, %xmm0
+; SSE41-NEXT:    pcmpeqq %xmm2, %xmm0
+; SSE41-NEXT:    blendvpd %xmm0, %xmm3, %xmm4
+; SSE41-NEXT:    movapd %xmm4, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v2i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v2i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX2-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; AVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v2i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX512F-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
 ; AVX512F-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512VL-NEXT:    vptestnmq %xmm2, %xmm2, %k1
 ; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1}
 ; AVX512VL-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v2i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
 ; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512VLBW-NEXT:    vptestnmq %xmm2, %xmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvq %xmm2, %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v2i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v2i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsllq %xmm2, %xmm0, %xmm3
+; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    psllq %xmm2, %xmm3
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
-; X32-SSE-NEXT:    psllq %xmm4, %xmm5
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm5 = xmm3[0],xmm5[1]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [64,0,64,0]
-; X32-SSE-NEXT:    psubq %xmm2, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE-NEXT:    psrlq %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; X32-SSE-NEXT:    psrlq %xmm3, %xmm1
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm4[0],xmm1[1]
-; X32-SSE-NEXT:    orpd %xmm5, %xmm1
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    pcmpeqd %xmm2, %xmm3
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
-; X32-SSE-NEXT:    pand %xmm3, %xmm2
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    pandn %xmm1, %xmm2
-; X32-SSE-NEXT:    por %xmm2, %xmm0
+; X32-SSE-NEXT:    movl $63, %eax
+; X32-SSE-NEXT:    movd %eax, %xmm3
+; X32-SSE-NEXT:    pand %xmm2, %xmm3
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,1,0,1]
+; X32-SSE-NEXT:    pxor %xmm4, %xmm4
+; X32-SSE-NEXT:    pcmpeqd %xmm2, %xmm4
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[1,0,3,2]
+; X32-SSE-NEXT:    pand %xmm4, %xmm5
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [64,0,64,0]
+; X32-SSE-NEXT:    psubq %xmm2, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm2
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm2
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm2[0],xmm1[1]
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    psllq %xmm3, %xmm2
+; X32-SSE-NEXT:    por %xmm1, %xmm2
+; X32-SSE-NEXT:    pand %xmm5, %xmm0
+; X32-SSE-NEXT:    pandn %xmm2, %xmm5
+; X32-SSE-NEXT:    por %xmm5, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <2 x i64> %amt, <2 x i64> undef, <2 x i32> zeroinitializer
   %res = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %splat)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    andps {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    xorps %xmm4, %xmm4
 ; SSE2-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm5
 ; SSE2-NEXT:    pslld %xmm4, %xmm5
 ; SSE2-NEXT:    movd %xmm2, %eax
 ; SSE2-NEXT:    movl $32, %ecx
 ; SSE2-NEXT:    subl %eax, %ecx
 ; SSE2-NEXT:    movd %ecx, %xmm4
 ; SSE2-NEXT:    psrld %xmm4, %xmm1
 ; SSE2-NEXT:    por %xmm5, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
 ; SSE2-NEXT:    pand %xmm2, %xmm0
 ; SSE2-NEXT:    pandn %xmm1, %xmm2
 ; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v4i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[0,0,0,0]
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm4[0],zero,xmm4[1],zero
-; SSE41-NEXT:    movdqa %xmm3, %xmm2
-; SSE41-NEXT:    pslld %xmm0, %xmm2
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm2[0],zero,xmm2[1],zero
+; SSE41-NEXT:    movdqa %xmm3, %xmm4
+; SSE41-NEXT:    pslld %xmm0, %xmm4
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [32,32,32,32]
-; SSE41-NEXT:    psubd %xmm4, %xmm0
+; SSE41-NEXT:    psubd %xmm2, %xmm0
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
 ; SSE41-NEXT:    psrld %xmm0, %xmm1
-; SSE41-NEXT:    por %xmm1, %xmm2
+; SSE41-NEXT:    por %xmm1, %xmm4
 ; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqd %xmm4, %xmm0
-; SSE41-NEXT:    blendvps %xmm0, %xmm3, %xmm2
-; SSE41-NEXT:    movaps %xmm2, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm2, %xmm0
+; SSE41-NEXT:    blendvps %xmm0, %xmm3, %xmm4
+; SSE41-NEXT:    movaps %xmm4, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX1-NEXT:    vpslld %xmm3, %xmm0, %xmm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX2-NEXT:    vpslld %xmm3, %xmm0, %xmm3
+; AVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX2-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
 ; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v4i32:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512F-NEXT:    vpslld %xmm3, %xmm0, %xmm3
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512F-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
 ; AVX512F-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
 ; AVX512F-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VL-NEXT:    vpslld %xmm3, %xmm0, %xmm3
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VL-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512VL-NEXT:    vptestnmd %xmm2, %xmm2, %k1
 ; AVX512VL-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1}
 ; AVX512VL-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512BW-NEXT:    vpslld %xmm3, %xmm0, %xmm3
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512BW-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VBMI2-NEXT:    vpslld %xmm3, %xmm0, %xmm3
+; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VBMI2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VBMI2-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
 ; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VLBW-NEXT:    vpslld %xmm3, %xmm0, %xmm3
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VLBW-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512VLBW-NEXT:    vptestnmd %xmm2, %xmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa32 %xmm0, %xmm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvd %xmm2, %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; XOPAVX1-NEXT:    vpslld %xmm3, %xmm0, %xmm3
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
 ; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; XOPAVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; XOPAVX2-NEXT:    vpslld %xmm3, %xmm0, %xmm3
+; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; XOPAVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; XOPAVX2-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    andps {{\.LCPI.*}}, %xmm2
 ; X32-SSE-NEXT:    pxor %xmm3, %xmm3
 ; X32-SSE-NEXT:    xorps %xmm4, %xmm4
 ; X32-SSE-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
 ; X32-SSE-NEXT:    pslld %xmm4, %xmm5
 ; X32-SSE-NEXT:    movd %xmm2, %eax
 ; X32-SSE-NEXT:    movl $32, %ecx
 ; X32-SSE-NEXT:    subl %eax, %ecx
 ; X32-SSE-NEXT:    movd %ecx, %xmm4
 ; X32-SSE-NEXT:    psrld %xmm4, %xmm1
 ; X32-SSE-NEXT:    por %xmm5, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm2
 ; X32-SSE-NEXT:    pand %xmm2, %xmm0
 ; X32-SSE-NEXT:    pandn %xmm1, %xmm2
 ; X32-SSE-NEXT:    por %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %amt, <4 x i32> undef, <4 x i32> zeroinitializer
   %res = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %splat)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @splatvar_funnnel_v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v8i16:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; SSE2-NEXT:    psubw %xmm2, %xmm3
-; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    pcmpeqw %xmm2, %xmm4
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm0, %xmm5
-; SSE2-NEXT:    psllw %xmm2, %xmm5
+; SSE2-NEXT:    movdqa %xmm0, %xmm4
+; SSE2-NEXT:    psllw %xmm2, %xmm4
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,0,0,0]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; SSE2-NEXT:    psubw %xmm2, %xmm3
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE2-NEXT:    psrlw %xmm3, %xmm1
-; SSE2-NEXT:    por %xmm5, %xmm1
-; SSE2-NEXT:    pand %xmm4, %xmm0
-; SSE2-NEXT:    pandn %xmm1, %xmm4
-; SSE2-NEXT:    por %xmm4, %xmm0
+; SSE2-NEXT:    por %xmm4, %xmm1
+; SSE2-NEXT:    pxor %xmm3, %xmm3
+; SSE2-NEXT:    pcmpeqw %xmm2, %xmm3
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pandn %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v8i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; SSE41-NEXT:    movdqa %xmm3, %xmm4
+; SSE41-NEXT:    psllw %xmm0, %xmm4
 ; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[0,0,0,0]
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
-; SSE41-NEXT:    movdqa %xmm3, %xmm2
-; SSE41-NEXT:    psllw %xmm0, %xmm2
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,0,0]
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [16,16,16,16,16,16,16,16]
-; SSE41-NEXT:    psubw %xmm4, %xmm0
+; SSE41-NEXT:    psubw %xmm2, %xmm0
 ; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
 ; SSE41-NEXT:    psrlw %xmm0, %xmm1
-; SSE41-NEXT:    por %xmm1, %xmm2
+; SSE41-NEXT:    por %xmm1, %xmm4
 ; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqw %xmm4, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    pcmpeqw %xmm2, %xmm0
+; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm4
+; SSE41-NEXT:    movdqa %xmm4, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v8i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
+; AVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
 ; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v8i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
 ; AVX512F-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
+; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VBMI2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
 ; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX512VLBW-NEXT:    vptestnmw %xmm2, %xmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu16 %xmm0, %xmm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvw %xmm2, %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; XOPAVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; XOPAVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; XOPAVX2-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
+; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; XOPAVX2-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v8i16:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
-; X32-SSE-NEXT:    psubw %xmm2, %xmm3
-; X32-SSE-NEXT:    pxor %xmm4, %xmm4
-; X32-SSE-NEXT:    pcmpeqw %xmm2, %xmm4
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
-; X32-SSE-NEXT:    psllw %xmm2, %xmm5
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
+; X32-SSE-NEXT:    psllw %xmm2, %xmm4
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,0,0,0]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; X32-SSE-NEXT:    psubw %xmm2, %xmm3
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; X32-SSE-NEXT:    psrlw %xmm3, %xmm1
-; X32-SSE-NEXT:    por %xmm5, %xmm1
-; X32-SSE-NEXT:    pand %xmm4, %xmm0
-; X32-SSE-NEXT:    pandn %xmm1, %xmm4
-; X32-SSE-NEXT:    por %xmm4, %xmm0
+; X32-SSE-NEXT:    por %xmm4, %xmm1
+; X32-SSE-NEXT:    pxor %xmm3, %xmm3
+; X32-SSE-NEXT:    pcmpeqw %xmm2, %xmm3
+; X32-SSE-NEXT:    pand %xmm3, %xmm0
+; X32-SSE-NEXT:    pandn %xmm1, %xmm3
+; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <8 x i16> %amt, <8 x i16> undef, <8 x i32> zeroinitializer
   %res = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %splat)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @splatvar_funnnel_v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE2-NEXT:    psubb %xmm2, %xmm3
-; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    pcmpeqb %xmm2, %xmm4
-; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0]
-; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm0, %xmm5
-; SSE2-NEXT:    psllw %xmm2, %xmm5
-; SSE2-NEXT:    pcmpeqd %xmm6, %xmm6
-; SSE2-NEXT:    psllw %xmm2, %xmm6
-; SSE2-NEXT:    pcmpeqd %xmm2, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,0,0]
-; SSE2-NEXT:    pand %xmm5, %xmm6
+; SSE2-NEXT:    movdqa %xmm2, %xmm3
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    psrlw %xmm3, %xmm1
-; SSE2-NEXT:    psrlw %xmm3, %xmm2
-; SSE2-NEXT:    psrlw $8, %xmm2
+; SSE2-NEXT:    movdqa %xmm0, %xmm4
+; SSE2-NEXT:    psllw %xmm3, %xmm4
+; SSE2-NEXT:    pcmpeqd %xmm5, %xmm5
+; SSE2-NEXT:    pcmpeqd %xmm6, %xmm6
+; SSE2-NEXT:    psllw %xmm3, %xmm6
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm6[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; SSE2-NEXT:    pand %xmm4, %xmm3
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE2-NEXT:    pand %xmm1, %xmm2
-; SSE2-NEXT:    por %xmm6, %xmm2
-; SSE2-NEXT:    pand %xmm4, %xmm0
-; SSE2-NEXT:    pandn %xmm2, %xmm4
-; SSE2-NEXT:    por %xmm4, %xmm0
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; SSE2-NEXT:    psubb %xmm2, %xmm4
+; SSE2-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm4[0]
+; SSE2-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    psrlw %xmm4, %xmm1
+; SSE2-NEXT:    psrlw %xmm4, %xmm5
+; SSE2-NEXT:    psrlw $8, %xmm5
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
+; SSE2-NEXT:    pand %xmm1, %xmm4
+; SSE2-NEXT:    por %xmm3, %xmm4
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    pcmpeqb %xmm2, %xmm1
+; SSE2-NEXT:    pand %xmm1, %xmm0
+; SSE2-NEXT:    pandn %xmm4, %xmm1
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm2
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm5 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    psllw %xmm5, %xmm4
+; SSE41-NEXT:    psllw %xmm0, %xmm4
+; SSE41-NEXT:    pcmpeqd %xmm5, %xmm5
 ; SSE41-NEXT:    pcmpeqd %xmm6, %xmm6
-; SSE41-NEXT:    pcmpeqd %xmm7, %xmm7
-; SSE41-NEXT:    psllw %xmm5, %xmm7
-; SSE41-NEXT:    pshufb %xmm0, %xmm7
-; SSE41-NEXT:    pand %xmm7, %xmm4
-; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE41-NEXT:    psubb %xmm2, %xmm5
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,zero,zero,zero,zero,xmm5[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    psrlw %xmm5, %xmm1
-; SSE41-NEXT:    psrlw %xmm5, %xmm6
-; SSE41-NEXT:    pshufb {{.*#+}} xmm6 = xmm6[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; SSE41-NEXT:    pand %xmm1, %xmm6
-; SSE41-NEXT:    por %xmm6, %xmm4
+; SSE41-NEXT:    psllw %xmm0, %xmm6
+; SSE41-NEXT:    pxor %xmm0, %xmm0
+; SSE41-NEXT:    pshufb %xmm0, %xmm6
+; SSE41-NEXT:    pand %xmm6, %xmm4
+; SSE41-NEXT:    pshufb %xmm0, %xmm2
+; SSE41-NEXT:    movdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; SSE41-NEXT:    psubb %xmm2, %xmm6
+; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    psrlw %xmm6, %xmm1
+; SSE41-NEXT:    psrlw %xmm6, %xmm5
+; SSE41-NEXT:    pshufb {{.*#+}} xmm5 = xmm5[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; SSE41-NEXT:    pand %xmm1, %xmm5
+; SSE41-NEXT:    por %xmm5, %xmm4
 ; SSE41-NEXT:    pcmpeqb %xmm2, %xmm0
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm4
 ; SSE41-NEXT:    movdqa %xmm4, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v16i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm5
-; AVX1-NEXT:    vpcmpeqd %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpsllw %xmm4, %xmm6, %xmm4
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpand %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX1-NEXT:    vpsubb %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,zero,zero,zero,zero,xmm5[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw %xmm5, %xmm6, %xmm5
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vpor %xmm1, %xmm4, %xmm1
-; AVX1-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm4
+; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
+; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
+; AVX1-NEXT:    vpshufb %xmm6, %xmm3, %xmm3
+; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpshufb %xmm6, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX1-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm2
 ; AVX1-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm3, %xmm0, %xmm4
 ; AVX2-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX2-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX2-NEXT:    vpbroadcastb %xmm3, %xmm3
 ; AVX2-NEXT:    vpand %xmm3, %xmm4, %xmm3
+; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
 ; AVX2-NEXT:    vpsrlw %xmm4, %xmm5, %xmm4
 ; AVX2-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX2-NEXT:    vpbroadcastb %xmm4, %xmm4
 ; AVX2-NEXT:    vpand %xmm4, %xmm1, %xmm1
 ; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsllvd %zmm4, %zmm3, %zmm3
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
 ; AVX512F-NEXT:    vpord %zmm1, %zmm3, %zmm1
 ; AVX512F-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm3, %zmm3
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm1, %zmm1
 ; AVX512VL-NEXT:    vpord %zmm1, %zmm3, %zmm1
 ; AVX512VL-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512BW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
 ; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm1, %ymm1
 ; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512VBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; AVX512VLBW-NEXT:    vpsllvw %ymm4, %ymm3, %ymm3
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VLBW-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VLBW-NEXT:    vpmovwb %ymm1, %xmm1
 ; AVX512VLBW-NEXT:    vptestnmb %xmm2, %xmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %xmm0, %xmm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; AVX512VLVBMI2-NEXT:    vpsllvw %ymm4, %ymm3, %ymm3
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLVBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VLVBMI2-NEXT:    vpsrlvw %ymm4, %ymm1, %ymm1
 ; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VLVBMI2-NEXT:    vpmovwb %ymm1, %xmm1
 ; AVX512VLVBMI2-NEXT:    vptestnmb %xmm2, %xmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %xmm0, %xmm1 {%k1}
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    vzeroupper
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX1:       # %bb.0:
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm4
 ; XOPAVX1-NEXT:    vpsubb {{.*}}(%rip), %xmm2, %xmm5
 ; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpor %xmm1, %xmm4, %xmm1
 ; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm0, %xmm3
 ; XOPAVX2-NEXT:    vpsubb {{.*}}(%rip), %xmm2, %xmm4
 ; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpblendvb %xmm2, %xmm0, %xmm1, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; X32-SSE-NEXT:    psubb %xmm2, %xmm3
-; X32-SSE-NEXT:    pxor %xmm4, %xmm4
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm4
-; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0]
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
-; X32-SSE-NEXT:    psllw %xmm2, %xmm5
-; X32-SSE-NEXT:    pcmpeqd %xmm6, %xmm6
-; X32-SSE-NEXT:    psllw %xmm2, %xmm6
-; X32-SSE-NEXT:    pcmpeqd %xmm2, %xmm2
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,0,0]
-; X32-SSE-NEXT:    pand %xmm5, %xmm6
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    psrlw %xmm3, %xmm1
-; X32-SSE-NEXT:    psrlw %xmm3, %xmm2
-; X32-SSE-NEXT:    psrlw $8, %xmm2
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
+; X32-SSE-NEXT:    psllw %xmm3, %xmm4
+; X32-SSE-NEXT:    pcmpeqd %xmm5, %xmm5
+; X32-SSE-NEXT:    pcmpeqd %xmm6, %xmm6
+; X32-SSE-NEXT:    psllw %xmm3, %xmm6
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm6[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; X32-SSE-NEXT:    pand %xmm4, %xmm3
 ; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; X32-SSE-NEXT:    pand %xmm1, %xmm2
-; X32-SSE-NEXT:    por %xmm6, %xmm2
-; X32-SSE-NEXT:    pand %xmm4, %xmm0
-; X32-SSE-NEXT:    pandn %xmm2, %xmm4
-; X32-SSE-NEXT:    por %xmm4, %xmm0
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; X32-SSE-NEXT:    psubb %xmm2, %xmm4
+; X32-SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm4[0]
+; X32-SSE-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; X32-SSE-NEXT:    psrlw %xmm4, %xmm1
+; X32-SSE-NEXT:    psrlw %xmm4, %xmm5
+; X32-SSE-NEXT:    psrlw $8, %xmm5
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
+; X32-SSE-NEXT:    pand %xmm1, %xmm4
+; X32-SSE-NEXT:    por %xmm3, %xmm4
+; X32-SSE-NEXT:    pxor %xmm1, %xmm1
+; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm1
+; X32-SSE-NEXT:    pand %xmm1, %xmm0
+; X32-SSE-NEXT:    pandn %xmm4, %xmm1
+; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <16 x i8> %amt, <16 x i8> undef, <16 x i32> zeroinitializer
   %res = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %splat)
   ret <16 x i8> %res
 }
 
