[X86] Freeze vXi8 shl(x,1) -> add(x,x) vector fold (PR50468)

[RKSimon]

Summary

We don't have any vXi8 shift instructions (other than on XOP which is handled separately), so replace the shl(x,1) ->add(x,x) fold with shl(x,1) ->add(freeze(x),freeze(x)) to avoid the undef issues identified in PR50468.

Split off from D106675 as I'm still looking at whether we can fix the vXi16/i32/i64 issues with the D106679 alternative.

[RKSimon]

Any objections to this?

[RKSimon]

Add comment explaining the need for freeze

[pengfei]

llvm/lib/Target/X86/X86ISelLowering.cpp

28729

Sorry, I don't quite understand the background. My question is why don't we check the oprand is a undef?
The affected tests seem all reasonable values, though they seem not have regressions.

[craig.topper]

llvm/lib/Target/X86/X86ISelLowering.cpp

28729

It’s not enough to check for undef at the time of the fold. You need to know that no future DAGCombine can make the input undef.

(shl X, 1) must produce an even number even if X is undef. computeKnownBits will look at the shift amount and tell downstream code that the lsb is 0 without looking at the left hand side. And the value may not be undef at the time computeKnownBits is called but could be simplified to undef later.

(add undef, undef) does not produce an even number. Register allocation is free to pick different registers for undef.

The freeze forces register allocation to use the same register.

[pengfei]

llvm/lib/Target/X86/X86ISelLowering.cpp

28729

Nice answer! I'm clear now. Thanks Craig.

[RKSimon]

Thanks for the explanation Craig! I always feel I confuse both myself and other people when I start talking about undef and poison......

Is everyone happy for this to go in?

[spatel]

LGTM

llvm/lib/Target/X86/X86ISelLowering.cpp

28729

Undef reasoning is complicated, and the explanation here is great. How about enshrining it as a code comment? :)

// R may be undef at run-time, but (shl R, 1) must be an even number (LSB must be 0).
// (add undef, undef) however can be any value. To make this safe, we must freeze R
// to ensure that register allocation uses the same register for an undefined value.
// This ensures that the result will still be even and preserves the original semantics.

[RKSimon]

Thanks - I'll add the explanation comment in the commit

llvm/lib/Target/X86/X86ISelLowering.cpp

     return ArithmeticShiftRight64(ShiftAmt);
 
   if (VT == MVT::v16i8 || (Subtarget.hasInt256() && VT == MVT::v32i8) ||
       (Subtarget.hasBWI() && VT == MVT::v64i8)) {
     unsigned NumElts = VT.getVectorNumElements();
     MVT ShiftVT = MVT::getVectorVT(MVT::i16, NumElts / 2);
 
     // Simple i8 add case
-    if (Op.getOpcode() == ISD::SHL && ShiftAmt == 1)
+    if (Op.getOpcode() == ISD::SHL && ShiftAmt == 1) {
+      R = DAG.getNode(ISD::FREEZE, dl, VT, R);
       return DAG.getNode(ISD::ADD, dl, VT, R, R);
+    }
 
     // ashr(R, 7)  === cmp_slt(R, 0)
     if (Op.getOpcode() == ISD::SRA && ShiftAmt == 7) {
       SDValue Zeros = DAG.getConstant(0, dl, VT);
       if (VT.is512BitVector()) {
         assert(VT == MVT::v64i8 && "Unexpected element type!");
         SDValue CMP = DAG.getSetCC(dl, MVT::v64i1, Zeros, R, ISD::SETGT);
         return DAG.getNode(ISD::SIGN_EXTEND, dl, VT, CMP);

   // The advantage is that the two shifts from the example would be
   // lowered as X86ISD::VSRLI nodes in parallel before blending.
   if (ConstantAmt && (VT == MVT::v8i16 || VT == MVT::v4i32 ||
                       (VT == MVT::v16i16 && Subtarget.hasInt256()))) {
     SDValue Amt1, Amt2;
     unsigned NumElts = VT.getVectorNumElements();
     SmallVector<int, 8> ShuffleMask;
     for (unsigned i = 0; i != NumElts; ++i) {
       SDValue A = Amt->getOperand(i);
       if (A.isUndef()) {
         ShuffleMask.push_back(SM_SentinelUndef);
         continue;
       }
       if (!Amt1 || Amt1 == A) {
         ShuffleMask.push_back(i);
         Amt1 = A;
         continue;

llvm/test/CodeGen/X86/bitreverse.ll

 ; X64-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; X64-NEXT:    por %xmm1, %xmm0
 ; X64-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; X64-NEXT:    pand %xmm0, %xmm1
 ; X64-NEXT:    psllw $2, %xmm1
 ; X64-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; X64-NEXT:    psrlw $2, %xmm0
 ; X64-NEXT:    por %xmm1, %xmm0
-; X64-NEXT:    movdqa {{.*#+}} xmm1 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; X64-NEXT:    movdqa {{.*#+}} xmm1 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; X64-NEXT:    pand %xmm0, %xmm1
-; X64-NEXT:    paddb %xmm1, %xmm1
+; X64-NEXT:    psrlw $1, %xmm1
 ; X64-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; X64-NEXT:    psrlw $1, %xmm0
+; X64-NEXT:    paddb %xmm0, %xmm0
 ; X64-NEXT:    por %xmm1, %xmm0
 ; X64-NEXT:    retq
 ;
 ; X86XOP-LABEL: test_bitreverse_v2i16:
 ; X86XOP:       # %bb.0:
 ; X86XOP-NEXT:    vpperm {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0, %xmm0
 ; X86XOP-NEXT:    retl
   %b = call <2 x i16> @llvm.bitreverse.v2i16(<2 x i16> %a)

llvm/test/CodeGen/X86/combine-bitreverse.ll

 ; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
 ; X86-NEXT:    por %xmm1, %xmm0
 ; X86-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; X86-NEXT:    pand %xmm0, %xmm1
 ; X86-NEXT:    psllw $2, %xmm1
 ; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
 ; X86-NEXT:    psrlw $2, %xmm0
 ; X86-NEXT:    por %xmm1, %xmm0
-; X86-NEXT:    movdqa {{.*#+}} xmm1 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; X86-NEXT:    movdqa {{.*#+}} xmm1 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; X86-NEXT:    pand %xmm0, %xmm1
-; X86-NEXT:    paddb %xmm1, %xmm1
+; X86-NEXT:    psrlw $1, %xmm1
 ; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-NEXT:    psrlw $1, %xmm0
+; X86-NEXT:    paddb %xmm0, %xmm0
 ; X86-NEXT:    por %xmm1, %xmm0
 ; X86-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: test_demandedbits_bitreverse:
 ; X64:       # %bb.0:
 ; X64-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
 ; X64-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]

llvm/test/CodeGen/X86/vector-bitreverse.ll

 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    pand %xmm0, %xmm1
 ; SSE2-NEXT:    psllw $2, %xmm1
 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    paddb %xmm1, %xmm1
+; SSE2-NEXT:    psrlw $1, %xmm1
 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v16i8:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm2
 ; SSSE3-NEXT:    pand %xmm1, %xmm2

 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    pand %xmm0, %xmm1
 ; SSE2-NEXT:    psllw $2, %xmm1
 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    paddb %xmm1, %xmm1
+; SSE2-NEXT:    psrlw $1, %xmm1
 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v8i16:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm2

 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    pand %xmm0, %xmm1
 ; SSE2-NEXT:    psllw $2, %xmm1
 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    paddb %xmm1, %xmm1
+; SSE2-NEXT:    psrlw $1, %xmm1
 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v4i32:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm2

 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    pand %xmm0, %xmm1
 ; SSE2-NEXT:    psllw $2, %xmm1
 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    pand %xmm0, %xmm1
-; SSE2-NEXT:    paddb %xmm1, %xmm1
+; SSE2-NEXT:    psrlw $1, %xmm1
 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v2i64:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8]
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm2

 ; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm4
 ; SSE2-NEXT:    pand %xmm3, %xmm4
 ; SSE2-NEXT:    psllw $2, %xmm4
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [204,204,204,204,204,204,204,204,204,204,204,204,204,204,204,204]
 ; SSE2-NEXT:    pand %xmm5, %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm4, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm6
 ; SSE2-NEXT:    pand %xmm4, %xmm6
-; SSE2-NEXT:    paddb %xmm6, %xmm6
-; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
+; SSE2-NEXT:    psrlw $1, %xmm6
+; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
 ; SSE2-NEXT:    pand %xmm7, %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm6, %xmm0
 ; SSE2-NEXT:    movdqa %xmm2, %xmm6
 ; SSE2-NEXT:    psllw $4, %xmm6
 ; SSE2-NEXT:    psrlw $4, %xmm2
 ; SSE2-NEXT:    pand %xmm1, %xmm2
 ; SSE2-NEXT:    pandn %xmm6, %xmm1
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    pand %xmm1, %xmm3
 ; SSE2-NEXT:    psllw $2, %xmm3
 ; SSE2-NEXT:    pand %xmm5, %xmm1
 ; SSE2-NEXT:    psrlw $2, %xmm1
 ; SSE2-NEXT:    por %xmm3, %xmm1
 ; SSE2-NEXT:    pand %xmm1, %xmm4
-; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    psrlw $1, %xmm4
 ; SSE2-NEXT:    pand %xmm7, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    por %xmm4, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v32i8:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm2
 ; SSSE3-NEXT:    pand %xmm4, %xmm2

 ; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm4
 ; SSE2-NEXT:    pand %xmm3, %xmm4
 ; SSE2-NEXT:    psllw $2, %xmm4
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [204,204,204,204,204,204,204,204,204,204,204,204,204,204,204,204]
 ; SSE2-NEXT:    pand %xmm5, %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm4, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm7
 ; SSE2-NEXT:    pand %xmm4, %xmm7
-; SSE2-NEXT:    paddb %xmm7, %xmm7
-; SSE2-NEXT:    movdqa {{.*#+}} xmm6 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
+; SSE2-NEXT:    psrlw $1, %xmm7
+; SSE2-NEXT:    movdqa {{.*#+}} xmm6 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
 ; SSE2-NEXT:    pand %xmm6, %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm7, %xmm0
 ; SSE2-NEXT:    movdqa %xmm2, %xmm7
 ; SSE2-NEXT:    psrlw $8, %xmm7
 ; SSE2-NEXT:    psllw $8, %xmm2
 ; SSE2-NEXT:    por %xmm7, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm7
 ; SSE2-NEXT:    psllw $4, %xmm7
 ; SSE2-NEXT:    psrlw $4, %xmm2
 ; SSE2-NEXT:    pand %xmm1, %xmm2
 ; SSE2-NEXT:    pandn %xmm7, %xmm1
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    pand %xmm1, %xmm3
 ; SSE2-NEXT:    psllw $2, %xmm3
 ; SSE2-NEXT:    pand %xmm5, %xmm1
 ; SSE2-NEXT:    psrlw $2, %xmm1
 ; SSE2-NEXT:    por %xmm3, %xmm1
 ; SSE2-NEXT:    pand %xmm1, %xmm4
-; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    psrlw $1, %xmm4
 ; SSE2-NEXT:    pand %xmm6, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    por %xmm4, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v16i16:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 = [1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; SSSE3-NEXT:    pshufb %xmm4, %xmm0
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]

 ; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm5
 ; SSE2-NEXT:    pand %xmm3, %xmm5
 ; SSE2-NEXT:    psllw $2, %xmm5
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm8 = [204,204,204,204,204,204,204,204,204,204,204,204,204,204,204,204]
 ; SSE2-NEXT:    pand %xmm8, %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm5, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm6
 ; SSE2-NEXT:    pand %xmm5, %xmm6
-; SSE2-NEXT:    paddb %xmm6, %xmm6
-; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
+; SSE2-NEXT:    psrlw $1, %xmm6
+; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
 ; SSE2-NEXT:    pand %xmm7, %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm6, %xmm0
 ; SSE2-NEXT:    movdqa %xmm2, %xmm6
 ; SSE2-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm4[8],xmm6[9],xmm4[9],xmm6[10],xmm4[10],xmm6[11],xmm4[11],xmm6[12],xmm4[12],xmm6[13],xmm4[13],xmm6[14],xmm4[14],xmm6[15],xmm4[15]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3],xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    packuswb %xmm6, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    psllw $4, %xmm4
 ; SSE2-NEXT:    psrlw $4, %xmm2
 ; SSE2-NEXT:    pand %xmm1, %xmm2
 ; SSE2-NEXT:    pandn %xmm4, %xmm1
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    pand %xmm1, %xmm3
 ; SSE2-NEXT:    psllw $2, %xmm3
 ; SSE2-NEXT:    pand %xmm8, %xmm1
 ; SSE2-NEXT:    psrlw $2, %xmm1
 ; SSE2-NEXT:    por %xmm3, %xmm1
 ; SSE2-NEXT:    pand %xmm1, %xmm5
-; SSE2-NEXT:    paddb %xmm5, %xmm5
+; SSE2-NEXT:    psrlw $1, %xmm5
 ; SSE2-NEXT:    pand %xmm7, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    por %xmm5, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v8i32:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 = [3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
 ; SSSE3-NEXT:    pshufb %xmm4, %xmm0
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]

 ; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm5
 ; SSE2-NEXT:    pand %xmm3, %xmm5
 ; SSE2-NEXT:    psllw $2, %xmm5
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm8 = [204,204,204,204,204,204,204,204,204,204,204,204,204,204,204,204]
 ; SSE2-NEXT:    pand %xmm8, %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm5, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm6
 ; SSE2-NEXT:    pand %xmm5, %xmm6
-; SSE2-NEXT:    paddb %xmm6, %xmm6
-; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
+; SSE2-NEXT:    psrlw $1, %xmm6
+; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
 ; SSE2-NEXT:    pand %xmm7, %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm6, %xmm0
 ; SSE2-NEXT:    movdqa %xmm2, %xmm6
 ; SSE2-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm4[8],xmm6[9],xmm4[9],xmm6[10],xmm4[10],xmm6[11],xmm4[11],xmm6[12],xmm4[12],xmm6[13],xmm4[13],xmm6[14],xmm4[14],xmm6[15],xmm4[15]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,3,0,1]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3],xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    packuswb %xmm6, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    psllw $4, %xmm4
 ; SSE2-NEXT:    psrlw $4, %xmm2
 ; SSE2-NEXT:    pand %xmm1, %xmm2
 ; SSE2-NEXT:    pandn %xmm4, %xmm1
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    pand %xmm1, %xmm3
 ; SSE2-NEXT:    psllw $2, %xmm3
 ; SSE2-NEXT:    pand %xmm8, %xmm1
 ; SSE2-NEXT:    psrlw $2, %xmm1
 ; SSE2-NEXT:    por %xmm3, %xmm1
 ; SSE2-NEXT:    pand %xmm1, %xmm5
-; SSE2-NEXT:    paddb %xmm5, %xmm5
+; SSE2-NEXT:    psrlw $1, %xmm5
 ; SSE2-NEXT:    pand %xmm7, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    por %xmm5, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v4i64:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 = [7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8]
 ; SSSE3-NEXT:    pshufb %xmm4, %xmm0
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]

 ; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm6
 ; SSE2-NEXT:    pand %xmm5, %xmm6
 ; SSE2-NEXT:    psllw $2, %xmm6
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm8 = [204,204,204,204,204,204,204,204,204,204,204,204,204,204,204,204]
 ; SSE2-NEXT:    pand %xmm8, %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm6, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm6 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm6 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm7
 ; SSE2-NEXT:    pand %xmm6, %xmm7
-; SSE2-NEXT:    paddb %xmm7, %xmm7
-; SSE2-NEXT:    movdqa {{.*#+}} xmm9 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
+; SSE2-NEXT:    psrlw $1, %xmm7
+; SSE2-NEXT:    movdqa {{.*#+}} xmm9 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
 ; SSE2-NEXT:    pand %xmm9, %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm7, %xmm0
 ; SSE2-NEXT:    movdqa %xmm1, %xmm7
 ; SSE2-NEXT:    psllw $4, %xmm7
 ; SSE2-NEXT:    movdqa %xmm3, %xmm4
 ; SSE2-NEXT:    pandn %xmm7, %xmm4
 ; SSE2-NEXT:    psrlw $4, %xmm1
 ; SSE2-NEXT:    pand %xmm3, %xmm1
 ; SSE2-NEXT:    por %xmm4, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm4
 ; SSE2-NEXT:    pand %xmm5, %xmm4
 ; SSE2-NEXT:    psllw $2, %xmm4
 ; SSE2-NEXT:    pand %xmm8, %xmm1
 ; SSE2-NEXT:    psrlw $2, %xmm1
 ; SSE2-NEXT:    por %xmm4, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm4
 ; SSE2-NEXT:    pand %xmm6, %xmm4
-; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    psrlw $1, %xmm4
 ; SSE2-NEXT:    pand %xmm9, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    por %xmm4, %xmm1
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    psllw $4, %xmm4
 ; SSE2-NEXT:    movdqa %xmm3, %xmm7
 ; SSE2-NEXT:    pandn %xmm4, %xmm7
 ; SSE2-NEXT:    psrlw $4, %xmm2
 ; SSE2-NEXT:    pand %xmm3, %xmm2
 ; SSE2-NEXT:    por %xmm7, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    pand %xmm5, %xmm4
 ; SSE2-NEXT:    psllw $2, %xmm4
 ; SSE2-NEXT:    pand %xmm8, %xmm2
 ; SSE2-NEXT:    psrlw $2, %xmm2
 ; SSE2-NEXT:    por %xmm4, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    pand %xmm6, %xmm4
-; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    psrlw $1, %xmm4
 ; SSE2-NEXT:    pand %xmm9, %xmm2
-; SSE2-NEXT:    psrlw $1, %xmm2
+; SSE2-NEXT:    paddb %xmm2, %xmm2
 ; SSE2-NEXT:    por %xmm4, %xmm2
 ; SSE2-NEXT:    movdqa %xmm10, %xmm4
 ; SSE2-NEXT:    psllw $4, %xmm4
 ; SSE2-NEXT:    psrlw $4, %xmm10
 ; SSE2-NEXT:    pand %xmm3, %xmm10
 ; SSE2-NEXT:    pandn %xmm4, %xmm3
 ; SSE2-NEXT:    por %xmm10, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm5
 ; SSE2-NEXT:    psllw $2, %xmm5
 ; SSE2-NEXT:    pand %xmm8, %xmm3
 ; SSE2-NEXT:    psrlw $2, %xmm3
 ; SSE2-NEXT:    por %xmm5, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm6
-; SSE2-NEXT:    paddb %xmm6, %xmm6
+; SSE2-NEXT:    psrlw $1, %xmm6
 ; SSE2-NEXT:    pand %xmm9, %xmm3
-; SSE2-NEXT:    psrlw $1, %xmm3
+; SSE2-NEXT:    paddb %xmm3, %xmm3
 ; SSE2-NEXT:    por %xmm6, %xmm3
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v64i8:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm5
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm8 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
 ; SSSE3-NEXT:    pand %xmm8, %xmm0

