[X86][SSE] Remove unnecessary bit-and in pshufb vector ctlz (PR39703)

[RKSimon]

Summary

SSE PSHUFB vector ctlz lowering works at the i4 nibble level. As detailed in PR39703, we were masking the lower nibble off but we only actually use it in the case where the upper nibble is known to be zero, making it safe to remove the mask and save an instruction.

[craig.topper]

LGTM

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

static SDValue LowerVectorCTLZInRegLUT(SDValue Op, const SDLoc &DL,
   // Begin by bitcasting the input to byte vector, then split those bytes
   // into lo/hi nibbles and use the PSHUFB LUT to perform CLTZ on each of them.
   // If the hi input nibble is zero then we add both results together, otherwise
   // we just take the hi result (by masking the lo result to zero before the
   // add).
   SDValue Op0 = DAG.getBitcast(CurrVT, Op.getOperand(0));
   SDValue Zero = getZeroVector(CurrVT, Subtarget, DAG, DL);
 
-  SDValue NibbleMask = DAG.getConstant(0xF, DL, CurrVT);
   SDValue NibbleShift = DAG.getConstant(0x4, DL, CurrVT);
-  SDValue Lo = DAG.getNode(ISD::AND, DL, CurrVT, Op0, NibbleMask);
+  SDValue Lo = Op0;
   SDValue Hi = DAG.getNode(ISD::SRL, DL, CurrVT, Op0, NibbleShift);
   SDValue HiZ;
   if (CurrVT.is512BitVector()) {
     MVT MaskVT = MVT::getVectorVT(MVT::i1, CurrVT.getVectorNumElements());
     HiZ = DAG.getSetCC(DL, MaskVT, Hi, Zero, ISD::SETEQ);
     HiZ = DAG.getNode(ISD::SIGN_EXTEND, DL, CurrVT, HiZ);
   } else {
     HiZ = DAG.getSetCC(DL, CurrVT, Hi, Zero, ISD::SETEQ);

llvm/trunk/test/CodeGen/X86/combine-srl.ll

; AVX-NEXT:    retq
   %3 = lshr <4 x i32> %2, <i32 5, i32 5, i32 5, i32 5>
   ret <4 x i32> %3
 }
 
 define <4 x i32> @combine_vec_lshr_lzcnt_bit1(<4 x i32> %x) {
 ; SSE-LABEL: combine_vec_lshr_lzcnt_bit1:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSE-NEXT:    movdqa %xmm2, %xmm3
+; SSE-NEXT:    pshufb %xmm0, %xmm3
 ; SSE-NEXT:    movdqa %xmm0, %xmm1
 ; SSE-NEXT:    psrlw $4, %xmm1
-; SSE-NEXT:    pxor %xmm2, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSE-NEXT:    movdqa %xmm3, %xmm4
-; SSE-NEXT:    pshufb %xmm1, %xmm3
-; SSE-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE-NEXT:    pand %xmm0, %xmm5
-; SSE-NEXT:    pshufb %xmm5, %xmm4
-; SSE-NEXT:    pand %xmm1, %xmm4
-; SSE-NEXT:    paddb %xmm4, %xmm3
-; SSE-NEXT:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    pxor %xmm4, %xmm4
+; SSE-NEXT:    pshufb %xmm1, %xmm2
+; SSE-NEXT:    pcmpeqb %xmm4, %xmm1
 ; SSE-NEXT:    pand %xmm3, %xmm1
-; SSE-NEXT:    psrlw $8, %xmm3
-; SSE-NEXT:    paddw %xmm1, %xmm3
-; SSE-NEXT:    pcmpeqw %xmm2, %xmm0
+; SSE-NEXT:    paddb %xmm2, %xmm1
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    pcmpeqb %xmm4, %xmm2
+; SSE-NEXT:    psrlw $8, %xmm2
+; SSE-NEXT:    pand %xmm1, %xmm2
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    paddw %xmm2, %xmm1
+; SSE-NEXT:    pcmpeqw %xmm4, %xmm0
 ; SSE-NEXT:    psrld $16, %xmm0
-; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    psrld $16, %xmm3
-; SSE-NEXT:    paddd %xmm3, %xmm0
-; SSE-NEXT:    psrld $5, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    psrld $16, %xmm1
+; SSE-NEXT:    paddd %xmm0, %xmm1
+; SSE-NEXT:    psrld $5, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: combine_vec_lshr_lzcnt_bit1:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm1
-; AVX-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX-NEXT:    vpcmpeqb %xmm2, %xmm1, %xmm3
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm4
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm5 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX-NEXT:    vpshufb %xmm4, %xmm5, %xmm4
-; AVX-NEXT:    vpand %xmm3, %xmm4, %xmm3
-; AVX-NEXT:    vpshufb %xmm1, %xmm5, %xmm1
-; AVX-NEXT:    vpaddb %xmm1, %xmm3, %xmm1
-; AVX-NEXT:    vpcmpeqb %xmm2, %xmm0, %xmm3
-; AVX-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX-NEXT:    vpand %xmm3, %xmm1, %xmm3
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
+; AVX-NEXT:    vpand %xmm5, %xmm2, %xmm2
+; AVX-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
+; AVX-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
+; AVX-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; AVX-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX-NEXT:    vpsrlw $8, %xmm1, %xmm1
-; AVX-NEXT:    vpaddw %xmm3, %xmm1, %xmm1
-; AVX-NEXT:    vpcmpeqw %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm0
 ; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpsrld $5, %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %1 = and <4 x i32> %x, <i32 4, i32 32, i32 64, i32 128>
   %2 = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %1, i1 0)

llvm/trunk/test/CodeGen/X86/prefer-avx256-lzcnt.ll

 ; AVX512F-NEXT:    retq
   %out = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %in, i1 false)
   ret <8 x i16> %out
 }
 
 define <16 x i8> @testv16i8(<16 x i8> %in) {
 ; AVX256-LABEL: testv16i8:
 ; AVX256:       # %bb.0:
-; AVX256-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX256-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX256-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX256-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX256-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX256-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
 ; AVX256-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX256-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX256-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX256-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm1
-; AVX256-NEXT:    vpand %xmm1, %xmm2, %xmm1
-; AVX256-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX256-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
+; AVX256-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX256-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX256-NEXT:    vpcmpeqb %xmm3, %xmm0, %xmm3
+; AVX256-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX256-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
+; AVX256-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
 ; AVX256-NEXT:    retq
 ;
 ; AVX512-LABEL: testv16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512-NEXT:    vpsubb {{.*}}(%rip), %xmm0, %xmm0

 ; AVX512-NEXT:    retq
   %out = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %in, i1 false)
   ret <16 x i16> %out
 }
 
 define <32 x i8> @testv32i8(<32 x i8> %in) {
 ; AVX256-LABEL: testv32i8:
 ; AVX256:       # %bb.0:
-; AVX256-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX256-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX256-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX256-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; AVX256-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX256-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; AVX256-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX256-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX256-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX256-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm1
-; AVX256-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; AVX256-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX256-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; AVX256-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX256-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX256-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; AVX256-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX256-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; AVX256-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
 ; AVX256-NEXT:    retq
 ;
 ; AVX512-LABEL: testv32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512-NEXT:    vplzcntd %zmm1, %zmm1
 ; AVX512-NEXT:    vpmovdb %zmm1, %xmm1

