[DAGCombiner] try to move splat after binop with splat constant

[spatel]

Summary

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

We do this in IR, and there's a similar fold for the case with 2 non-constant operands just above the code diff in this patch.

This was discussed in D79718, and the extra shuffle in the test (llvm/test/CodeGen/X86/vector-fshl-128.ll::sink_splatvar) where it was noticed disappears because demanded elements analysis is no longer blocked. The large majority of the test diffs seem to be benign code scheduling changes, but I do see another type of win: moving the splat later allows binop narrowing in some cases. I don't see any obvious regressions; those were avoided on x86 and ARM with the INSERT_VECTOR_ELT restriction.

[craig.topper]

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

853–854

Something weird happened here I think. We appear to be loading a constant as an immediate and moving it into a vector. Then we AND it with something that was just ANDed with a constant pool. Could the two ANDs using a single constant pool?

[spatel]

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

853–854

I think the root cause is that we don't have a combine for:
ZERO_EXTEND_VECTOR_INREG --> BITCAST
...if we know all of the high bits are already zero.

That's visible in existing tests - for example in the SSE41 or AVX1 output for this test, we don't need to pand+pmovzxwq.

I have a draft of that patch, and it causes a massive amount of test diffs...taking a look now.

[spatel]

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

853–854

Oh wait...I got my vector elements mixed up. This is just a bizarre build vector legalization:

Legalizing: t83: v8i16 = BUILD_VECTOR Constant:i16<-1>, Constant:i16<0>, Constant:i16<0>, Constant:i16<0>, undef:i16, undef:i16, undef:i16, undef:i16
Trying custom legalization
Creating constant: t84: i32 = Constant<65535>
Creating new node: t85: v4i32 = scalar_to_vector Constant:i32<65535>
Creating constant: t86: i32 = Constant<0>
Creating new node: t87: v4i32 = BUILD_VECTOR Constant:i32<0>, Constant:i32<0>, Constant:i32<0>, Constant:i32<0>
Creating new node: t88: v4i32 = vector_shuffle<4,1,2,3> t87, t85
Creating new node: t89: v8i16 = bitcast t88

[spatel]

Rebased - x86 tests diffs are altered after D80131 / rGfa038e03504c.

[spatel]

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

853–854

So we still end up with and-of-and or some variant of that here even with the updated build vector lowering.
I might've been staring at this too long, but I'm not sure how to solve it (and not sure if it's worth the effort).
Here's what I see happening:

1. turn the raw IR into a rotl
2. expand the rotl including masking of the shift amount
3. convert generic vector shifts to x86 shifts by splatted scalar amount
4. end up with 2 'and' with target constants of different element widths:

    t63: i64 = X86ISD::WrapperRIP TargetConstantPool:i64<<8 x i16> <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>> 0
  t61: v8i16,ch = load<(load 16 from constant-pool)> t0, t63, undef:i64
t48: v8i16 = and t4, t61
          t95: i64 = X86ISD::WrapperRIP TargetConstantPool:i64<<4 x i32> <i32 65535, i32 0, i32 0, i32 0>> 0
        t96: v8i16,ch = load<(load 16 from constant-pool)> t0, t95, undef:i64
      t80: v8i16 = and t48, t96

[craig.topper]

Ok lets go with this. LGTM

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

SDValue DAGCombiner::SimplifyVBinOp(SDNode *N) {
   assert(N->getValueType(0).isVector() &&
          "SimplifyVBinOp only works on vectors!");
 
   SDValue LHS = N->getOperand(0);
   SDValue RHS = N->getOperand(1);
   SDValue Ops[] = {LHS, RHS};
   EVT VT = N->getValueType(0);
   unsigned Opcode = N->getOpcode();
+  SDNodeFlags Flags = N->getFlags();
 
   // See if we can constant fold the vector operation.
   if (SDValue Fold = DAG.FoldConstantVectorArithmetic(
           Opcode, SDLoc(LHS), LHS.getValueType(), Ops, N->getFlags()))
     return Fold;
 