 ; SSE2-NEXT:    movdqa {{.*#+}} xmm10 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm6
 ; SSE2-NEXT:    pand %xmm10, %xmm6
 ; SSE2-NEXT:    psllw $2, %xmm6
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm8 = [204,204,204,204,204,204,204,204,204,204,204,204,204,204,204,204]
 ; SSE2-NEXT:    pand %xmm8, %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm6, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm6 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm6 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm7
 ; SSE2-NEXT:    pand %xmm6, %xmm7
-; SSE2-NEXT:    paddb %xmm7, %xmm7
-; SSE2-NEXT:    movdqa {{.*#+}} xmm9 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
+; SSE2-NEXT:    psrlw $1, %xmm7
+; SSE2-NEXT:    movdqa {{.*#+}} xmm9 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
 ; SSE2-NEXT:    pand %xmm9, %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm7, %xmm0
 ; SSE2-NEXT:    movdqa %xmm1, %xmm7
 ; SSE2-NEXT:    psrlw $8, %xmm7
 ; SSE2-NEXT:    psllw $8, %xmm1
 ; SSE2-NEXT:    por %xmm7, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm7
 ; SSE2-NEXT:    psllw $4, %xmm7
 ; SSE2-NEXT:    movdqa %xmm3, %xmm5
 ; SSE2-NEXT:    pandn %xmm7, %xmm5
 ; SSE2-NEXT:    psrlw $4, %xmm1
 ; SSE2-NEXT:    pand %xmm3, %xmm1
 ; SSE2-NEXT:    por %xmm5, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm5
 ; SSE2-NEXT:    pand %xmm10, %xmm5
 ; SSE2-NEXT:    psllw $2, %xmm5
 ; SSE2-NEXT:    pand %xmm8, %xmm1
 ; SSE2-NEXT:    psrlw $2, %xmm1
 ; SSE2-NEXT:    por %xmm5, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm5
 ; SSE2-NEXT:    pand %xmm6, %xmm5
-; SSE2-NEXT:    paddb %xmm5, %xmm5
+; SSE2-NEXT:    psrlw $1, %xmm5
 ; SSE2-NEXT:    pand %xmm9, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    por %xmm5, %xmm1
 ; SSE2-NEXT:    movdqa %xmm2, %xmm5
 ; SSE2-NEXT:    psrlw $8, %xmm5
 ; SSE2-NEXT:    psllw $8, %xmm2
 ; SSE2-NEXT:    por %xmm5, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm5
 ; SSE2-NEXT:    psllw $4, %xmm5
 ; SSE2-NEXT:    movdqa %xmm3, %xmm7
 ; SSE2-NEXT:    pandn %xmm5, %xmm7
 ; SSE2-NEXT:    psrlw $4, %xmm2
 ; SSE2-NEXT:    pand %xmm3, %xmm2
 ; SSE2-NEXT:    por %xmm7, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm5
 ; SSE2-NEXT:    pand %xmm10, %xmm5
 ; SSE2-NEXT:    psllw $2, %xmm5
 ; SSE2-NEXT:    pand %xmm8, %xmm2
 ; SSE2-NEXT:    psrlw $2, %xmm2
 ; SSE2-NEXT:    por %xmm5, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm5
 ; SSE2-NEXT:    pand %xmm6, %xmm5
-; SSE2-NEXT:    paddb %xmm5, %xmm5
+; SSE2-NEXT:    psrlw $1, %xmm5
 ; SSE2-NEXT:    pand %xmm9, %xmm2
-; SSE2-NEXT:    psrlw $1, %xmm2
+; SSE2-NEXT:    paddb %xmm2, %xmm2
 ; SSE2-NEXT:    por %xmm5, %xmm2
 ; SSE2-NEXT:    movdqa %xmm4, %xmm5
 ; SSE2-NEXT:    psrlw $8, %xmm5
 ; SSE2-NEXT:    psllw $8, %xmm4
 ; SSE2-NEXT:    por %xmm5, %xmm4
 ; SSE2-NEXT:    movdqa %xmm4, %xmm5
 ; SSE2-NEXT:    psllw $4, %xmm5
 ; SSE2-NEXT:    psrlw $4, %xmm4
 ; SSE2-NEXT:    pand %xmm3, %xmm4
 ; SSE2-NEXT:    pandn %xmm5, %xmm3
 ; SSE2-NEXT:    por %xmm4, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm10
 ; SSE2-NEXT:    psllw $2, %xmm10
 ; SSE2-NEXT:    pand %xmm8, %xmm3
 ; SSE2-NEXT:    psrlw $2, %xmm3
 ; SSE2-NEXT:    por %xmm10, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm6
-; SSE2-NEXT:    paddb %xmm6, %xmm6
+; SSE2-NEXT:    psrlw $1, %xmm6
 ; SSE2-NEXT:    pand %xmm9, %xmm3
-; SSE2-NEXT:    psrlw $1, %xmm3
+; SSE2-NEXT:    paddb %xmm3, %xmm3
 ; SSE2-NEXT:    por %xmm6, %xmm3
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v32i16:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm5
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm8 = [1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]

 ; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm7
 ; SSE2-NEXT:    pand %xmm5, %xmm7
 ; SSE2-NEXT:    psllw $2, %xmm7
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm8 = [204,204,204,204,204,204,204,204,204,204,204,204,204,204,204,204]
 ; SSE2-NEXT:    pand %xmm8, %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm7, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm6
 ; SSE2-NEXT:    pand %xmm7, %xmm6
-; SSE2-NEXT:    paddb %xmm6, %xmm6
-; SSE2-NEXT:    movdqa {{.*#+}} xmm9 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
+; SSE2-NEXT:    psrlw $1, %xmm6
+; SSE2-NEXT:    movdqa {{.*#+}} xmm9 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
 ; SSE2-NEXT:    pand %xmm9, %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm6, %xmm0
 ; SSE2-NEXT:    movdqa %xmm1, %xmm6
 ; SSE2-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm10[8],xmm6[9],xmm10[9],xmm6[10],xmm10[10],xmm6[11],xmm10[11],xmm6[12],xmm10[12],xmm6[13],xmm10[13],xmm6[14],xmm10[14],xmm6[15],xmm10[15]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm10[0],xmm1[1],xmm10[1],xmm1[2],xmm10[2],xmm1[3],xmm10[3],xmm1[4],xmm10[4],xmm1[5],xmm10[5],xmm1[6],xmm10[6],xmm1[7],xmm10[7]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    packuswb %xmm6, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm6
 ; SSE2-NEXT:    psllw $4, %xmm6
 ; SSE2-NEXT:    movdqa %xmm3, %xmm4
 ; SSE2-NEXT:    pandn %xmm6, %xmm4
 ; SSE2-NEXT:    psrlw $4, %xmm1
 ; SSE2-NEXT:    pand %xmm3, %xmm1
 ; SSE2-NEXT:    por %xmm4, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm4
 ; SSE2-NEXT:    pand %xmm5, %xmm4
 ; SSE2-NEXT:    psllw $2, %xmm4
 ; SSE2-NEXT:    pand %xmm8, %xmm1
 ; SSE2-NEXT:    psrlw $2, %xmm1
 ; SSE2-NEXT:    por %xmm4, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm4
 ; SSE2-NEXT:    pand %xmm7, %xmm4
-; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    psrlw $1, %xmm4
 ; SSE2-NEXT:    pand %xmm9, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    por %xmm4, %xmm1
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm10[8],xmm4[9],xmm10[9],xmm4[10],xmm10[10],xmm4[11],xmm10[11],xmm4[12],xmm10[12],xmm4[13],xmm10[13],xmm4[14],xmm10[14],xmm4[15],xmm10[15]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm10[0],xmm2[1],xmm10[1],xmm2[2],xmm10[2],xmm2[3],xmm10[3],xmm2[4],xmm10[4],xmm2[5],xmm10[5],xmm2[6],xmm10[6],xmm2[7],xmm10[7]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    packuswb %xmm4, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    psllw $4, %xmm4
 ; SSE2-NEXT:    movdqa %xmm3, %xmm6
 ; SSE2-NEXT:    pandn %xmm4, %xmm6
 ; SSE2-NEXT:    psrlw $4, %xmm2
 ; SSE2-NEXT:    pand %xmm3, %xmm2
 ; SSE2-NEXT:    por %xmm6, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    pand %xmm5, %xmm4
 ; SSE2-NEXT:    psllw $2, %xmm4
 ; SSE2-NEXT:    pand %xmm8, %xmm2
 ; SSE2-NEXT:    psrlw $2, %xmm2
 ; SSE2-NEXT:    por %xmm4, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    pand %xmm7, %xmm4
-; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    psrlw $1, %xmm4
 ; SSE2-NEXT:    pand %xmm9, %xmm2
-; SSE2-NEXT:    psrlw $1, %xmm2
+; SSE2-NEXT:    paddb %xmm2, %xmm2
 ; SSE2-NEXT:    por %xmm4, %xmm2
 ; SSE2-NEXT:    movdqa %xmm11, %xmm4
 ; SSE2-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm10[8],xmm4[9],xmm10[9],xmm4[10],xmm10[10],xmm4[11],xmm10[11],xmm4[12],xmm10[12],xmm4[13],xmm10[13],xmm4[14],xmm10[14],xmm4[15],xmm10[15]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3],xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm11[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    packuswb %xmm4, %xmm6
 ; SSE2-NEXT:    movdqa %xmm6, %xmm4
 ; SSE2-NEXT:    psllw $4, %xmm4
 ; SSE2-NEXT:    psrlw $4, %xmm6
 ; SSE2-NEXT:    pand %xmm3, %xmm6
 ; SSE2-NEXT:    pandn %xmm4, %xmm3
 ; SSE2-NEXT:    por %xmm6, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm5
 ; SSE2-NEXT:    psllw $2, %xmm5
 ; SSE2-NEXT:    pand %xmm8, %xmm3
 ; SSE2-NEXT:    psrlw $2, %xmm3
 ; SSE2-NEXT:    por %xmm5, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm7
-; SSE2-NEXT:    paddb %xmm7, %xmm7
+; SSE2-NEXT:    psrlw $1, %xmm7
 ; SSE2-NEXT:    pand %xmm9, %xmm3
-; SSE2-NEXT:    psrlw $1, %xmm3
+; SSE2-NEXT:    paddb %xmm3, %xmm3
 ; SSE2-NEXT:    por %xmm7, %xmm3
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v16i32:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm5
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm8 = [3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]

 ; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm7
 ; SSE2-NEXT:    pand %xmm5, %xmm7
 ; SSE2-NEXT:    psllw $2, %xmm7
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm8 = [204,204,204,204,204,204,204,204,204,204,204,204,204,204,204,204]
 ; SSE2-NEXT:    pand %xmm8, %xmm0
 ; SSE2-NEXT:    psrlw $2, %xmm0
 ; SSE2-NEXT:    por %xmm7, %xmm0
-; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
+; SSE2-NEXT:    movdqa {{.*#+}} xmm7 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm6
 ; SSE2-NEXT:    pand %xmm7, %xmm6
-; SSE2-NEXT:    paddb %xmm6, %xmm6
-; SSE2-NEXT:    movdqa {{.*#+}} xmm9 = [170,170,170,170,170,170,170,170,170,170,170,170,170,170,170,170]
+; SSE2-NEXT:    psrlw $1, %xmm6
+; SSE2-NEXT:    movdqa {{.*#+}} xmm9 = [85,85,85,85,85,85,85,85,85,85,85,85,85,85,85,85]
 ; SSE2-NEXT:    pand %xmm9, %xmm0
-; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    paddb %xmm0, %xmm0
 ; SSE2-NEXT:    por %xmm6, %xmm0
 ; SSE2-NEXT:    movdqa %xmm1, %xmm6
 ; SSE2-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm10[8],xmm6[9],xmm10[9],xmm6[10],xmm10[10],xmm6[11],xmm10[11],xmm6[12],xmm10[12],xmm6[13],xmm10[13],xmm6[14],xmm10[14],xmm6[15],xmm10[15]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,3,0,1]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm10[0],xmm1[1],xmm10[1],xmm1[2],xmm10[2],xmm1[3],xmm10[3],xmm1[4],xmm10[4],xmm1[5],xmm10[5],xmm1[6],xmm10[6],xmm1[7],xmm10[7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]

 ; SSE2-NEXT:    movdqa %xmm1, %xmm4
 ; SSE2-NEXT:    pand %xmm5, %xmm4
 ; SSE2-NEXT:    psllw $2, %xmm4
 ; SSE2-NEXT:    pand %xmm8, %xmm1
 ; SSE2-NEXT:    psrlw $2, %xmm1
 ; SSE2-NEXT:    por %xmm4, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm4
 ; SSE2-NEXT:    pand %xmm7, %xmm4
-; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    psrlw $1, %xmm4
 ; SSE2-NEXT:    pand %xmm9, %xmm1
-; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    por %xmm4, %xmm1
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm10[8],xmm4[9],xmm10[9],xmm4[10],xmm10[10],xmm4[11],xmm10[11],xmm4[12],xmm10[12],xmm4[13],xmm10[13],xmm4[14],xmm10[14],xmm4[15],xmm10[15]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm10[0],xmm2[1],xmm10[1],xmm2[2],xmm10[2],xmm2[3],xmm10[3],xmm2[4],xmm10[4],xmm2[5],xmm10[5],xmm2[6],xmm10[6],xmm2[7],xmm10[7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]

 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    pand %xmm5, %xmm4
 ; SSE2-NEXT:    psllw $2, %xmm4
 ; SSE2-NEXT:    pand %xmm8, %xmm2
 ; SSE2-NEXT:    psrlw $2, %xmm2
 ; SSE2-NEXT:    por %xmm4, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; SSE2-NEXT:    pand %xmm7, %xmm4
-; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    psrlw $1, %xmm4
 ; SSE2-NEXT:    pand %xmm9, %xmm2
-; SSE2-NEXT:    psrlw $1, %xmm2
+; SSE2-NEXT:    paddb %xmm2, %xmm2
 ; SSE2-NEXT:    por %xmm4, %xmm2
 ; SSE2-NEXT:    movdqa %xmm11, %xmm4
 ; SSE2-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm10[8],xmm4[9],xmm10[9],xmm4[10],xmm10[10],xmm4[11],xmm10[11],xmm4[12],xmm10[12],xmm4[13],xmm10[13],xmm4[14],xmm10[14],xmm4[15],xmm10[15]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,3,0,1]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3],xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm6 = xmm11[2,3,0,1]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[3,2,1,0,4,5,6,7]
 ; SSE2-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,7,6,5,4]
 ; SSE2-NEXT:    packuswb %xmm4, %xmm6
 ; SSE2-NEXT:    movdqa %xmm6, %xmm4
 ; SSE2-NEXT:    psllw $4, %xmm4
 ; SSE2-NEXT:    psrlw $4, %xmm6
 ; SSE2-NEXT:    pand %xmm3, %xmm6
 ; SSE2-NEXT:    pandn %xmm4, %xmm3
 ; SSE2-NEXT:    por %xmm6, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm5
 ; SSE2-NEXT:    psllw $2, %xmm5
 ; SSE2-NEXT:    pand %xmm8, %xmm3
 ; SSE2-NEXT:    psrlw $2, %xmm3
 ; SSE2-NEXT:    por %xmm5, %xmm3
 ; SSE2-NEXT:    pand %xmm3, %xmm7
-; SSE2-NEXT:    paddb %xmm7, %xmm7
+; SSE2-NEXT:    psrlw $1, %xmm7
 ; SSE2-NEXT:    pand %xmm9, %xmm3
-; SSE2-NEXT:    psrlw $1, %xmm3
+; SSE2-NEXT:    paddb %xmm3, %xmm3
 ; SSE2-NEXT:    por %xmm7, %xmm3
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: test_bitreverse_v8i64:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm5
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm8 = [7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8]

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

 ; SSE2-NEXT:    por %xmm2, %xmm4
 ; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
 ; SSE2-NEXT:    pand %xmm2, %xmm4
 ; SSE2-NEXT:    pandn %xmm3, %xmm2
 ; SSE2-NEXT:    por %xmm4, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    paddb %xmm2, %xmm3
+; SSE2-NEXT:    psrlw $7, %xmm3
+; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-NEXT:    psrlw $7, %xmm4
-; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4
+; SSE2-NEXT:    paddb %xmm2, %xmm4
 ; SSE2-NEXT:    por %xmm3, %xmm4
 ; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm0
 ; SSE2-NEXT:    pand %xmm0, %xmm4
 ; SSE2-NEXT:    pandn %xmm2, %xmm0
 ; SSE2-NEXT:    por %xmm4, %xmm0
 ; SSE2-NEXT:    retq
 ;

 ; SSE41-NEXT:    movdqa %xmm1, %xmm3
 ; SSE41-NEXT:    psllw $2, %xmm3
 ; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
 ; SSE41-NEXT:    por %xmm0, %xmm3
 ; SSE41-NEXT:    paddb %xmm2, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    paddb %xmm1, %xmm0
+; SSE41-NEXT:    psrlw $7, %xmm0
+; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    psrlw $7, %xmm3
-; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
+; SSE41-NEXT:    paddb %xmm1, %xmm3
 ; SSE41-NEXT:    por %xmm0, %xmm3
 ; SSE41-NEXT:    paddb %xmm2, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: var_funnnel_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm2
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
 ; AVX-NEXT:    vpsllw $4, %xmm0, %xmm3
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpor %xmm2, %xmm3, %xmm2
 ; AVX-NEXT:    vpsllw $5, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $6, %xmm0, %xmm2
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
 ; AVX-NEXT:    vpsllw $2, %xmm0, %xmm3
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpor %xmm2, %xmm3, %xmm2
 ; AVX-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm2
-; AVX-NEXT:    vpsrlw $7, %xmm0, %xmm3
-; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
-; AVX-NEXT:    vpor %xmm3, %xmm2, %xmm2
+; AVX-NEXT:    vpsrlw $7, %xmm0, %xmm2
+; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
+; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm3
+; AVX-NEXT:    vpor %xmm2, %xmm3, %xmm2
 ; AVX-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512F-NEXT:    vpand %xmm2, %xmm1, %xmm3