llvm/trunk/test/CodeGen/X86/vector-lzcnt-128.ll

 ; SSE3-NEXT:    paddb %xmm1, %xmm2
 ; SSE3-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE3-NEXT:    pxor %xmm0, %xmm0
 ; SSE3-NEXT:    psadbw %xmm2, %xmm0
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: testv2i64:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm1, %xmm4
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSSE3-NEXT:    movdqa %xmm2, %xmm3
+; SSSE3-NEXT:    pshufb %xmm0, %xmm3
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    psrlw $4, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    pshufb %xmm1, %xmm3
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSSE3-NEXT:    pand %xmm4, %xmm1
-; SSSE3-NEXT:    paddb %xmm3, %xmm1
-; SSSE3-NEXT:    movdqa %xmm0, %xmm3
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm3
-; SSSE3-NEXT:    psrlw $8, %xmm3
-; SSSE3-NEXT:    pand %xmm1, %xmm3
+; SSSE3-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSSE3-NEXT:    pxor %xmm4, %xmm4
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSSE3-NEXT:    pand %xmm3, %xmm1
+; SSSE3-NEXT:    paddb %xmm2, %xmm1
+; SSSE3-NEXT:    movdqa %xmm0, %xmm2
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm2
+; SSSE3-NEXT:    psrlw $8, %xmm2
+; SSSE3-NEXT:    pand %xmm1, %xmm2
 ; SSSE3-NEXT:    psrlw $8, %xmm1
-; SSSE3-NEXT:    paddw %xmm3, %xmm1
-; SSSE3-NEXT:    movdqa %xmm0, %xmm3
-; SSSE3-NEXT:    pcmpeqw %xmm2, %xmm3
-; SSSE3-NEXT:    psrld $16, %xmm3
-; SSSE3-NEXT:    pand %xmm1, %xmm3
+; SSSE3-NEXT:    paddw %xmm2, %xmm1
+; SSSE3-NEXT:    movdqa %xmm0, %xmm2
+; SSSE3-NEXT:    pcmpeqw %xmm4, %xmm2
+; SSSE3-NEXT:    psrld $16, %xmm2
+; SSSE3-NEXT:    pand %xmm1, %xmm2
 ; SSSE3-NEXT:    psrld $16, %xmm1
-; SSSE3-NEXT:    paddd %xmm3, %xmm1
-; SSSE3-NEXT:    pcmpeqd %xmm2, %xmm0
+; SSSE3-NEXT:    paddd %xmm2, %xmm1
+; SSSE3-NEXT:    pcmpeqd %xmm4, %xmm0
 ; SSSE3-NEXT:    psrlq $32, %xmm0
 ; SSSE3-NEXT:    pand %xmm1, %xmm0
 ; SSSE3-NEXT:    psrlq $32, %xmm1
 ; SSSE3-NEXT:    paddq %xmm0, %xmm1
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: testv2i64:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm1, %xmm4
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    pshufb %xmm0, %xmm3
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $4, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pshufb %xmm1, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSE41-NEXT:    pand %xmm4, %xmm1
-; SSE41-NEXT:    paddb %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm3
-; SSE41-NEXT:    psrlw $8, %xmm3
-; SSE41-NEXT:    pand %xmm1, %xmm3
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    pxor %xmm4, %xmm4
+; SSE41-NEXT:    pshufb %xmm1, %xmm2
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSE41-NEXT:    pand %xmm3, %xmm1
+; SSE41-NEXT:    paddb %xmm2, %xmm1
+; SSE41-NEXT:    movdqa %xmm0, %xmm2
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm2
+; SSE41-NEXT:    psrlw $8, %xmm2
+; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    paddw %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pcmpeqw %xmm2, %xmm3
-; SSE41-NEXT:    psrld $16, %xmm3
-; SSE41-NEXT:    pand %xmm1, %xmm3
+; SSE41-NEXT:    paddw %xmm2, %xmm1
+; SSE41-NEXT:    movdqa %xmm0, %xmm2
+; SSE41-NEXT:    pcmpeqw %xmm4, %xmm2
+; SSE41-NEXT:    psrld $16, %xmm2
+; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    paddd %xmm3, %xmm1
-; SSE41-NEXT:    pcmpeqd %xmm2, %xmm0
+; SSE41-NEXT:    paddd %xmm2, %xmm1
+; SSE41-NEXT:    pcmpeqd %xmm4, %xmm0
 ; SSE41-NEXT:    psrlq $32, %xmm0
 ; SSE41-NEXT:    pand %xmm1, %xmm0
 ; SSE41-NEXT:    psrlq $32, %xmm1
 ; SSE41-NEXT:    paddq %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: testv2i64:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
 ; AVX-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
 ; AVX-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm2
 ; AVX-NEXT:    vpsrld $16, %xmm2, %xmm2
 ; AVX-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
 ; AVX-NEXT:    vpcmpeqd %xmm4, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpsrlq $32, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv2i64:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX512VLBWDQ-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrld $16, %xmm2, %xmm2

 ; AVX512CD-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512CD-NEXT:    vplzcntq %zmm0, %zmm0
 ; AVX512CD-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512CD-NEXT:    vzeroupper
 ; AVX512CD-NEXT:    retq
 ;
 ; X32-SSE-LABEL: testv2i64:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm4
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
+; X32-SSE-NEXT:    pshufb %xmm0, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psrlw $4, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm3
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm1
-; X32-SSE-NEXT:    pand %xmm4, %xmm1
-; X32-SSE-NEXT:    paddb %xmm3, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm3
-; X32-SSE-NEXT:    psrlw $8, %xmm3
-; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    pxor %xmm4, %xmm4
+; X32-SSE-NEXT:    pshufb %xmm1, %xmm2
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
+; X32-SSE-NEXT:    paddb %xmm2, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm2
+; X32-SSE-NEXT:    psrlw $8, %xmm2
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
 ; X32-SSE-NEXT:    psrlw $8, %xmm1
-; X32-SSE-NEXT:    paddw %xmm3, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    pcmpeqw %xmm2, %xmm3
-; X32-SSE-NEXT:    psrld $16, %xmm3
-; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    paddw %xmm2, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    pcmpeqw %xmm4, %xmm2
+; X32-SSE-NEXT:    psrld $16, %xmm2
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
 ; X32-SSE-NEXT:    psrld $16, %xmm1
-; X32-SSE-NEXT:    paddd %xmm3, %xmm1
-; X32-SSE-NEXT:    pcmpeqd %xmm2, %xmm0
+; X32-SSE-NEXT:    paddd %xmm2, %xmm1
+; X32-SSE-NEXT:    pcmpeqd %xmm4, %xmm0
 ; X32-SSE-NEXT:    psrlq $32, %xmm0
 ; X32-SSE-NEXT:    pand %xmm1, %xmm0
 ; X32-SSE-NEXT:    psrlq $32, %xmm1
 ; X32-SSE-NEXT:    paddq %xmm0, %xmm1
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
 
   %out = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %in, i1 0)

 ; SSE3-NEXT:    paddb %xmm1, %xmm2
 ; SSE3-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE3-NEXT:    pxor %xmm0, %xmm0
 ; SSE3-NEXT:    psadbw %xmm2, %xmm0
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: testv2i64u:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm1, %xmm4
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSSE3-NEXT:    movdqa %xmm2, %xmm3
+; SSSE3-NEXT:    pshufb %xmm0, %xmm3
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    psrlw $4, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    pshufb %xmm1, %xmm3
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSSE3-NEXT:    pand %xmm4, %xmm1
-; SSSE3-NEXT:    paddb %xmm3, %xmm1
-; SSSE3-NEXT:    movdqa %xmm0, %xmm3
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm3
-; SSSE3-NEXT:    psrlw $8, %xmm3
-; SSSE3-NEXT:    pand %xmm1, %xmm3
+; SSSE3-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSSE3-NEXT:    pxor %xmm4, %xmm4
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSSE3-NEXT:    pand %xmm3, %xmm1
+; SSSE3-NEXT:    paddb %xmm2, %xmm1
+; SSSE3-NEXT:    movdqa %xmm0, %xmm2
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm2
+; SSSE3-NEXT:    psrlw $8, %xmm2
+; SSSE3-NEXT:    pand %xmm1, %xmm2
 ; SSSE3-NEXT:    psrlw $8, %xmm1
-; SSSE3-NEXT:    paddw %xmm3, %xmm1
-; SSSE3-NEXT:    movdqa %xmm0, %xmm3
-; SSSE3-NEXT:    pcmpeqw %xmm2, %xmm3
-; SSSE3-NEXT:    psrld $16, %xmm3
-; SSSE3-NEXT:    pand %xmm1, %xmm3
+; SSSE3-NEXT:    paddw %xmm2, %xmm1
+; SSSE3-NEXT:    movdqa %xmm0, %xmm2
+; SSSE3-NEXT:    pcmpeqw %xmm4, %xmm2
+; SSSE3-NEXT:    psrld $16, %xmm2
+; SSSE3-NEXT:    pand %xmm1, %xmm2
 ; SSSE3-NEXT:    psrld $16, %xmm1
-; SSSE3-NEXT:    paddd %xmm3, %xmm1
-; SSSE3-NEXT:    pcmpeqd %xmm2, %xmm0
+; SSSE3-NEXT:    paddd %xmm2, %xmm1
+; SSSE3-NEXT:    pcmpeqd %xmm4, %xmm0
 ; SSSE3-NEXT:    psrlq $32, %xmm0
 ; SSSE3-NEXT:    pand %xmm1, %xmm0
 ; SSSE3-NEXT:    psrlq $32, %xmm1
 ; SSSE3-NEXT:    paddq %xmm0, %xmm1
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: testv2i64u:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm1, %xmm4
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    pshufb %xmm0, %xmm3
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $4, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pshufb %xmm1, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSE41-NEXT:    pand %xmm4, %xmm1
-; SSE41-NEXT:    paddb %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm3
-; SSE41-NEXT:    psrlw $8, %xmm3
-; SSE41-NEXT:    pand %xmm1, %xmm3
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    pxor %xmm4, %xmm4
+; SSE41-NEXT:    pshufb %xmm1, %xmm2
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSE41-NEXT:    pand %xmm3, %xmm1
+; SSE41-NEXT:    paddb %xmm2, %xmm1
+; SSE41-NEXT:    movdqa %xmm0, %xmm2
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm2
+; SSE41-NEXT:    psrlw $8, %xmm2
+; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    paddw %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pcmpeqw %xmm2, %xmm3
-; SSE41-NEXT:    psrld $16, %xmm3
-; SSE41-NEXT:    pand %xmm1, %xmm3
+; SSE41-NEXT:    paddw %xmm2, %xmm1
+; SSE41-NEXT:    movdqa %xmm0, %xmm2
+; SSE41-NEXT:    pcmpeqw %xmm4, %xmm2
+; SSE41-NEXT:    psrld $16, %xmm2
+; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    psrld $16, %xmm1
-; SSE41-NEXT:    paddd %xmm3, %xmm1
-; SSE41-NEXT:    pcmpeqd %xmm2, %xmm0
+; SSE41-NEXT:    paddd %xmm2, %xmm1
+; SSE41-NEXT:    pcmpeqd %xmm4, %xmm0
 ; SSE41-NEXT:    psrlq $32, %xmm0
 ; SSE41-NEXT:    pand %xmm1, %xmm0
 ; SSE41-NEXT:    psrlq $32, %xmm1
 ; SSE41-NEXT:    paddq %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: testv2i64u:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
 ; AVX-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
 ; AVX-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm2
 ; AVX-NEXT:    vpsrld $16, %xmm2, %xmm2
 ; AVX-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
 ; AVX-NEXT:    vpcmpeqd %xmm4, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpsrlq $32, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv2i64u:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX512VLBWDQ-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrld $16, %xmm2, %xmm2