 ;
 ; Constant Shifts

llvm/test/CodeGen/X86/vector-fshl-256.ll

 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <4 x i64> @splatvar_funnnel_v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v4i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovddup {{.*#+}} xmm2 = xmm2[0,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
 ; AVX1-NEXT:    vpsllq %xmm2, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm4
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [64,64]
-; AVX1-NEXT:    vpsubq %xmm4, %xmm5, %xmm6
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; AVX1-NEXT:    vpsrlq %xmm6, %xmm7, %xmm6
-; AVX1-NEXT:    vpsubq %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpsrlq %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm2 = xmm2[0,0]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
+; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; AVX1-NEXT:    vpsrlq %xmm4, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
 ; AVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
 ; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; AVX1-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
+; AVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
 ; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v4i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
 ; AVX512F-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VL-NEXT:    vptestnmq %ymm2, %ymm2, %k1
 ; AVX512VL-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1}
 ; AVX512VL-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v4i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
+; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
 ; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
 ; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VLBW-NEXT:    vptestnmq %ymm2, %ymm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa64 %ymm0, %ymm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshldvq %ymm2, %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vmovddup {{.*#+}} xmm2 = xmm2[0,0]
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vpsllq %xmm2, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpsllq %xmm2, %xmm0, %xmm4
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [64,64]
-; XOPAVX1-NEXT:    vpsubq %xmm4, %xmm5, %xmm6
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; XOPAVX1-NEXT:    vpsrlq %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpsrlq %xmm5, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vmovddup {{.*#+}} xmm2 = xmm2[0,0]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
+; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpsrlq %xmm4, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
 ; XOPAVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
 ; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
+; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpsrlq %xmm4, %ymm1, %ymm1
 ; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
 ; XOPAVX2-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <4 x i64> %amt, <4 x i64> undef, <4 x i32> zeroinitializer
   %res = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %splat)
   ret <4 x i64> %res
 }
 
 define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; AVX1-NEXT:    vpslld %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm4, %xmm5, %xmm6
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm6 = xmm6[0],zero,xmm6[1],zero
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; AVX1-NEXT:    vpsrld %xmm6, %xmm7, %xmm6
-; AVX1-NEXT:    vpsubd %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm5[0],zero,xmm5[1],zero
-; AVX1-NEXT:    vpsrld %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero
+; AVX1-NEXT:    vpslld %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm4
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
+; AVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; AVX1-NEXT:    vpsrld %xmm4, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
 ; AVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
 ; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; AVX1-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX2-NEXT:    vpslld %xmm3, %ymm0, %ymm3
+; AVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX2-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
 ; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v8i32:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512F-NEXT:    vpslld %xmm3, %ymm0, %ymm3
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512F-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
 ; AVX512F-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to8}, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VL-NEXT:    vpslld %xmm3, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VL-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VL-NEXT:    vptestnmd %ymm2, %ymm2, %k1
 ; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1}
 ; AVX512VL-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512BW-NEXT:    vpslld %xmm3, %ymm0, %ymm3
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512BW-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VBMI2-NEXT:    vpslld %xmm3, %ymm0, %ymm3
+; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VBMI2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VBMI2-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
 ; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
 ; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to8}, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VLBW-NEXT:    vpslld %xmm3, %ymm0, %ymm3
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VLBW-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VLBW-NEXT:    vptestnmd %ymm2, %ymm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshldvd %ymm2, %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; XOPAVX1-NEXT:    vpslld %xmm3, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpslld %xmm3, %xmm0, %xmm3
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [32,32,32,32]
-; XOPAVX1-NEXT:    vpsubd %xmm4, %xmm5, %xmm6
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm6 = xmm6[0],zero,xmm6[1],zero
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; XOPAVX1-NEXT:    vpsrld %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm5[0],zero,xmm5[1],zero
-; XOPAVX1-NEXT:    vpsrld %xmm5, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero
+; XOPAVX1-NEXT:    vpslld %xmm4, %xmm3, %xmm3
+; XOPAVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm4
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
+; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpsrld %xmm4, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
 ; XOPAVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
 ; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; XOPAVX2-NEXT:    vpslld %xmm3, %ymm0, %ymm3
+; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; XOPAVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; XOPAVX2-NEXT:    vpsrld %xmm4, %ymm1, %ymm1
 ; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
 ; XOPAVX2-NEXT:    vblendvps %ymm2, %ymm0, %ymm1, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <8 x i32> %amt, <8 x i32> undef, <8 x i32> zeroinitializer
   %res = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %splat)
   ret <8 x i32> %res
 }
 
 define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v16i16:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX1-NEXT:    vpsllw %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm4
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
 ; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; AVX1-NEXT:    vpsllw %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; AVX1-NEXT:    vpsubw %xmm4, %xmm5, %xmm6
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; AVX1-NEXT:    vpsrlw %xmm6, %xmm7, %xmm6
-; AVX1-NEXT:    vpsubw %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm5, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
+; AVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
 ; AVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
 ; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; AVX1-NEXT:    vandnps %ymm1, %ymm2, %ymm1
 ; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
+; AVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v16i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX512BW-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
+; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VBMI2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX512VBMI2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
 ; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX512VLBW-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VLBW-NEXT:    vptestnmw %ymm2, %ymm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu16 %ymm0, %ymm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshldvw %ymm2, %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; XOPAVX1-NEXT:    vpsllw %xmm4, %xmm3, %xmm3
+; XOPAVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm4
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
 ; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
 ; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; XOPAVX1-NEXT:    vpsllw %xmm3, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm3
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; XOPAVX1-NEXT:    vpsubw %xmm4, %xmm5, %xmm6
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
-; XOPAVX1-NEXT:    vpsrlw %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero
-; XOPAVX1-NEXT:    vpsrlw %xmm5, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
+; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
 ; XOPAVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
 ; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; XOPAVX2-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
+; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; XOPAVX2-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; XOPAVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
 ; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <16 x i16> %amt, <16 x i16> undef, <16 x i32> zeroinitializer
   %res = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %splat)
   ret <16 x i16> %res
 }
 
 define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm8, %xmm8, %xmm8
-; AVX1-NEXT:    vpshufb %xmm8, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX1-NEXT:    vpsllw %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpcmpeqd %xmm9, %xmm9, %xmm9
-; AVX1-NEXT:    vpsllw %xmm4, %xmm9, %xmm7
-; AVX1-NEXT:    vpshufb %xmm8, %xmm7, %xmm7
-; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
+; AVX1-NEXT:    vpsllw %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpsllw %xmm4, %xmm5, %xmm6
+; AVX1-NEXT:    vpxor %xmm7, %xmm7, %xmm7
+; AVX1-NEXT:    vpshufb %xmm7, %xmm6, %xmm6
+; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm4, %ymm4
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm10 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX1-NEXT:    vpsubb %xmm5, %xmm10, %xmm3
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
+; AVX1-NEXT:    vpshufb %xmm7, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX1-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
-; AVX1-NEXT:    vpsrlw %xmm3, %xmm6, %xmm6
-; AVX1-NEXT:    vpsrlw %xmm3, %xmm9, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpshufb %xmm7, %xmm3, %xmm3
-; AVX1-NEXT:    vpand %xmm3, %xmm6, %xmm3
-; AVX1-NEXT:    vpsubb %xmm2, %xmm10, %xmm6
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm6, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw %xmm6, %xmm9, %xmm6
-; AVX1-NEXT:    vpshufb %xmm7, %xmm6, %xmm6
-; AVX1-NEXT:    vpand %xmm6, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm1, %ymm4, %ymm1
-; AVX1-NEXT:    vpcmpeqb %xmm8, %xmm5, %xmm3
-; AVX1-NEXT:    vpcmpeqb %xmm8, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm6, %xmm6
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX1-NEXT:    vpand %xmm5, %xmm6, %xmm6
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
+; AVX1-NEXT:    vorps %ymm1, %ymm3, %ymm1
+; AVX1-NEXT:    vpcmpeqb %xmm7, %xmm2, %xmm2
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; AVX1-NEXT:    vandnps %ymm1, %ymm2, %ymm1
 ; AVX1-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX2-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
 ; AVX2-NEXT:    vpsllw %xmm3, %ymm5, %ymm3
 ; AVX2-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX2-NEXT:    vpand %ymm3, %ymm4, %ymm3
+; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX2-NEXT:    vpsrlw %xmm4, %ymm5, %ymm4
 ; AVX2-NEXT:    vpsrlw $8, %ymm4, %ymm4
 ; AVX2-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX2-NEXT:    vpand %ymm4, %ymm1, %ymm1
 ; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX512F-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm5, %ymm3
 ; AVX512F-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX512F-NEXT:    vpand %ymm3, %ymm4, %ymm3
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpsrlw %xmm4, %ymm5, %ymm4
 ; AVX512F-NEXT:    vpsrlw $8, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512F-NEXT:    vpand %ymm4, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX512VL-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm5, %ymm3
 ; AVX512VL-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm4, %ymm3
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm5, %ymm4
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512BW-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
 ; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512BW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512BW-NEXT:    vpmovwb %zmm1, %ymm1
 ; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
 ; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm1, %ymm1
 ; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512VBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
 ; AVX512VLBW-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLBW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VLBW-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VLBW-NEXT:    vpmovwb %zmm1, %ymm1
 ; AVX512VLBW-NEXT:    vptestnmb %ymm2, %ymm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %ymm0, %ymm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
 ; AVX512VLVBMI2-NEXT:    vpsllvw %zmm4, %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLVBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
 ; AVX512VLVBMI2-NEXT:    vpsrlvw %zmm4, %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpmovwb %zmm1, %ymm1
 ; AVX512VLVBMI2-NEXT:    vptestnmb %ymm2, %ymm2, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %ymm0, %ymm1 {%k1}
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpxor %xmm8, %xmm8, %xmm8
-; XOPAVX1-NEXT:    vpshufb %xmm8, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; XOPAVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm5
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm5, %ymm4
+; XOPAVX1-NEXT:    vpsubb {{.*}}(%rip), %xmm2, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
+; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vorps %ymm1, %ymm4, %ymm1
+; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm6
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm6, %ymm5
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX1-NEXT:    vpsubb %xmm6, %xmm4, %xmm7
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
-; XOPAVX1-NEXT:    vpshlb %xmm7, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpsubb %xmm6, %xmm2, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; XOPAVX1-NEXT:    vorps %ymm1, %ymm5, %ymm1
-; XOPAVX1-NEXT:    vpcomeqb %xmm8, %xmm4, %xmm3
-; XOPAVX1-NEXT:    vpcomeqb %xmm8, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm4
 ; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm0, %xmm4
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; XOPAVX2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
 ; XOPAVX2-NEXT:    vpxor %xmm6, %xmm6, %xmm6

llvm/test/CodeGen/X86/vector-fshl-512.ll

 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <8 x i64> @splatvar_funnnel_v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v8i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512F-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; AVX512F-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsrlq %xmm4, %zmm1, %zmm1
 ; AVX512F-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
 ; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsrlq %xmm4, %zmm1, %zmm1
 ; AVX512VL-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VL-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512VL-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
 ; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsrlq %xmm4, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshldvq %zmm2, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsrlq %xmm4, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VLBW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvq %zmm2, %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <8 x i64> %amt, <8 x i64> undef, <8 x i32> zeroinitializer
   %res = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %splat)
   ret <8 x i64> %res
 }
 
 define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v16i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512F-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; AVX512F-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512F-NEXT:    vpslld %xmm3, %zmm0, %zmm3
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512F-NEXT:    vpsrld %xmm4, %zmm1, %zmm1
 ; AVX512F-NEXT:    vpord %zmm1, %zmm3, %zmm1
 ; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
 ; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VL-NEXT:    vpslld %xmm3, %zmm0, %zmm3
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VL-NEXT:    vpsrld %xmm4, %zmm1, %zmm1
 ; AVX512VL-NEXT:    vpord %zmm1, %zmm3, %zmm1
 ; AVX512VL-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512VL-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
 ; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512BW-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512BW-NEXT:    vpslld %xmm3, %zmm0, %zmm3
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512BW-NEXT:    vpsrld %xmm4, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpord %zmm1, %zmm3, %zmm1
 ; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshldvd %zmm2, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VLBW-NEXT:    vpslld %xmm3, %zmm0, %zmm3
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VLBW-NEXT:    vpsrld %xmm4, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpord %zmm1, %zmm3, %zmm1
 ; AVX512VLBW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa32 %zmm0, %zmm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvd %zmm2, %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <16 x i32> %amt, <16 x i32> undef, <16 x i32> zeroinitializer
   %res = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %splat)
   ret <16 x i32> %res
 }
 
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm4, %ymm4
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm5, %ymm0, %ymm6
+; AVX512F-NEXT:    vpbroadcastw %xmm4, %ymm4
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm7 = [16,16,16,16,16,16,16,16]
 ; AVX512F-NEXT:    vpsubw %xmm4, %xmm7, %xmm7
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm7 = xmm7[0],zero,zero,zero,xmm7[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm7, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpor %ymm2, %ymm6, %ymm2
 ; AVX512F-NEXT:    vpxor %xmm6, %xmm6, %xmm6
 ; AVX512F-NEXT:    vpcmpeqw %ymm6, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpblendvb %ymm4, %ymm0, %ymm2, %ymm0
 ; AVX512F-NEXT:    vpsllw %xmm5, %ymm1, %ymm2
 ; AVX512F-NEXT:    vpsrlw %xmm7, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpor %ymm3, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpblendvb %ymm4, %ymm1, %ymm2, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm4, %ymm4
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm5, %ymm0, %ymm6
+; AVX512VL-NEXT:    vpbroadcastw %xmm4, %ymm4
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm7 = [16,16,16,16,16,16,16,16]
 ; AVX512VL-NEXT:    vpsubw %xmm4, %xmm7, %xmm7
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm7 = xmm7[0],zero,zero,zero,xmm7[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm7, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpor %ymm2, %ymm6, %ymm2
 ; AVX512VL-NEXT:    vpxor %xmm6, %xmm6, %xmm6
 ; AVX512VL-NEXT:    vpcmpeqw %ymm6, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm0, %ymm2, %ymm0
 ; AVX512VL-NEXT:    vpsllw %xmm5, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpsrlw %xmm7, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpor %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %zmm2
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm3, %zmm0, %zmm3
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshldvw %zmm2, %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm3, %zmm0, %zmm3
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VLBW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu16 %zmm0, %zmm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshldvw %zmm2, %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <32 x i16> %amt, <32 x i16> undef, <32 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %splat)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm5 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm5, %ymm0, %ymm6
 ; AVX512F-NEXT:    vpcmpeqd %ymm9, %ymm9, %ymm9
 ; AVX512F-NEXT:    vpsllw %xmm5, %ymm9, %ymm8
 ; AVX512F-NEXT:    vpbroadcastb %xmm8, %ymm8
 ; AVX512F-NEXT:    vpand %ymm8, %ymm6, %ymm6
+; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm4, %xmm7, %xmm7
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm7 = xmm7[0],zero,zero,zero,zero,zero,zero,zero,xmm7[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm7, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsrlw %xmm7, %ymm9, %ymm9
 ; AVX512F-NEXT:    vpsrlw $8, %ymm9, %ymm9
 ; AVX512F-NEXT:    vpbroadcastb %xmm9, %ymm9
 ; AVX512F-NEXT:    vpand %ymm9, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpor %ymm2, %ymm6, %ymm2
 ; AVX512F-NEXT:    vpxor %xmm6, %xmm6, %xmm6
 ; AVX512F-NEXT:    vpcmpeqb %ymm6, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpblendvb %ymm4, %ymm0, %ymm2, %ymm0
 ; AVX512F-NEXT:    vpsllw %xmm5, %ymm1, %ymm2
 ; AVX512F-NEXT:    vpand %ymm8, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsrlw %xmm7, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpand %ymm9, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpor %ymm3, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpblendvb %ymm4, %ymm1, %ymm2, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm5 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm5, %ymm0, %ymm6
 ; AVX512VL-NEXT:    vpcmpeqd %ymm9, %ymm9, %ymm9
 ; AVX512VL-NEXT:    vpsllw %xmm5, %ymm9, %ymm8
 ; AVX512VL-NEXT:    vpbroadcastb %xmm8, %ymm8
 ; AVX512VL-NEXT:    vpand %ymm8, %ymm6, %ymm6
+; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VL-NEXT:    vpsubb %xmm4, %xmm7, %xmm7
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm7 = xmm7[0],zero,zero,zero,zero,zero,zero,zero,xmm7[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm7, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpsrlw %xmm7, %ymm9, %ymm9
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm9, %ymm9
 ; AVX512VL-NEXT:    vpbroadcastb %xmm9, %ymm9
 ; AVX512VL-NEXT:    vpand %ymm9, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpor %ymm2, %ymm6, %ymm2
 ; AVX512VL-NEXT:    vpxor %xmm6, %xmm6, %xmm6
 ; AVX512VL-NEXT:    vpcmpeqb %ymm6, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm0, %ymm2, %ymm0
 ; AVX512VL-NEXT:    vpsllw %xmm5, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm8, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpsrlw %xmm7, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm9, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpor %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm3, %zmm0, %zmm4
 ; AVX512BW-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5
 ; AVX512BW-NEXT:    vpsllw %xmm3, %zmm5, %zmm3
 ; AVX512BW-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512BW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512BW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512BW-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsrlw %xmm4, %zmm5, %zmm4
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm4, %zmm4
 ; AVX512BW-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512BW-NEXT:    vpandq %zmm4, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsllw %xmm3, %zmm0, %zmm4
 ; AVX512VBMI2-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5
 ; AVX512VBMI2-NEXT:    vpsllw %xmm3, %zmm5, %zmm3
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
+; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vpsrlw %xmm4, %zmm5, %zmm4
 ; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpandq %zmm4, %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm3, %zmm0, %zmm4
 ; AVX512VLBW-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5
 ; AVX512VLBW-NEXT:    vpsllw %xmm3, %zmm5, %zmm3
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpsrlw %xmm4, %zmm5, %zmm4
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpandq %zmm4, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VLBW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLVBMI2-NEXT:    vpsllw %xmm3, %zmm0, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5
 ; AVX512VLVBMI2-NEXT:    vpsllw %xmm3, %zmm5, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
+; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLVBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLVBMI2-NEXT:    vpsrlw %xmm4, %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpsrlw %xmm4, %zmm5, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpandq %zmm4, %zmm1, %zmm1