 ; X86-SSE2-NEXT:    por %xmm2, %xmm4
 ; X86-SSE2-NEXT:    paddb %xmm1, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm2
 ; X86-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
 ; X86-SSE2-NEXT:    pand %xmm2, %xmm4
 ; X86-SSE2-NEXT:    pandn %xmm3, %xmm2
 ; X86-SSE2-NEXT:    por %xmm4, %xmm2
 ; X86-SSE2-NEXT:    movdqa %xmm2, %xmm3
-; X86-SSE2-NEXT:    paddb %xmm2, %xmm3
+; X86-SSE2-NEXT:    psrlw $7, %xmm3
+; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm3
 ; X86-SSE2-NEXT:    movdqa %xmm2, %xmm4
-; X86-SSE2-NEXT:    psrlw $7, %xmm4
-; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm4
+; X86-SSE2-NEXT:    paddb %xmm2, %xmm4
 ; X86-SSE2-NEXT:    por %xmm3, %xmm4
 ; X86-SSE2-NEXT:    paddb %xmm1, %xmm1
 ; X86-SSE2-NEXT:    pcmpgtb %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pand %xmm0, %xmm4
 ; X86-SSE2-NEXT:    pandn %xmm2, %xmm0
 ; X86-SSE2-NEXT:    por %xmm4, %xmm0
 ; X86-SSE2-NEXT:    retl
   %res = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %x, <16 x i8> %x, <16 x i8> %amt)

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

 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrlw $6, %ymm0, %ymm2
 ; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX2-NEXT:    vpsllw $2, %ymm0, %ymm3
 ; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
 ; AVX2-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
-; AVX2-NEXT:    vpsrlw $7, %ymm0, %ymm3
-; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
-; AVX2-NEXT:    vpor %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpsrlw $7, %ymm0, %ymm2
+; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
+; AVX2-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
 ; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpsllw $5, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw $6, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
 ; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
-; AVX512F-NEXT:    vpsrlw $7, %ymm0, %ymm3
-; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
-; AVX512F-NEXT:    vpor %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsrlw $7, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
+; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm3

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

 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpsubb %zmm1, %zmm2, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX512BW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsllw $5, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm4
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm3
+; AVX512BW-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
 ; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vpmovb2m %zmm2, %k2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k2}
-; AVX512BW-NEXT:    vpsrlw $2, %zmm2, %zmm5
-; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
-; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
-; AVX512BW-NEXT:    vpsrlw $1, %zmm2, %zmm5
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k2
+; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm3
+; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k2}
+; AVX512BW-NEXT:    vpsllw $2, %zmm3, %zmm5
 ; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
+; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
-; AVX512BW-NEXT:    vpandq %zmm3, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
+; AVX512BW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512BW-NEXT:    vpsubb %zmm1, %zmm4, %zmm1
+; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsllw $5, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm3
-; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm2
+; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k2
-; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm1
 ; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
-; AVX512BW-NEXT:    vpsllw $2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsrlw $2, %zmm0, %zmm1
 ; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm1
-; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512BW-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VLBW-NEXT:    vpsubb %zmm1, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsllw $5, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm4
+; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm1, %zmm3
+; AVX512VLBW-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k2
-; AVX512VLBW-NEXT:    vpsrlw $4, %zmm0, %zmm2
-; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k2}
-; AVX512VLBW-NEXT:    vpsrlw $2, %zmm2, %zmm5
-; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
-; AVX512VLBW-NEXT:    vpsrlw $1, %zmm2, %zmm5
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k2
+; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm3
+; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k2}
+; AVX512VLBW-NEXT:    vpsllw $2, %zmm3, %zmm5
 ; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
-; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
+; AVX512VLBW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512VLBW-NEXT:    vpsubb %zmm1, %zmm4, %zmm1
+; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpsllw $5, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm2
+; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k2
-; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsrlw $4, %zmm0, %zmm1
 ; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
-; AVX512VLBW-NEXT:    vpsllw $2, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsrlw $2, %zmm0, %zmm1
 ; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm1
-; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VLBW-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VBMI2-NEXT:    vpsubb %zmm1, %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpsllw $5, %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm4
+; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm3
+; AVX512VBMI2-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k2
-; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm2
-; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k2}
-; AVX512VBMI2-NEXT:    vpsrlw $2, %zmm2, %zmm5
-; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
-; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm2, %zmm5
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k2
+; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
+; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k2}
+; AVX512VBMI2-NEXT:    vpsllw $2, %zmm3, %zmm5
 ; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
 ; AVX512VBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
-; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
+; AVX512VBMI2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512VBMI2-NEXT:    vpsubb %zmm1, %zmm4, %zmm1
+; AVX512VBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vpsllw $5, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpaddb %zmm1, %zmm1, %zmm3
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VBMI2-NEXT:    vpaddb %zmm1, %zmm1, %zmm2
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm1, %k2
-; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm1
+; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
-; AVX512VBMI2-NEXT:    vpsllw $2, %zmm0, %zmm1
+; AVX512VBMI2-NEXT:    vpsrlw $2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm1
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512VLVBMI2-NEXT:    vpsubb %zmm1, %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm4
+; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm3
+; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k2
-; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm2
-; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k2}
-; AVX512VLVBMI2-NEXT:    vpsrlw $2, %zmm2, %zmm5
-; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
-; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm2, %zmm5
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k2
+; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
+; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k2}
+; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm3, %zmm5
 ; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
-; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
+; AVX512VLVBMI2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512VLVBMI2-NEXT:    vpsubb %zmm1, %zmm4, %zmm1
+; AVX512VLVBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm1, %zmm1, %zmm3
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm1, %zmm1, %zmm2
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm1, %k2
-; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm1
+; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
-; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm0, %zmm1
+; AVX512VLVBMI2-NEXT:    vpsrlw $2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm1
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VLVBMI2-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vporq %zmm0, %zmm3, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %x, <64 x i8> %amt)
   ret <64 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts
 ;

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

 ; X86-SSE2-NEXT:    retl
   %res = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %x, <8 x i16> %y, <8 x i16> %amt)
   ret <8 x i16> %res
 }
 
 define <16 x i8> @var_funnnel_v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %amt) nounwind {
 ; SSE2-LABEL: var_funnnel_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; SSE2-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-NEXT:    pandn %xmm5, %xmm4
-; SSE2-NEXT:    psllw $5, %xmm4
+; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; SSE2-NEXT:    movdqa %xmm2, %xmm5
+; SSE2-NEXT:    pand %xmm4, %xmm5
+; SSE2-NEXT:    psllw $5, %xmm5
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
-; SSE2-NEXT:    paddb %xmm0, %xmm0
+; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
 ; SSE2-NEXT:    movdqa %xmm6, %xmm7
-; SSE2-NEXT:    pandn %xmm0, %xmm7
-; SSE2-NEXT:    psllw $4, %xmm0
-; SSE2-NEXT:    pand %xmm6, %xmm0
-; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT:    por %xmm7, %xmm0
-; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    pandn %xmm1, %xmm7
+; SSE2-NEXT:    psrlw $4, %xmm1
+; SSE2-NEXT:    pand %xmm6, %xmm1
+; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE2-NEXT:    por %xmm7, %xmm1
+; SSE2-NEXT:    paddb %xmm5, %xmm5
 ; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
+; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
 ; SSE2-NEXT:    movdqa %xmm6, %xmm7
-; SSE2-NEXT:    pandn %xmm0, %xmm7
-; SSE2-NEXT:    psllw $2, %xmm0
-; SSE2-NEXT:    pand %xmm6, %xmm0
-; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; SSE2-NEXT:    por %xmm7, %xmm0
-; SSE2-NEXT:    paddb %xmm4, %xmm4
+; SSE2-NEXT:    pandn %xmm1, %xmm7
+; SSE2-NEXT:    psrlw $2, %xmm1
+; SSE2-NEXT:    pand %xmm6, %xmm1
+; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
+; SSE2-NEXT:    por %xmm7, %xmm1
+; SSE2-NEXT:    paddb %xmm5, %xmm5
 ; SSE2-NEXT:    pxor %xmm6, %xmm6
-; SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
-; SSE2-NEXT:    movdqa %xmm6, %xmm4
-; SSE2-NEXT:    pandn %xmm0, %xmm4
-; SSE2-NEXT:    paddb %xmm0, %xmm0
-; SSE2-NEXT:    pand %xmm6, %xmm0
-; SSE2-NEXT:    pand %xmm5, %xmm2
-; SSE2-NEXT:    psllw $5, %xmm2
-; SSE2-NEXT:    pxor %xmm5, %xmm5
-; SSE2-NEXT:    pcmpgtb %xmm2, %xmm5
-; SSE2-NEXT:    movdqa %xmm5, %xmm6
-; SSE2-NEXT:    pandn %xmm1, %xmm6
-; SSE2-NEXT:    psrlw $4, %xmm1
-; SSE2-NEXT:    pand %xmm5, %xmm1
+; SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
+; SSE2-NEXT:    movdqa %xmm6, %xmm5
+; SSE2-NEXT:    pandn %xmm1, %xmm5
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    pand %xmm6, %xmm1
 ; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT:    por %xmm6, %xmm1
+; SSE2-NEXT:    por %xmm5, %xmm1
+; SSE2-NEXT:    pandn %xmm4, %xmm2
+; SSE2-NEXT:    psllw $5, %xmm2
+; SSE2-NEXT:    pxor %xmm4, %xmm4
+; SSE2-NEXT:    pcmpgtb %xmm2, %xmm4
+; SSE2-NEXT:    paddb %xmm0, %xmm0
+; SSE2-NEXT:    movdqa %xmm4, %xmm5
+; SSE2-NEXT:    pandn %xmm0, %xmm5
+; SSE2-NEXT:    psllw $4, %xmm0
+; SSE2-NEXT:    pand %xmm4, %xmm0
+; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT:    por %xmm5, %xmm0
 ; SSE2-NEXT:    paddb %xmm2, %xmm2
-; SSE2-NEXT:    pxor %xmm5, %xmm5
-; SSE2-NEXT:    pcmpgtb %xmm2, %xmm5
-; SSE2-NEXT:    movdqa %xmm5, %xmm6
-; SSE2-NEXT:    pandn %xmm1, %xmm6
-; SSE2-NEXT:    psrlw $2, %xmm1
-; SSE2-NEXT:    pand %xmm5, %xmm1
-; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT:    por %xmm6, %xmm1
+; SSE2-NEXT:    pxor %xmm4, %xmm4
+; SSE2-NEXT:    pcmpgtb %xmm2, %xmm4
+; SSE2-NEXT:    movdqa %xmm4, %xmm5
+; SSE2-NEXT:    pandn %xmm0, %xmm5
+; SSE2-NEXT:    psllw $2, %xmm0
+; SSE2-NEXT:    pand %xmm4, %xmm0
+; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; SSE2-NEXT:    por %xmm5, %xmm0
 ; SSE2-NEXT:    paddb %xmm2, %xmm2
 ; SSE2-NEXT:    pcmpgtb %xmm2, %xmm3
 ; SSE2-NEXT:    movdqa %xmm3, %xmm2
-; SSE2-NEXT:    pandn %xmm1, %xmm2
-; SSE2-NEXT:    psrlw $1, %xmm1
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
-; SSE2-NEXT:    por %xmm2, %xmm1
-; SSE2-NEXT:    por %xmm4, %xmm1
-; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    pandn %xmm0, %xmm2
+; SSE2-NEXT:    por %xmm1, %xmm2
+; SSE2-NEXT:    paddb %xmm0, %xmm0
+; SSE2-NEXT:    pand %xmm3, %xmm0
+; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_funnnel_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm2, %xmm3
 ; SSE41-NEXT:    movdqa %xmm0, %xmm2
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; SSE41-NEXT:    movdqa %xmm3, %xmm0