 ; AVX512CD-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512CD-NEXT:    vplzcntq %zmm0, %zmm0
 ; AVX512CD-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512CD-NEXT:    vzeroupper
 ; AVX512CD-NEXT:    retq
 ;
 ; X32-SSE-LABEL: testv2i64u:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm4
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
+; X32-SSE-NEXT:    pshufb %xmm0, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psrlw $4, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm3
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm1
-; X32-SSE-NEXT:    pand %xmm4, %xmm1
-; X32-SSE-NEXT:    paddb %xmm3, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm3
-; X32-SSE-NEXT:    psrlw $8, %xmm3
-; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    pxor %xmm4, %xmm4
+; X32-SSE-NEXT:    pshufb %xmm1, %xmm2
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
+; X32-SSE-NEXT:    paddb %xmm2, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm2
+; X32-SSE-NEXT:    psrlw $8, %xmm2
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
 ; X32-SSE-NEXT:    psrlw $8, %xmm1
-; X32-SSE-NEXT:    paddw %xmm3, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    pcmpeqw %xmm2, %xmm3
-; X32-SSE-NEXT:    psrld $16, %xmm3
-; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    paddw %xmm2, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    pcmpeqw %xmm4, %xmm2
+; X32-SSE-NEXT:    psrld $16, %xmm2
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
 ; X32-SSE-NEXT:    psrld $16, %xmm1
-; X32-SSE-NEXT:    paddd %xmm3, %xmm1
-; X32-SSE-NEXT:    pcmpeqd %xmm2, %xmm0
+; X32-SSE-NEXT:    paddd %xmm2, %xmm1
+; X32-SSE-NEXT:    pcmpeqd %xmm4, %xmm0
 ; X32-SSE-NEXT:    psrlq $32, %xmm0
 ; X32-SSE-NEXT:    pand %xmm1, %xmm0
 ; X32-SSE-NEXT:    psrlq $32, %xmm1
 ; X32-SSE-NEXT:    paddq %xmm0, %xmm1
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
 
   %out = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %in, i1 -1)

 ; SSE3-NEXT:    psadbw %xmm1, %xmm2
 ; SSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; SSE3-NEXT:    psadbw %xmm1, %xmm0
 ; SSE3-NEXT:    packuswb %xmm2, %xmm0
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: testv4i32:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm1, %xmm4
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSSE3-NEXT:    movdqa %xmm2, %xmm3
+; SSSE3-NEXT:    pshufb %xmm0, %xmm3
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    psrlw $4, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    pshufb %xmm1, %xmm3
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSSE3-NEXT:    pand %xmm4, %xmm1
-; SSSE3-NEXT:    paddb %xmm3, %xmm1
-; SSSE3-NEXT:    movdqa %xmm0, %xmm3
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm3
-; SSSE3-NEXT:    psrlw $8, %xmm3
-; SSSE3-NEXT:    pand %xmm1, %xmm3
+; SSSE3-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSSE3-NEXT:    pxor %xmm4, %xmm4
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSSE3-NEXT:    pand %xmm3, %xmm1
+; SSSE3-NEXT:    paddb %xmm2, %xmm1
+; SSSE3-NEXT:    movdqa %xmm0, %xmm2
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm2
+; SSSE3-NEXT:    psrlw $8, %xmm2
+; SSSE3-NEXT:    pand %xmm1, %xmm2
 ; SSSE3-NEXT:    psrlw $8, %xmm1
-; SSSE3-NEXT:    paddw %xmm3, %xmm1
-; SSSE3-NEXT:    pcmpeqw %xmm2, %xmm0
+; SSSE3-NEXT:    paddw %xmm2, %xmm1
+; SSSE3-NEXT:    pcmpeqw %xmm4, %xmm0
 ; SSSE3-NEXT:    psrld $16, %xmm0
 ; SSSE3-NEXT:    pand %xmm1, %xmm0
 ; SSSE3-NEXT:    psrld $16, %xmm1
 ; SSSE3-NEXT:    paddd %xmm0, %xmm1
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: testv4i32:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm1, %xmm4
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    pshufb %xmm0, %xmm3
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $4, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pshufb %xmm1, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSE41-NEXT:    pand %xmm4, %xmm1
-; SSE41-NEXT:    paddb %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm3
-; SSE41-NEXT:    psrlw $8, %xmm3
-; SSE41-NEXT:    pand %xmm1, %xmm3
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    pxor %xmm4, %xmm4
+; SSE41-NEXT:    pshufb %xmm1, %xmm2
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSE41-NEXT:    pand %xmm3, %xmm1
+; SSE41-NEXT:    paddb %xmm2, %xmm1
+; SSE41-NEXT:    movdqa %xmm0, %xmm2
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm2
+; SSE41-NEXT:    psrlw $8, %xmm2
+; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    paddw %xmm3, %xmm1
-; SSE41-NEXT:    pcmpeqw %xmm2, %xmm0
+; SSE41-NEXT:    paddw %xmm2, %xmm1
+; SSE41-NEXT:    pcmpeqw %xmm4, %xmm0
 ; SSE41-NEXT:    psrld $16, %xmm0
 ; SSE41-NEXT:    pand %xmm1, %xmm0
 ; SSE41-NEXT:    psrld $16, %xmm1
 ; SSE41-NEXT:    paddd %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: testv4i32:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
 ; AVX-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
 ; AVX-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm0
 ; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv4i32:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX512VLBWDQ-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
 ; AVX512VLBWDQ-NEXT:    vpsrld $16, %xmm0, %xmm0

 ; AVX512CD-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512CD-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512CD-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512CD-NEXT:    vzeroupper
 ; AVX512CD-NEXT:    retq
 ;
 ; X32-SSE-LABEL: testv4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm4
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
+; X32-SSE-NEXT:    pshufb %xmm0, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psrlw $4, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm3
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm1
-; X32-SSE-NEXT:    pand %xmm4, %xmm1
-; X32-SSE-NEXT:    paddb %xmm3, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm3
-; X32-SSE-NEXT:    psrlw $8, %xmm3
-; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    pxor %xmm4, %xmm4
+; X32-SSE-NEXT:    pshufb %xmm1, %xmm2
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
+; X32-SSE-NEXT:    paddb %xmm2, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm2
+; X32-SSE-NEXT:    psrlw $8, %xmm2
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
 ; X32-SSE-NEXT:    psrlw $8, %xmm1
-; X32-SSE-NEXT:    paddw %xmm3, %xmm1
-; X32-SSE-NEXT:    pcmpeqw %xmm2, %xmm0
+; X32-SSE-NEXT:    paddw %xmm2, %xmm1
+; X32-SSE-NEXT:    pcmpeqw %xmm4, %xmm0
 ; X32-SSE-NEXT:    psrld $16, %xmm0
 ; X32-SSE-NEXT:    pand %xmm1, %xmm0
 ; X32-SSE-NEXT:    psrld $16, %xmm1
 ; X32-SSE-NEXT:    paddd %xmm0, %xmm1
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
 
   %out = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %in, i1 0)

 ; SSE3-NEXT:    psadbw %xmm1, %xmm2
 ; SSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; SSE3-NEXT:    psadbw %xmm1, %xmm0
 ; SSE3-NEXT:    packuswb %xmm2, %xmm0
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: testv4i32u:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm1, %xmm4
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSSE3-NEXT:    movdqa %xmm2, %xmm3
+; SSSE3-NEXT:    pshufb %xmm0, %xmm3
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    psrlw $4, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    pshufb %xmm1, %xmm3
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSSE3-NEXT:    pand %xmm4, %xmm1
-; SSSE3-NEXT:    paddb %xmm3, %xmm1
-; SSSE3-NEXT:    movdqa %xmm0, %xmm3
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm3
-; SSSE3-NEXT:    psrlw $8, %xmm3
-; SSSE3-NEXT:    pand %xmm1, %xmm3
+; SSSE3-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSSE3-NEXT:    pxor %xmm4, %xmm4
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSSE3-NEXT:    pand %xmm3, %xmm1
+; SSSE3-NEXT:    paddb %xmm2, %xmm1
+; SSSE3-NEXT:    movdqa %xmm0, %xmm2
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm2
+; SSSE3-NEXT:    psrlw $8, %xmm2
+; SSSE3-NEXT:    pand %xmm1, %xmm2
 ; SSSE3-NEXT:    psrlw $8, %xmm1
-; SSSE3-NEXT:    paddw %xmm3, %xmm1
-; SSSE3-NEXT:    pcmpeqw %xmm2, %xmm0
+; SSSE3-NEXT:    paddw %xmm2, %xmm1
+; SSSE3-NEXT:    pcmpeqw %xmm4, %xmm0
 ; SSSE3-NEXT:    psrld $16, %xmm0
 ; SSSE3-NEXT:    pand %xmm1, %xmm0
 ; SSSE3-NEXT:    psrld $16, %xmm1
 ; SSSE3-NEXT:    paddd %xmm0, %xmm1
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: testv4i32u:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm1, %xmm4
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    pshufb %xmm0, %xmm3
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $4, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pshufb %xmm1, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSE41-NEXT:    pand %xmm4, %xmm1
-; SSE41-NEXT:    paddb %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm3
-; SSE41-NEXT:    psrlw $8, %xmm3
-; SSE41-NEXT:    pand %xmm1, %xmm3
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    pxor %xmm4, %xmm4
+; SSE41-NEXT:    pshufb %xmm1, %xmm2
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSE41-NEXT:    pand %xmm3, %xmm1
+; SSE41-NEXT:    paddb %xmm2, %xmm1
+; SSE41-NEXT:    movdqa %xmm0, %xmm2
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm2
+; SSE41-NEXT:    psrlw $8, %xmm2
+; SSE41-NEXT:    pand %xmm1, %xmm2
 ; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    paddw %xmm3, %xmm1
-; SSE41-NEXT:    pcmpeqw %xmm2, %xmm0
+; SSE41-NEXT:    paddw %xmm2, %xmm1
+; SSE41-NEXT:    pcmpeqw %xmm4, %xmm0
 ; SSE41-NEXT:    psrld $16, %xmm0
 ; SSE41-NEXT:    pand %xmm1, %xmm0
 ; SSE41-NEXT:    psrld $16, %xmm1
 ; SSE41-NEXT:    paddd %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: testv4i32u:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
 ; AVX-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
 ; AVX-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm0
 ; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddd %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv4i32u:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX512VLBWDQ-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpand %xmm2, %xmm1, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqw %xmm4, %xmm0, %xmm0
 ; AVX512VLBWDQ-NEXT:    vpsrld $16, %xmm0, %xmm0