   // Move unary shuffles with identical masks after a vector binop:
   // VBinOp (shuffle A, Undef, Mask), (shuffle B, Undef, Mask))
   //   --> shuffle (VBinOp A, B), Undef, Mask
   // This does not require type legality checks because we are creating the
   // same types of operations that are in the original sequence. We do have to
   // restrict ops like integer div that have immediate UB (eg, div-by-zero)
   // though. This code is adapted from the identical transform in instcombine.
   if (Opcode != ISD::UDIV && Opcode != ISD::SDIV &&
       Opcode != ISD::UREM && Opcode != ISD::SREM &&
       Opcode != ISD::UDIVREM && Opcode != ISD::SDIVREM) {
     auto *Shuf0 = dyn_cast<ShuffleVectorSDNode>(LHS);
     auto *Shuf1 = dyn_cast<ShuffleVectorSDNode>(RHS);
     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());
+                                     RHS.getOperand(0), Flags);
       SDValue UndefV = LHS.getOperand(1);
       return DAG.getVectorShuffle(VT, DL, NewBinOp, UndefV, Shuf0->getMask());
     }
+
+    // Try to sink a splat shuffle after a binop with a uniform constant.
+    // This is limited to cases where neither the shuffle nor the constant have
+    // undefined elements because that could be poison-unsafe or inhibit
+    // demanded elements analysis. It is further limited to not change a splat
+    // of an inserted scalar because that may be optimized better by
+    // load-folding or other target-specific behaviors.
+    if (isConstOrConstSplat(RHS) && Shuf0 && is_splat(Shuf0->getMask()) &&
+        Shuf0->hasOneUse() && Shuf0->getOperand(1).isUndef() &&
+        Shuf0->getOperand(0).getOpcode() != ISD::INSERT_VECTOR_ELT) {
+      // binop (splat X), (splat C) --> splat (binop X, C)
+      SDLoc DL(N);
+      SDValue X = Shuf0->getOperand(0);
+      SDValue NewBinOp = DAG.getNode(Opcode, DL, VT, X, RHS, Flags);
+      return DAG.getVectorShuffle(VT, DL, NewBinOp, DAG.getUNDEF(VT),
+                                  Shuf0->getMask());
+    }
+    if (isConstOrConstSplat(LHS) && Shuf1 && is_splat(Shuf1->getMask()) &&
+        Shuf1->hasOneUse() && Shuf1->getOperand(1).isUndef() &&
+        Shuf1->getOperand(0).getOpcode() != ISD::INSERT_VECTOR_ELT) {
+      // binop (splat C), (splat X) --> splat (binop C, X)
+      SDLoc DL(N);
+      SDValue X = Shuf1->getOperand(0);
+      SDValue NewBinOp = DAG.getNode(Opcode, DL, VT, LHS, X, Flags);
+      return DAG.getVectorShuffle(VT, DL, NewBinOp, DAG.getUNDEF(VT),
+                                  Shuf1->getMask());
+    }
   }
 
   // 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/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:    movdqa {{.*#+}} xmm4 = [64,64]
 ; SSE2-NEXT:    psubq %xmm2, %xmm4
 ; SSE2-NEXT:    psrlq %xmm4, %xmm1
 ; SSE2-NEXT:    por %xmm3, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; 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:    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:    pshufd {{.*#+}} xmm1 = xmm2[0,1,0,1]
 ; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqq %xmm4, %xmm0
-; SSE41-NEXT:    blendvpd %xmm0, %xmm3, %xmm2
-; SSE41-NEXT:    movapd %xmm2, %xmm0
+; SSE41-NEXT:    pcmpeqq %xmm1, %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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; AVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX512F-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX512BW-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX512VBMI2-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpsrlq %xmm4, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    pshufd {{.*#+}} xmm3 = xmm2[0,1,0,1]
+; X32-SSE-NEXT:    pxor %xmm4, %xmm4
+; X32-SSE-NEXT:    pcmpeqd %xmm3, %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 %xmm3, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm3
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
+; X32-SSE-NEXT:    psrlq %xmm4, %xmm1
+; X32-SSE-NEXT:    movsd {{.*#+}} xmm1 = xmm3[0],xmm1[1]
 ; 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:    por %xmm1, %xmm3
+; X32-SSE-NEXT:    pand %xmm5, %xmm0
+; X32-SSE-NEXT:    pandn %xmm3, %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:    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:    pshufd {{.*#+}} xmm1 = xmm2[0,0,0,0]
 ; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqd %xmm4, %xmm0
-; SSE41-NEXT:    blendvps %xmm0, %xmm3, %xmm2
-; SSE41-NEXT:    movaps %xmm2, %xmm0
+; SSE41-NEXT:    pcmpeqd %xmm1, %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:    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:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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:    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:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} xmm4 = 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:    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:    pshufd {{.*#+}} xmm2 = xmm4[0,0,0,0]
+; 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:    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:    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:    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:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,0,0,0]
 ; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pcmpeqw %xmm4, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    pcmpeqw %xmm1, %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:    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:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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 {{.*#+}} xmm4 = 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:    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:    pshufd {{.*#+}} xmm2 = xmm4[0,0,0,0]
+; 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:    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:    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:    pshuflw {{.*#+}} xmm1 = xmm2[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    pcmpeqb %xmm1, %xmm2
+; SSE2-NEXT:    pand %xmm2, %xmm0
+; SSE2-NEXT:    pandn %xmm4, %xmm2
+; SSE2-NEXT:    por %xmm2, %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:    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:    pshufb %xmm0, %xmm2
 ; 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:    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:    vpshufb %xmm6, %xmm2, %xmm2
+; 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:    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:    vpbroadcastb %xmm2, %xmm2
 ; 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:    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:    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:    pshuflw {{.*#+}} xmm1 = xmm2[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:    pcmpeqb %xmm1, %xmm2
+; X32-SSE-NEXT:    pand %xmm2, %xmm0
+; X32-SSE-NEXT:    pandn %xmm4, %xmm2
+; X32-SSE-NEXT:    por %xmm2, %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
 }
 
 ; CGP should allow a cross-block splat shift amount to be seen in SDAG.
 ; PR37426 - https://bugs.llvm.org/show_bug.cgi?id=37426