 ; AVX512VLVBMI2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512VLVBMI2-NEXT:    vzeroupper
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOP-LABEL: var_funnnel_v16i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vmovdqa {{.*#+}} xmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; XOP-NEXT:    vpandn %xmm3, %xmm2, %xmm4
+; XOP-NEXT:    vpand %xmm3, %xmm2, %xmm4
+; XOP-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; XOP-NEXT:    vpsubb %xmm4, %xmm5, %xmm4
+; XOP-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
+; XOP-NEXT:    vpandn %xmm3, %xmm2, %xmm2
 ; XOP-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
-; XOP-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
-; XOP-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; XOP-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; XOP-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
-; XOP-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
+; XOP-NEXT:    vpshlb %xmm2, %xmm0, %xmm0
 ; XOP-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X86-SSE2-LABEL: var_funnnel_v16i8:
 ; X86-SSE2:       # %bb.0:
-; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm5 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; X86-SSE2-NEXT:    movdqa %xmm2, %xmm4
-; X86-SSE2-NEXT:    pandn %xmm5, %xmm4
-; X86-SSE2-NEXT:    psllw $5, %xmm4
+; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; X86-SSE2-NEXT:    movdqa %xmm2, %xmm5
+; X86-SSE2-NEXT:    pand %xmm4, %xmm5
+; X86-SSE2-NEXT:    psllw $5, %xmm5
 ; X86-SSE2-NEXT:    pxor %xmm3, %xmm3
 ; X86-SSE2-NEXT:    pxor %xmm6, %xmm6
-; X86-SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
-; X86-SSE2-NEXT:    paddb %xmm0, %xmm0
+; X86-SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
 ; X86-SSE2-NEXT:    movdqa %xmm6, %xmm7
-; X86-SSE2-NEXT:    pandn %xmm0, %xmm7
-; X86-SSE2-NEXT:    psllw $4, %xmm0
-; X86-SSE2-NEXT:    pand %xmm6, %xmm0
-; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE2-NEXT:    por %xmm7, %xmm0
-; X86-SSE2-NEXT:    paddb %xmm4, %xmm4
+; X86-SSE2-NEXT:    pandn %xmm1, %xmm7
+; X86-SSE2-NEXT:    psrlw $4, %xmm1
+; X86-SSE2-NEXT:    pand %xmm6, %xmm1
+; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
+; X86-SSE2-NEXT:    por %xmm7, %xmm1
+; X86-SSE2-NEXT:    paddb %xmm5, %xmm5
 ; X86-SSE2-NEXT:    pxor %xmm6, %xmm6
-; X86-SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
+; X86-SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
 ; X86-SSE2-NEXT:    movdqa %xmm6, %xmm7
-; X86-SSE2-NEXT:    pandn %xmm0, %xmm7
-; X86-SSE2-NEXT:    psllw $2, %xmm0
-; X86-SSE2-NEXT:    pand %xmm6, %xmm0
-; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
-; X86-SSE2-NEXT:    por %xmm7, %xmm0
-; X86-SSE2-NEXT:    paddb %xmm4, %xmm4
+; X86-SSE2-NEXT:    pandn %xmm1, %xmm7
+; X86-SSE2-NEXT:    psrlw $2, %xmm1
+; X86-SSE2-NEXT:    pand %xmm6, %xmm1
+; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
+; X86-SSE2-NEXT:    por %xmm7, %xmm1
+; X86-SSE2-NEXT:    paddb %xmm5, %xmm5
 ; X86-SSE2-NEXT:    pxor %xmm6, %xmm6
-; X86-SSE2-NEXT:    pcmpgtb %xmm4, %xmm6
-; X86-SSE2-NEXT:    movdqa %xmm6, %xmm4
-; X86-SSE2-NEXT:    pandn %xmm0, %xmm4
-; X86-SSE2-NEXT:    paddb %xmm0, %xmm0
-; X86-SSE2-NEXT:    pand %xmm6, %xmm0
-; X86-SSE2-NEXT:    pand %xmm5, %xmm2
-; X86-SSE2-NEXT:    psllw $5, %xmm2
-; X86-SSE2-NEXT:    pxor %xmm5, %xmm5
-; X86-SSE2-NEXT:    pcmpgtb %xmm2, %xmm5
-; X86-SSE2-NEXT:    movdqa %xmm5, %xmm6
-; X86-SSE2-NEXT:    pandn %xmm1, %xmm6
-; X86-SSE2-NEXT:    psrlw $4, %xmm1
-; X86-SSE2-NEXT:    pand %xmm5, %xmm1
+; X86-SSE2-NEXT:    pcmpgtb %xmm5, %xmm6
+; X86-SSE2-NEXT:    movdqa %xmm6, %xmm5
+; X86-SSE2-NEXT:    pandn %xmm1, %xmm5
+; X86-SSE2-NEXT:    psrlw $1, %xmm1
+; X86-SSE2-NEXT:    pand %xmm6, %xmm1
 ; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
-; X86-SSE2-NEXT:    por %xmm6, %xmm1
+; X86-SSE2-NEXT:    por %xmm5, %xmm1
+; X86-SSE2-NEXT:    pandn %xmm4, %xmm2
+; X86-SSE2-NEXT:    psllw $5, %xmm2
+; X86-SSE2-NEXT:    pxor %xmm4, %xmm4
+; X86-SSE2-NEXT:    pcmpgtb %xmm2, %xmm4
+; X86-SSE2-NEXT:    paddb %xmm0, %xmm0
+; X86-SSE2-NEXT:    movdqa %xmm4, %xmm5
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm5
+; X86-SSE2-NEXT:    psllw $4, %xmm0
+; X86-SSE2-NEXT:    pand %xmm4, %xmm0
+; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
+; X86-SSE2-NEXT:    por %xmm5, %xmm0
 ; X86-SSE2-NEXT:    paddb %xmm2, %xmm2
-; X86-SSE2-NEXT:    pxor %xmm5, %xmm5
-; X86-SSE2-NEXT:    pcmpgtb %xmm2, %xmm5
-; X86-SSE2-NEXT:    movdqa %xmm5, %xmm6
-; X86-SSE2-NEXT:    pandn %xmm1, %xmm6
-; X86-SSE2-NEXT:    psrlw $2, %xmm1
-; X86-SSE2-NEXT:    pand %xmm5, %xmm1
-; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
-; X86-SSE2-NEXT:    por %xmm6, %xmm1
+; X86-SSE2-NEXT:    pxor %xmm4, %xmm4
+; X86-SSE2-NEXT:    pcmpgtb %xmm2, %xmm4
+; X86-SSE2-NEXT:    movdqa %xmm4, %xmm5
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm5
+; X86-SSE2-NEXT:    psllw $2, %xmm0
+; X86-SSE2-NEXT:    pand %xmm4, %xmm0
+; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
+; X86-SSE2-NEXT:    por %xmm5, %xmm0
 ; X86-SSE2-NEXT:    paddb %xmm2, %xmm2
 ; X86-SSE2-NEXT:    pcmpgtb %xmm2, %xmm3
 ; X86-SSE2-NEXT:    movdqa %xmm3, %xmm2
-; X86-SSE2-NEXT:    pandn %xmm1, %xmm2
-; X86-SSE2-NEXT:    psrlw $1, %xmm1
-; X86-SSE2-NEXT:    pand %xmm3, %xmm1
-; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
-; X86-SSE2-NEXT:    por %xmm2, %xmm1
-; X86-SSE2-NEXT:    por %xmm4, %xmm1
-; X86-SSE2-NEXT:    por %xmm1, %xmm0
+; X86-SSE2-NEXT:    pandn %xmm0, %xmm2
+; X86-SSE2-NEXT:    por %xmm1, %xmm2
+; X86-SSE2-NEXT:    paddb %xmm0, %xmm0
+; X86-SSE2-NEXT:    pand %xmm3, %xmm0
+; X86-SSE2-NEXT:    por %xmm2, %xmm0
 ; X86-SSE2-NEXT:    retl
   %res = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %x, <16 x i8> %y, <16 x i8> %amt)
   ret <16 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts
 ;