 ; AVX512CD-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512CD-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512CD-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512CD-NEXT:    vzeroupper
 ; AVX512CD-NEXT:    retq
 ;
 ; X32-SSE-LABEL: testv4i32u:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm4
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
+; X32-SSE-NEXT:    pshufb %xmm0, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psrlw $4, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm3
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm1
-; X32-SSE-NEXT:    pand %xmm4, %xmm1
-; X32-SSE-NEXT:    paddb %xmm3, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm3
-; X32-SSE-NEXT:    psrlw $8, %xmm3
-; X32-SSE-NEXT:    pand %xmm1, %xmm3
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    pxor %xmm4, %xmm4
+; X32-SSE-NEXT:    pshufb %xmm1, %xmm2
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
+; X32-SSE-NEXT:    paddb %xmm2, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm2
+; X32-SSE-NEXT:    psrlw $8, %xmm2
+; X32-SSE-NEXT:    pand %xmm1, %xmm2
 ; X32-SSE-NEXT:    psrlw $8, %xmm1
-; X32-SSE-NEXT:    paddw %xmm3, %xmm1
-; X32-SSE-NEXT:    pcmpeqw %xmm2, %xmm0
+; X32-SSE-NEXT:    paddw %xmm2, %xmm1
+; X32-SSE-NEXT:    pcmpeqw %xmm4, %xmm0
 ; X32-SSE-NEXT:    psrld $16, %xmm0
 ; X32-SSE-NEXT:    pand %xmm1, %xmm0
 ; X32-SSE-NEXT:    psrld $16, %xmm1
 ; X32-SSE-NEXT:    paddd %xmm0, %xmm1
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
 
   %out = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %in, i1 -1)

 ; SSE3-NEXT:    movdqa %xmm2, %xmm0
 ; SSE3-NEXT:    psllw $8, %xmm0
 ; SSE3-NEXT:    paddb %xmm2, %xmm0
 ; SSE3-NEXT:    psrlw $8, %xmm0
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: testv8i16:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm1, %xmm4
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSSE3-NEXT:    movdqa %xmm2, %xmm3
+; SSSE3-NEXT:    pshufb %xmm0, %xmm3
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    psrlw $4, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    pshufb %xmm1, %xmm3
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSSE3-NEXT:    pand %xmm4, %xmm1
-; SSSE3-NEXT:    paddb %xmm3, %xmm1
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm0
+; SSSE3-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSSE3-NEXT:    pxor %xmm4, %xmm4
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSSE3-NEXT:    pand %xmm3, %xmm1
+; SSSE3-NEXT:    paddb %xmm2, %xmm1
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm0
 ; SSSE3-NEXT:    psrlw $8, %xmm0
 ; SSSE3-NEXT:    pand %xmm1, %xmm0
 ; SSSE3-NEXT:    psrlw $8, %xmm1
 ; SSSE3-NEXT:    paddw %xmm0, %xmm1
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: testv8i16:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm1, %xmm4
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    pshufb %xmm0, %xmm3
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $4, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pshufb %xmm1, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSE41-NEXT:    pand %xmm4, %xmm1
-; SSE41-NEXT:    paddb %xmm3, %xmm1
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm0
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    pxor %xmm4, %xmm4
+; SSE41-NEXT:    pshufb %xmm1, %xmm2
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSE41-NEXT:    pand %xmm3, %xmm1
+; SSE41-NEXT:    paddb %xmm2, %xmm1
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pand %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    paddw %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: testv8i16:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv8i16:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX512VLBWDQ-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm0
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX512VLBWDQ-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
 ; AVX512VLBWDQ-NEXT:    retq
 ;

 ; AVX512CD-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512CD-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512CD-NEXT:    vpsubw {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512CD-NEXT:    vzeroupper
 ; AVX512CD-NEXT:    retq
 ;
 ; X32-SSE-LABEL: testv8i16:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm4
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
+; X32-SSE-NEXT:    pshufb %xmm0, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psrlw $4, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm3
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm1
-; X32-SSE-NEXT:    pand %xmm4, %xmm1
-; X32-SSE-NEXT:    paddb %xmm3, %xmm1
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    pxor %xmm4, %xmm4
+; X32-SSE-NEXT:    pshufb %xmm1, %xmm2
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
+; X32-SSE-NEXT:    paddb %xmm2, %xmm1
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm0
 ; X32-SSE-NEXT:    psrlw $8, %xmm0
 ; X32-SSE-NEXT:    pand %xmm1, %xmm0
 ; X32-SSE-NEXT:    psrlw $8, %xmm1
 ; X32-SSE-NEXT:    paddw %xmm0, %xmm1
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %out = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %in, i1 0)
   ret <8 x i16> %out

 ; SSE3-NEXT:    movdqa %xmm2, %xmm0
 ; SSE3-NEXT:    psllw $8, %xmm0
 ; SSE3-NEXT:    paddb %xmm2, %xmm0
 ; SSE3-NEXT:    psrlw $8, %xmm0
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: testv8i16u:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm4
-; SSSE3-NEXT:    pshufb %xmm1, %xmm4
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSSE3-NEXT:    movdqa %xmm2, %xmm3
+; SSSE3-NEXT:    pshufb %xmm0, %xmm3
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    psrlw $4, %xmm1
-; SSSE3-NEXT:    pand %xmm2, %xmm1
-; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    pshufb %xmm1, %xmm3
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSSE3-NEXT:    pand %xmm4, %xmm1
-; SSSE3-NEXT:    paddb %xmm3, %xmm1
-; SSSE3-NEXT:    pcmpeqb %xmm2, %xmm0
+; SSSE3-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSSE3-NEXT:    pxor %xmm4, %xmm4
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSSE3-NEXT:    pand %xmm3, %xmm1
+; SSSE3-NEXT:    paddb %xmm2, %xmm1
+; SSSE3-NEXT:    pcmpeqb %xmm4, %xmm0
 ; SSSE3-NEXT:    psrlw $8, %xmm0
 ; SSSE3-NEXT:    pand %xmm1, %xmm0
 ; SSSE3-NEXT:    psrlw $8, %xmm1
 ; SSSE3-NEXT:    paddw %xmm0, %xmm1
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: testv8i16u:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSE41-NEXT:    movdqa %xmm3, %xmm4
-; SSE41-NEXT:    pshufb %xmm1, %xmm4
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; SSE41-NEXT:    movdqa %xmm2, %xmm3
+; SSE41-NEXT:    pshufb %xmm0, %xmm3
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $4, %xmm1
-; SSE41-NEXT:    pand %xmm2, %xmm1
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pshufb %xmm1, %xmm3
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm1
-; SSE41-NEXT:    pand %xmm4, %xmm1
-; SSE41-NEXT:    paddb %xmm3, %xmm1
-; SSE41-NEXT:    pcmpeqb %xmm2, %xmm0
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    pxor %xmm4, %xmm4
+; SSE41-NEXT:    pshufb %xmm1, %xmm2
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm1
+; SSE41-NEXT:    pand %xmm3, %xmm1
+; SSE41-NEXT:    paddb %xmm2, %xmm1
+; SSE41-NEXT:    pcmpeqb %xmm4, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
 ; SSE41-NEXT:    pand %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    paddw %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: testv8i16u:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv8i16u:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm4
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm4, %xmm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm1, %xmm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm3, %xmm5
 ; AVX512VLBWDQ-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm4, %xmm0, %xmm0
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX512VLBWDQ-NEXT:    vpand %xmm0, %xmm1, %xmm0
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %xmm0, %xmm1, %xmm0
 ; AVX512VLBWDQ-NEXT:    retq
 ;