llvm/test/CodeGen/X86/vector-fshl-rot-128.ll

 ; SSE2-NEXT:    subl %eax, %ecx
 ; SSE2-NEXT:    movd %ecx, %xmm1
 ; SSE2-NEXT:    psrld %xmm1, %xmm0
 ; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v4i32:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
 ; SSE41-NEXT:    pslld %xmm2, %xmm3
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [32,32,32,32]
 ; SSE41-NEXT:    psubd %xmm1, %xmm2
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero
 ; SSE41-NEXT:    psrld %xmm1, %xmm0
 ; SSE41-NEXT:    por %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vpslld %xmm2, %xmm0, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [32,32,32,32]
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [31,31,31,31]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpslld %xmm2, %xmm0, %xmm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [32,32,32,32]
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq
; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %amt, <4 x i32> undef, <4 x i32> zeroinitializer
   %res = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> %splat)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @splatvar_funnnel_v8i16(<8 x i16> %x, <8 x i16> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v8i16:
 ; SSE2:       # %bb.0:
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
 ; SSE2-NEXT:    psubw %xmm1, %xmm2
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    psllw %xmm1, %xmm3
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE2-NEXT:    psrlw %xmm2, %xmm0
 ; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v8i16:
 ; SSE41:       # %bb.0:
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
 ; SSE41-NEXT:    psllw %xmm2, %xmm3
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
 ; SSE41-NEXT:    psubw %xmm1, %xmm2
 ; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm1 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; SSE41-NEXT:    psrlw %xmm1, %xmm0
 ; SSE41-NEXT:    por %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v8i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX1-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_funnnel_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX512-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vprotw %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotw %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v8i16:
 ; X32-SSE:       # %bb.0:
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
 ; X32-SSE-NEXT:    psubw %xmm1, %xmm2
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
 ; X32-SSE-NEXT:    psllw %xmm1, %xmm3
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; X32-SSE-NEXT:    psrlw %xmm2, %xmm0
 ; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <8 x i16> %amt, <8 x i16> undef, <8 x i32> zeroinitializer
   %res = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %x, <8 x i16> %x, <8 x i16> %splat)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @splatvar_funnnel_v16i8(<16 x i8> %x, <16 x i8> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; SSE2-NEXT:    psubb %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm1, %xmm3
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psllw %xmm3, %xmm1

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE2-NEXT:    pand %xmm0, %xmm2
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v16i8:
 ; SSE41:       # %bb.0:
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    pxor %xmm3, %xmm3
 ; SSE41-NEXT:    pshufb %xmm3, %xmm1
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm4 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    psllw %xmm4, %xmm2
 ; SSE41-NEXT:    pcmpeqd %xmm5, %xmm5
 ; SSE41-NEXT:    pcmpeqd %xmm6, %xmm6
 ; SSE41-NEXT:    psllw %xmm4, %xmm6
 ; SSE41-NEXT:    pshufb %xmm3, %xmm6
 ; SSE41-NEXT:    pand %xmm6, %xmm2
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; SSE41-NEXT:    psubb %xmm1, %xmm3
 ; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm1 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    psrlw %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw %xmm1, %xmm5
 ; SSE41-NEXT:    pshufb {{.*#+}} xmm5 = xmm5[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
 ; SSE41-NEXT:    pand %xmm0, %xmm5
 ; SSE41-NEXT:    por %xmm5, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v16i8:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm4
 ; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpand %xmm2, %xmm4, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX1-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm5, %xmm1
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm0, %xmm3
 ; AVX2-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
 ; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX2-NEXT:    vpand %xmm2, %xmm3, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm3, %xmm1

 ; XOPAVX2-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotb %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; X32-SSE-NEXT:    psubb %xmm1, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psllw %xmm3, %xmm1

llvm/test/CodeGen/X86/vector-fshl-rot-512.ll

; AVX512-NEXT:    retq
   %splat = shufflevector <16 x i32> %amt, <16 x i32> undef, <16 x i32> zeroinitializer
   %res = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %x, <16 x i32> %x, <16 x i32> %splat)
   ret <16 x i32> %res
 }
 
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
 ; AVX512F-NEXT:    vpsubw %xmm2, %xmm5, %xmm2
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm0, %ymm4, %ymm0
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm1, %ymm3
 ; AVX512F-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
 ; AVX512VL-NEXT:    vpsubw %xmm2, %xmm5, %xmm2
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm0, %ymm4, %ymm0
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm1, %ymm3
; AVX512VLBW-NEXT:    retq
   %splat = shufflevector <32 x i16> %amt, <32 x i16> undef, <32 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %x, <32 x i16> %x, <32 x i16> %splat)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX512F-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm5, %ymm6
 ; AVX512F-NEXT:    vpbroadcastb %xmm6, %ymm6
 ; AVX512F-NEXT:    vpand %ymm6, %ymm4, %ymm4
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm2, %xmm7, %xmm2
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw %xmm2, %ymm5, %ymm5
 ; AVX512F-NEXT:    vpsrlw $8, %ymm5, %ymm5
 ; AVX512F-NEXT:    vpbroadcastb %xmm5, %ymm5
 ; AVX512F-NEXT:    vpand %ymm5, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm0, %ymm4, %ymm0
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm1, %ymm3
 ; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpand %ymm5, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX512VL-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm5, %ymm6
 ; AVX512VL-NEXT:    vpbroadcastb %xmm6, %ymm6
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VL-NEXT:    vpsubb %xmm2, %xmm7, %xmm2

llvm/test/CodeGen/X86/vector-fshr-128.ll

 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <2 x i64> @splatvar_funnnel_v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v2i64:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    movdqa %xmm1, %xmm3
 ; SSE2-NEXT:    psrlq %xmm2, %xmm3
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [64,64]
 ; SSE2-NEXT:    psubq %xmm2, %xmm4
 ; SSE2-NEXT:    psllq %xmm4, %xmm0
 ; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    pcmpeqd %xmm2, %xmm3
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
 ; SSE2-NEXT:    pand %xmm3, %xmm2
 ; SSE2-NEXT:    pand %xmm2, %xmm1
 ; SSE2-NEXT:    pandn %xmm0, %xmm2
 ; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v2i64:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrlq %xmm2, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [64,64]
 ; SSE41-NEXT:    psubq %xmm2, %xmm4
 ; SSE41-NEXT:    psllq %xmm4, %xmm3
 ; SSE41-NEXT:    por %xmm0, %xmm3
 ; SSE41-NEXT:    pxor %xmm0, %xmm0
 ; SSE41-NEXT:    pcmpeqq %xmm2, %xmm0
 ; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm3
 ; SSE41-NEXT:    movapd %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v2i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v2i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX2-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
+; AVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v2i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vptestnmq %xmm2, %xmm2, %k1
 ; AVX512VL-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1}
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v2i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
+; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vptestnmq %xmm2, %xmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvq %xmm2, %xmm0, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v2i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v2i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsrlq %xmm2, %xmm1, %xmm3
+; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
-; X32-SSE-NEXT:    psrlq %xmm2, %xmm3
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
-; X32-SSE-NEXT:    psrlq %xmm4, %xmm5
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm5 = xmm3[0],xmm5[1]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [64,0,64,0]
-; X32-SSE-NEXT:    psubq %xmm2, %xmm3
+; X32-SSE-NEXT:    movl $63, %eax
+; X32-SSE-NEXT:    movd %eax, %xmm3
+; X32-SSE-NEXT:    pand %xmm2, %xmm3
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm3[0,1,0,1]
+; X32-SSE-NEXT:    pxor %xmm5, %xmm5
+; X32-SSE-NEXT:    pcmpeqd %xmm4, %xmm5
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[1,0,3,2]
+; X32-SSE-NEXT:    pand %xmm5, %xmm2
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm5 = [64,0,64,0]
+; X32-SSE-NEXT:    psubq %xmm4, %xmm5
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
-; X32-SSE-NEXT:    psllq %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; X32-SSE-NEXT:    psllq %xmm3, %xmm0
+; X32-SSE-NEXT:    psllq %xmm5, %xmm4
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,3,0,1]
+; X32-SSE-NEXT:    psllq %xmm5, %xmm0
 ; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm4[0],xmm0[1]
-; X32-SSE-NEXT:    orpd %xmm5, %xmm0
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    pcmpeqd %xmm2, %xmm3
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[1,0,3,2]
-; X32-SSE-NEXT:    pand %xmm3, %xmm2
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
+; X32-SSE-NEXT:    psrlq %xmm3, %xmm4
+; X32-SSE-NEXT:    por %xmm0, %xmm4
 ; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pandn %xmm0, %xmm2
+; X32-SSE-NEXT:    pandn %xmm4, %xmm2
 ; X32-SSE-NEXT:    por %xmm1, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <2 x i64> %amt, <2 x i64> undef, <2 x i32> zeroinitializer
   %res = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x, <2 x i64> %y, <2 x i64> %splat)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    andps {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    xorps %xmm4, %xmm4
 ; SSE2-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
 ; SSE2-NEXT:    movdqa %xmm1, %xmm5
 ; SSE2-NEXT:    psrld %xmm4, %xmm5
 ; SSE2-NEXT:    movd %xmm2, %eax
 ; SSE2-NEXT:    movl $32, %ecx
 ; SSE2-NEXT:    subl %eax, %ecx
 ; SSE2-NEXT:    movd %ecx, %xmm4
 ; SSE2-NEXT:    pslld %xmm4, %xmm0
 ; SSE2-NEXT:    por %xmm5, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
 ; SSE2-NEXT:    pand %xmm2, %xmm1
 ; SSE2-NEXT:    pandn %xmm0, %xmm2
 ; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v4i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm2[0],zero,xmm2[1],zero
 ; SSE41-NEXT:    movdqa %xmm1, %xmm4
 ; SSE41-NEXT:    psrld %xmm0, %xmm4
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [32,32,32,32]
 ; SSE41-NEXT:    psubd %xmm2, %xmm0
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
 ; SSE41-NEXT:    pslld %xmm0, %xmm3
 ; SSE41-NEXT:    por %xmm4, %xmm3
 ; SSE41-NEXT:    pxor %xmm0, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm2, %xmm0
 ; SSE41-NEXT:    blendvps %xmm0, %xmm1, %xmm3
 ; SSE41-NEXT:    movaps %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX1-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX2-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
+; AVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX2-NEXT:    vpslld %xmm4, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v4i32:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512F-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512F-NEXT:    vpslld %xmm4, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VL-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VL-NEXT:    vpslld %xmm4, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vptestnmd %xmm2, %xmm2, %k1
 ; AVX512VL-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512BW-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512BW-NEXT:    vpslld %xmm4, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VBMI2-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
+; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VBMI2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VBMI2-NEXT:    vpslld %xmm4, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VLBW-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VLBW-NEXT:    vpslld %xmm4, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vptestnmd %xmm2, %xmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa32 %xmm1, %xmm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvd %xmm2, %xmm0, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; XOPAVX1-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
 ; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; XOPAVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [31,31,31,31]
 ; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; XOPAVX2-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
+; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; XOPAVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; XOPAVX2-NEXT:    vpslld %xmm4, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vblendvps %xmm2, %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    andps {{\.LCPI.*}}, %xmm2
 ; X32-SSE-NEXT:    pxor %xmm3, %xmm3
 ; X32-SSE-NEXT:    xorps %xmm4, %xmm4
 ; X32-SSE-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
 ; X32-SSE-NEXT:    psrld %xmm4, %xmm5
 ; X32-SSE-NEXT:    movd %xmm2, %eax
 ; X32-SSE-NEXT:    movl $32, %ecx
 ; X32-SSE-NEXT:    subl %eax, %ecx
 ; X32-SSE-NEXT:    movd %ecx, %xmm4
 ; X32-SSE-NEXT:    pslld %xmm4, %xmm0
 ; X32-SSE-NEXT:    por %xmm5, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm2
 ; X32-SSE-NEXT:    pand %xmm2, %xmm1
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm2
 ; X32-SSE-NEXT:    por %xmm1, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %amt, <4 x i32> undef, <4 x i32> zeroinitializer
   %res = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %splat)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @splatvar_funnnel_v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v8i16:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; SSE2-NEXT:    psubw %xmm3, %xmm4
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
+; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    movdqa %xmm1, %xmm4
+; SSE2-NEXT:    psrlw %xmm2, %xmm4
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
+; SSE2-NEXT:    psubw %xmm3, %xmm2
+; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
+; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    psllw %xmm2, %xmm0
+; SSE2-NEXT:    por %xmm4, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
 ; SSE2-NEXT:    pcmpeqw %xmm3, %xmm2
-; SSE2-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1]
-; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    psrlw %xmm3, %xmm5
-; SSE2-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm4[0,1]
-; SSE2-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    psllw %xmm4, %xmm0
-; SSE2-NEXT:    por %xmm5, %xmm0
 ; SSE2-NEXT:    pand %xmm2, %xmm1
 ; SSE2-NEXT:    pandn %xmm0, %xmm2
 ; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v8i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,0,0]
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; SSE41-NEXT:    movdqa %xmm1, %xmm4
 ; SSE41-NEXT:    psrlw %xmm0, %xmm4
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,0,0]
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [16,16,16,16,16,16,16,16]
 ; SSE41-NEXT:    psubw %xmm2, %xmm0
 ; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
 ; SSE41-NEXT:    psllw %xmm0, %xmm3
 ; SSE41-NEXT:    por %xmm4, %xmm3
 ; SSE41-NEXT:    pxor %xmm0, %xmm0
 ; SSE41-NEXT:    pcmpeqw %xmm2, %xmm0
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm1, %xmm3
 ; SSE41-NEXT:    movdqa %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v8i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX1-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
+; AVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v8i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
+; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VBMI2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vptestnmw %xmm2, %xmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu16 %xmm1, %xmm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvw %xmm2, %xmm0, %xmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %xmm1, %xmm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; XOPAVX1-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; XOPAVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; XOPAVX2-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
+; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; XOPAVX2-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v8i16:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm3
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
-; X32-SSE-NEXT:    psubw %xmm3, %xmm4
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
+; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
+; X32-SSE-NEXT:    psrlw %xmm2, %xmm4
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
+; X32-SSE-NEXT:    psubw %xmm3, %xmm2
+; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
+; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; X32-SSE-NEXT:    psllw %xmm2, %xmm0
+; X32-SSE-NEXT:    por %xmm4, %xmm0
 ; X32-SSE-NEXT:    pxor %xmm2, %xmm2
 ; X32-SSE-NEXT:    pcmpeqw %xmm3, %xmm2
-; X32-SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,1]
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
-; X32-SSE-NEXT:    psrlw %xmm3, %xmm5
-; X32-SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm4[0,1]
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    psllw %xmm4, %xmm0
-; X32-SSE-NEXT:    por %xmm5, %xmm0
 ; X32-SSE-NEXT:    pand %xmm2, %xmm1
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm2
 ; X32-SSE-NEXT:    por %xmm1, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <8 x i16> %amt, <8 x i16> undef, <8 x i32> zeroinitializer
   %res = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %splat)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @splatvar_funnnel_v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm3
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; SSE2-NEXT:    psubb %xmm3, %xmm4
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    pcmpeqb %xmm3, %xmm2
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm3
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    psrlw %xmm3, %xmm5
+; SSE2-NEXT:    movdqa %xmm1, %xmm4
+; SSE2-NEXT:    psrlw %xmm3, %xmm4
+; SSE2-NEXT:    pcmpeqd %xmm5, %xmm5
 ; SSE2-NEXT:    pcmpeqd %xmm6, %xmm6
 ; SSE2-NEXT:    psrlw %xmm3, %xmm6
-; SSE2-NEXT:    pcmpeqd %xmm3, %xmm3
 ; SSE2-NEXT:    psrlw $8, %xmm6
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,0,0]
-; SSE2-NEXT:    pand %xmm5, %xmm6
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm6[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; SSE2-NEXT:    pand %xmm4, %xmm3
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; SSE2-NEXT:    psubb %xmm2, %xmm4
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm4[0]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE2-NEXT:    psllw %xmm4, %xmm0
-; SSE2-NEXT:    psllw %xmm4, %xmm3
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; SSE2-NEXT:    pand %xmm0, %xmm3
-; SSE2-NEXT:    por %xmm6, %xmm3
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm3, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    psllw %xmm4, %xmm5
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
+; SSE2-NEXT:    pand %xmm0, %xmm4
+; SSE2-NEXT:    por %xmm3, %xmm4
+; SSE2-NEXT:    pxor %xmm0, %xmm0
+; SSE2-NEXT:    pcmpeqb %xmm2, %xmm0
+; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    pandn %xmm4, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pshufb %xmm0, %xmm2
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    movdqa %xmm1, %xmm5
-; SSE41-NEXT:    psrlw %xmm4, %xmm5
+; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    movdqa %xmm1, %xmm4
+; SSE41-NEXT:    psrlw %xmm0, %xmm4
+; SSE41-NEXT:    pcmpeqd %xmm5, %xmm5
 ; SSE41-NEXT:    pcmpeqd %xmm6, %xmm6
-; SSE41-NEXT:    pcmpeqd %xmm7, %xmm7
-; SSE41-NEXT:    psrlw %xmm4, %xmm7
-; SSE41-NEXT:    pshufb {{.*#+}} xmm7 = xmm7[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; SSE41-NEXT:    pand %xmm5, %xmm7
+; SSE41-NEXT:    psrlw %xmm0, %xmm6
+; SSE41-NEXT:    pshufb {{.*#+}} xmm6 = xmm6[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; SSE41-NEXT:    pand %xmm4, %xmm6
+; SSE41-NEXT:    pxor %xmm0, %xmm0
+; SSE41-NEXT:    pshufb %xmm0, %xmm2
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; SSE41-NEXT:    psubb %xmm2, %xmm4
 ; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    psllw %xmm4, %xmm3
-; SSE41-NEXT:    psllw %xmm4, %xmm6
-; SSE41-NEXT:    pshufb %xmm0, %xmm6
-; SSE41-NEXT:    pand %xmm6, %xmm3
-; SSE41-NEXT:    por %xmm7, %xmm3
+; SSE41-NEXT:    psllw %xmm4, %xmm5
+; SSE41-NEXT:    pshufb %xmm0, %xmm5
+; SSE41-NEXT:    pand %xmm5, %xmm3
+; SSE41-NEXT:    por %xmm6, %xmm3
 ; SSE41-NEXT:    pcmpeqb %xmm2, %xmm0
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm1, %xmm3
 ; SSE41-NEXT:    movdqa %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v16i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm5
-; AVX1-NEXT:    vpcmpeqd %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm6, %xmm4
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpand %xmm4, %xmm5, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX1-NEXT:    vpsubb %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,zero,zero,zero,zero,xmm5[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpsllw %xmm5, %xmm6, %xmm5
-; AVX1-NEXT:    vpshufb %xmm3, %xmm5, %xmm5
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrlw %xmm3, %xmm1, %xmm4
+; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX1-NEXT:    vpand %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX1-NEXT:    vpsubb %xmm2, %xmm6, %xmm6
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsllw %xmm6, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpshufb %xmm4, %xmm5, %xmm5
 ; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpor %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqb %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm3, %xmm1, %xmm4
 ; AVX2-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX2-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
 ; AVX2-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX2-NEXT:    vpbroadcastb %xmm3, %xmm3
 ; AVX2-NEXT:    vpand %xmm3, %xmm4, %xmm3
+; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
 ; AVX2-NEXT:    vpbroadcastb %xmm4, %xmm4
 ; AVX2-NEXT:    vpand %xmm4, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlvd %zmm4, %zmm3, %zmm3
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
 ; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512F-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpord %zmm3, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm3 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero,xmm2[8],zero,zero,zero,xmm2[9],zero,zero,zero,xmm2[10],zero,zero,zero,xmm2[11],zero,zero,zero,xmm2[12],zero,zero,zero,xmm2[13],zero,zero,zero,xmm2[14],zero,zero,zero,xmm2[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlvd %zmm4, %zmm3, %zmm3
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero,xmm4[8],zero,zero,zero,xmm4[9],zero,zero,zero,xmm4[10],zero,zero,zero,xmm4[11],zero,zero,zero,xmm4[12],zero,zero,zero,xmm4[13],zero,zero,zero,xmm4[14],zero,zero,zero,xmm4[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllvd %zmm4, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpord %zmm3, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512BW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
 ; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
 ; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512VBMI2-NEXT:    vzeroupper
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; AVX512VLBW-NEXT:    vpsrlvw %ymm4, %ymm3, %ymm3
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512VLBW-NEXT:    vpsllvw %ymm4, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLBW-NEXT:    vptestnmb %xmm2, %xmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %xmm1, %xmm0 {%k1}
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; AVX512VLVBMI2-NEXT:    vpsrlvw %ymm4, %ymm3, %ymm3
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLVBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512VLVBMI2-NEXT:    vpsllvw %ymm4, %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLVBMI2-NEXT:    vptestnmb %xmm2, %xmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %xmm1, %xmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vzeroupper
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX1:       # %bb.0:
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm4
 ; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm1, %xmm4
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm5, %xmm5
 ; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpor %xmm4, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm3, %xmm4
 ; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm1, %xmm4
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm5, %xmm5
 ; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpor %xmm4, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,0,0]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm3
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; X32-SSE-NEXT:    psubb %xmm3, %xmm4
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pcmpeqb %xmm3, %xmm2
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
-; X32-SSE-NEXT:    psrlw %xmm3, %xmm5
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
+; X32-SSE-NEXT:    psrlw %xmm3, %xmm4
+; X32-SSE-NEXT:    pcmpeqd %xmm5, %xmm5
 ; X32-SSE-NEXT:    pcmpeqd %xmm6, %xmm6
 ; X32-SSE-NEXT:    psrlw %xmm3, %xmm6
-; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm3
 ; X32-SSE-NEXT:    psrlw $8, %xmm6
 ; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,0,0]
-; X32-SSE-NEXT:    pand %xmm5, %xmm6
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm6[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; X32-SSE-NEXT:    pand %xmm4, %xmm3
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; X32-SSE-NEXT:    psubb %xmm2, %xmm4
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm4[0]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; X32-SSE-NEXT:    psllw %xmm4, %xmm0
-; X32-SSE-NEXT:    psllw %xmm4, %xmm3
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; X32-SSE-NEXT:    pand %xmm0, %xmm3
-; X32-SSE-NEXT:    por %xmm6, %xmm3
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pandn %xmm3, %xmm2
-; X32-SSE-NEXT:    por %xmm1, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
+; X32-SSE-NEXT:    psllw %xmm4, %xmm5
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
+; X32-SSE-NEXT:    pand %xmm0, %xmm4
+; X32-SSE-NEXT:    por %xmm3, %xmm4
+; X32-SSE-NEXT:    pxor %xmm0, %xmm0
+; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm0
+; X32-SSE-NEXT:    pand %xmm0, %xmm1
+; X32-SSE-NEXT:    pandn %xmm4, %xmm0
+; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <16 x i8> %amt, <16 x i8> undef, <16 x i32> zeroinitializer
   %res = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %splat)
   ret <16 x i8> %res
 }
 
 ;
 ; Constant Shifts

llvm/test/CodeGen/X86/vector-fshr-256.ll

 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <4 x i64> @splatvar_funnnel_v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v4i64:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovddup {{.*#+}} xmm2 = xmm2[0,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
 ; AVX1-NEXT:    vpsrlq %xmm2, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm4
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [64,64]
-; AVX1-NEXT:    vpsubq %xmm4, %xmm5, %xmm6
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; AVX1-NEXT:    vpsllq %xmm6, %xmm7, %xmm6
-; AVX1-NEXT:    vpsubq %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpsllq %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
+; AVX1-NEXT:    vmovddup {{.*#+}} xmm2 = xmm2[0,0]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
+; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; AVX1-NEXT:    vpsllq %xmm4, %xmm5, %xmm5
+; AVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
 ; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqq %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; AVX1-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
+; AVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v4i64:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vptestnmq %ymm2, %ymm2, %k1
 ; AVX512VL-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1}
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v4i64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
+; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vptestnmq %ymm2, %ymm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvq %ymm2, %ymm0, %ymm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vmovddup {{.*#+}} xmm2 = xmm2[0,0]
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vpsrlq %xmm2, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpsrlq %xmm2, %xmm1, %xmm4
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [64,64]
-; XOPAVX1-NEXT:    vpsubq %xmm4, %xmm5, %xmm6
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; XOPAVX1-NEXT:    vpsllq %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpsllq %xmm5, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vmovddup {{.*#+}} xmm2 = xmm2[0,0]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
+; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; XOPAVX1-NEXT:    vpsllq %xmm4, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpcomeqq %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [63,63,63,63]
-; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
+; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpsllq %xmm4, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcmpeqq %ymm3, %ymm2, %ymm2
 ; XOPAVX2-NEXT:    vblendvpd %ymm2, %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <4 x i64> %amt, <4 x i64> undef, <4 x i32> zeroinitializer
   %res = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %x, <4 x i64> %y, <4 x i64> %splat)
   ret <4 x i64> %res
 }
 
 define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX1-NEXT:    vpsrld %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm4, %xmm5, %xmm6
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm6 = xmm6[0],zero,xmm6[1],zero
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; AVX1-NEXT:    vpslld %xmm6, %xmm7, %xmm6
-; AVX1-NEXT:    vpsubd %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm5[0],zero,xmm5[1],zero
-; AVX1-NEXT:    vpslld %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero
+; AVX1-NEXT:    vpsrld %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm4
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
+; AVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; AVX1-NEXT:    vpslld %xmm4, %xmm5, %xmm5
+; AVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
 ; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; AVX1-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX2-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
+; AVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX2-NEXT:    vpslld %xmm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v8i32:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512F-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512F-NEXT:    vpslld %xmm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to8}, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VL-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VL-NEXT:    vpslld %xmm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vptestnmd %ymm2, %ymm2, %k1
 ; AVX512VL-NEXT:    vmovdqa32 %ymm1, %ymm0 {%k1}
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512BW-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512BW-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512BW-NEXT:    vpslld %xmm4, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; AVX512VBMI2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VBMI2-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
+; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VBMI2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VBMI2-NEXT:    vpslld %xmm4, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to8}, %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VLBW-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VLBW-NEXT:    vpslld %xmm4, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vptestnmd %ymm2, %ymm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa32 %ymm1, %ymm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvd %ymm2, %ymm0, %ymm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; XOPAVX1-NEXT:    vpsrld %xmm3, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpsrld %xmm3, %xmm1, %xmm3
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [32,32,32,32]
-; XOPAVX1-NEXT:    vpsubd %xmm4, %xmm5, %xmm6
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm6 = xmm6[0],zero,xmm6[1],zero
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; XOPAVX1-NEXT:    vpslld %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm5[0],zero,xmm5[1],zero
-; XOPAVX1-NEXT:    vpslld %xmm5, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero
+; XOPAVX1-NEXT:    vpsrld %xmm4, %xmm3, %xmm3
+; XOPAVX1-NEXT:    vpsrld %xmm4, %xmm1, %xmm4
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [32,32,32,32]
+; XOPAVX1-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; XOPAVX1-NEXT:    vpslld %xmm4, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpslld %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpcomeqd %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [31,31,31,31,31,31,31,31]
-; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; XOPAVX2-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
+; XOPAVX2-NEXT:    vpbroadcastd %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; XOPAVX2-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; XOPAVX2-NEXT:    vpslld %xmm4, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcmpeqd %ymm3, %ymm2, %ymm2
 ; XOPAVX2-NEXT:    vblendvps %ymm2, %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <8 x i32> %amt, <8 x i32> undef, <8 x i32> zeroinitializer
   %res = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %x, <8 x i32> %y, <8 x i32> %splat)
   ret <8 x i32> %res
 }
 
 define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v16i16:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm4
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
 ; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX1-NEXT:    vpsrlw %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; AVX1-NEXT:    vpsubw %xmm4, %xmm5, %xmm6
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; AVX1-NEXT:    vpsllw %xmm6, %xmm7, %xmm6
-; AVX1-NEXT:    vpsubw %xmm2, %xmm5, %xmm5
-; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
+; AVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; AVX1-NEXT:    vpsllw %xmm4, %xmm5, %xmm5
+; AVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
 ; AVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
 ; AVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
+; AVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v16i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512F-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VL-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
+; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VBMI2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VBMI2-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vptestnmw %ymm2, %ymm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu16 %ymm1, %ymm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvw %ymm2, %ymm0, %ymm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa %ymm1, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; XOPAVX1-NEXT:    vpsrlw %xmm4, %xmm3, %xmm3
+; XOPAVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm4
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
 ; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
 ; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; XOPAVX1-NEXT:    vpsrlw %xmm3, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpsrlw %xmm3, %xmm1, %xmm3
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm3, %ymm3
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; XOPAVX1-NEXT:    vpsubw %xmm4, %xmm5, %xmm6
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
-; XOPAVX1-NEXT:    vpsllw %xmm6, %xmm7, %xmm6
-; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm5, %xmm5
-; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero
-; XOPAVX1-NEXT:    vpsllw %xmm5, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
+; XOPAVX1-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; XOPAVX1-NEXT:    vpsllw %xmm4, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpsllw %xmm4, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpcomeqw %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm0, %ymm1, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; XOPAVX2-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
+; XOPAVX2-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; XOPAVX2-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; XOPAVX2-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX2-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
 ; XOPAVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <16 x i16> %amt, <16 x i16> undef, <16 x i32> zeroinitializer
   %res = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %x, <16 x i16> %y, <16 x i16> %splat)
   ret <16 x i16> %res
 }
 
 define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
-; AVX1-NEXT:    vpcmpeqd %xmm8, %xmm8, %xmm8
-; AVX1-NEXT:    vpsrlw %xmm4, %xmm8, %xmm7
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpsrlw %xmm4, %xmm5, %xmm6
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsrlw %xmm4, %xmm1, %xmm4
-; AVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
-; AVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm4, %ymm9
-; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX1-NEXT:    vpsubb %xmm5, %xmm7, %xmm6
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; AVX1-NEXT:    vpsllw %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpsllw %xmm6, %xmm8, %xmm6
-; AVX1-NEXT:    vpshufb %xmm3, %xmm6, %xmm6
 ; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm4
-; AVX1-NEXT:    vpsubb %xmm2, %xmm7, %xmm6
+; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm4, %ymm3
+; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX1-NEXT:    vpsubb %xmm2, %xmm6, %xmm6
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm7
+; AVX1-NEXT:    vpsllw %xmm6, %xmm7, %xmm7
+; AVX1-NEXT:    vpsllw %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpshufb %xmm4, %xmm5, %xmm5
+; AVX1-NEXT:    vpand %xmm5, %xmm7, %xmm7
 ; AVX1-NEXT:    vpsllw %xmm6, %xmm0, %xmm0
-; AVX1-NEXT:    vpsllw %xmm6, %xmm8, %xmm6
-; AVX1-NEXT:    vpshufb %xmm3, %xmm6, %xmm6
-; AVX1-NEXT:    vpand %xmm6, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm9, %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpeqb %xmm3, %xmm5, %xmm4
-; AVX1-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; AVX1-NEXT:    vpand %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm7, %ymm0, %ymm0
+; AVX1-NEXT:    vorps %ymm3, %ymm0, %ymm0
+; AVX1-NEXT:    vpcmpeqb %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; AVX1-NEXT:    vandnps %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm3, %ymm1, %ymm4
 ; AVX2-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
 ; AVX2-NEXT:    vpsrlw %xmm3, %ymm5, %ymm3
 ; AVX2-NEXT:    vpsrlw $8, %ymm3, %ymm3
 ; AVX2-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX2-NEXT:    vpand %ymm3, %ymm4, %ymm3
+; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsllw %xmm4, %ymm5, %ymm4
 ; AVX2-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm3, %ymm1, %ymm4
 ; AVX512F-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
 ; AVX512F-NEXT:    vpsrlw %xmm3, %ymm5, %ymm3
 ; AVX512F-NEXT:    vpsrlw $8, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX512F-NEXT:    vpand %ymm3, %ymm4, %ymm3
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsllw %xmm4, %ymm5, %ymm4
 ; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512F-NEXT:    vpand %ymm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm1, %ymm4
 ; AVX512VL-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
 ; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm5, %ymm3
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm4, %ymm3
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VL-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsllw %xmm4, %ymm5, %ymm4
 ; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512BW-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512BW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
 ; AVX512VBMI2-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
 ; AVX512VBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
 ; AVX512VLBW-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLBW-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
 ; AVX512VLBW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VLBW-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLBW-NEXT:    vptestnmb %ymm2, %ymm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %ymm1, %ymm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero
-; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero,ymm2[16],zero,ymm2[17],zero,ymm2[18],zero,ymm2[19],zero,ymm2[20],zero,ymm2[21],zero,ymm2[22],zero,ymm2[23],zero,ymm2[24],zero,ymm2[25],zero,ymm2[26],zero,ymm2[27],zero,ymm2[28],zero,ymm2[29],zero,ymm2[30],zero,ymm2[31],zero
 ; AVX512VLVBMI2-NEXT:    vpsrlvw %zmm4, %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} ymm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLVBMI2-NEXT:    vpsubb %ymm2, %ymm4, %ymm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm4 = ymm4[0],zero,ymm4[1],zero,ymm4[2],zero,ymm4[3],zero,ymm4[4],zero,ymm4[5],zero,ymm4[6],zero,ymm4[7],zero,ymm4[8],zero,ymm4[9],zero,ymm4[10],zero,ymm4[11],zero,ymm4[12],zero,ymm4[13],zero,ymm4[14],zero,ymm4[15],zero,ymm4[16],zero,ymm4[17],zero,ymm4[18],zero,ymm4[19],zero,ymm4[20],zero,ymm4[21],zero,ymm4[22],zero,ymm4[23],zero,ymm4[24],zero,ymm4[25],zero,ymm4[26],zero,ymm4[27],zero,ymm4[28],zero,ymm4[29],zero,ymm4[30],zero,ymm4[31],zero
 ; AVX512VLVBMI2-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512VLVBMI2-NEXT:    vpsllvw %zmm4, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vptestnmb %ymm2, %ymm2, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %ymm1, %ymm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; XOPAVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; XOPAVX1-NEXT:    vpsubb %xmm4, %xmm3, %xmm5
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
-; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm6, %xmm1, %xmm6
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm6, %ymm5
-; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; XOPAVX1-NEXT:    vpsubb %xmm4, %xmm8, %xmm7
+; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm3, %xmm4
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm5, %xmm5
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm1, %xmm4
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm4, %ymm4
+; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm5, %xmm5
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm7, %xmm6, %xmm6
-; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm8, %xmm7
-; XOPAVX1-NEXT:    vpshlb %xmm7, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm6, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vorps %ymm5, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vorps %ymm4, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    vpcomeqb %xmm3, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vpcmov %ymm2, %ymm0, %ymm1, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
-; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; XOPAVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpsubb %xmm3, %xmm4, %xmm3
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm5
 ; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm5, %xmm3
 ; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm1, %xmm4
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3

llvm/test/CodeGen/X86/vector-fshr-512.ll

 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <8 x i64> @splatvar_funnnel_v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v8i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512F-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; AVX512F-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsllq %xmm4, %zmm0, %zmm0
 ; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512F-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsllq %xmm4, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v8i64:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsllq %xmm4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshrdvq %zmm2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip){1to8}, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsllq %xmm4, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vptestnmq %zmm2, %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i64:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastq %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvq %zmm2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <8 x i64> %amt, <8 x i64> undef, <8 x i32> zeroinitializer
   %res = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %splat)
   ret <8 x i64> %res
 }
 
 define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v16i32:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512F-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; AVX512F-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512F-NEXT:    vpsrld %xmm3, %zmm1, %zmm3
+; AVX512F-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512F-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512F-NEXT:    vpslld %xmm4, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpord %zmm3, %zmm0, %zmm0
 ; AVX512F-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512VL-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VL-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VL-NEXT:    vpsrld %xmm3, %zmm1, %zmm3
+; AVX512VL-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VL-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VL-NEXT:    vpslld %xmm4, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpord %zmm3, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512VL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512BW-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512BW-NEXT:    vpsrld %xmm3, %zmm1, %zmm3
+; AVX512BW-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512BW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512BW-NEXT:    vpslld %xmm4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpord %zmm3, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshrdvd %zmm2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vpandd {{.*}}(%rip){1to16}, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
+; AVX512VLBW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VLBW-NEXT:    vpsrld %xmm3, %zmm1, %zmm3
+; AVX512VLBW-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpbroadcastd {{.*#+}} xmm4 = [32,32,32,32]
 ; AVX512VLBW-NEXT:    vpsubd %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm4[0],zero,xmm4[1],zero
 ; AVX512VLBW-NEXT:    vpslld %xmm4, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpord %zmm3, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vptestnmd %zmm2, %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i32:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastd %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvd %zmm2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <16 x i32> %amt, <16 x i32> undef, <16 x i32> zeroinitializer
   %res = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %splat)
   ret <16 x i32> %res
 }
 
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm4, %ymm4
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm5, %ymm2, %ymm6
+; AVX512F-NEXT:    vpbroadcastw %xmm4, %ymm4
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm7 = [16,16,16,16,16,16,16,16]
 ; AVX512F-NEXT:    vpsubw %xmm4, %xmm7, %xmm7
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm7 = xmm7[0],zero,zero,zero,xmm7[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm7, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm6, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpxor %xmm6, %xmm6, %xmm6
 ; AVX512F-NEXT:    vpcmpeqw %ymm6, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw %xmm5, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpsllw %xmm7, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm2, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm4, %ymm3, %ymm1, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm4, %ymm4
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm5, %ymm2, %ymm6
+; AVX512VL-NEXT:    vpbroadcastw %xmm4, %ymm4
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm7 = [16,16,16,16,16,16,16,16]
 ; AVX512VL-NEXT:    vpsubw %xmm4, %xmm7, %xmm7
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm7 = xmm7[0],zero,zero,zero,xmm7[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm7, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm6, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpxor %xmm6, %xmm6, %xmm6
 ; AVX512VL-NEXT:    vpcmpeqw %ymm6, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlw %xmm5, %ymm3, %ymm2
 ; AVX512VL-NEXT:    vpsllw %xmm7, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm2, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw %xmm2, %zmm2
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsrlw %xmm3, %zmm1, %zmm3
+; AVX512BW-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512BW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpshrdvw %zmm2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsrlw %xmm3, %zmm1, %zmm3
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [16,16,16,16,16,16,16,16]
 ; AVX512VLBW-NEXT:    vpsubw %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vptestnmw %zmm2, %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu16 %zmm1, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vpbroadcastw %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpshrdvw %zmm2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <32 x i16> %amt, <32 x i16> undef, <32 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %splat)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm5 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm5, %ymm2, %ymm6
 ; AVX512F-NEXT:    vpcmpeqd %ymm9, %ymm9, %ymm9
 ; AVX512F-NEXT:    vpsrlw %xmm5, %ymm9, %ymm8
 ; AVX512F-NEXT:    vpsrlw $8, %ymm8, %ymm8
 ; AVX512F-NEXT:    vpbroadcastb %xmm8, %ymm8
 ; AVX512F-NEXT:    vpand %ymm8, %ymm6, %ymm6
+; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm4, %xmm7, %xmm7
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm7 = xmm7[0],zero,zero,zero,zero,zero,zero,zero,xmm7[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm7, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsllw %xmm7, %ymm9, %ymm9
 ; AVX512F-NEXT:    vpbroadcastb %xmm9, %ymm9
 ; AVX512F-NEXT:    vpand %ymm9, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm6, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpxor %xmm6, %xmm6, %xmm6
 ; AVX512F-NEXT:    vpcmpeqb %ymm6, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw %xmm5, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpand %ymm8, %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsllw %xmm7, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpand %ymm9, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm2, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm4, %ymm3, %ymm1, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm5 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm5, %ymm2, %ymm6
 ; AVX512VL-NEXT:    vpcmpeqd %ymm9, %ymm9, %ymm9
 ; AVX512VL-NEXT:    vpsrlw %xmm5, %ymm9, %ymm8
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm8, %ymm8
 ; AVX512VL-NEXT:    vpbroadcastb %xmm8, %ymm8
 ; AVX512VL-NEXT:    vpand %ymm8, %ymm6, %ymm6
+; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VL-NEXT:    vpsubb %xmm4, %xmm7, %xmm7
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm7 = xmm7[0],zero,zero,zero,zero,zero,zero,zero,xmm7[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm7, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsllw %xmm7, %ymm9, %ymm9
 ; AVX512VL-NEXT:    vpbroadcastb %xmm9, %ymm9
 ; AVX512VL-NEXT:    vpand %ymm9, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm6, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpxor %xmm6, %xmm6, %xmm6
 ; AVX512VL-NEXT:    vpcmpeqb %ymm6, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlw %xmm5, %ymm3, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm8, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpsllw %xmm7, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpand %ymm9, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm2, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512BW-NEXT:    vpsrlw %xmm3, %zmm1, %zmm4
 ; AVX512BW-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5
 ; AVX512BW-NEXT:    vpsrlw %xmm3, %zmm5, %zmm3
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512BW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512BW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsllw %xmm4, %zmm5, %zmm4
 ; AVX512BW-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512BW-NEXT:    vpandq %zmm4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsrlw %xmm3, %zmm1, %zmm4
 ; AVX512VBMI2-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5
 ; AVX512VBMI2-NEXT:    vpsrlw %xmm3, %zmm5, %zmm3
 ; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
+; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VBMI2-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vpsllw %xmm4, %zmm5, %zmm4
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpandq %zmm4, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsrlw %xmm3, %zmm1, %zmm4
 ; AVX512VLBW-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5
 ; AVX512VLBW-NEXT:    vpsrlw %xmm3, %zmm5, %zmm3
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLBW-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpsllw %xmm4, %zmm5, %zmm4
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpandq %zmm4, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLVBMI2-NEXT:    vpsrlw %xmm3, %zmm1, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpternlogd $255, %zmm5, %zmm5, %zmm5
 ; AVX512VLVBMI2-NEXT:    vpsrlw %xmm3, %zmm5, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
+; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VLVBMI2-NEXT:    vpsubb %xmm2, %xmm4, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm4[0],zero,zero,zero,zero,zero,zero,zero,xmm4[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLVBMI2-NEXT:    vpsllw %xmm4, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vpsllw %xmm4, %zmm5, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpandq %zmm4, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0

llvm/test/CodeGen/X86/vector-fshr-rot-128.ll

 ; AVX512VLBW-LABEL: splatvar_funnnel_v2i64:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vprorvq %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v2i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vprotq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v2i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v2i64:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [63,0,63,0]
 ; X32-SSE-NEXT:    pxor %xmm3, %xmm3

 ; SSE2-NEXT:    subl %eax, %ecx
 ; SSE2-NEXT:    movd %ecx, %xmm1
 ; SSE2-NEXT:    psrld %xmm1, %xmm0
 ; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v4i32:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE41-NEXT:    pxor %xmm2, %xmm2
 ; SSE41-NEXT:    psubd %xmm1, %xmm2
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,0,0,0]
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pslld %xmm1, %xmm3
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32,32,32,32]
-; SSE41-NEXT:    psubd %xmm2, %xmm1
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; SSE41-NEXT:    pslld %xmm2, %xmm3
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [32,32,32,32]
+; SSE41-NEXT:    psubd %xmm1, %xmm2
+; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero
 ; SSE41-NEXT:    psrld %xmm1, %xmm0
 ; SSE41-NEXT:    por %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vpslld %xmm2, %xmm0, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [32,32,32,32]
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [31,31,31,31]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpslld %xmm2, %xmm0, %xmm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [32,32,32,32]
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq

 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i32:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vprorvd %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v4i32:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movd %xmm1, %eax
 ; X32-SSE-NEXT:    negl %eax
 ; X32-SSE-NEXT:    andl $31, %eax
; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %amt, <4 x i32> undef, <4 x i32> zeroinitializer
   %res = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> %splat)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @splatvar_funnnel_v8i16(<8 x i16> %x, <8 x i16> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v8i16:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
 ; SSE2-NEXT:    psubw %xmm1, %xmm2
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [16,16,16,16,16,16,16,16]
-; SSE2-NEXT:    psubw %xmm2, %xmm1
-; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
-; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    psllw %xmm2, %xmm3
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
+; SSE2-NEXT:    psubw %xmm1, %xmm2
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    psrlw %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
+; SSE2-NEXT:    psllw %xmm1, %xmm3
+; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
+; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    psrlw %xmm2, %xmm0
 ; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v8i16:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE41-NEXT:    pxor %xmm2, %xmm2
 ; SSE41-NEXT:    psubw %xmm1, %xmm2
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm1 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    psllw %xmm1, %xmm3
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [16,16,16,16,16,16,16,16]
-; SSE41-NEXT:    psubw %xmm2, %xmm1
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; SSE41-NEXT:    psllw %xmm2, %xmm3
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
+; SSE41-NEXT:    psubw %xmm1, %xmm2
+; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm1 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; SSE41-NEXT:    psrlw %xmm1, %xmm0
 ; SSE41-NEXT:    por %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v8i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX1-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_funnnel_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
 ; AVX512-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX512-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vprotw %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotw %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v8i16:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; X32-SSE-NEXT:    pxor %xmm2, %xmm2
 ; X32-SSE-NEXT:    psubw %xmm1, %xmm2
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [16,16,16,16,16,16,16,16]
-; X32-SSE-NEXT:    psubw %xmm2, %xmm1
-; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    psllw %xmm2, %xmm3
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
+; X32-SSE-NEXT:    psubw %xmm1, %xmm2
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    psrlw %xmm1, %xmm0
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
+; X32-SSE-NEXT:    psllw %xmm1, %xmm3
+; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
+; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; X32-SSE-NEXT:    psrlw %xmm2, %xmm0
 ; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <8 x i16> %amt, <8 x i16> undef, <8 x i32> zeroinitializer
   %res = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x, <8 x i16> %x, <8 x i16> %splat)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @splatvar_funnnel_v16i8(<16 x i8> %x, <16 x i8> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
 ; SSE2-NEXT:    psubb %xmm1, %xmm2
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; SSE2-NEXT:    psubb %xmm2, %xmm3
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1

 ; SSE2-NEXT:    pand %xmm0, %xmm2
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pshufb %xmm2, %xmm1
 ; SSE41-NEXT:    pxor %xmm3, %xmm3
 ; SSE41-NEXT:    psubb %xmm1, %xmm3
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm3
+; SSE41-NEXT:    pshufb %xmm2, %xmm3
 ; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm4 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psllw %xmm4, %xmm1
 ; SSE41-NEXT:    pcmpeqd %xmm5, %xmm5
 ; SSE41-NEXT:    pcmpeqd %xmm6, %xmm6
 ; SSE41-NEXT:    psllw %xmm4, %xmm6
 ; SSE41-NEXT:    pshufb %xmm2, %xmm6
 ; SSE41-NEXT:    pand %xmm6, %xmm1
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; SSE41-NEXT:    psubb %xmm3, %xmm2
 ; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    psrlw %xmm2, %xmm0
 ; SSE41-NEXT:    psrlw %xmm2, %xmm5
 ; SSE41-NEXT:    pshufb {{.*#+}} xmm5 = xmm5[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
 ; SSE41-NEXT:    pand %xmm0, %xmm5
 ; SSE41-NEXT:    por %xmm5, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_funnnel_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm4
 ; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpand %xmm2, %xmm4, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX1-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm5, %xmm1
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm0, %xmm3
 ; AVX2-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
 ; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX2-NEXT:    vpand %xmm2, %xmm3, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm3, %xmm1

 ; AVX512VLBW-NEXT:    vpor %ymm3, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vprotb %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotb %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; X32-SSE-NEXT:    pxor %xmm2, %xmm2
 ; X32-SSE-NEXT:    psubb %xmm1, %xmm2
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; X32-SSE-NEXT:    psubb %xmm2, %xmm3
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1

llvm/test/CodeGen/X86/vector-fshr-rot-256.ll

 ; AVX512VLBW-LABEL: splatvar_funnnel_v4i64:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpbroadcastq %xmm1, %ymm1
 ; AVX512VLBW-NEXT:    vprorvq %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vmovddup {{.*#+}} xmm1 = xmm1[0,0]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vprotq %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vprotq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v4i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %ymm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsubq %ymm1, %ymm2, %ymm1
+; XOPAVX2-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %ymm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
 ; XOPAVX2-NEXT:    vprotq %xmm2, %xmm3, %xmm2
 ; XOPAVX2-NEXT:    vprotq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <4 x i64> %amt, <4 x i64> undef, <4 x i32> zeroinitializer
   %res = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %x, <4 x i64> %x, <4 x i64> %splat)
   ret <4 x i64> %res
 }
 
 define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpslld %xmm2, %xmm3, %xmm4
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [32,32,32,32]
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm5, %xmm1
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm3, %xmm3
 ; AVX1-NEXT:    vpor %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpslld %xmm2, %xmm0, %xmm2
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm1, %ymm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %ymm1
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [31,31,31,31]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpslld %xmm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [32,32,32,32]
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpsrld %xmm1, %ymm0, %ymm0

 ; AVX512VLBW-LABEL: splatvar_funnnel_v8i32:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpbroadcastd %xmm1, %ymm1
 ; AVX512VLBW-NEXT:    vprorvd %ymm1, %ymm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vprotd %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v8i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %ymm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsubd %ymm1, %ymm2, %ymm1
+; XOPAVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %ymm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
 ; XOPAVX2-NEXT:    vprotd %xmm2, %xmm3, %xmm2
 ; XOPAVX2-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <8 x i32> %amt, <8 x i32> undef, <8 x i32> zeroinitializer
   %res = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %x, <8 x i32> %x, <8 x i32> %splat)
   ret <8 x i32> %res
 }
 
 define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm3, %xmm4
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
 ; AVX1-NEXT:    vpsubw %xmm1, %xmm5, %xmm1
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm3, %xmm3
 ; AVX1-NEXT:    vpor %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm1, %ymm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %ymm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_funnnel_v16i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastw %xmm1, %ymm1
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %ymm1
 ; AVX512-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsllw %xmm2, %ymm0, %ymm2
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX512-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vprotw %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vprotw %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v16i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %ymm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsubw %ymm1, %ymm2, %ymm1
+; XOPAVX2-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %ymm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
 ; XOPAVX2-NEXT:    vprotw %xmm2, %xmm3, %xmm2
 ; XOPAVX2-NEXT:    vprotw %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <16 x i16> %amt, <16 x i16> undef, <16 x i32> zeroinitializer
   %res = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %x, <16 x i16> %x, <16 x i16> %splat)
   ret <16 x i16> %res
 }
 
 define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %amt) nounwind {
 ; AVX1-LABEL: splatvar_funnnel_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm4, %xmm5
 ; AVX1-NEXT:    vpcmpeqd %xmm6, %xmm6, %xmm6
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm6, %xmm7
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm7, %xmm2
 ; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm5

 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm6, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_funnnel_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
 ; AVX2-NEXT:    vpcmpeqd %ymm4, %ymm4, %ymm4
 ; AVX2-NEXT:    vpsllw %xmm2, %ymm4, %ymm2
 ; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX2-NEXT:    vpand %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrlw %xmm1, %ymm4, %ymm1
 ; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpcmpeqd %ymm4, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpsllw %xmm2, %ymm4, %ymm2
 ; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512F-NEXT:    vpand %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw %xmm1, %ymm4, %ymm1
 ; AVX512F-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; AVX512VL-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpcmpeqd %ymm4, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpsllw %xmm2, %ymm4, %ymm2
 ; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm2, %ymm3, %ymm2
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]

 ; AVX512VLBW-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vprotb %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vprotb %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsubb %ymm1, %ymm2, %ymm1
+; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
 ; XOPAVX2-NEXT:    vprotb %xmm2, %xmm3, %xmm2
 ; XOPAVX2-NEXT:    vprotb %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %splat = shufflevector <32 x i8> %amt, <32 x i8> undef, <32 x i32> zeroinitializer
   %res = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %x, <32 x i8> %x, <32 x i8> %splat)

llvm/test/CodeGen/X86/vector-fshr-rot-512.ll

; AVX512-NEXT:    retq
   %splat = shufflevector <16 x i32> %amt, <16 x i32> undef, <16 x i32> zeroinitializer
   %res = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %x, <16 x i32> %x, <16 x i32> %splat)
   ret <16 x i32> %res
 }
 
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
 ; AVX512F-NEXT:    vpsubw %xmm2, %xmm5, %xmm2
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm0, %ymm4, %ymm0
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm1, %ymm3
 ; AVX512F-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
 ; AVX512VL-NEXT:    vpsubw %xmm2, %xmm5, %xmm2
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlw %xmm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm0, %ymm4, %ymm0
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm1, %ymm3
; AVX512VLBW-NEXT:    retq
   %splat = shufflevector <32 x i16> %amt, <32 x i16> undef, <32 x i32> zeroinitializer
   %res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %x, <32 x i16> %splat)
   ret <32 x i16> %res
 }
 
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX512F-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm5, %ymm6
 ; AVX512F-NEXT:    vpbroadcastb %xmm6, %ymm6
 ; AVX512F-NEXT:    vpand %ymm6, %ymm4, %ymm4
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm2, %xmm7, %xmm2
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsrlw %xmm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw %xmm2, %ymm5, %ymm5
 ; AVX512F-NEXT:    vpsrlw $8, %ymm5, %ymm5
 ; AVX512F-NEXT:    vpbroadcastb %xmm5, %ymm5
 ; AVX512F-NEXT:    vpand %ymm5, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm0, %ymm4, %ymm0
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm1, %ymm3
 ; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpsrlw %xmm2, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpand %ymm5, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
 ; AVX512VL-NEXT:    vpcmpeqd %ymm5, %ymm5, %ymm5
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm5, %ymm6
 ; AVX512VL-NEXT:    vpbroadcastb %xmm6, %ymm6
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm7 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512VL-NEXT:    vpsubb %xmm2, %xmm7, %xmm2

llvm/test/CodeGen/X86/vector-reduce-umax-widen.ll

 
 ;
 ; vXi16
 ;
 
 define i16 @test_v2i16(<2 x i16> %a0) {
 ; SSE2-LABEL: test_v2i16:
 ; SSE2:       # %bb.0:
+; SSE2-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i16:
 ; SSE41:       # %bb.0:

 ; SSE2-LABEL: test_v4i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v4i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i16:
 ; SSE41:       # %bb.0:

llvm/test/CodeGen/X86/vector-reduce-umax.ll

 
 ;
 ; vXi32
 ;
 
 define i32 @test_v2i32(<2 x i32> %a0) {
 ; SSE2-LABEL: test_v2i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [4294967295,0,4294967295,0]
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    pxor %xmm2, %xmm3
 ; SSE2-NEXT:    pxor %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-NEXT:    pcmpgtd %xmm3, %xmm4
+; SSE2-NEXT:    movdqa %xmm3, %xmm4
+; SSE2-NEXT:    pcmpgtd %xmm2, %xmm4
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
 ; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; SSE2-NEXT:    pand %xmm5, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
 ; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm3
-; SSE2-NEXT:    por %xmm1, %xmm3
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pandn %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm0, %xmm3
 ; SSE2-NEXT:    movd %xmm3, %eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor %xmm1, %xmm1
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
 ; SSE41-NEXT:    movdqa %xmm1, %xmm3
 ; SSE41-NEXT:    pxor %xmm0, %xmm3
 ; SSE41-NEXT:    pxor %xmm2, %xmm0
 ; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm4
-; SSE41-NEXT:    pcmpgtd %xmm0, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,0,2,2]
-; SSE41-NEXT:    pand %xmm4, %xmm0
-; SSE41-NEXT:    por %xmm3, %xmm0
+; SSE41-NEXT:    pcmpgtd %xmm0, %xmm4
+; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
+; SSE41-NEXT:    pcmpeqd %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE41-NEXT:    pand %xmm5, %xmm0
+; SSE41-NEXT:    por %xmm4, %xmm0
 ; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm2
 ; SSE41-NEXT:    movd %xmm2, %eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v2i32:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm2, %xmm1
-; AVX1-NEXT:    vblendvpd %xmm1, %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
+; AVX1-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v2i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX2-NEXT:    vpcmpgtq %xmm0, %xmm2, %xmm1
-; AVX2-NEXT:    vblendvpd %xmm1, %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX2-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
+; AVX2-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v2i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpblendd {{.*#+}} xmm2 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX512BW-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX512BW-NEXT:    vpmaxuq %zmm0, %zmm2, %zmm0
+; AVX512BW-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512BW-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v2i32:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX512VL-NEXT:    vpmaxuq %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512VL-NEXT:    vpmaxuq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
 ; AVX512VL-NEXT:    retq
   %1 = call i32 @llvm.experimental.vector.reduce.umax.i32.v2i32(<2 x i32> %a0)
   ret i32 %1
 }
 
 define i32 @test_v4i32(<4 x i32> %a0) {
 ; SSE2-LABEL: test_v4i32:

 
 ;
 ; vXi16
 ;
 
 define i16 @test_v2i16(<2 x i16> %a0) {
 ; SSE2-LABEL: test_v2i16:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [65535,0,0,0,65535,0,0,0]
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    por %xmm0, %xmm2
+; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm3, %xmm4
 ; SSE2-NEXT:    pcmpgtd %xmm2, %xmm4
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
 ; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; SSE2-NEXT:    pand %xmm5, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
 ; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm3
-; SSE2-NEXT:    por %xmm1, %xmm3
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pandn %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm0, %xmm3
 ; SSE2-NEXT:    movd %xmm3, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor %xmm1, %xmm1
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm2 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
 ; SSE41-NEXT:    movdqa %xmm1, %xmm3
 ; SSE41-NEXT:    por %xmm0, %xmm3
 ; SSE41-NEXT:    por %xmm2, %xmm0
 ; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm4
-; SSE41-NEXT:    pcmpgtd %xmm0, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,0,2,2]
-; SSE41-NEXT:    pand %xmm4, %xmm0
-; SSE41-NEXT:    por %xmm3, %xmm0
+; SSE41-NEXT:    pcmpgtd %xmm0, %xmm4
+; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
+; SSE41-NEXT:    pcmpeqd %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE41-NEXT:    pand %xmm5, %xmm0
+; SSE41-NEXT:    por %xmm4, %xmm0
 ; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm2
 ; SSE41-NEXT:    movd %xmm2, %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i16:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm2
-; AVX-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
+; AVX-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v2i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX512BW-NEXT:    vpmaxuq %zmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
+; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512BW-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v2i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX512VL-NEXT:    vpmaxuq %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
+; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512VL-NEXT:    vpmaxuq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
 ; AVX512VL-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v2i16(<2 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v4i16(<4 x i16> %a0) {
 ; SSE2-LABEL: test_v4i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [65535,0,65535,0,65535,0,65535,0]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pand %xmm1, %xmm0
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    pcmpgtd %xmm2, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm0
 ; SSE2-NEXT:    pandn %xmm2, %xmm3
 ; SSE2-NEXT:    por %xmm0, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    pcmpgtd %xmm0, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movd %xmm2, %eax
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,1,1]
+; SSE2-NEXT:    movdqa %xmm3, %xmm1
+; SSE2-NEXT:    pcmpgtd %xmm0, %xmm1
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    pandn %xmm0, %xmm1
+; SSE2-NEXT:    por %xmm3, %xmm1
+; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v4i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor %xmm1, %xmm1
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0],xmm1[1],xmm2[2],xmm1[3,4,5,6,7]
 ; SSE41-NEXT:    pmaxud %xmm0, %xmm2
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4],xmm1[5],xmm2[6],xmm1[7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; SSE41-NEXT:    pmaxud %xmm1, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4],xmm1[5],xmm2[6],xmm1[7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
+; SSE41-NEXT:    pmaxud %xmm2, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i16:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendw {{.*#+}} xmm2 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3,4,5,6,7]
 ; AVX-NEXT:    vpmaxud %xmm0, %xmm2, %xmm0
-; AVX-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX-NEXT:    vpmaxud %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX-NEXT:    vpmaxud %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpblendw {{.*#+}} xmm2 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3,4,5,6,7]
 ; AVX512-NEXT:    vpmaxud %xmm0, %xmm2, %xmm0
-; AVX512-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX512-NEXT:    vpmaxud %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX512-NEXT:    vpmaxud %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v4i16(<4 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v8i16(<8 x i16> %a0) {
 ; SSE2-LABEL: test_v8i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i16:
 ; SSE41:       # %bb.0:

 
 ;
 ; vXi8
 ;
 
 define i8 @test_v2i8(<2 x i8> %a0) {
 ; SSE2-LABEL: test_v2i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    por %xmm0, %xmm2
+; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm3, %xmm4
 ; SSE2-NEXT:    pcmpgtd %xmm2, %xmm4
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
 ; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; SSE2-NEXT:    pand %xmm5, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
 ; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm3
-; SSE2-NEXT:    por %xmm1, %xmm3
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pandn %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm0, %xmm3
 ; SSE2-NEXT:    movd %xmm3, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm2 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [2147483648,2147483648]
 ; SSE41-NEXT:    movdqa %xmm1, %xmm4
 ; SSE41-NEXT:    por %xmm3, %xmm4
 ; SSE41-NEXT:    por %xmm2, %xmm3
 ; SSE41-NEXT:    movdqa %xmm4, %xmm0
-; SSE41-NEXT:    pcmpeqd %xmm3, %xmm0
-; SSE41-NEXT:    pcmpgtd %xmm3, %xmm4
-; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,0,2,2]
-; SSE41-NEXT:    pand %xmm3, %xmm0
-; SSE41-NEXT:    por %xmm4, %xmm0
+; SSE41-NEXT:    pcmpgtd %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[0,0,2,2]
+; SSE41-NEXT:    pcmpeqd %xmm4, %xmm3
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
+; SSE41-NEXT:    pand %xmm5, %xmm3
+; SSE41-NEXT:    por %xmm3, %xmm0
 ; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm2
 ; SSE41-NEXT:    pextrb $0, %xmm2, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm2
-; AVX-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
+; AVX-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v2i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512BW-NEXT:    vpmaxuq %zmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512BW-NEXT:    vpmaxuq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v2i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpmaxuq %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512VL-NEXT:    vpmaxuq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
 ; SSE2-LABEL: test_v4i8:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pand %xmm1, %xmm0
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    pcmpgtd %xmm2, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm0
 ; SSE2-NEXT:    pandn %xmm2, %xmm3
 ; SSE2-NEXT:    por %xmm0, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-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]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    pcmpgtd %xmm0, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movd %xmm2, %eax
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,1,1]
+; SSE2-NEXT:    movdqa %xmm3, %xmm1
+; SSE2-NEXT:    pcmpgtd %xmm0, %xmm1
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    pandn %xmm0, %xmm1
+; SSE2-NEXT:    por %xmm3, %xmm1
+; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v4i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[8],zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    pmaxud %xmm2, %xmm0
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    pmaxud %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    pand %xmm1, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; SSE41-NEXT:    pmaxud %xmm0, %xmm1
+; SSE41-NEXT:    pextrb $0, %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v4i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm1 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vandps %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpmaxud %xmm0, %xmm2, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpmaxud %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX1-NEXT:    vpmaxud %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v4i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [255,255,255,255]
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpmaxud %xmm0, %xmm2, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpmaxud %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX2-NEXT:    vpmaxud %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [255,255,255,255]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512-NEXT:    vpmaxud %xmm0, %xmm2, %xmm0
-; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512-NEXT:    vpmaxud %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX512-NEXT:    vpmaxud %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
 ; SSE2-LABEL: test_v8i8:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pand %xmm1, %xmm0
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
 ; SSE2-NEXT:    pand %xmm1, %xmm2
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm2, %xmm0
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-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]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    pand %xmm1, %xmm0
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
+; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    pmaxuw %xmm2, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    pmaxuw %xmm2, %xmm0
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm0
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    pmaxuw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    pand %xmm1, %xmm0
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE41-NEXT:    pmaxuw %xmm0, %xmm1
+; SSE41-NEXT:    pextrb $0, %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v8i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpmaxuw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpmaxuw %xmm0, %xmm2, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm0
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpmaxuw %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v8i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastw {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpmaxuw %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpmaxuw %xmm0, %xmm2, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm0
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpmaxuw %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
+; AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastw {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512-NEXT:    vpmaxuw %xmm0, %xmm2, %xmm0
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512-NEXT:    vpmaxuw %xmm0, %xmm2, %xmm0
-; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm0
-; AVX512-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512-NEXT:    vpmaxuw %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
+; AVX512-NEXT:    vpbroadcastd %xmm1, %xmm1
+; AVX512-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {

llvm/test/CodeGen/X86/vector-reduce-umin-widen.ll

 
 ;
 ; vXi16
 ;
 
 define i16 @test_v2i16(<2 x i16> %a0) {
 ; SSE2-LABEL: test_v2i16:
 ; SSE2:       # %bb.0:
+; SSE2-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE2-NEXT:    pxor %xmm2, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i16:
 ; SSE41:       # %bb.0:

 ; SSE2-LABEL: test_v4i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v4i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i16:
 ; SSE41:       # %bb.0:

llvm/test/CodeGen/X86/vector-reduce-umin.ll

 
 ;
 ; vXi32
 ;
 
 define i32 @test_v2i32(<2 x i32> %a0) {
 ; SSE2-LABEL: test_v2i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [4294967295,0,4294967295,0]
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    pxor %xmm2, %xmm3
-; SSE2-NEXT:    pxor %xmm0, %xmm2
+; SSE2-NEXT:    pxor %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    pcmpgtd %xmm3, %xmm4
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
 ; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; SSE2-NEXT:    pand %xmm5, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
 ; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm3
-; SSE2-NEXT:    por %xmm1, %xmm3
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pandn %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm0, %xmm3
 ; SSE2-NEXT:    movd %xmm3, %eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor %xmm1, %xmm1
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [2147483648,2147483648]
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    pxor %xmm3, %xmm0
-; SSE41-NEXT:    pxor %xmm2, %xmm3
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm4
-; SSE41-NEXT:    pcmpgtd %xmm0, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,0,2,2]
-; SSE41-NEXT:    pand %xmm4, %xmm0
-; SSE41-NEXT:    por %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
+; SSE41-NEXT:    movdqa %xmm1, %xmm3
+; SSE41-NEXT:    pxor %xmm0, %xmm3
+; SSE41-NEXT:    pxor %xmm2, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    pcmpgtd %xmm3, %xmm4
+; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
+; SSE41-NEXT:    pcmpeqd %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE41-NEXT:    pand %xmm5, %xmm0
+; SSE41-NEXT:    por %xmm4, %xmm0
 ; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm2
 ; SSE41-NEXT:    movd %xmm2, %eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v2i32:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm0, %xmm1
-; AVX1-NEXT:    vblendvpd %xmm1, %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX1-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm2
+; AVX1-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v2i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm2 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX2-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX2-NEXT:    vpcmpgtq %xmm2, %xmm0, %xmm1
-; AVX2-NEXT:    vblendvpd %xmm1, %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX2-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm2
+; AVX2-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v2i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpblendd {{.*#+}} xmm2 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX512BW-NEXT:    vpunpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX512BW-NEXT:    vpminuq %zmm0, %zmm2, %zmm0
+; AVX512BW-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512BW-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v2i32:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; AVX512VL-NEXT:    vpminuq %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512VL-NEXT:    vpminuq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
 ; AVX512VL-NEXT:    retq
   %1 = call i32 @llvm.experimental.vector.reduce.umin.i32.v2i32(<2 x i32> %a0)
   ret i32 %1
 }
 
 define i32 @test_v4i32(<4 x i32> %a0) {
 ; SSE2-LABEL: test_v4i32:

 
 ;
 ; vXi16
 ;
 
 define i16 @test_v2i16(<2 x i16> %a0) {
 ; SSE2-LABEL: test_v2i16:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [65535,0,0,0,65535,0,0,0]
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    por %xmm0, %xmm2
+; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    pcmpgtd %xmm3, %xmm4
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
 ; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; SSE2-NEXT:    pand %xmm5, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
 ; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm3
-; SSE2-NEXT:    por %xmm1, %xmm3
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pandn %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm0, %xmm3
 ; SSE2-NEXT:    movd %xmm3, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor %xmm1, %xmm1
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm2 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [2147483648,2147483648]
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    por %xmm3, %xmm0
-; SSE41-NEXT:    por %xmm2, %xmm3
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm4
-; SSE41-NEXT:    pcmpgtd %xmm0, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,0,2,2]
-; SSE41-NEXT:    pand %xmm4, %xmm0
-; SSE41-NEXT:    por %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
+; SSE41-NEXT:    movdqa %xmm1, %xmm3
+; SSE41-NEXT:    por %xmm0, %xmm3
+; SSE41-NEXT:    por %xmm2, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    pcmpgtd %xmm3, %xmm4
+; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
+; SSE41-NEXT:    pcmpeqd %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE41-NEXT:    pand %xmm5, %xmm0
+; SSE41-NEXT:    por %xmm4, %xmm0
 ; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm2
 ; SSE41-NEXT:    movd %xmm2, %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i16:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm2
+; AVX-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v2i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX512BW-NEXT:    vpminuq %zmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
+; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512BW-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v2i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX512VL-NEXT:    vpminuq %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
+; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512VL-NEXT:    vpminuq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
 ; AVX512VL-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umin.i16.v2i16(<2 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v4i16(<4 x i16> %a0) {
 ; SSE2-LABEL: test_v4i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [65535,0,65535,0,65535,0,65535,0]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pand %xmm1, %xmm0
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm3
 ; SSE2-NEXT:    pcmpgtd %xmm0, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm0
 ; SSE2-NEXT:    pandn %xmm2, %xmm3
 ; SSE2-NEXT:    por %xmm0, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pcmpgtd %xmm1, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movd %xmm2, %eax
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,1,1]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    pcmpgtd %xmm3, %xmm1
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    pandn %xmm0, %xmm1
+; SSE2-NEXT:    por %xmm3, %xmm1
+; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v4i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor %xmm1, %xmm1
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0],xmm1[1],xmm2[2],xmm1[3,4,5,6,7]
 ; SSE41-NEXT:    pminud %xmm0, %xmm2
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4],xmm1[5],xmm2[6],xmm1[7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
-; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; SSE41-NEXT:    pminud %xmm1, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0],xmm1[1],xmm2[2],xmm1[3],xmm2[4],xmm1[5],xmm2[6],xmm1[7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,1,1]
+; SSE41-NEXT:    pminud %xmm2, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i16:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendw {{.*#+}} xmm2 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3,4,5,6,7]
 ; AVX-NEXT:    vpminud %xmm0, %xmm2, %xmm0
-; AVX-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX-NEXT:    vpminud %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX-NEXT:    vpminud %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX512-NEXT:    vpblendw {{.*#+}} xmm2 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
 ; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; AVX512-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3,4,5,6,7]
 ; AVX512-NEXT:    vpminud %xmm0, %xmm2, %xmm0
-; AVX512-NEXT:    vpblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX512-NEXT:    vpminud %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3],xmm0[4],xmm1[5],xmm0[6],xmm1[7]
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX512-NEXT:    vpminud %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umin.i16.v4i16(<4 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v8i16(<8 x i16> %a0) {
 ; SSE2-LABEL: test_v8i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:

 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i16:
 ; SSE41:       # %bb.0:

 
 ;
 ; vXi8
 ;
 
 define i8 @test_v2i8(<2 x i8> %a0) {
 ; SSE2-LABEL: test_v2i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648]
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    por %xmm0, %xmm2
+; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    pcmpgtd %xmm3, %xmm4
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
 ; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
 ; SSE2-NEXT:    pand %xmm5, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[1,1,3,3]
 ; SSE2-NEXT:    por %xmm2, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm3
-; SSE2-NEXT:    por %xmm1, %xmm3
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    pandn %xmm1, %xmm3
+; SSE2-NEXT:    por %xmm0, %xmm3
 ; SSE2-NEXT:    movd %xmm3, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm2 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [2147483648,2147483648]
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    por %xmm3, %xmm0
-; SSE41-NEXT:    por %xmm2, %xmm3
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm4
-; SSE41-NEXT:    pcmpgtd %xmm0, %xmm3
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,0,2,2]
-; SSE41-NEXT:    pand %xmm4, %xmm0
-; SSE41-NEXT:    por %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]
+; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [2147483648,2147483648]
+; SSE41-NEXT:    movdqa %xmm1, %xmm3
+; SSE41-NEXT:    por %xmm0, %xmm3
+; SSE41-NEXT:    por %xmm2, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    pcmpgtd %xmm3, %xmm4
+; SSE41-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,2,2]
+; SSE41-NEXT:    pcmpeqd %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
+; SSE41-NEXT:    pand %xmm5, %xmm0
+; SSE41-NEXT:    por %xmm4, %xmm0
 ; SSE41-NEXT:    blendvpd %xmm0, %xmm1, %xmm2
 ; SSE41-NEXT:    pextrb $0, %xmm2, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vblendvpd %xmm2, %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX-NEXT:    vpcmpgtq %xmm0, %xmm1, %xmm2
+; AVX-NEXT:    vblendvpd %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v2i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512BW-NEXT:    vpminuq %zmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512BW-NEXT:    vpminuq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v2i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpminuq %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX512VL-NEXT:    vpminuq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umin.i8.v2i8(<2 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v4i8(<4 x i8> %a0) {
 ; SSE2-LABEL: test_v4i8:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pand %xmm1, %xmm0
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm3
 ; SSE2-NEXT:    pcmpgtd %xmm0, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm0
 ; SSE2-NEXT:    pandn %xmm2, %xmm3
 ; SSE2-NEXT:    por %xmm0, %xmm3
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,3]
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-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]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pcmpgtd %xmm1, %xmm2
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pandn %xmm0, %xmm2
-; SSE2-NEXT:    por %xmm1, %xmm2
-; SSE2-NEXT:    movd %xmm2, %eax
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,1,1]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    pcmpgtd %xmm3, %xmm1
+; SSE2-NEXT:    pand %xmm1, %xmm3
+; SSE2-NEXT:    pandn %xmm0, %xmm1
+; SSE2-NEXT:    por %xmm3, %xmm1
+; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v4i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[8],zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    pminud %xmm2, %xmm0
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    pminud %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    pand %xmm1, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; SSE41-NEXT:    pminud %xmm0, %xmm1
+; SSE41-NEXT:    pextrb $0, %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v4i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm1 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
 ; AVX1-NEXT:    vandps %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpminud %xmm0, %xmm2, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpminud %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX1-NEXT:    vpminud %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v4i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [255,255,255,255]
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpminud %xmm0, %xmm2, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpminud %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX2-NEXT:    vpminud %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v4i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm1 = [255,255,255,255]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512-NEXT:    vpminud %xmm0, %xmm2, %xmm0
-; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512-NEXT:    vpminud %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX512-NEXT:    vpminud %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umin.i8.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
 ; SSE2-LABEL: test_v8i8:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pand %xmm1, %xmm0
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    pminsw %xmm0, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
 ; SSE2-NEXT:    pand %xmm1, %xmm2
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE2-NEXT:    pminsw %xmm2, %xmm0
-; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    psrld $16, %xmm0
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-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]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE2-NEXT:    pminsw %xmm1, %xmm0
-; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    pand %xmm1, %xmm0
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE2-NEXT:    pminsw %xmm0, %xmm1
+; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    pminuw %xmm2, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    pminuw %xmm2, %xmm0
-; SSE41-NEXT:    pand %xmm0, %xmm1
-; SSE41-NEXT:    psrld $16, %xmm0
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    pminuw %xmm1, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    pand %xmm1, %xmm0
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE41-NEXT:    pminuw %xmm0, %xmm1
+; SSE41-NEXT:    pextrb $0, %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v8i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpminuw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpminuw %xmm0, %xmm2, %xmm0
-; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm0
-; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpminuw %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; AVX1-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v8i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastw {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpminuw %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpminuw %xmm0, %xmm2, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm0
-; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpminuw %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
+; AVX2-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpbroadcastw {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512-NEXT:    vpminuw %xmm0, %xmm2, %xmm0
 ; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4],zero,xmm0[6],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512-NEXT:    vpminuw %xmm0, %xmm2, %xmm0
-; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm1
-; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm0
-; AVX512-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512-NEXT:    vpminuw %xmm0, %xmm1, %xmm0
+; AVX512-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm0[1,1,2,3,4,5,6,7]
+; AVX512-NEXT:    vpbroadcastd %xmm1, %xmm1
+; AVX512-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umin.i8.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {

llvm/test/CodeGen/X86/vector-rotate-128.ll

 ; SSE2-NEXT:    subl %eax, %ecx
 ; SSE2-NEXT:    movd %ecx, %xmm1
 ; SSE2-NEXT:    psrld %xmm1, %xmm0
 ; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_rotate_v4i32:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
 ; SSE41-NEXT:    pslld %xmm2, %xmm3
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [32,32,32,32]
 ; SSE41-NEXT:    psubd %xmm1, %xmm2
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero
 ; SSE41-NEXT:    psrld %xmm1, %xmm0
 ; SSE41-NEXT:    por %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_rotate_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vpslld %xmm2, %xmm0, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [32,32,32,32]
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_rotate_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [31,31,31,31]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpslld %xmm2, %xmm0, %xmm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [32,32,32,32]
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq
; X32-SSE-NEXT:    retl
   %lshr = lshr <4 x i32> %a, %splat32
   %or = or <4 x i32> %shl, %lshr
   ret <4 x i32> %or
 }
 
 define <8 x i16> @splatvar_rotate_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
 ; SSE2-LABEL: splatvar_rotate_v8i16:
 ; SSE2:       # %bb.0:
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
 ; SSE2-NEXT:    psubw %xmm1, %xmm2
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    psllw %xmm1, %xmm3
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE2-NEXT:    psrlw %xmm2, %xmm0
 ; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_rotate_v8i16:
 ; SSE41:       # %bb.0:
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3
 ; SSE41-NEXT:    psllw %xmm2, %xmm3
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
 ; SSE41-NEXT:    psubw %xmm1, %xmm2
 ; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm1 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; SSE41-NEXT:    psrlw %xmm1, %xmm0
 ; SSE41-NEXT:    por %xmm3, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_rotate_v8i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX1-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_rotate_v8i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_rotate_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX512-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_rotate_v8i16:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; XOPAVX1-NEXT:    vprotw %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_rotate_v8i16:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotw %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_rotate_v8i16:
 ; X32-SSE:       # %bb.0:
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [16,16,16,16,16,16,16,16]
 ; X32-SSE-NEXT:    psubw %xmm1, %xmm2
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
 ; X32-SSE-NEXT:    psllw %xmm1, %xmm3
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0,1]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; X32-SSE-NEXT:    psrlw %xmm2, %xmm0
 ; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <8 x i16> %b, <8 x i16> undef, <8 x i32> zeroinitializer
   %splat16 = sub <8 x i16> <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>, %splat
   %shl = shl <8 x i16> %a, %splat
   %lshr = lshr <8 x i16> %a, %splat16
   %or = or <8 x i16> %shl, %lshr
   ret <8 x i16> %or
 }
 
 define <16 x i8> @splatvar_rotate_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
 ; SSE2-LABEL: splatvar_rotate_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; SSE2-NEXT:    psubb %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm1, %xmm3
 ; SSE2-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
 ; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psllw %xmm3, %xmm1

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE2-NEXT:    pand %xmm0, %xmm2
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_rotate_v16i8:
 ; SSE41:       # %bb.0:
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    pxor %xmm3, %xmm3
 ; SSE41-NEXT:    pshufb %xmm3, %xmm1
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm4 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    psllw %xmm4, %xmm2
 ; SSE41-NEXT:    pcmpeqd %xmm5, %xmm5
 ; SSE41-NEXT:    pcmpeqd %xmm6, %xmm6
 ; SSE41-NEXT:    psllw %xmm4, %xmm6
 ; SSE41-NEXT:    pshufb %xmm3, %xmm6
 ; SSE41-NEXT:    pand %xmm6, %xmm2
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; SSE41-NEXT:    psubb %xmm1, %xmm3
 ; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm1 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    psrlw %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw %xmm1, %xmm5
 ; SSE41-NEXT:    pshufb {{.*#+}} xmm5 = xmm5[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
 ; SSE41-NEXT:    pand %xmm0, %xmm5
 ; SSE41-NEXT:    por %xmm5, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_rotate_v16i8:
 ; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm0, %xmm4
 ; AVX1-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX1-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpand %xmm2, %xmm4, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX1-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm5, %xmm1
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_rotate_v16i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm0, %xmm3
 ; AVX2-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
 ; AVX2-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX2-NEXT:    vpand %xmm2, %xmm3, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm3, %xmm1

 ; XOPAVX2-LABEL: splatvar_rotate_v16i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotb %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_rotate_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; X32-SSE-NEXT:    psubb %xmm1, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
 ; X32-SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0]
 ; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psllw %xmm3, %xmm1

llvm/test/CodeGen/X86/vector-shift-ashr-128.ll

 ; AVX-NEXT:    vpsrlq %xmm1, %xmm2, %xmm2
 ; AVX-NEXT:    vpsrlq %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpxor %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    vpsubq %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v2i64:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vpshaq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v2i64:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpshaq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v2i64:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512-NEXT:    vpsraq %xmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0

 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v16i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpshab %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v16i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpshab %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: splatvar_shift_v16i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero

llvm/test/CodeGen/X86/vector-shift-ashr-sub128-widen.ll

 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [32896,32896,32896,32896,32896,32896,32896,32896]
 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v8i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
 ; XOPAVX1-NEXT:    vpshab %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v8i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpshab %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: splatvar_shift_v8i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero

 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [32896,32896,32896,32896,32896,32896,32896,32896]
 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v4i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; XOPAVX1-NEXT:    vpshab %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v4i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpshab %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: splatvar_shift_v4i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero

 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [32896,32896,32896,32896,32896,32896,32896,32896]
 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: splatvar_shift_v2i8:
 ; XOP:       # %bb.0:
-; XOP-NEXT:    insertq {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7,u,u,u,u,u,u,u,u]
 ; XOP-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOP-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOP-NEXT:    insertq {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7,u,u,u,u,u,u,u,u]
 ; XOP-NEXT:    vpshab %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: splatvar_shift_v2i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovsxbd %xmm0, %zmm0
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero

llvm/test/CodeGen/X86/vector-shift-ashr-sub128.ll

 define <2 x i32> @splatvar_shift_v2i32(<2 x i32> %a, <2 x i32> %b) nounwind {
 ; SSE2-LABEL: splatvar_shift_v2i32:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    psllq $32, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,3,2,3]
 ; SSE2-NEXT:    psrad $31, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,1]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,0,1,1]
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    psrlq %xmm0, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
+; SSE2-NEXT:    psrlq %xmm3, %xmm2
+; SSE2-NEXT:    movsd {{.*#+}} xmm2 = xmm1[0],xmm2[1]
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [9223372036854775808,9223372036854775808]
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    psrlq %xmm0, %xmm3
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,0,1]
-; SSE2-NEXT:    psrlq %xmm4, %xmm1
-; SSE2-NEXT:    movsd {{.*#+}} xmm1 = xmm3[0],xmm1[1]
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    psrlq %xmm0, %xmm3
-; SSE2-NEXT:    psrlq %xmm4, %xmm2
-; SSE2-NEXT:    movsd {{.*#+}} xmm2 = xmm3[0],xmm2[1]
+; SSE2-NEXT:    movdqa %xmm1, %xmm4
+; SSE2-NEXT:    psrlq %xmm0, %xmm4
+; SSE2-NEXT:    psrlq %xmm3, %xmm1
+; SSE2-NEXT:    movsd {{.*#+}} xmm1 = xmm4[0],xmm1[1]
 ; SSE2-NEXT:    xorpd %xmm1, %xmm2
 ; SSE2-NEXT:    psubq %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_shift_v2i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2

 ; AVX1-NEXT:    vpsubq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v2i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpsllq $32, %xmm0, %xmm2
 ; AVX2-NEXT:    vpsrad $31, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3]
-; AVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; AVX2-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
 ; AVX2-NEXT:    vpsrlvq %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsubq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v2i32:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpsllq $32, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpshaq {{.*}}(%rip), %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
 ; XOPAVX1-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vpshaq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v2i32:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpsllq $32, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpshaq {{.*}}(%rip), %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; XOPAVX2-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
 ; XOPAVX2-NEXT:    vpshaq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v2i32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsllq $32, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsraq $32, %zmm0, %zmm0
-; AVX512-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; AVX512-NEXT:    vpsravq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v2i32:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllq $32, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsraq $32, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512VL-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; AVX512VL-NEXT:    vpsravq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v2i32:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    psllq $32, %xmm0
; X32-SSE-NEXT:    retl
   ret <2 x i32> %shift
 }
 
 define <4 x i16> @splatvar_shift_v4i16(<4 x i16> %a, <4 x i16> %b) nounwind {
 ; SSE2-LABEL: splatvar_shift_v4i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pslld $16, %xmm0
 ; SSE2-NEXT:    psrad $16, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[0,0,0,0]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm1[0,0,0,0,4,5,6,7]
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[2,3,3,3,4,5,6,7]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm3
 ; SSE2-NEXT:    psrad %xmm1, %xmm3
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,1,1,1,4,5,6,7]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psrad %xmm4, %xmm1
 ; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[2,3,3,3,4,5,6,7]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm4
 ; SSE2-NEXT:    psrad %xmm3, %xmm4
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,1,1,1,4,5,6,7]
 ; SSE2-NEXT:    psrad %xmm2, %xmm0
 ; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1]
 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,3]
 ; SSE2-NEXT:    movaps %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_shift_v4i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pslld $16, %xmm0
 ; SSE41-NEXT:    psrad $16, %xmm0
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0],xmm2[1],xmm3[2],xmm2[3],xmm3[4],xmm2[5],xmm3[6],xmm2[7]
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
+; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm2 = xmm1[2,3,3,3,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm0, %xmm3
+; SSE41-NEXT:    psrad %xmm2, %xmm3
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[2,3,3,3,4,5,6,7]
 ; SSE41-NEXT:    movdqa %xmm0, %xmm5
 ; SSE41-NEXT:    psrad %xmm4, %xmm5
-; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm0, %xmm6
-; SSE41-NEXT:    psrad %xmm4, %xmm6
-; SSE41-NEXT:    pblendw {{.*#+}} xmm6 = xmm5[0,1,2,3],xmm6[4,5,6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrad %xmm1, %xmm2
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm3[0,1,1,1,4,5,6,7]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm5 = xmm3[0,1,2,3],xmm5[4,5,6,7]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,1,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm0, %xmm3
+; SSE41-NEXT:    psrad %xmm1, %xmm3
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,1,1,1,4,5,6,7]
 ; SSE41-NEXT:    psrad %xmm1, %xmm0
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm6[2,3],xmm0[4,5],xmm6[6,7]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3],xmm0[4,5],xmm5[6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_shift_v4i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpslld $16, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrad $16, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm2[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsrad %xmm3, %xmm0, %xmm3
-; AVX1-NEXT:    vpsrlq $32, %xmm1, %xmm4
+; AVX1-NEXT:    vpsrlq $32, %xmm2, %xmm4
 ; AVX1-NEXT:    vpsrad %xmm4, %xmm0, %xmm4
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; AVX1-NEXT:    vpsrad %xmm2, %xmm0, %xmm2
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
-; AVX1-NEXT:    vpsrad %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX1-NEXT:    vpsrad %xmm1, %xmm0, %xmm1
+; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
+; AVX1-NEXT:    vpsrad %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v4i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpslld $16, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrad $16, %xmm0, %xmm0
-; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
-; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; AVX2-NEXT:    vpsravd %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v4i16:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpslld $16, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpsrad $16, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
+; XOPAVX1-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
 ; XOPAVX1-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vpshad %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v4i16:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpslld $16, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpsrad $16, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
-; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; XOPAVX2-NEXT:    vpsravd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v4i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpslld $16, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsrad $16, %xmm0, %xmm0
-; AVX512-NEXT:    vpbroadcastd %xmm1, %xmm1
-; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
+; AVX512-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; AVX512-NEXT:    vpsravd %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v4i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpslld $16, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrad $16, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpbroadcastd %xmm1, %xmm1
+; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; AVX512VL-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; AVX512VL-NEXT:    vpsravd %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v4i16:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    pslld $16, %xmm0
 ; X32-SSE-NEXT:    psrad $16, %xmm0
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[0,0,0,0]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm1[0,0,0,0,4,5,6,7]
+; X32-SSE-NEXT:    pxor %xmm1, %xmm1
+; X32-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
 ; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[2,3,3,3,4,5,6,7]
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
 ; X32-SSE-NEXT:    psrad %xmm1, %xmm3
 ; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,1,1,1,4,5,6,7]
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psrad %xmm4, %xmm1
 ; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]

 ;
 ; AVX2-LABEL: splatvar_shift_v2i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpsllq $48, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsrad $31, %xmm0, %xmm2
 ; AVX2-NEXT:    vpsrad $16, %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3]
-; AVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; AVX2-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [9223372036854775808,9223372036854775808]
 ; AVX2-NEXT:    vpsrlvq %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsubq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq

 ; XOPAVX1-NEXT:    vpshaq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v2i16:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpsllq $48, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpshaq {{.*}}(%rip), %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; XOPAVX2-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
 ; XOPAVX2-NEXT:    vpshaq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v2i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsllq $48, %xmm0, %xmm0
 ; AVX512-NEXT:    vpsraq $48, %zmm0, %zmm0
-; AVX512-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; AVX512-NEXT:    vpsravq %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v2i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllq $48, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsraq $48, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; AVX512VL-NEXT:    vpsravq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v2i16:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    psllq $48, %xmm0

llvm/test/CodeGen/X86/vector-shift-lshr-128.ll

 ; AVX2-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v16i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v16i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: splatvar_shift_v16i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero

llvm/test/CodeGen/X86/vector-shift-lshr-sub128-widen.ll

 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v8i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
 ; XOPAVX1-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v8i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: splatvar_shift_v8i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero

 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v4i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7]
 ; XOPAVX1-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v4i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: splatvar_shift_v4i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero

 ; AVX2-NEXT:    vpsrlw %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: splatvar_shift_v2i8:
 ; XOP:       # %bb.0:
-; XOP-NEXT:    insertq {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7,u,u,u,u,u,u,u,u]
 ; XOP-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOP-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOP-NEXT:    insertq {{.*#+}} xmm1 = xmm1[0,0,2,3,4,5,6,7,u,u,u,u,u,u,u,u]
 ; XOP-NEXT:    vpshlb %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: splatvar_shift_v2i8:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512DQ-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero

llvm/test/CodeGen/X86/vector-shift-lshr-sub128.ll

 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <2 x i32> @splatvar_shift_v2i32(<2 x i32> %a, <2 x i32> %b) nounwind {
 ; SSE2-LABEL: splatvar_shift_v2i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [4294967295,0,4294967295,0]
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
-; SSE2-NEXT:    pand %xmm2, %xmm1
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm2
 ; SSE2-NEXT:    psrlq %xmm1, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; SSE2-NEXT:    psrlq %xmm1, %xmm0
 ; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1]
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_shift_v2i32:

 ; AVX1-NEXT:    vpsrlq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v2i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3]
-; AVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; AVX2-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v2i32:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7]
 ; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
 ; XOPAVX1-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vpshlq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v2i32:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3]
-; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; XOPAVX2-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v2i32:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3]
-; AVX512-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX512-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; AVX512-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v2i32:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3]
-; AVX512VL-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512VL-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; AVX512VL-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v2i32:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [4294967295,0,4294967295,0]
 ; X32-SSE-NEXT:    pand %xmm2, %xmm0
 ; X32-SSE-NEXT:    pand %xmm1, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
 ; X32-SSE-NEXT:    psrlq %xmm2, %xmm3
 ; X32-SSE-NEXT:    pxor %xmm2, %xmm2
 ; X32-SSE-NEXT:    movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
 ; X32-SSE-NEXT:    psrlq %xmm2, %xmm0
 ; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <2 x i32> %b, <2 x i32> undef, <2 x i32> zeroinitializer
   %shift = lshr <2 x i32> %a, %splat
   ret <2 x i32> %shift
 }
 
 define <4 x i16> @splatvar_shift_v4i16(<4 x i16> %a, <4 x i16> %b) nounwind {
 ; SSE2-LABEL: splatvar_shift_v4i16:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [65535,0,65535,0,65535,0,65535,0]
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[0,0,0,0]
-; SSE2-NEXT:    pand %xmm2, %xmm3
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm3[2,3,3,3,4,5,6,7]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    psrld %xmm1, %xmm2
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[0,1,1,1,4,5,6,7]
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm1[0,0,0,0,4,5,6,7]
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[2,3,3,3,4,5,6,7]
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
+; SSE2-NEXT:    psrld %xmm1, %xmm3
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,1,1,1,4,5,6,7]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psrld %xmm4, %xmm1
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[2,3,0,1]
+; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[2,3,3,3,4,5,6,7]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm4
 ; SSE2-NEXT:    psrld %xmm3, %xmm4
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,1,1,1,4,5,6,7]
 ; SSE2-NEXT:    psrld %xmm2, %xmm0
 ; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1]
 ; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,3]
 ; SSE2-NEXT:    movaps %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_shift_v4i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0],xmm2[1],xmm3[2],xmm2[3],xmm3[4],xmm2[5],xmm3[6],xmm2[7]
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm0, %xmm5
+; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[0,0,0,0,4,5,6,7]
+; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[2,3,3,3,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    psrld %xmm1, %xmm3
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm1[2,3,3,3,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm2, %xmm5
 ; SSE41-NEXT:    psrld %xmm4, %xmm5
-; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[2,3,3,3,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm0, %xmm6
-; SSE41-NEXT:    psrld %xmm4, %xmm6
-; SSE41-NEXT:    pblendw {{.*#+}} xmm6 = xmm5[0,1,2,3],xmm6[4,5,6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrld %xmm1, %xmm2
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm3[0,1,1,1,4,5,6,7]
-; SSE41-NEXT:    psrld %xmm1, %xmm0
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm6[2,3],xmm0[4,5],xmm6[6,7]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm5 = xmm3[0,1,2,3],xmm5[4,5,6,7]
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,1,1,1,4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    psrld %xmm0, %xmm3
+; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[0,1,1,1,4,5,6,7]
+; SSE41-NEXT:    psrld %xmm0, %xmm2
+; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm5[2,3],xmm2[4,5],xmm5[6,7]
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_shift_v4i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrld %xmm3, %xmm0, %xmm3
-; AVX1-NEXT:    vpsrlq $32, %xmm1, %xmm4
+; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm4 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsrld %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm5
+; AVX1-NEXT:    vpsrld %xmm5, %xmm0, %xmm5
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm5[0,1,2,3],xmm4[4,5,6,7]
+; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm2 = xmm3[2],xmm2[2],xmm3[3],xmm2[3]
 ; AVX1-NEXT:    vpsrld %xmm2, %xmm0, %xmm2
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm4[2,3],xmm0[4,5],xmm4[6,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v4i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7]
-; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
-; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; AVX2-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v4i16:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7]
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
+; XOPAVX1-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; XOPAVX1-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vpshld %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v4i16:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7]
-; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
-; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; XOPAVX2-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v4i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7]
-; AVX512-NEXT:    vpbroadcastd %xmm1, %xmm1
-; AVX512-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
+; AVX512-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; AVX512-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v4i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7]
-; AVX512VL-NEXT:    vpbroadcastd %xmm1, %xmm1
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
+; AVX512VL-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; AVX512VL-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v4i16:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [65535,0,65535,0,65535,0,65535,0]
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[0,0,0,0]
-; X32-SSE-NEXT:    pand %xmm2, %xmm3
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm3[2,3,3,3,4,5,6,7]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
-; X32-SSE-NEXT:    psrld %xmm1, %xmm2
-; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[0,1,1,1,4,5,6,7]
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm1[0,0,0,0,4,5,6,7]
+; X32-SSE-NEXT:    pxor %xmm1, %xmm1
+; X32-SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[2,3,3,3,4,5,6,7]
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
+; X32-SSE-NEXT:    psrld %xmm1, %xmm3
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,1,1,1,4,5,6,7]
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psrld %xmm4, %xmm1
-; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[2,3,0,1]
+; X32-SSE-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm3[0]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
 ; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[2,3,3,3,4,5,6,7]
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
 ; X32-SSE-NEXT:    psrld %xmm3, %xmm4
 ; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,1,1,1,4,5,6,7]
 ; X32-SSE-NEXT:    psrld %xmm2, %xmm0
 ; X32-SSE-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm4[1]
 ; X32-SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,3]
 ; X32-SSE-NEXT:    movaps %xmm1, %xmm0

 ; AVX1-NEXT:    vpsrlq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v2i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7]
-; AVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; AVX2-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v2i16:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7]
 ; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; XOPAVX1-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vpshlq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v2i16:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7]
-; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; XOPAVX2-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v2i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7]
-; AVX512-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; AVX512-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v2i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7]
-; AVX512VL-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; AVX512VL-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v2i16:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [65535,0,0,0,65535,0,0,0]
 ; X32-SSE-NEXT:    pand %xmm2, %xmm0

 ; AVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $1, %xmm0, %xmm1
 ; AVX1-NEXT:    vpaddw %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v8i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero
-; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX2-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31]
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; XOPAVX1-LABEL: splatvar_shift_v8i8:
-; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero
-; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
-; XOPAVX1-NEXT:    vpshlw %xmm1, %xmm0, %xmm0
-; XOPAVX1-NEXT:    retq
-;
-; XOPAVX2-LABEL: splatvar_shift_v8i8:
-; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; XOPAVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; XOPAVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero
-; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
-; XOPAVX2-NEXT:    vpshlw %xmm1, %xmm0, %xmm0
-; XOPAVX2-NEXT:    retq
+; XOP-LABEL: splatvar_shift_v8i8:
+; XOP:       # %bb.0:
+; XOP-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; XOP-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero
+; XOP-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; XOP-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; XOP-NEXT:    vpshlw %xmm1, %xmm0, %xmm0
+; XOP-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: splatvar_shift_v8i8:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; AVX512DQ-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX512DQ-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero
-; AVX512DQ-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512DQ-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX512DQ-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512DQ-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_shift_v8i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; AVX512BW-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero
 ; AVX512BW-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQVL-LABEL: splatvar_shift_v8i8:
 ; AVX512DQVL:       # %bb.0:
-; AVX512DQVL-NEXT:    vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; AVX512DQVL-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX512DQVL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero
-; AVX512DQVL-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512DQVL-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX512DQVL-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX512DQVL-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
 ; AVX512DQVL-NEXT:    vpmovdw %ymm0, %xmm0
 ; AVX512DQVL-NEXT:    vzeroupper
 ; AVX512DQVL-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: splatvar_shift_v8i8:
 ; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vpbroadcastw {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; AVX512BWVL-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX512BWVL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero,xmm1[0],zero
 ; AVX512BWVL-NEXT:    vpsrlvw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v8i8:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]

 ; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,1,1,1,4,5,6,7]
 ; SSE41-NEXT:    psrld %xmm1, %xmm0
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7]
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3],xmm0[4,5],xmm5[6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_shift_v4i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vbroadcastss {{.*#+}} xmm2 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
-; AVX1-NEXT:    vandps %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero
 ; AVX1-NEXT:    vpsrldq {{.*#+}} xmm2 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsrld %xmm2, %xmm0, %xmm2
 ; AVX1-NEXT:    vpsrlq $32, %xmm1, %xmm3
 ; AVX1-NEXT:    vpsrld %xmm3, %xmm0, %xmm3
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
 ; AVX1-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm3 = xmm1[2],xmm3[2],xmm1[3],xmm3[3]
 ; AVX1-NEXT:    vpsrld %xmm3, %xmm0, %xmm3
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4,5,6,7]
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v4i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255]
-; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero
 ; AVX2-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v4i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vbroadcastss {{.*#+}} xmm2 = [3.57331108E-43,3.57331108E-43,3.57331108E-43,3.57331108E-43]
-; XOPAVX1-NEXT:    vandps %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero
-; XOPAVX1-NEXT:    vxorps %xmm2, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vpshld %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v4i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255]
-; XOPAVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero
 ; XOPAVX2-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v4i8:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255]
-; AVX512-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero
 ; AVX512-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v4i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [255,255,255,255]
-; AVX512VL-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero,xmm1[0],zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v4i8:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
 ; X32-SSE-NEXT:    pand %xmm2, %xmm0

 ; SSE41-NEXT:    psrlq %xmm1, %xmm2
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; SSE41-NEXT:    psrlq %xmm1, %xmm0
 ; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_shift_v2i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovddup {{.*#+}} xmm2 = [1.2598673968951787E-321,1.2598673968951787E-321]
-; AVX1-NEXT:    # xmm2 = mem[0,0]
-; AVX1-NEXT:    vandps %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vpsrlq %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; AVX1-NEXT:    vpsrlq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v2i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq {{.*#+}} xmm2 = [255,255]
-; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v2i8:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vmovddup {{.*#+}} xmm2 = [1.2598673968951787E-321,1.2598673968951787E-321]
-; XOPAVX1-NEXT:    # xmm2 = mem[0,0]
-; XOPAVX1-NEXT:    vandps %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero
-; XOPAVX1-NEXT:    vxorps %xmm2, %xmm2, %xmm2
+; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
 ; XOPAVX1-NEXT:    vpshlq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v2i8:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq {{.*#+}} xmm2 = [255,255]
-; XOPAVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero
 ; XOPAVX2-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v2i8:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastq {{.*#+}} xmm2 = [255,255]
-; AVX512-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero
 ; AVX512-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v2i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} xmm2 = [255,255]
-; AVX512VL-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsrlvq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v2i8:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
 ; X32-SSE-NEXT:    pand %xmm2, %xmm0

llvm/test/CodeGen/X86/vector-shift-shl-sub128.ll

 
 ;
 ; Uniform Variable Shifts
 ;
 
 define <2 x i32> @splatvar_shift_v2i32(<2 x i32> %a, <2 x i32> %b) nounwind {
 ; SSE2-LABEL: splatvar_shift_v2i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm2
 ; SSE2-NEXT:    psllq %xmm1, %xmm2
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; SSE2-NEXT:    psllq %xmm1, %xmm0
 ; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1]
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_shift_v2i32:

 ; AVX1-NEXT:    vpsllq %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; AVX1-NEXT:    vpsllq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v2i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; AVX2-NEXT:    vpsllvq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v2i32:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
 ; XOPAVX1-NEXT:    vpshlq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v2i32:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; XOPAVX2-NEXT:    vpsllvq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v2i32:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; AVX512-NEXT:    vpsllvq %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v2i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512VL-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; AVX512VL-NEXT:    vpsllvq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v2i32:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [4294967295,0,4294967295,0]

 ; AVX1-NEXT:    vpslld $23, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vcvttps2dq %xmm1, %xmm1
 ; AVX1-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v4i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
-; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; AVX2-NEXT:    vpsllvd %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v4i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; XOPAVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
+; XOPAVX1-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; XOPAVX1-NEXT:    vpshld %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v4i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
-; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; XOPAVX2-NEXT:    vpsllvd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v4i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastd %xmm1, %xmm1
-; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
+; AVX512-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
 ; AVX512-NEXT:    vpsllvd %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v4i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastd %xmm1, %xmm1
+; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7]
+; AVX512VL-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; AVX512VL-NEXT:    vpsllvd %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v4i16:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; X32-SSE-NEXT:    pslld $23, %xmm1
 ; X32-SSE-NEXT:    paddd {{\.LCPI.*}}, %xmm1

 ; AVX1-NEXT:    vpsllq %xmm1, %xmm0, %xmm2
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
 ; AVX1-NEXT:    vpsllq %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_shift_v2i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; AVX2-NEXT:    vpsllvq %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v2i16:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; XOPAVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; XOPAVX1-NEXT:    vpshlq %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: splatvar_shift_v2i16:
 ; XOPAVX2:       # %bb.0:
-; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; XOPAVX2-NEXT:    vpsllvq %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_shift_v2i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; AVX512-NEXT:    vpsllvq %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_shift_v2i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastq %xmm1, %xmm1
+; AVX512VL-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
 ; AVX512VL-NEXT:    vpsllvq %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_shift_v2i16:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]