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

 ; AVX512VLVBMI2-NEXT:    vpsllvw %zmm2, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VLVBMI2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_funnnel_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vmovaps {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; XOPAVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm4
+; XOPAVX1-NEXT:    vandps %ymm3, %ymm2, %ymm4
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm4, %xmm5
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; XOPAVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm6
-; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm6, %xmm5
-; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm0, %xmm0
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm0, %ymm0
-; XOPAVX1-NEXT:    vandps %ymm3, %ymm2, %ymm2
+; XOPAVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
+; XOPAVX1-NEXT:    vpsubb %xmm5, %xmm6, %xmm5
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm7
+; XOPAVX1-NEXT:    vpshlb %xmm5, %xmm7, %xmm5
+; XOPAVX1-NEXT:    vpsubb %xmm4, %xmm6, %xmm4
+; XOPAVX1-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm5, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vandnps %ymm3, %ymm2, %ymm2
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
-; XOPAVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; XOPAVX1-NEXT:    vpsubb %xmm3, %xmm4, %xmm3
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
-; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm5, %xmm3
-; XOPAVX1-NEXT:    vpsubb %xmm2, %xmm4, %xmm2
-; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
-; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; XOPAVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
+; XOPAVX1-NEXT:    vpshlb %xmm3, %xmm4, %xmm3
+; XOPAVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm0
+; XOPAVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    vorps %ymm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: var_funnnel_v32i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; XOPAVX2-NEXT:    vpandn %ymm3, %ymm2, %ymm4
+; XOPAVX2-NEXT:    vpand %ymm3, %ymm2, %ymm4
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm4, %xmm5
-; XOPAVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
-; 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:    vpand %ymm3, %ymm2, %ymm2
+; XOPAVX2-NEXT:    vpxor %xmm6, %xmm6, %xmm6
+; XOPAVX2-NEXT:    vpsubb %xmm5, %xmm6, %xmm5
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm1, %xmm7
+; XOPAVX2-NEXT:    vpshlb %xmm5, %xmm7, %xmm5
+; XOPAVX2-NEXT:    vpsubb %xmm4, %xmm6, %xmm4
+; XOPAVX2-NEXT:    vpshlb %xmm4, %xmm1, %xmm1
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm5, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpandn %ymm3, %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, %xmm2
-; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
-; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
+; XOPAVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm0
+; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm4
+; XOPAVX2-NEXT:    vpshlb %xmm3, %xmm4, %xmm3
+; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm0, %xmm0
+; XOPAVX2-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
   %res = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %x, <32 x i8> %y, <32 x i8> %amt)
   ret <32 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts

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

 ; SSE2-NEXT:    por %xmm2, %xmm4
 ; SSE2-NEXT:    paddb %xmm3, %xmm3
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
 ; SSE2-NEXT:    pcmpgtb %xmm3, %xmm2
 ; SSE2-NEXT:    pand %xmm2, %xmm4
 ; SSE2-NEXT:    pandn %xmm1, %xmm2
 ; SSE2-NEXT:    por %xmm4, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm1
-; SSE2-NEXT:    paddb %xmm2, %xmm1
+; SSE2-NEXT:    psrlw $7, %xmm1
+; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-NEXT:    psrlw $7, %xmm4
-; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4
+; SSE2-NEXT:    paddb %xmm2, %xmm4
 ; SSE2-NEXT:    por %xmm1, %xmm4
 ; SSE2-NEXT:    paddb %xmm3, %xmm3
 ; SSE2-NEXT:    pcmpgtb %xmm3, %xmm0
 ; SSE2-NEXT:    pand %xmm0, %xmm4
 ; SSE2-NEXT:    pandn %xmm2, %xmm0
 ; SSE2-NEXT:    por %xmm4, %xmm0
 ; SSE2-NEXT:    retq
 ;

 ; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
 ; SSE41-NEXT:    movdqa %xmm2, %xmm3
 ; SSE41-NEXT:    psllw $2, %xmm3
 ; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
 ; SSE41-NEXT:    por %xmm1, %xmm3
 ; SSE41-NEXT:    paddb %xmm0, %xmm0
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm1
-; SSE41-NEXT:    paddb %xmm2, %xmm1
+; SSE41-NEXT:    psrlw $7, %xmm1
+; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1
 ; SSE41-NEXT:    movdqa %xmm2, %xmm3
-; SSE41-NEXT:    psrlw $7, %xmm3
-; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
+; SSE41-NEXT:    paddb %xmm2, %xmm3
 ; SSE41-NEXT:    por %xmm1, %xmm3
 ; SSE41-NEXT:    paddb %xmm0, %xmm0
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: var_funnnel_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm2
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
 ; AVX-NEXT:    vpsllw $4, %xmm0, %xmm3
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpor %xmm2, %xmm3, %xmm2
 ; AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
 ; AVX-NEXT:    vpsllw $5, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $6, %xmm0, %xmm2
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
 ; AVX-NEXT:    vpsllw $2, %xmm0, %xmm3
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpor %xmm2, %xmm3, %xmm2
 ; AVX-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm2
-; AVX-NEXT:    vpsrlw $7, %xmm0, %xmm3
-; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
-; AVX-NEXT:    vpor %xmm3, %xmm2, %xmm2
+; AVX-NEXT:    vpsrlw $7, %xmm0, %xmm2
+; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
+; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm3
+; AVX-NEXT:    vpor %xmm2, %xmm3, %xmm2
 ; AVX-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v16i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512F-NEXT:    vpand %xmm2, %xmm1, %xmm3

 ; X86-SSE2-NEXT:    por %xmm2, %xmm4
 ; X86-SSE2-NEXT:    paddb %xmm3, %xmm3
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm2
 ; X86-SSE2-NEXT:    pcmpgtb %xmm3, %xmm2
 ; X86-SSE2-NEXT:    pand %xmm2, %xmm4
 ; X86-SSE2-NEXT:    pandn %xmm1, %xmm2
 ; X86-SSE2-NEXT:    por %xmm4, %xmm2
 ; X86-SSE2-NEXT:    movdqa %xmm2, %xmm1
-; X86-SSE2-NEXT:    paddb %xmm2, %xmm1
+; X86-SSE2-NEXT:    psrlw $7, %xmm1
+; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
 ; X86-SSE2-NEXT:    movdqa %xmm2, %xmm4
-; X86-SSE2-NEXT:    psrlw $7, %xmm4
-; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm4
+; X86-SSE2-NEXT:    paddb %xmm2, %xmm4
 ; X86-SSE2-NEXT:    por %xmm1, %xmm4
 ; X86-SSE2-NEXT:    paddb %xmm3, %xmm3
 ; X86-SSE2-NEXT:    pcmpgtb %xmm3, %xmm0
 ; X86-SSE2-NEXT:    pand %xmm0, %xmm4
 ; X86-SSE2-NEXT:    pandn %xmm2, %xmm0
 ; X86-SSE2-NEXT:    por %xmm4, %xmm0
 ; X86-SSE2-NEXT:    retl
   %res = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %x, <16 x i8> %x, <16 x i8> %amt)

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

 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrlw $6, %ymm0, %ymm2
 ; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX2-NEXT:    vpsllw $2, %ymm0, %ymm3
 ; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
 ; AVX2-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
-; AVX2-NEXT:    vpsrlw $7, %ymm0, %ymm3
-; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
-; AVX2-NEXT:    vpor %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpsrlw $7, %ymm0, %ymm2
+; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
+; AVX2-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
 ; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpsubb %ymm1, %ymm3, %ymm1
 ; AVX512F-NEXT:    vpsllw $5, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw $6, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
 ; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
-; AVX512F-NEXT:    vpsrlw $7, %ymm0, %ymm3
-; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
-; AVX512F-NEXT:    vpor %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsrlw $7, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
+; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm3