 ; AVX512CD-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512CD-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512CD-NEXT:    vpsubw {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512CD-NEXT:    vzeroupper
 ; AVX512CD-NEXT:    retq
 ;
 ; X32-SSE-LABEL: testv8i16u:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm4
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
+; X32-SSE-NEXT:    pshufb %xmm0, %xmm3
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    psrlw $4, %xmm1
-; X32-SSE-NEXT:    pand %xmm2, %xmm1
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pshufb %xmm1, %xmm3
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm1
-; X32-SSE-NEXT:    pand %xmm4, %xmm1
-; X32-SSE-NEXT:    paddb %xmm3, %xmm1
-; X32-SSE-NEXT:    pcmpeqb %xmm2, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    pxor %xmm4, %xmm4
+; X32-SSE-NEXT:    pshufb %xmm1, %xmm2
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm1
+; X32-SSE-NEXT:    pand %xmm3, %xmm1
+; X32-SSE-NEXT:    paddb %xmm2, %xmm1
+; X32-SSE-NEXT:    pcmpeqb %xmm4, %xmm0
 ; X32-SSE-NEXT:    psrlw $8, %xmm0
 ; X32-SSE-NEXT:    pand %xmm1, %xmm0
 ; X32-SSE-NEXT:    psrlw $8, %xmm1
 ; X32-SSE-NEXT:    paddw %xmm0, %xmm1
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %out = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %in, i1 -1)
   ret <8 x i16> %out

 ; SSE3-NEXT:    movdqa %xmm3, %xmm0
 ; SSE3-NEXT:    psrlw $4, %xmm0
 ; SSE3-NEXT:    paddb %xmm3, %xmm0
 ; SSE3-NEXT:    pand %xmm2, %xmm0
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: testv16i8:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm3
-; SSSE3-NEXT:    pand %xmm2, %xmm3
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSSE3-NEXT:    movdqa %xmm1, %xmm4
-; SSSE3-NEXT:    pshufb %xmm3, %xmm4
+; SSSE3-NEXT:    movdqa %xmm1, %xmm2
+; SSSE3-NEXT:    pshufb %xmm0, %xmm2
 ; SSSE3-NEXT:    psrlw $4, %xmm0
-; SSSE3-NEXT:    pand %xmm2, %xmm0
-; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    pcmpeqb %xmm0, %xmm2
-; SSSE3-NEXT:    pand %xmm4, %xmm2
+; SSSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSSE3-NEXT:    pxor %xmm3, %xmm3
+; SSSE3-NEXT:    pcmpeqb %xmm0, %xmm3
+; SSSE3-NEXT:    pand %xmm2, %xmm3
 ; SSSE3-NEXT:    pshufb %xmm0, %xmm1
-; SSSE3-NEXT:    paddb %xmm2, %xmm1
+; SSSE3-NEXT:    paddb %xmm3, %xmm1
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: testv16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand %xmm2, %xmm3
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    pshufb %xmm3, %xmm4
+; SSE41-NEXT:    movdqa %xmm1, %xmm2
+; SSE41-NEXT:    pshufb %xmm0, %xmm2
 ; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pcmpeqb %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm4, %xmm2
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    pcmpeqb %xmm0, %xmm3
+; SSE41-NEXT:    pand %xmm2, %xmm3
 ; SSE41-NEXT:    pshufb %xmm0, %xmm1
-; SSE41-NEXT:    paddb %xmm2, %xmm1
+; SSE41-NEXT:    paddb %xmm3, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: testv16i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
 ; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm1
-; AVX-NEXT:    vpand %xmm1, %xmm2, %xmm1
-; AVX-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX-NEXT:    vpcmpeqb %xmm3, %xmm0, %xmm3
+; AVX-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv16i8:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX512VLBWDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm1
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm2, %xmm1
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX512VLBWDQ-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VLBWDQ-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm3, %xmm0, %xmm3
+; AVX512VLBWDQ-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
+; AVX512VLBWDQ-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
 ; AVX512VLBWDQ-NEXT:    retq
 ;
 ; AVX512-LABEL: testv16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512-NEXT:    vpsubb {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
 ;
 ; X32-SSE-LABEL: testv16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    pand %xmm2, %xmm3
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE-NEXT:    pshufb %xmm3, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm2
+; X32-SSE-NEXT:    pshufb %xmm0, %xmm2
 ; X32-SSE-NEXT:    psrlw $4, %xmm0
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pcmpeqb %xmm0, %xmm2
-; X32-SSE-NEXT:    pand %xmm4, %xmm2
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    pxor %xmm3, %xmm3
+; X32-SSE-NEXT:    pcmpeqb %xmm0, %xmm3
+; X32-SSE-NEXT:    pand %xmm2, %xmm3
 ; X32-SSE-NEXT:    pshufb %xmm0, %xmm1
-; X32-SSE-NEXT:    paddb %xmm2, %xmm1
+; X32-SSE-NEXT:    paddb %xmm3, %xmm1
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %out = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %in, i1 0)
   ret <16 x i8> %out
 }
 
 define <16 x i8> @testv16i8u(<16 x i8> %in) nounwind {
 ; SSE2-LABEL: testv16i8u:

 ; SSE3-NEXT:    movdqa %xmm3, %xmm0
 ; SSE3-NEXT:    psrlw $4, %xmm0
 ; SSE3-NEXT:    paddb %xmm3, %xmm0
 ; SSE3-NEXT:    pand %xmm2, %xmm0
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: testv16i8u:
 ; SSSE3:       # %bb.0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSSE3-NEXT:    movdqa %xmm0, %xmm3
-; SSSE3-NEXT:    pand %xmm2, %xmm3
 ; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSSE3-NEXT:    movdqa %xmm1, %xmm4
-; SSSE3-NEXT:    pshufb %xmm3, %xmm4
+; SSSE3-NEXT:    movdqa %xmm1, %xmm2
+; SSSE3-NEXT:    pshufb %xmm0, %xmm2
 ; SSSE3-NEXT:    psrlw $4, %xmm0
-; SSSE3-NEXT:    pand %xmm2, %xmm0
-; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    pcmpeqb %xmm0, %xmm2
-; SSSE3-NEXT:    pand %xmm4, %xmm2
+; SSSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSSE3-NEXT:    pxor %xmm3, %xmm3
+; SSSE3-NEXT:    pcmpeqb %xmm0, %xmm3
+; SSSE3-NEXT:    pand %xmm2, %xmm3
 ; SSSE3-NEXT:    pshufb %xmm0, %xmm1
-; SSSE3-NEXT:    paddb %xmm2, %xmm1
+; SSSE3-NEXT:    paddb %xmm3, %xmm1
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: testv16i8u:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    pand %xmm2, %xmm3
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; SSE41-NEXT:    movdqa %xmm1, %xmm4
-; SSE41-NEXT:    pshufb %xmm3, %xmm4
+; SSE41-NEXT:    movdqa %xmm1, %xmm2
+; SSE41-NEXT:    pshufb %xmm0, %xmm2
 ; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pcmpeqb %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm4, %xmm2
+; SSE41-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    pcmpeqb %xmm0, %xmm3
+; SSE41-NEXT:    pand %xmm2, %xmm3
 ; SSE41-NEXT:    pshufb %xmm0, %xmm1
-; SSE41-NEXT:    paddb %xmm2, %xmm1
+; SSE41-NEXT:    paddb %xmm3, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: testv16i8u:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
 ; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm1
-; AVX-NEXT:    vpand %xmm1, %xmm2, %xmm1
-; AVX-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX-NEXT:    vpcmpeqb %xmm3, %xmm0, %xmm3
+; AVX-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv16i8u:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm0, %xmm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX512VLBWDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm1
-; AVX512VLBWDQ-NEXT:    vpand %xmm1, %xmm2, %xmm1
-; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
-; AVX512VLBWDQ-NEXT:    vpaddb %xmm0, %xmm1, %xmm0
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX512VLBWDQ-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %xmm3, %xmm0, %xmm3
+; AVX512VLBWDQ-NEXT:    vpand %xmm3, %xmm2, %xmm2
+; AVX512VLBWDQ-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
+; AVX512VLBWDQ-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
 ; AVX512VLBWDQ-NEXT:    retq
 ;
 ; AVX512-LABEL: testv16i8u:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512-NEXT:    vpsubb {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
 ;
 ; X32-SSE-LABEL: testv16i8u:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    pand %xmm2, %xmm3
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE-NEXT:    pshufb %xmm3, %xmm4
+; X32-SSE-NEXT:    movdqa %xmm1, %xmm2
+; X32-SSE-NEXT:    pshufb %xmm0, %xmm2
 ; X32-SSE-NEXT:    psrlw $4, %xmm0
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pcmpeqb %xmm0, %xmm2
-; X32-SSE-NEXT:    pand %xmm4, %xmm2
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    pxor %xmm3, %xmm3
+; X32-SSE-NEXT:    pcmpeqb %xmm0, %xmm3
+; X32-SSE-NEXT:    pand %xmm2, %xmm3
 ; X32-SSE-NEXT:    pshufb %xmm0, %xmm1
-; X32-SSE-NEXT:    paddb %xmm2, %xmm1
+; X32-SSE-NEXT:    paddb %xmm3, %xmm1
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %out = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %in, i1 -1)
   ret <16 x i8> %out
 }
 
 define <2 x i64> @foldv2i64() nounwind {
 ; SSE-LABEL: foldv2i64:

llvm/trunk/test/CodeGen/X86/vector-lzcnt-256.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX --check-prefix=AVX2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX512VL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+avx512dq | FileCheck %s --check-prefix=X64 --check-prefix=AVX512VLBWDQ
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+avx512cd,+avx512vl | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX512 --check-prefix=AVX512VLCD
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+avx512cd | FileCheck %s --check-prefix=X64 --check-prefix=NOBW --check-prefix=AVX512 --check-prefix=AVX512CD
 ;
 ; Just one 32-bit run to make sure we do reasonable things for i64 lzcnt.
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X32-AVX
 
 define <4 x i64> @testv4i64(<4 x i64> %in) nounwind {
 ; AVX1-LABEL: testv4i64:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpshufb %xmm1, %xmm4, %xmm5
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm4
 ; AVX1-NEXT:    vpsrlw $4, %xmm2, %xmm1
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm6
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm6
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm6, %xmm7
-; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
-; AVX1-NEXT:    vpshufb %xmm6, %xmm4, %xmm6
-; AVX1-NEXT:    vpaddb %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
+; AVX1-NEXT:    vpshufb %xmm6, %xmm3, %xmm6
+; AVX1-NEXT:    vpaddb %xmm6, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm2, %xmm6
 ; AVX1-NEXT:    vpsrlw $8, %xmm6, %xmm6
-; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm6
-; AVX1-NEXT:    vpsrlw $8, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddw %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm6
+; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddw %xmm6, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm2, %xmm6
 ; AVX1-NEXT:    vpsrld $16, %xmm6, %xmm6
-; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm6
-; AVX1-NEXT:    vpsrld $16, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddd %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm6
+; AVX1-NEXT:    vpsrld $16, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddd %xmm6, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlq $32, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm2
-; AVX1-NEXT:    vpsrlq $32, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddq %xmm2, %xmm5, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm5
-; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm5
+; AVX1-NEXT:    vpand %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpsrlq $32, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddq %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm4
 ; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm6
-; AVX1-NEXT:    vpand %xmm3, %xmm6, %xmm3
-; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm3, %xmm6
-; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpaddb %xmm3, %xmm5, %xmm3
+; AVX1-NEXT:    vpand %xmm5, %xmm6, %xmm5
+; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm5, %xmm6
+; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vpshufb %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddb %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm4
 ; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm4
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm4
 ; AVX1-NEXT:    vpsrld $16, %xmm4, %xmm4
 ; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm4
 ; AVX1-NEXT:    vpsrld $16, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm1
 ; AVX1-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: testv4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX2-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX2-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX2-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX2-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX2-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
 ; AVX2-NEXT:    vpsrld $16, %ymm2, %ymm2
 ; AVX2-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX2-NEXT:    vpsrld $16, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
 ; AVX2-NEXT:    vpcmpeqd %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrlq $32, %ymm0, %ymm0
 ; AVX2-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpsrlq $32, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddq %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VL-LABEL: testv4i64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsrld $16, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpsrld $16, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqd %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlq $32, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    vpsrlq $32, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddq %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv4i64:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VLBWDQ-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrld $16, %ymm2, %ymm2

 ; AVX512CD:       # %bb.0:
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512CD-NEXT:    vplzcntq %zmm0, %zmm0
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512CD-NEXT:    retq
 ;
 ; X32-AVX-LABEL: testv4i64:
 ; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; X32-AVX-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X32-AVX-NEXT:    vpand {{\.LCPI.*}}, %ymm3, %ymm3
 ; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; X32-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
 ; X32-AVX-NEXT:    vpsrld $16, %ymm2, %ymm2
; X32-AVX-NEXT:    retl
   %out = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %in, i1 0)
   ret <4 x i64> %out
 }
 
 define <4 x i64> @testv4i64u(<4 x i64> %in) nounwind {
 ; AVX1-LABEL: testv4i64u:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpshufb %xmm1, %xmm4, %xmm5
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vpshufb %xmm2, %xmm3, %xmm4
 ; AVX1-NEXT:    vpsrlw $4, %xmm2, %xmm1
-; AVX1-NEXT:    vpand %xmm3, %xmm1, %xmm6
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm5, %xmm1, %xmm6
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm6, %xmm7
-; AVX1-NEXT:    vpand %xmm7, %xmm5, %xmm5
-; AVX1-NEXT:    vpshufb %xmm6, %xmm4, %xmm6
-; AVX1-NEXT:    vpaddb %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpand %xmm7, %xmm4, %xmm4
+; AVX1-NEXT:    vpshufb %xmm6, %xmm3, %xmm6
+; AVX1-NEXT:    vpaddb %xmm6, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm2, %xmm6
 ; AVX1-NEXT:    vpsrlw $8, %xmm6, %xmm6
-; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm6
-; AVX1-NEXT:    vpsrlw $8, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddw %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm6
+; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddw %xmm6, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm2, %xmm6
 ; AVX1-NEXT:    vpsrld $16, %xmm6, %xmm6
-; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm6
-; AVX1-NEXT:    vpsrld $16, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddd %xmm6, %xmm5, %xmm5
+; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm6
+; AVX1-NEXT:    vpsrld $16, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddd %xmm6, %xmm4, %xmm4
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlq $32, %xmm2, %xmm2
-; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm2
-; AVX1-NEXT:    vpsrlq $32, %xmm5, %xmm5
-; AVX1-NEXT:    vpaddq %xmm2, %xmm5, %xmm2
-; AVX1-NEXT:    vpand %xmm3, %xmm0, %xmm5
-; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm5
+; AVX1-NEXT:    vpand %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpsrlq $32, %xmm4, %xmm4
+; AVX1-NEXT:    vpaddq %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpshufb %xmm0, %xmm3, %xmm4
 ; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm6
-; AVX1-NEXT:    vpand %xmm3, %xmm6, %xmm3
-; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm3, %xmm6
-; AVX1-NEXT:    vpand %xmm6, %xmm5, %xmm5
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpaddb %xmm3, %xmm5, %xmm3
+; AVX1-NEXT:    vpand %xmm5, %xmm6, %xmm5
+; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm5, %xmm6
+; AVX1-NEXT:    vpand %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vpshufb %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddb %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm4
 ; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
 ; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm4
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqw %xmm1, %xmm0, %xmm4
 ; AVX1-NEXT:    vpsrld $16, %xmm4, %xmm4
 ; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm4
 ; AVX1-NEXT:    vpsrld $16, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddd %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm1
 ; AVX1-NEXT:    vpaddq %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: testv4i64u:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX2-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX2-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX2-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX2-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX2-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
 ; AVX2-NEXT:    vpsrld $16, %ymm2, %ymm2
 ; AVX2-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX2-NEXT:    vpsrld $16, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
 ; AVX2-NEXT:    vpcmpeqd %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrlq $32, %ymm0, %ymm0
 ; AVX2-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpsrlq $32, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddq %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VL-LABEL: testv4i64u:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsrld $16, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpsrld $16, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqd %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlq $32, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    vpsrlq $32, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddq %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv4i64u:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VLBWDQ-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrld $16, %ymm2, %ymm2

 ; AVX512CD:       # %bb.0:
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512CD-NEXT:    vplzcntq %zmm0, %zmm0
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512CD-NEXT:    retq
 ;
 ; X32-AVX-LABEL: testv4i64u:
 ; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; X32-AVX-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X32-AVX-NEXT:    vpand {{\.LCPI.*}}, %ymm3, %ymm3
 ; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; X32-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
 ; X32-AVX-NEXT:    vpsrld $16, %ymm2, %ymm2
; X32-AVX-NEXT:    retl
   %out = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %in, i1 -1)
   ret <4 x i64> %out
 }
 
 define <8 x i32> @testv8i32(<8 x i32> %in) nounwind {
 ; AVX1-LABEL: testv8i32:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm5
-; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm5
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm5, %xmm7
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm4, %xmm7
 ; AVX1-NEXT:    vpand %xmm7, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm5
-; AVX1-NEXT:    vpaddb %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm1, %xmm5
-; AVX1-NEXT:    vpsrlw $8, %xmm5, %xmm5
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm5
+; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm4
+; AVX1-NEXT:    vpaddb %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm1, %xmm4
+; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
+; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm4
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX1-NEXT:    vpaddw %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqw %xmm6, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpsrld $16, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm3
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm5
-; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm2
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm5
+; AVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm4, %xmm5
 ; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm0, %xmm3
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm3
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpcmpeqw %xmm6, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpsrld $16, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddd %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: testv8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX2-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX2-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX2-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX2-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX2-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrld $16, %ymm0, %ymm0
 ; AVX2-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpsrld $16, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VL-LABEL: testv8i32:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrld $16, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    vpsrld $16, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv8i32:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VLBWDQ-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
 ; AVX512VLBWDQ-NEXT:    vpsrld $16, %ymm0, %ymm0

 ; AVX512CD:       # %bb.0:
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512CD-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512CD-NEXT:    retq
 ;
 ; X32-AVX-LABEL: testv8i32:
 ; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; X32-AVX-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X32-AVX-NEXT:    vpand {{\.LCPI.*}}, %ymm3, %ymm3
 ; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; X32-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
 ; X32-AVX-NEXT:    vpsrld $16, %ymm0, %ymm0
 ; X32-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; X32-AVX-NEXT:    vpsrld $16, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; X32-AVX-NEXT:    retl
 
   %out = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %in, i1 0)
   ret <8 x i32> %out
 }
 
 define <8 x i32> @testv8i32u(<8 x i32> %in) nounwind {
 ; AVX1-LABEL: testv8i32u:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm5
-; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm5
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm5, %xmm7
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm4, %xmm7
 ; AVX1-NEXT:    vpand %xmm7, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm5
-; AVX1-NEXT:    vpaddb %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm1, %xmm5
-; AVX1-NEXT:    vpsrlw $8, %xmm5, %xmm5
-; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm5
+; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm4
+; AVX1-NEXT:    vpaddb %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm1, %xmm4
+; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm4
+; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm4
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX1-NEXT:    vpaddw %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpaddw %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqw %xmm6, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrld $16, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpsrld $16, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddd %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm3
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm5
-; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm2
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm5
+; AVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm4, %xmm5
 ; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm0, %xmm3
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm3
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpcmpeqw %xmm6, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpsrld $16, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddd %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: testv8i32u:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX2-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX2-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX2-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX2-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX2-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrld $16, %ymm0, %ymm0
 ; AVX2-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpsrld $16, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VL-LABEL: testv8i32u:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrld $16, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    vpsrld $16, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv8i32u:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VLBWDQ-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
 ; AVX512VLBWDQ-NEXT:    vpsrld $16, %ymm0, %ymm0

 ; AVX512CD:       # %bb.0:
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512CD-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512CD-NEXT:    retq
 ;
 ; X32-AVX-LABEL: testv8i32u:
 ; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; X32-AVX-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X32-AVX-NEXT:    vpand {{\.LCPI.*}}, %ymm3, %ymm3
 ; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; X32-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
 ; X32-AVX-NEXT:    vpsrld $16, %ymm0, %ymm0
 ; X32-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; X32-AVX-NEXT:    vpsrld $16, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
 ; X32-AVX-NEXT:    retl
 
   %out = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %in, i1 -1)
   ret <8 x i32> %out
 }
 
 define <16 x i16> @testv16i16(<16 x i16> %in) nounwind {
 ; AVX1-LABEL: testv16i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm5
-; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm5
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm5, %xmm7
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm4, %xmm7
 ; AVX1-NEXT:    vpand %xmm7, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm5
-; AVX1-NEXT:    vpaddb %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm4
+; AVX1-NEXT:    vpaddb %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddw %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm3
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm5
-; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm2
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm5
+; AVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm4, %xmm5
 ; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: testv16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX2-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX2-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX2-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX2-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VL-LABEL: testv16i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv16i16:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VLBWDQ-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512VLBWDQ-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
 ; AVX512VLBWDQ-NEXT:    retq
 ;
 ; AVX512-LABEL: testv16i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512-NEXT:    vpsubw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512-NEXT:    retq
 ;
 ; X32-AVX-LABEL: testv16i16:
 ; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; X32-AVX-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X32-AVX-NEXT:    vpand {{\.LCPI.*}}, %ymm3, %ymm3
 ; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; X32-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
 ; X32-AVX-NEXT:    retl
   %out = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %in, i1 0)
   ret <16 x i16> %out
 }
 
 define <16 x i16> @testv16i16u(<16 x i16> %in) nounwind {
 ; AVX1-LABEL: testv16i16u:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm5
-; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm5
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm4
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
 ; AVX1-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm5, %xmm7
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm4, %xmm7
 ; AVX1-NEXT:    vpand %xmm7, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm5, %xmm4, %xmm5
-; AVX1-NEXT:    vpaddb %xmm5, %xmm3, %xmm3
+; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm4
+; AVX1-NEXT:    vpaddb %xmm4, %xmm3, %xmm3
 ; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX1-NEXT:    vpand %xmm1, %xmm3, %xmm1
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
 ; AVX1-NEXT:    vpaddw %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm3
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm5
-; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm2
-; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm2, %xmm5
+; AVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm3
+; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm4
+; AVX1-NEXT:    vpand %xmm5, %xmm4, %xmm4
+; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm4, %xmm5
 ; AVX1-NEXT:    vpand %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpshufb %xmm4, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpeqb %xmm6, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpaddw %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: testv16i16u:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX2-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX2-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX2-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX2-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX2-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX2-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX2-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VL-LABEL: testv16i16u:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VL-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv16i16u:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VLBWDQ-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; AVX512VLBWDQ-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512VLBWDQ-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX512VLBWDQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VLBWDQ-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
 ; AVX512VLBWDQ-NEXT:    retq
 ;
 ; AVX512-LABEL: testv16i16u:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512-NEXT:    vpsubw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512-NEXT:    retq
 ;
 ; X32-AVX-LABEL: testv16i16u:
 ; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm4
-; X32-AVX-NEXT:    vpand %ymm1, %ymm4, %ymm1
+; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X32-AVX-NEXT:    vpand {{\.LCPI.*}}, %ymm3, %ymm3
 ; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm1, %ymm5
+; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
 ; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
+; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
 ; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; X32-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; X32-AVX-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
 ; X32-AVX-NEXT:    retl
   %out = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %in, i1 -1)
   ret <16 x i16> %out
 }
 
 define <32 x i8> @testv32i8(<32 x i8> %in) nounwind {
 ; AVX1-LABEL: testv32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm3
 ; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
 ; AVX1-NEXT:    vpcmpeqb %xmm5, %xmm1, %xmm6
 ; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm1, %xmm4, %xmm1
+; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm3
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm3
 ; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqb %xmm5, %xmm0, %xmm2
-; AVX1-NEXT:    vpand %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpshufb %xmm0, %xmm4, %xmm0
-; AVX1-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
+; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqb %xmm5, %xmm0, %xmm4
+; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm0
+; AVX1-NEXT:    vpaddb %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: testv32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm1
-; AVX2-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX2-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VL-LABEL: testv32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm1
-; AVX512VL-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; AVX512VL-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512VL-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv32i8:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512VLBWDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm1
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512VLBWDQ-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; AVX512VLBWDQ-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
 ; AVX512VLBWDQ-NEXT:    retq
 ;
 ; AVX512-LABEL: testv32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512-NEXT:    vplzcntd %zmm1, %zmm1
 ; AVX512-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm2 = [24,24,24,24,24,24,24,24,24,24,24,24,24,24,24,24]
 ; AVX512-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
 ; AVX512-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512-NEXT:    vpsubb %xmm2, %xmm0, %xmm0
 ; AVX512-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX512-NEXT:    retq
 ;
 ; X32-AVX-LABEL: testv32i8:
 ; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; X32-AVX-NEXT:    vpand {{\.LCPI.*}}, %ymm0, %ymm0
+; X32-AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; X32-AVX-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; X32-AVX-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; X32-AVX-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
 ; X32-AVX-NEXT:    retl
   %out = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %in, i1 0)
   ret <32 x i8> %out
 }
 
 define <32 x i8> @testv32i8u(<32 x i8> %in) nounwind {
 ; AVX1-LABEL: testv32i8u:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm3
 ; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm1
-; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX1-NEXT:    vpand %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
 ; AVX1-NEXT:    vpcmpeqb %xmm5, %xmm1, %xmm6
 ; AVX1-NEXT:    vpand %xmm6, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufb %xmm1, %xmm4, %xmm1
+; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm3
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm3
 ; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm0
-; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpcmpeqb %xmm5, %xmm0, %xmm2
-; AVX1-NEXT:    vpand %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpshufb %xmm0, %xmm4, %xmm0
-; AVX1-NEXT:    vpaddb %xmm0, %xmm2, %xmm0
+; AVX1-NEXT:    vpand %xmm4, %xmm0, %xmm0
+; AVX1-NEXT:    vpcmpeqb %xmm5, %xmm0, %xmm4
+; AVX1-NEXT:    vpand %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpshufb %xmm0, %xmm2, %xmm0
+; AVX1-NEXT:    vpaddb %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: testv32i8u:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX2-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX2-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm1
-; AVX2-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; AVX2-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX2-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; AVX2-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX2-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512VL-LABEL: testv32i8u:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VL-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VL-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm1
-; AVX512VL-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; AVX512VL-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512VL-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX512VL-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512VLBWDQ-LABEL: testv32i8u:
 ; AVX512VLBWDQ:       # %bb.0:
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; AVX512VLBWDQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; AVX512VLBWDQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512VLBWDQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm1
-; AVX512VLBWDQ-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512VLBWDQ-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VLBWDQ-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX512VLBWDQ-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; AVX512VLBWDQ-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; AVX512VLBWDQ-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; AVX512VLBWDQ-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
 ; AVX512VLBWDQ-NEXT:    retq
 ;
 ; AVX512-LABEL: testv32i8u:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmovzxbd {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
 ; AVX512-NEXT:    vplzcntd %zmm1, %zmm1
 ; AVX512-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512-NEXT:    vmovdqa {{.*#+}} xmm2 = [24,24,24,24,24,24,24,24,24,24,24,24,24,24,24,24]
 ; AVX512-NEXT:    vpsubb %xmm2, %xmm1, %xmm1
 ; AVX512-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
 ; AVX512-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512-NEXT:    vpsubb %xmm2, %xmm0, %xmm0
 ; AVX512-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX512-NEXT:    retq
 ;
 ; X32-AVX-LABEL: testv32i8u:
 ; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; X32-AVX-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
 ; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpand %ymm1, %ymm2, %ymm1
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; X32-AVX-NEXT:    vpand {{\.LCPI.*}}, %ymm0, %ymm0
+; X32-AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; X32-AVX-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; X32-AVX-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; X32-AVX-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
 ; X32-AVX-NEXT:    retl
   %out = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %in, i1 -1)
   ret <32 x i8> %out
 }
 
 define <4 x i64> @foldv4i64() nounwind {
 ; X64-LABEL: foldv4i64:
 ; X64:       # %bb.0:

llvm/trunk/test/CodeGen/X86/vector-lzcnt-512.ll

 ; AVX512CDBW-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512CDBW-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512CDBW-NEXT:    vpsubw %ymm2, %ymm0, %ymm0
 ; AVX512CDBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512CDBW-NEXT:    retq
 ;
 ; AVX512BW-LABEL: testv32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm4
-; AVX512BW-NEXT:    vptestnmb %zmm1, %zmm4, %k0
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm1, %zmm2
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm3
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vptestnmb %zmm4, %zmm3, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm5
 ; AVX512BW-NEXT:    vpandq %zmm5, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm4, %zmm1
-; AVX512BW-NEXT:    vpshufb %zmm1, %zmm3, %zmm1
+; AVX512BW-NEXT:    vpandq %zmm4, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpshufb %zmm3, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpaddb %zmm1, %zmm2, %zmm1
 ; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpandq %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpaddw %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: testv32i16:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm3
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
-; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm5
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm5, %ymm5
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm2, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm4
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm5, %ymm4, %ymm4
 ; AVX512DQ-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm5, %ymm7
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm4, %ymm7
 ; AVX512DQ-NEXT:    vpand %ymm7, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
-; AVX512DQ-NEXT:    vpaddb %ymm5, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm2, %ymm4
+; AVX512DQ-NEXT:    vpaddb %ymm4, %ymm3, %ymm3
 ; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpand %ymm0, %ymm3, %ymm0
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm3, %ymm3
 ; AVX512DQ-NEXT:    vpaddw %ymm0, %ymm3, %ymm0
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
-; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm5
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm5, %ymm2
-; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm2, %ymm5
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm2, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm4
+; AVX512DQ-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm4, %ymm5
 ; AVX512DQ-NEXT:    vpand %ymm5, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm4, %ymm2
+; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm2, %ymm2
 ; AVX512DQ-NEXT:    vpaddb %ymm2, %ymm3, %ymm2
 ; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm1, %ymm1
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512DQ-NEXT:    vpand %ymm1, %ymm2, %ymm1
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512DQ-NEXT:    vpaddw %ymm1, %ymm2, %ymm1
 ; AVX512DQ-NEXT:    retq
   %out = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %in, i1 0)

 ; AVX512CDBW-NEXT:    vplzcntd %zmm0, %zmm0
 ; AVX512CDBW-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512CDBW-NEXT:    vpsubw %ymm2, %ymm0, %ymm0
 ; AVX512CDBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512CDBW-NEXT:    retq
 ;
 ; AVX512BW-LABEL: testv32i16u:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm4
-; AVX512BW-NEXT:    vptestnmb %zmm1, %zmm4, %k0
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm1, %zmm2
+; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm3
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vptestnmb %zmm4, %zmm3, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm5
 ; AVX512BW-NEXT:    vpandq %zmm5, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm4, %zmm1
-; AVX512BW-NEXT:    vpshufb %zmm1, %zmm3, %zmm1
+; AVX512BW-NEXT:    vpandq %zmm4, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpshufb %zmm3, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpaddb %zmm1, %zmm2, %zmm1
 ; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm0, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpandq %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpaddw %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: testv32i16u:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm3
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
-; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm5
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm5, %ymm5
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm2, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm4
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm5 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm5, %ymm4, %ymm4
 ; AVX512DQ-NEXT:    vpxor %xmm6, %xmm6, %xmm6
-; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm5, %ymm7
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm4, %ymm7
 ; AVX512DQ-NEXT:    vpand %ymm7, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm5, %ymm4, %ymm5
-; AVX512DQ-NEXT:    vpaddb %ymm5, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm2, %ymm4
+; AVX512DQ-NEXT:    vpaddb %ymm4, %ymm3, %ymm3
 ; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpand %ymm0, %ymm3, %ymm0
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm3, %ymm3
 ; AVX512DQ-NEXT:    vpaddw %ymm0, %ymm3, %ymm0
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
-; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm5
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm5, %ymm2
-; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm2, %ymm5
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm2, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm4
+; AVX512DQ-NEXT:    vpand %ymm5, %ymm4, %ymm4
+; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm4, %ymm5
 ; AVX512DQ-NEXT:    vpand %ymm5, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm2, %ymm4, %ymm2
+; AVX512DQ-NEXT:    vpshufb %ymm4, %ymm2, %ymm2
 ; AVX512DQ-NEXT:    vpaddb %ymm2, %ymm3, %ymm2
 ; AVX512DQ-NEXT:    vpcmpeqb %ymm6, %ymm1, %ymm1
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512DQ-NEXT:    vpand %ymm1, %ymm2, %ymm1
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512DQ-NEXT:    vpaddw %ymm1, %ymm2, %ymm1
 ; AVX512DQ-NEXT:    retq
   %out = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %in, i1 -1)

 ; AVX512CDBW-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512CDBW-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
 ; AVX512CDBW-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; AVX512CDBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512CDBW-NEXT:    retq
 ;
 ; AVX512BW-LABEL: testv64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm1, %zmm2
 ; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestnmb %zmm1, %zmm0, %k0
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vptestnmb %zmm3, %zmm0, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm4
 ; AVX512BW-NEXT:    vpandq %zmm4, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpaddb %zmm0, %zmm2, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: testv64i8:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm3
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm2, %ymm3
 ; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm4, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpxor %xmm5, %xmm5, %xmm5
 ; AVX512DQ-NEXT:    vpcmpeqb %ymm5, %ymm0, %ymm6
 ; AVX512DQ-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm2, %ymm0
 ; AVX512DQ-NEXT:    vpaddb %ymm0, %ymm3, %ymm0
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm2, %ymm3
 ; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm2
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm3, %ymm2
-; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
-; AVX512DQ-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm4, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm4
+; AVX512DQ-NEXT:    vpand %ymm4, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vpaddb %ymm1, %ymm3, %ymm1
 ; AVX512DQ-NEXT:    retq
   %out = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %in, i1 0)
   ret <64 x i8> %out
 }
 
 define <64 x i8> @testv64i8u(<64 x i8> %in) nounwind {
 ; AVX512CD-LABEL: testv64i8u:
 ; AVX512CD:       # %bb.0:

 ; AVX512CDBW-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512CDBW-NEXT:    vpsubb %xmm3, %xmm0, %xmm0
 ; AVX512CDBW-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm0
 ; AVX512CDBW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512CDBW-NEXT:    retq
 ;
 ; AVX512BW-LABEL: testv64i8u:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512BW-NEXT:    vpshufb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm1, %zmm2
 ; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm0
-; AVX512BW-NEXT:    vptestnmb %zmm1, %zmm0, %k0
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512BW-NEXT:    vptestnmb %zmm3, %zmm0, %k0
 ; AVX512BW-NEXT:    vpmovm2b %k0, %zmm4
 ; AVX512BW-NEXT:    vpandq %zmm4, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpshufb %zmm0, %zmm3, %zmm0
+; AVX512BW-NEXT:    vpandq %zmm3, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpshufb %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpaddb %zmm0, %zmm2, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: testv64i8u:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm3
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm2, %ymm3
 ; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512DQ-NEXT:    vpand %ymm4, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpxor %xmm5, %xmm5, %xmm5
 ; AVX512DQ-NEXT:    vpcmpeqb %ymm5, %ymm0, %ymm6
 ; AVX512DQ-NEXT:    vpand %ymm6, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
+; AVX512DQ-NEXT:    vpshufb %ymm0, %ymm2, %ymm0
 ; AVX512DQ-NEXT:    vpaddb %ymm0, %ymm3, %ymm0
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm3
-; AVX512DQ-NEXT:    vpshufb %ymm3, %ymm4, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm2, %ymm3
 ; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm2
-; AVX512DQ-NEXT:    vpand %ymm2, %ymm3, %ymm2
-; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm4, %ymm1
-; AVX512DQ-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vpand %ymm4, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpcmpeqb %ymm5, %ymm1, %ymm4
+; AVX512DQ-NEXT:    vpand %ymm4, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpshufb %ymm1, %ymm2, %ymm1
+; AVX512DQ-NEXT:    vpaddb %ymm1, %ymm3, %ymm1
 ; AVX512DQ-NEXT:    retq
   %out = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %in, i1 -1)
   ret <64 x i8> %out
 }
 
 declare <8 x i64> @llvm.ctlz.v8i64(<8 x i64>, i1)
 declare <16 x i32> @llvm.ctlz.v16i32(<16 x i32>, i1)
 declare <32 x i16> @llvm.ctlz.v32i16(<32 x i16>, i1)
 declare <64 x i8> @llvm.ctlz.v64i8(<64 x i8>, i1)