 ; SSE2-NEXT:    movdqu %xmm2, 1024(%rdi,%rax)
 ; SSE2-NEXT:    addq $16, %rax
 ; SSE2-NEXT:    jne .LBB8_1
 ; SSE2-NEXT:  # %bb.2: # %end
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: sink_splatvar:
 ; SSE41:       # %bb.0: # %entry
-; SSE41-NEXT:    movd %esi, %xmm0
+; SSE41-NEXT:    movd %esi, %xmm1
 ; SSE41-NEXT:    movq $-1024, %rax # imm = 0xFC00
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,0,0,0]
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [32,32,32,32]
 ; SSE41-NEXT:    psubd %xmm1, %xmm0
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; SSE41-NEXT:    .p2align 4, 0x90
 ; SSE41-NEXT:  .LBB8_1: # %loop
 ; SSE41-NEXT:    # =>This Inner Loop Header: Depth=1
 ; SSE41-NEXT:    movdqu 1024(%rdi,%rax), %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm3
 ; SSE41-NEXT:    psrld %xmm0, %xmm3
 ; SSE41-NEXT:    pslld %xmm1, %xmm2
 ; SSE41-NEXT:    por %xmm3, %xmm2
 ; SSE41-NEXT:    movdqu %xmm2, 1024(%rdi,%rax)
 ; SSE41-NEXT:    addq $16, %rax
 ; SSE41-NEXT:    jne .LBB8_1
 ; SSE41-NEXT:  # %bb.2: # %end
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: sink_splatvar:
 ; AVX1:       # %bb.0: # %entry
 ; AVX1-NEXT:    vmovd %esi, %xmm0
 ; AVX1-NEXT:    movq $-1024, %rax # imm = 0xFC00
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm0 = [32,32,32,32]
 ; AVX1-NEXT:    vpsubd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    .p2align 4, 0x90
 ; AVX1-NEXT:  .LBB8_1: # %loop
 ; AVX1-NEXT:    # =>This Inner Loop Header: Depth=1

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:    vpermilps {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; 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:    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:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
+; 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:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpermilps {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; 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:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
+; 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:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsllq %xmm2, %ymm0, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    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:    vextractf128 $1, %ymm0, %xmm3
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; 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:    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, %ymm0, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    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, %ymm0, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VL-NEXT:    vpslld %xmm3, %ymm0, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    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, %ymm0, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    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, %ymm0, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VLBW-NEXT:    vpslld %xmm3, %ymm0, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    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:    vextractf128 $1, %ymm0, %xmm3
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; 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:    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, %ymm0, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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 %xmm5, %xmm8, %xmm3
-; AVX1-NEXT:    vpcmpeqb %xmm2, %xmm8, %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 %xmm5, %xmm5, %xmm5
 ; AVX2-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX2-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX2-NEXT:    vpand %ymm3, %ymm4, %ymm3
 ; 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, %xmm5, %xmm4
 ; AVX2-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX2-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX2-NEXT:    vpand %ymm4, %ymm1, %ymm1
 ; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; 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 %xmm5, %xmm5, %xmm5
 ; AVX512F-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX512F-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX512F-NEXT:    vpand %ymm3, %ymm4, %ymm3
 ; 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, %xmm5, %xmm4
 ; AVX512F-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512F-NEXT:    vpand %ymm4, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; 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 %xmm5, %xmm5, %xmm5
 ; AVX512VL-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX512VL-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm4, %ymm3
 ; 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, %xmm5, %xmm4
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; 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:    vextracti128 $1, %ymm2, %xmm4
-; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm3, %xmm3
+; XOPAVX2-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; XOPAVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
+; XOPAVX2-NEXT:    vpshlb %xmm2, %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
 ; XOPAVX2-NEXT:    vpsubb %xmm5, %xmm6, %xmm5
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm7

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 {{.*#+}} xmm3 = [63,63]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
 ; 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:    vpbroadcastq %xmm2, %zmm2
 ; 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:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
 ; 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:    vpbroadcastq %xmm2, %zmm2
 ; 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 {{.*#+}} xmm3 = [63,63]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
 ; 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:    vpbroadcastq %xmm2, %zmm2
 ; 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:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpsllq %xmm2, %zmm0, %zmm3
 ; 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:    vpbroadcastq %xmm2, %zmm2
 ; 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 {{.*#+}} 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, %zmm0, %zmm3
 ; 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:    vpbroadcastd %xmm2, %zmm2
 ; 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:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VL-NEXT:    vpslld %xmm3, %zmm0, %zmm3
 ; 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:    vpbroadcastd %xmm2, %zmm2
 ; 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 {{.*#+}} 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, %zmm0, %zmm3
 ; 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:    vpbroadcastd %xmm2, %zmm2
 ; 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:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VLBW-NEXT:    vpslld %xmm3, %zmm0, %zmm3
 ; 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:    vpbroadcastd %xmm2, %zmm2
 ; 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:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT:    vpsllw %xmm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm4
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
 ; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512F-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512F-NEXT:    vpsllw %xmm3, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm4
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %xmm4, %xmm5, %xmm6
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm7
-; AVX512F-NEXT:    vpsrlw %xmm6, %ymm7, %ymm6
-; AVX512F-NEXT:    vpsubw %xmm2, %xmm5, %xmm5
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm5, %ymm1, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm6, %zmm1, %zmm1
+; 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:    vextracti64x4 $1, %zmm1, %ymm5
+; AVX512F-NEXT:    vpsrlw %xmm4, %ymm5, %ymm5
+; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm5, %zmm1, %zmm1
 ; AVX512F-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm2
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
 ; AVX512F-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsllw %xmm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm4
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
 ; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512VL-NEXT:    vpsllw %xmm3, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm4
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm4, %xmm5, %xmm6
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm7
-; AVX512VL-NEXT:    vpsrlw %xmm6, %ymm7, %ymm6
-; AVX512VL-NEXT:    vpsubw %xmm2, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm5, %ymm1, %ymm1
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm6, %zmm1, %zmm1
+; 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:    vextracti64x4 $1, %zmm1, %ymm5
+; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm5, %zmm1, %zmm1
 ; AVX512VL-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpcmpeqw %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm2
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
 ; AVX512VL-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
 ; 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:    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:    vpbroadcastw %xmm2, %zmm2
 ; 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:    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:    vpbroadcastw %xmm2, %zmm2
 ; 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 %xmm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512F-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512F-NEXT:    vpsllw %xmm3, %ymm4, %ymm4
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT:    vpsllw %xmm4, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512F-NEXT:    vpsllw %xmm3, %xmm5, %xmm6
+; AVX512F-NEXT:    vpsllw %xmm4, %xmm5, %xmm6
 ; AVX512F-NEXT:    vpbroadcastb %xmm6, %ymm6
-; AVX512F-NEXT:    vpand %ymm6, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm4
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %xmm4, %xmm6, %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:    vextracti64x4 $1, %zmm1, %ymm8
-; AVX512F-NEXT:    vpsrlw %xmm7, %ymm8, %ymm8
-; AVX512F-NEXT:    vpsrlw %xmm7, %xmm5, %xmm7
-; AVX512F-NEXT:    vpsrlw $8, %xmm7, %xmm7
-; AVX512F-NEXT:    vpbroadcastb %xmm7, %ymm7
-; AVX512F-NEXT:    vpand %ymm7, %ymm8, %ymm7
-; AVX512F-NEXT:    vpsubb %xmm2, %xmm6, %xmm6
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpsrlw %xmm6, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsrlw %xmm6, %xmm5, %xmm5
+; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm4
+; AVX512F-NEXT:    vpand %ymm6, %ymm4, %ymm4
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
+; 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:    vextracti64x4 $1, %zmm1, %ymm6
+; AVX512F-NEXT:    vpsrlw %xmm4, %ymm6, %ymm6
+; AVX512F-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
 ; AVX512F-NEXT:    vpsrlw $8, %xmm5, %xmm5
 ; AVX512F-NEXT:    vpbroadcastb %xmm5, %ymm5
+; AVX512F-NEXT:    vpand %ymm5, %ymm6, %ymm6
+; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpand %ymm5, %ymm1, %ymm1
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm7, %zmm1, %zmm1
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm6, %zmm1, %zmm1
 ; AVX512F-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512F-NEXT:    vpcmpeqb %ymm3, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm2
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
 ; AVX512F-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm4
-; AVX512VL-NEXT:    vpsllw %xmm3, %ymm4, %ymm4
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VL-NEXT:    vpsllw %xmm4, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpsllw %xmm3, %xmm5, %xmm6
+; AVX512VL-NEXT:    vpsllw %xmm4, %xmm5, %xmm6
 ; AVX512VL-NEXT:    vpbroadcastb %xmm6, %ymm6
-; AVX512VL-NEXT:    vpand %ymm6, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm4
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm4, %xmm6, %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:    vextracti64x4 $1, %zmm1, %ymm8
-; AVX512VL-NEXT:    vpsrlw %xmm7, %ymm8, %ymm8
-; AVX512VL-NEXT:    vpsrlw %xmm7, %xmm5, %xmm7
-; AVX512VL-NEXT:    vpsrlw $8, %xmm7, %xmm7
-; AVX512VL-NEXT:    vpbroadcastb %xmm7, %ymm7
-; AVX512VL-NEXT:    vpand %ymm7, %ymm8, %ymm7
-; AVX512VL-NEXT:    vpsubb %xmm2, %xmm6, %xmm6
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpsrlw %xmm6, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsrlw %xmm6, %xmm5, %xmm5
+; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm4
+; AVX512VL-NEXT:    vpand %ymm6, %ymm4, %ymm4
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
+; 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:    vextracti64x4 $1, %zmm1, %ymm6
+; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm6, %ymm6
+; AVX512VL-NEXT:    vpsrlw %xmm4, %xmm5, %xmm5
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm5, %xmm5
 ; AVX512VL-NEXT:    vpbroadcastb %xmm5, %ymm5
+; AVX512VL-NEXT:    vpand %ymm5, %ymm6, %ymm6
+; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm1, %ymm1
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm7, %zmm1, %zmm1
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm6, %zmm1, %zmm1
 ; AVX512VL-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm2
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
 ; AVX512VL-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm0
 ; 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:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX512BW-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX512BW-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512BW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
 ; 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, %xmm5, %xmm4
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512BW-NEXT:    vpandq %zmm4, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; 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:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX512VBMI2-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
 ; 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, %xmm5, %xmm4
 ; AVX512VBMI2-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpandq %zmm4, %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
+; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; 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:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX512VLBW-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
 ; 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, %xmm5, %xmm4
 ; AVX512VLBW-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpandq %zmm4, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; 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:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX512VLVBMI2-NEXT:    vpsllw %xmm3, %xmm5, %xmm3
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
 ; 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, %xmm5, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpandq %zmm4, %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
+; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <64 x i8> %amt, <64 x i8> undef, <64 x i32> zeroinitializer
   %res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %splat)
   ret <64 x i8> %res
 }

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:    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:    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:    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
; 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:    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 = [65535,0,0,0]
+; SSE2-NEXT:    pand %xmm1, %xmm2
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
+; SSE2-NEXT:    psllw %xmm2, %xmm3
 ; 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:    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:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
-; 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:    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
+; AVX-LABEL: splatvar_funnnel_v8i16:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; AVX-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpor %xmm0, %xmm2, %xmm0
+; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_funnnel_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpand {{.*}}(%rip), %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

 ; 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:    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 = [65535,0,0,0]
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
+; X32-SSE-NEXT:    psllw %xmm2, %xmm3
 ; 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:    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:    pmovzxbq {{.*#+}} xmm3 = 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:    psllw %xmm3, %xmm2
+; SSE41-NEXT:    pcmpeqd %xmm4, %xmm4
 ; 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:    psllw %xmm3, %xmm5
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    pshufb %xmm3, %xmm5
+; SSE41-NEXT:    pand %xmm5, %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:    psrlw %xmm1, %xmm4
+; SSE41-NEXT:    pshufb {{.*#+}} xmm4 = xmm4[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; SSE41-NEXT:    pand %xmm0, %xmm4
+; SSE41-NEXT:    por %xmm4, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %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:    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:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpshufb %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vpand %xmm2, %xmm3, %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:    vpsrlw %xmm1, %xmm4, %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:    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]

 ; 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-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:    movdqa {{.*#+}} xmm4 = [64,64]
 ; SSE2-NEXT:    psubq %xmm2, %xmm4
 ; SSE2-NEXT:    psllq %xmm4, %xmm0
 ; SSE2-NEXT:    por %xmm3, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; 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:    movdqa {{.*#+}} xmm4 = [64,64]
 ; SSE41-NEXT:    psubq %xmm2, %xmm4
 ; SSE41-NEXT:    psllq %xmm4, %xmm3
 ; SSE41-NEXT:    por %xmm0, %xmm3
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512F-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX512F-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VL-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512BW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512BW-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VBMI2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VBMI2-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX512VBMI2-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX512VBMI2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; AVX512VLBW-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; AVX512VLBW-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX512VLBW-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; XOPAVX1-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; XOPAVX1-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
 ; 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:    vmovdqa {{.*#+}} xmm4 = [64,64]
 ; XOPAVX2-NEXT:    vpsubq %xmm2, %xmm4, %xmm4
 ; XOPAVX2-NEXT:    vpsllq %xmm4, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vpbroadcastq %xmm2, %xmm2
 ; 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:    pshufd {{.*#+}} xmm4 = xmm2[0,1,0,1]
+; X32-SSE-NEXT:    pxor %xmm5, %xmm5
+; X32-SSE-NEXT:    pcmpeqd %xmm4, %xmm5
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[1,0,3,2]
+; X32-SSE-NEXT:    pand %xmm5, %xmm3
+; 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:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pandn %xmm0, %xmm2
-; X32-SSE-NEXT:    por %xmm1, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
+; X32-SSE-NEXT:    psrlq %xmm2, %xmm4
+; X32-SSE-NEXT:    por %xmm0, %xmm4
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
+; X32-SSE-NEXT:    pandn %xmm4, %xmm3
+; X32-SSE-NEXT:    por %xmm1, %xmm3
+; X32-SSE-NEXT:    movdqa %xmm3, %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:    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:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; 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:    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:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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:    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:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; 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 {{.*#+}} 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:    vpbroadcastd %xmm2, %xmm2
 ; 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:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    pcmpeqw %xmm3, %xmm2
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; SSE2-NEXT:    psubw %xmm2, %xmm3
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = 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, %xmm5
+; SSE2-NEXT:    psrlw %xmm2, %xmm5
 ; 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:    psllw %xmm3, %xmm0
 ; SSE2-NEXT:    por %xmm5, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,0,0,0]
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    pcmpeqw %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_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:    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:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,0,0]
 ; 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:    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:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    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:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,2,3,4,5,6,7]
+; XOPAVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; 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:    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:    vpbroadcastw %xmm2, %xmm2
 ; 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:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pcmpeqw %xmm3, %xmm2
+; 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:    pshuflw {{.*#+}} xmm4 = 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, %xmm5
+; X32-SSE-NEXT:    psrlw %xmm2, %xmm5
 ; 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:    psllw %xmm3, %xmm0
 ; X32-SSE-NEXT:    por %xmm5, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,0,0,0]
+; X32-SSE-NEXT:    pxor %xmm2, %xmm2
+; X32-SSE-NEXT:    pcmpeqw %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 <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:    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:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,0,0]
+; 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:    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:    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:    movdqa {{.*#+}} xmm0 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; SSE41-NEXT:    psubb %xmm2, %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:    psllw %xmm0, %xmm3
+; SSE41-NEXT:    psllw %xmm0, %xmm5
+; SSE41-NEXT:    pxor %xmm0, %xmm0
+; SSE41-NEXT:    pshufb %xmm0, %xmm5
+; SSE41-NEXT:    pand %xmm5, %xmm3
+; SSE41-NEXT:    por %xmm6, %xmm3
+; SSE41-NEXT:    pshufb %xmm0, %xmm2
 ; 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:    vpand %xmm5, %xmm0, %xmm0
-; AVX1-NEXT:    vpor %xmm4, %xmm0, %xmm0
-; 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:    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:    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:    vpsllw %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufb %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vpcmpeqb %xmm5, %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:    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:    vpbroadcastb %xmm2, %xmm2
 ; 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:    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:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,2,3,4,5,6,7]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,0,0]
+; 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:    vpermilps {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; 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:    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:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
+; 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:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
+; AVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
+; AVX512VBMI2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512VBMI2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    vpermilps {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm2, %ymm2
-; XOPAVX1-NEXT:    vandps {{.*}}(%rip), %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; 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:    vpshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
+; 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:    vpbroadcastq {{.*#+}} xmm3 = [63,63]
+; XOPAVX2-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsrlq %xmm2, %ymm1, %ymm3
 ; 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:    vpbroadcastq %xmm2, %ymm2
 ; 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:    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:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; 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:    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, %ymm1, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    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, %ymm1, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VL-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    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, %ymm1, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    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, %ymm1, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VLBW-NEXT:    vpsrld %xmm3, %ymm1, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    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:    vextractf128 $1, %ymm1, %xmm3
+; XOPAVX1-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; 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:    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, %ymm1, %ymm3
 ; 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:    vpbroadcastd %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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:    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:    vpbroadcastw %xmm2, %ymm2
 ; 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 %ymm0, %ymm9, %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 %xmm5, %xmm5, %xmm5
 ; AVX2-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
 ; AVX2-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX2-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX2-NEXT:    vpand %ymm3, %ymm4, %ymm3
 ; 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, %xmm5, %xmm4
 ; AVX2-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX2-NEXT:    vpand %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; 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 %xmm5, %xmm5, %xmm5
 ; AVX512F-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
 ; AVX512F-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX512F-NEXT:    vpand %ymm3, %ymm4, %ymm3
 ; 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, %xmm5, %xmm4
 ; AVX512F-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512F-NEXT:    vpand %ymm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; 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 %xmm5, %xmm5, %xmm5
 ; AVX512VL-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpbroadcastb %xmm3, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm4, %ymm3
 ; 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, %xmm5, %xmm4
 ; AVX512VL-NEXT:    vpbroadcastb %xmm4, %ymm4
 ; AVX512VL-NEXT:    vpand %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
 ; 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
+; XOPAVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; XOPAVX2-NEXT:    vpsubb %xmm2, %xmm3, %xmm3
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm4
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm4, %xmm4
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm1, %xmm3
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm4, %ymm3, %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:    vextracti128 $1, %ymm0, %xmm6
 ; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
 ; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vpor %ymm3, %ymm0, %ymm0

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 {{.*#+}} xmm3 = [63,63]
+; AVX512F-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
 ; 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:    vpbroadcastq %xmm2, %zmm2
 ; 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:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
 ; 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:    vpbroadcastq %xmm2, %zmm2
 ; 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 {{.*#+}} xmm3 = [63,63]
+; AVX512BW-NEXT:    vpand %xmm3, %xmm2, %xmm2
 ; AVX512BW-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
 ; 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:    vpbroadcastq %xmm2, %zmm2
 ; 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:    vpandq {{.*}}(%rip){1to2}, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpsrlq %xmm2, %zmm1, %zmm3
 ; 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:    vpbroadcastq %xmm2, %zmm2
 ; 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 {{.*#+}} 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, %zmm1, %zmm3
 ; 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:    vpbroadcastd %xmm2, %zmm2
 ; 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:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VL-NEXT:    vpsrld %xmm3, %zmm1, %zmm3
 ; 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:    vpbroadcastd %xmm2, %zmm2
 ; 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 {{.*#+}} 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, %zmm1, %zmm3
 ; 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:    vpbroadcastd %xmm2, %zmm2
 ; 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:    vpandd {{.*}}(%rip){1to4}, %xmm2, %xmm2
 ; AVX512VLBW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero
 ; AVX512VLBW-NEXT:    vpsrld %xmm3, %zmm1, %zmm3
 ; 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:    vpbroadcastd %xmm2, %zmm2
 ; 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:    vextracti64x4 $1, %zmm1, %ymm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512F-NEXT:    vpsrlw %xmm4, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm4
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
 ; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512F-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
-; AVX512F-NEXT:    vpsrlw %xmm3, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm4
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; AVX512F-NEXT:    vpsubw %xmm4, %xmm5, %xmm6
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm7
-; AVX512F-NEXT:    vpsllw %xmm6, %ymm7, %ymm6
-; AVX512F-NEXT:    vpsubw %xmm2, %xmm5, %xmm5
-; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm5, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm6, %zmm0, %zmm0
+; 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:    vextracti64x4 $1, %zmm0, %ymm5
+; AVX512F-NEXT:    vpsllw %xmm4, %ymm5, %ymm5
+; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm5, %zmm0, %zmm0
 ; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm3
 ; AVX512F-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512F-NEXT:    vpcmpeqw %ymm0, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpcmpeqw %ymm0, %ymm2, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpternlogq $202, %zmm3, %zmm1, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm4
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
 ; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
-; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm4
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm5 = [16,16,16,16,16,16,16,16]
-; AVX512VL-NEXT:    vpsubw %xmm4, %xmm5, %xmm6
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm7
-; AVX512VL-NEXT:    vpsllw %xmm6, %ymm7, %ymm6
-; AVX512VL-NEXT:    vpsubw %xmm2, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm5, %ymm0, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm6, %zmm0, %zmm0
+; 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:    vextracti64x4 $1, %zmm0, %ymm5
+; AVX512VL-NEXT:    vpsllw %xmm4, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm5, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm3
 ; AVX512VL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpcmpeqw %ymm0, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpcmpeqw %ymm0, %ymm2, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpternlogq $202, %zmm3, %zmm1, %zmm0
 ; 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:    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:    vpbroadcastw %xmm2, %zmm2
 ; 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:    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:    vpbroadcastw %xmm2, %zmm2
 ; 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 %xmm2, %ymm2
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512F-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
-; AVX512F-NEXT:    vpsrlw %xmm3, %ymm4, %ymm4
+; AVX512F-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
+; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512F-NEXT:    vpsrlw %xmm4, %ymm3, %ymm3
 ; AVX512F-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512F-NEXT:    vpsrlw %xmm3, %xmm5, %xmm6
+; AVX512F-NEXT:    vpsrlw %xmm4, %xmm5, %xmm6
 ; AVX512F-NEXT:    vpsrlw $8, %xmm6, %xmm6
 ; AVX512F-NEXT:    vpbroadcastb %xmm6, %ymm6
-; AVX512F-NEXT:    vpand %ymm6, %ymm4, %ymm4
-; AVX512F-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
 ; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm2, %ymm4
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512F-NEXT:    vpsubb %xmm4, %xmm6, %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:    vextracti64x4 $1, %zmm0, %ymm8
-; AVX512F-NEXT:    vpsllw %xmm7, %ymm8, %ymm8
-; AVX512F-NEXT:    vpsllw %xmm7, %xmm5, %xmm7
-; AVX512F-NEXT:    vpbroadcastb %xmm7, %ymm7
-; AVX512F-NEXT:    vpand %ymm7, %ymm8, %ymm7
-; AVX512F-NEXT:    vpsubb %xmm2, %xmm6, %xmm6
-; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512F-NEXT:    vpsllw %xmm6, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsllw %xmm6, %xmm5, %xmm5
+; AVX512F-NEXT:    vpsrlw %xmm4, %ymm1, %ymm4
+; AVX512F-NEXT:    vpand %ymm6, %ymm4, %ymm4
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
+; 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:    vextracti64x4 $1, %zmm0, %ymm6
+; AVX512F-NEXT:    vpsllw %xmm4, %ymm6, %ymm6
+; AVX512F-NEXT:    vpsllw %xmm4, %xmm5, %xmm5
 ; AVX512F-NEXT:    vpbroadcastb %xmm5, %ymm5
+; AVX512F-NEXT:    vpand %ymm5, %ymm6, %ymm6
+; AVX512F-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpand %ymm5, %ymm0, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm7, %zmm0, %zmm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm6, %zmm0, %zmm0
 ; AVX512F-NEXT:    vporq %zmm3, %zmm0, %zmm3
 ; AVX512F-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512F-NEXT:    vpcmpeqb %ymm0, %ymm4, %ymm4
 ; AVX512F-NEXT:    vpcmpeqb %ymm0, %ymm2, %ymm0
-; AVX512F-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
+; AVX512F-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512F-NEXT:    vpternlogq $202, %zmm3, %zmm1, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm2, %zmm2
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm4
-; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm4, %ymm4
+; AVX512VL-NEXT:    vextracti64x4 $1, %zmm1, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
+; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm4 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpcmpeqd %xmm5, %xmm5, %xmm5
-; AVX512VL-NEXT:    vpsrlw %xmm3, %xmm5, %xmm6
+; AVX512VL-NEXT:    vpsrlw %xmm4, %xmm5, %xmm6
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm6, %xmm6
 ; AVX512VL-NEXT:    vpbroadcastb %xmm6, %ymm6
-; AVX512VL-NEXT:    vpand %ymm6, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpsrlw %xmm3, %ymm1, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm2, %ymm4
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm6 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VL-NEXT:    vpsubb %xmm4, %xmm6, %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:    vextracti64x4 $1, %zmm0, %ymm8
-; AVX512VL-NEXT:    vpsllw %xmm7, %ymm8, %ymm8
-; AVX512VL-NEXT:    vpsllw %xmm7, %xmm5, %xmm7
-; AVX512VL-NEXT:    vpbroadcastb %xmm7, %ymm7
-; AVX512VL-NEXT:    vpand %ymm7, %ymm8, %ymm7
-; AVX512VL-NEXT:    vpsubb %xmm2, %xmm6, %xmm6
-; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm6 = xmm6[0],zero,zero,zero,zero,zero,zero,zero,xmm6[1],zero,zero,zero,zero,zero,zero,zero
-; AVX512VL-NEXT:    vpsllw %xmm6, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpsllw %xmm6, %xmm5, %xmm5
+; AVX512VL-NEXT:    vpsrlw %xmm4, %ymm1, %ymm4
+; AVX512VL-NEXT:    vpand %ymm6, %ymm4, %ymm4
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
+; 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:    vextracti64x4 $1, %zmm0, %ymm6
+; AVX512VL-NEXT:    vpsllw %xmm4, %ymm6, %ymm6
+; AVX512VL-NEXT:    vpsllw %xmm4, %xmm5, %xmm5
 ; AVX512VL-NEXT:    vpbroadcastb %xmm5, %ymm5
+; AVX512VL-NEXT:    vpand %ymm5, %ymm6, %ymm6
+; AVX512VL-NEXT:    vpsllw %xmm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm0, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm7, %zmm0, %zmm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm6, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm3
 ; AVX512VL-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpcmpeqb %ymm0, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpcmpeqb %ymm0, %ymm2, %ymm0
-; AVX512VL-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
+; AVX512VL-NEXT:    vinserti64x4 $1, %ymm0, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vpternlogq $202, %zmm3, %zmm1, %zmm0
 ; 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:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX512BW-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX512BW-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512BW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
 ; 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, %xmm5, %xmm4
 ; AVX512BW-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512BW-NEXT:    vpandq %zmm4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; 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:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX512VBMI2-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
 ; AVX512VBMI2-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
 ; 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, %xmm5, %xmm4
 ; AVX512VBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpandq %zmm4, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; 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:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX512VLBW-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
 ; AVX512VLBW-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm4, %zmm3
 ; 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, %xmm5, %xmm4
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpandq %zmm4, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; 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:    vpcmpeqd %xmm5, %xmm5, %xmm5
 ; AVX512VLVBMI2-NEXT:    vpsrlw %xmm3, %xmm5, %xmm3
 ; AVX512VLVBMI2-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm4, %zmm3
 ; 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, %xmm5, %xmm4
 ; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpandq %zmm4, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vptestnmb %zmm2, %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    retq
   %splat = shufflevector <64 x i8> %amt, <64 x i8> undef, <64 x i32> zeroinitializer
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %splat)
   ret <64 x i8> %res
 }
 

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:    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:    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:    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:    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

 ; 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 = [65535,0,0,0]
+; SSE2-NEXT:    pand %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
+; SSE2-NEXT:    psllw %xmm1, %xmm3
 ; 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:    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:    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:    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:    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:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
-; 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:    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
+; AVX-LABEL: splatvar_funnnel_v8i16:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; AVX-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpor %xmm0, %xmm2, %xmm0
+; AVX-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:    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 = [65535,0,0,0]
+; X32-SSE-NEXT:    pand %xmm2, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
+; X32-SSE-NEXT:    psllw %xmm1, %xmm3
 ; 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:    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:    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:    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:    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:    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:    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

 ; 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:    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:    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:    vextracti128 $1, %ymm1, %xmm2
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; XOPAVX2-NEXT:    vprotq %xmm2, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vpsubq %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; XOPAVX2-NEXT:    vprotq %xmm1, %xmm2, %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, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; 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:    vextracti128 $1, %ymm1, %xmm2
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; XOPAVX2-NEXT:    vprotd %xmm2, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; XOPAVX2-NEXT:    vprotd %xmm1, %xmm2, %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, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; 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, %xmm1
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; 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:    vextracti128 $1, %ymm1, %xmm2
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; XOPAVX2-NEXT:    vprotw %xmm2, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; XOPAVX2-NEXT:    vprotw %xmm1, %xmm2, %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, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; 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 %xmm4, %xmm4, %xmm4
 ; AVX2-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
 ; 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, %xmm4, %xmm1
 ; AVX2-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; 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, %xmm1
 ; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; 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 %xmm4, %xmm4, %xmm4
 ; AVX512F-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
 ; 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, %xmm4, %xmm1
 ; AVX512F-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; 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, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; AVX512VL-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; 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 %xmm4, %xmm4, %xmm4
 ; AVX512VL-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
 ; 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:    vextracti128 $1, %ymm1, %xmm2
-; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; XOPAVX2-NEXT:    vprotb %xmm2, %xmm3, %xmm2
+; XOPAVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
+; XOPAVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; XOPAVX2-NEXT:    vprotb %xmm1, %xmm2, %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)
   ret <32 x i8> %res
 }
 

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:    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:    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:    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
; X32-SSE-NEXT:    retl
   %shl = shl <4 x i32> %a, %splat
   %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:    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 = [65535,0,0,0]
+; SSE2-NEXT:    pand %xmm1, %xmm2
+; SSE2-NEXT:    movdqa %xmm0, %xmm3
+; SSE2-NEXT:    psllw %xmm2, %xmm3
 ; 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:    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:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
-; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
-; 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:    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
+; AVX-LABEL: splatvar_rotate_v8i16:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
+; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX-NEXT:    vpsllw %xmm2, %xmm0, %xmm2
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
+; AVX-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpor %xmm0, %xmm2, %xmm0
+; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: splatvar_rotate_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpand {{.*}}(%rip), %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

 ; 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:    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 = [65535,0,0,0]
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
+; X32-SSE-NEXT:    psllw %xmm2, %xmm3
 ; 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:    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:    pmovzxbq {{.*#+}} xmm3 = 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:    psllw %xmm3, %xmm2
+; SSE41-NEXT:    pcmpeqd %xmm4, %xmm4
 ; 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:    psllw %xmm3, %xmm5
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    pshufb %xmm3, %xmm5
+; SSE41-NEXT:    pand %xmm5, %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:    psrlw %xmm1, %xmm4
+; SSE41-NEXT:    pshufb {{.*#+}} xmm4 = xmm4[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
+; SSE41-NEXT:    pand %xmm0, %xmm4
+; SSE41-NEXT:    por %xmm4, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_rotate_v16i8:
 ; AVX1:       # %bb.0:
-; 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:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsllw %xmm2, %xmm0, %xmm3
+; AVX1-NEXT:    vpcmpeqd %xmm4, %xmm4, %xmm4
+; AVX1-NEXT:    vpsllw %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX1-NEXT:    vpshufb %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vpand %xmm2, %xmm3, %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:    vpsrlw %xmm1, %xmm4, %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:    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]

 ; 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-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