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

 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vporq %zmm3, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm3
-; AVX512BW-NEXT:    vpsllw $5, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
+; AVX512BW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpsubb %zmm1, %zmm2, %zmm2
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm4
 ; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vpmovb2m %zmm3, %k2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm3
-; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
-; AVX512BW-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k2}
-; AVX512BW-NEXT:    vpsrlw $2, %zmm3, %zmm5
-; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
-; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512BW-NEXT:    vpsrlw $1, %zmm3, %zmm5
+; AVX512BW-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k2}
+; AVX512BW-NEXT:    vpsllw $2, %zmm2, %zmm5
 ; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
+; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512BW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512BW-NEXT:    vpsubb %zmm1, %zmm4, %zmm1
-; AVX512BW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsllw $5, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm2
-; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm3
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k2
-; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm1
 ; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
-; AVX512BW-NEXT:    vpsllw $2, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpsrlw $2, %zmm0, %zmm1
 ; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm1
-; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512BW-NEXT:    vporq %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512BW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512BW-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: var_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm1, %zmm3
-; AVX512VLBW-NEXT:    vpsllw $5, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
+; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpsubb %zmm1, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm4
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k2
-; AVX512VLBW-NEXT:    vpsrlw $4, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k2}
-; AVX512VLBW-NEXT:    vpsrlw $2, %zmm3, %zmm5
-; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VLBW-NEXT:    vpsrlw $1, %zmm3, %zmm5
+; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k2}
+; AVX512VLBW-NEXT:    vpsllw $2, %zmm2, %zmm5
 ; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VLBW-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBW-NEXT:    vpsubb %zmm1, %zmm4, %zmm1
-; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
+; AVX512VLBW-NEXT:    vpandq %zmm3, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpsllw $5, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm2
-; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm3
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k2
-; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsrlw $4, %zmm0, %zmm1
 ; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
-; AVX512VLBW-NEXT:    vpsllw $2, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpsrlw $2, %zmm0, %zmm1
 ; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm1
-; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VLBW-NEXT:    vporq %zmm0, %zmm3, %zmm0
+; AVX512VLBW-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512VLBW-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: var_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
-; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX512VBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm3
-; AVX512VBMI2-NEXT:    vpsllw $5, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
+; AVX512VBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpsubb %zmm1, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm4
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k2
-; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm3
-; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k2}
-; AVX512VBMI2-NEXT:    vpsrlw $2, %zmm3, %zmm5
-; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm3, %zmm5
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k2}
+; AVX512VBMI2-NEXT:    vpsllw $2, %zmm2, %zmm5
 ; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
 ; AVX512VBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VBMI2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VBMI2-NEXT:    vpsubb %zmm1, %zmm4, %zmm1
-; AVX512VBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
+; AVX512VBMI2-NEXT:    vpandq %zmm3, %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vpsllw $5, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpaddb %zmm1, %zmm1, %zmm2
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VBMI2-NEXT:    vpaddb %zmm1, %zmm1, %zmm3
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm1, %k2
-; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm1
+; AVX512VBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
-; AVX512VBMI2-NEXT:    vpsllw $2, %zmm0, %zmm1
+; AVX512VBMI2-NEXT:    vpsrlw $2, %zmm0, %zmm1
 ; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm1
-; AVX512VBMI2-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    vporq %zmm0, %zmm3, %zmm0
+; AVX512VBMI2-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512VBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VBMI2-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: var_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
-; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
-; AVX512VLVBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm3
-; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm4
+; AVX512VLVBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpsubb %zmm1, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
+; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k2
-; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm3
-; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k2}
-; AVX512VLVBMI2-NEXT:    vpsrlw $2, %zmm3, %zmm5
-; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm3, %zmm5
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k2
+; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm2
+; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k2}
+; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm2, %zmm5
 ; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm5
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm2 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm4, %zmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm5, %zmm3 {%k1}
-; AVX512VLVBMI2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLVBMI2-NEXT:    vpsubb %zmm1, %zmm4, %zmm1
-; AVX512VLVBMI2-NEXT:    vpandq %zmm2, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
+; AVX512VLVBMI2-NEXT:    vpandq %zmm3, %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpsllw $5, %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm1, %zmm1, %zmm2
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm1, %zmm1, %zmm3
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm1, %k2
-; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm1
+; AVX512VLVBMI2-NEXT:    vpsrlw $4, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
-; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm0, %zmm1
+; AVX512VLVBMI2-NEXT:    vpsrlw $2, %zmm0, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm1
-; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VLVBMI2-NEXT:    vporq %zmm0, %zmm3, %zmm0
+; AVX512VLVBMI2-NEXT:    vpsrlw $1, %zmm0, %zmm1
+; AVX512VLVBMI2-NEXT:    vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm3, %k1
+; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512VLVBMI2-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %x, <64 x i8> %amt)
   ret <64 x i8> %res
 }
 
 ;
 ; Uniform Variable Shifts
 ;

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

 ; SSE2-NEXT:    por %xmm2, %xmm4
 ; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
 ; SSE2-NEXT:    pand %xmm2, %xmm4
 ; SSE2-NEXT:    pandn %xmm3, %xmm2
 ; SSE2-NEXT:    por %xmm4, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    paddb %xmm2, %xmm3
+; SSE2-NEXT:    psrlw $7, %xmm3
+; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
 ; SSE2-NEXT:    movdqa %xmm2, %xmm4
-; SSE2-NEXT:    psrlw $7, %xmm4
-; SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm4
+; SSE2-NEXT:    paddb %xmm2, %xmm4
 ; SSE2-NEXT:    por %xmm3, %xmm4
 ; SSE2-NEXT:    paddb %xmm1, %xmm1
 ; SSE2-NEXT:    pcmpgtb %xmm1, %xmm0
 ; SSE2-NEXT:    pand %xmm0, %xmm4
 ; SSE2-NEXT:    pandn %xmm2, %xmm0
 ; SSE2-NEXT:    por %xmm4, %xmm0
 ; SSE2-NEXT:    retq
 ;

 ; SSE41-NEXT:    movdqa %xmm1, %xmm3
 ; SSE41-NEXT:    psllw $2, %xmm3
 ; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
 ; SSE41-NEXT:    por %xmm0, %xmm3
 ; SSE41-NEXT:    paddb %xmm2, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    paddb %xmm1, %xmm0
+; SSE41-NEXT:    psrlw $7, %xmm0
+; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
 ; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    psrlw $7, %xmm3
-; SSE41-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3
+; SSE41-NEXT:    paddb %xmm1, %xmm3
 ; SSE41-NEXT:    por %xmm0, %xmm3
 ; SSE41-NEXT:    paddb %xmm2, %xmm2
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: var_rotate_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm2
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
 ; AVX-NEXT:    vpsllw $4, %xmm0, %xmm3
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpor %xmm2, %xmm3, %xmm2
 ; AVX-NEXT:    vpsllw $5, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $6, %xmm0, %xmm2
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
 ; AVX-NEXT:    vpsllw $2, %xmm0, %xmm3
 ; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpor %xmm2, %xmm3, %xmm2
 ; AVX-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm2
-; AVX-NEXT:    vpsrlw $7, %xmm0, %xmm3
-; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm3, %xmm3
-; AVX-NEXT:    vpor %xmm3, %xmm2, %xmm2
+; AVX-NEXT:    vpsrlw $7, %xmm0, %xmm2
+; AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
+; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm3
+; AVX-NEXT:    vpor %xmm2, %xmm3, %xmm2
 ; AVX-NEXT:    vpaddb %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpblendvb %xmm1, %xmm2, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_rotate_v16i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
 ; AVX512F-NEXT:    vpsubb %xmm1, %xmm2, %xmm2

 ; X86-SSE2-NEXT:    por %xmm2, %xmm4
 ; X86-SSE2-NEXT:    paddb %xmm1, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm2
 ; X86-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2
 ; X86-SSE2-NEXT:    pand %xmm2, %xmm4
 ; X86-SSE2-NEXT:    pandn %xmm3, %xmm2
 ; X86-SSE2-NEXT:    por %xmm4, %xmm2
 ; X86-SSE2-NEXT:    movdqa %xmm2, %xmm3
-; X86-SSE2-NEXT:    paddb %xmm2, %xmm3
+; X86-SSE2-NEXT:    psrlw $7, %xmm3
+; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm3
 ; X86-SSE2-NEXT:    movdqa %xmm2, %xmm4
-; X86-SSE2-NEXT:    psrlw $7, %xmm4
-; X86-SSE2-NEXT:    pand {{\.?LCPI[0-9]+_[0-9]+}}, %xmm4
+; X86-SSE2-NEXT:    paddb %xmm2, %xmm4
 ; X86-SSE2-NEXT:    por %xmm3, %xmm4
 ; X86-SSE2-NEXT:    paddb %xmm1, %xmm1
 ; X86-SSE2-NEXT:    pcmpgtb %xmm1, %xmm0
 ; X86-SSE2-NEXT:    pand %xmm0, %xmm4
 ; X86-SSE2-NEXT:    pandn %xmm2, %xmm0
 ; X86-SSE2-NEXT:    por %xmm4, %xmm0
 ; X86-SSE2-NEXT:    retl
   %b8 = sub <16 x i8> <i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8>, %b

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

 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrlw $6, %ymm0, %ymm2
 ; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX2-NEXT:    vpsllw $2, %ymm0, %ymm3
 ; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
 ; AVX2-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
-; AVX2-NEXT:    vpsrlw $7, %ymm0, %ymm3
-; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
-; AVX2-NEXT:    vpor %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpsrlw $7, %ymm0, %ymm2
+; AVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
+; AVX2-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: var_rotate_v32i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
 ; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpsllw $5, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw $6, %ymm0, %ymm2
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
 ; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm3
 ; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
 ; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
-; AVX512F-NEXT:    vpsrlw $7, %ymm0, %ymm3
-; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm3
-; AVX512F-NEXT:    vpor %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpsrlw $7, %ymm0, %ymm2
+; AVX512F-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
+; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_rotate_v32i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm3