[VectorLegalizer] Expansion of CTLZ using CTPOP when possible

[RKSimon]

Summary

This patch avoids scalarization of CTLZ by instead expanding to use CTPOP (from "Hacker's Delight") when the necessary operations are available.

This also adds the necessary cost models for X86 SSE2 targets (the main beneficiary) to ensure vectorization only happens when its useful.

[craig.topper]

lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp

1070–1071

Isn't this ternary operator unneeded now?

[RKSimon]

Simplified ExpandCTTZ_ZERO_UNDEF since we know its ISD::CTTZ_ZERO_UNDEF - thanks Craig

[craig.topper]

lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp

1056

Do we need to protect non-power of 2 bit widths here? I think the scalar equivalent does protect that.

[RKSimon]

Limit vector expansion to power-of-2 sized elements

[craig.topper]

LGTM

lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp

class VectorLegalizer {
   /// \brief Implement vselect in terms of XOR, AND, OR when blend is not
   /// supported by the target.
   SDValue ExpandVSELECT(SDValue Op);
   SDValue ExpandSELECT(SDValue Op);
   SDValue ExpandLoad(SDValue Op);
   SDValue ExpandStore(SDValue Op);
   SDValue ExpandFNEG(SDValue Op);
   SDValue ExpandBITREVERSE(SDValue Op);
-  SDValue ExpandCTLZ_CTTZ_ZERO_UNDEF(SDValue Op);
+  SDValue ExpandCTLZ(SDValue Op);
+  SDValue ExpandCTTZ_ZERO_UNDEF(SDValue Op);
 
   /// \brief Implements vector promotion.
   ///
   /// This is essentially just bitcasting the operands to a different type and
   /// bitcasting the result back to the original type.
   SDValue Promote(SDValue Op);
 
   /// \brief Implements [SU]INT_TO_FP vector promotion.
SDValue VectorLegalizer::Expand(SDValue Op) {
   case ISD::UINT_TO_FP:
     return ExpandUINT_TO_FLOAT(Op);
   case ISD::FNEG:
     return ExpandFNEG(Op);
   case ISD::SETCC:
     return UnrollVSETCC(Op);
   case ISD::BITREVERSE:
     return ExpandBITREVERSE(Op);
+  case ISD::CTLZ:
   case ISD::CTLZ_ZERO_UNDEF:
+    return ExpandCTLZ(Op);
   case ISD::CTTZ_ZERO_UNDEF:
-    return ExpandCTLZ_CTTZ_ZERO_UNDEF(Op);
+    return ExpandCTTZ_ZERO_UNDEF(Op);
   default:
     return DAG.UnrollVectorOp(Op.getNode());
   }
 }
 
 SDValue VectorLegalizer::ExpandSELECT(SDValue Op) {
   // Lower a select instruction where the condition is a scalar and the
   // operands are vectors. Lower this select to VSELECT and implement it
if (TLI.isOperationLegalOrCustom(ISD::FSUB, Op.getValueType())) {
     SDValue Zero = DAG.getConstantFP(-0.0, DL, Op.getValueType());
     // TODO: If FNEG had fast-math-flags, they'd get propagated to this FSUB.
     return DAG.getNode(ISD::FSUB, DL, Op.getValueType(),
                        Zero, Op.getOperand(0));
   }
   return DAG.UnrollVectorOp(Op.getNode());
 }
 
-SDValue VectorLegalizer::ExpandCTLZ_CTTZ_ZERO_UNDEF(SDValue Op) {
+SDValue VectorLegalizer::ExpandCTLZ(SDValue Op) {
+  EVT VT = Op.getValueType();
+
+  // If the non-ZERO_UNDEF version is supported we can use that instead.
+  if (Op.getOpcode() == ISD::CTLZ_ZERO_UNDEF &&
+      TLI.isOperationLegalOrCustom(ISD::CTLZ, VT)) {
+    SDLoc DL(Op);
+    return DAG.getNode(ISD::CTLZ, DL, Op.getValueType(), Op.getOperand(0));
+  }
+
+  // If CTPOP is available we can lower with a CTPOP based method:
+  // u16 ctlz(u16 x) {
+  //   x |= (x >> 1);
+  //   x |= (x >> 2);
+  //   x |= (x >> 4);
+  //   x |= (x >> 8);
+  //   return ctpop(~x);
+  // }
+  // Ref: "Hacker's Delight" by Henry Warren
+  if (TLI.isOperationLegalOrCustom(ISD::CTPOP, VT) &&
+      TLI.isOperationLegalOrCustom(ISD::SRL, VT) &&
+      TLI.isOperationLegalOrCustomOrPromote(ISD::OR, VT) &&
+      TLI.isOperationLegalOrCustomOrPromote(ISD::XOR, VT)) {
+    SDLoc DL(Op);
+    SDValue Res = Op.getOperand(0);
+    unsigned NumBitsPerElt = VT.getScalarSizeInBits();
+    EVT ShiftTy = TLI.getShiftAmountTy(VT, DAG.getDataLayout());
+
+    for (unsigned i = 1; i != NumBitsPerElt; i *= 2)
+      Res = DAG.getNode(
+          ISD::OR, DL, VT, Res,
+          DAG.getNode(ISD::SRL, DL, VT, Res, DAG.getConstant(i, DL, ShiftTy)));
+
+    Res = DAG.getNOT(DL, Res, VT);
+    return DAG.getNode(ISD::CTPOP, DL, VT, Res);
+  }
+
+  // Otherwise go ahead and unroll.
+  return DAG.UnrollVectorOp(Op.getNode());
+}
+
+SDValue VectorLegalizer::ExpandCTTZ_ZERO_UNDEF(SDValue Op) {
   // If the non-ZERO_UNDEF version is supported we can use that instead.
-  unsigned Opc = Op.getOpcode() == ISD::CTLZ_ZERO_UNDEF ? ISD::CTLZ : ISD::CTTZ;
-  if (TLI.isOperationLegalOrCustom(Opc, Op.getValueType())) {
+  if (TLI.isOperationLegalOrCustom(ISD::CTTZ, Op.getValueType())) {
     SDLoc DL(Op);
-    return DAG.getNode(Opc, DL, Op.getValueType(), Op.getOperand(0));
+    return DAG.getNode(ISD::CTTZ, DL, Op.getValueType(), Op.getOperand(0));
   }
 
   // Otherwise go ahead and unroll.
   return DAG.UnrollVectorOp(Op.getNode());
 }
 
 SDValue VectorLegalizer::UnrollVSETCC(SDValue Op) {
   EVT VT = Op.getValueType();

lib/Target/X86/X86TargetTransformInfo.cpp

static const CostTblEntry SSSE3CostTbl[] = {
     { ISD::CTTZ,       MVT::v4i32,  14 },
     { ISD::CTTZ,       MVT::v8i16,  12 },
     { ISD::CTTZ,       MVT::v16i8,   9 }
   };
   static const CostTblEntry SSE2CostTbl[] = {
     { ISD::BSWAP,      MVT::v2i64,   7 },
     { ISD::BSWAP,      MVT::v4i32,   7 },
     { ISD::BSWAP,      MVT::v8i16,   7 },
-    /* ISD::CTLZ - currently scalarized pre-SSSE3 */
+    { ISD::CTLZ,       MVT::v2i64,  25 },
+    { ISD::CTLZ,       MVT::v4i32,  26 },
+    { ISD::CTLZ,       MVT::v8i16,  20 },
+    { ISD::CTLZ,       MVT::v16i8,  17 },
     { ISD::CTPOP,      MVT::v2i64,  12 },
     { ISD::CTPOP,      MVT::v4i32,  15 },
     { ISD::CTPOP,      MVT::v8i16,  13 },
     { ISD::CTPOP,      MVT::v16i8,  10 },
     { ISD::CTTZ,       MVT::v2i64,  14 },
     { ISD::CTTZ,       MVT::v4i32,  18 },
     { ISD::CTTZ,       MVT::v8i16,  16 },
     { ISD::CTTZ,       MVT::v16i8,  13 }

test/Analysis/CostModel/X86/ctbits-cost.ll

 
 declare <4 x i64> @llvm.ctlz.v4i64(<4 x i64>, i1)
 declare <8 x i32> @llvm.ctlz.v8i32(<8 x i32>, i1)
 declare <16 x i16> @llvm.ctlz.v16i16(<16 x i16>, i1)
 declare <32 x i8> @llvm.ctlz.v32i8(<32 x i8>, i1)
 
 define <2 x i64> @var_ctlz_v2i64(<2 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v2i64':
-; SSE2: Found an estimated cost of 6 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 25 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 23 for instruction:   %ctlz
 ; AVX: Found an estimated cost of 23 for instruction:   %ctlz
   %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 0)
   ret <2 x i64> %ctlz
 }
 
 define <2 x i64> @var_ctlz_v2i64u(<2 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v2i64u':
-; SSE2: Found an estimated cost of 6 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 25 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 23 for instruction:   %ctlz
 ; AVX: Found an estimated cost of 23 for instruction:   %ctlz
   %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 1)
   ret <2 x i64> %ctlz
 }
 
 define <4 x i64> @var_ctlz_v4i64(<4 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v4i64':
-; SSE2: Found an estimated cost of 12 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 50 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 46 for instruction:   %ctlz
 ; AVX1: Found an estimated cost of 46 for instruction:   %ctlz
 ; AVX2: Found an estimated cost of 23 for instruction:   %ctlz
   %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 0)
   ret <4 x i64> %ctlz
 }
 
 define <4 x i64> @var_ctlz_v4i64u(<4 x i64> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v4i64u':
-; SSE2: Found an estimated cost of 12 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 50 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 46 for instruction:   %ctlz
 ; AVX1: Found an estimated cost of 46 for instruction:   %ctlz
 ; AVX2: Found an estimated cost of 23 for instruction:   %ctlz
   %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 1)
   ret <4 x i64> %ctlz
 }
 
 define <4 x i32> @var_ctlz_v4i32(<4 x i32> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v4i32':
-; SSE2: Found an estimated cost of 12 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 26 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 18 for instruction:   %ctlz
 ; AVX: Found an estimated cost of 18 for instruction:   %ctlz
   %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 0)
   ret <4 x i32> %ctlz
 }
 
 define <4 x i32> @var_ctlz_v4i32u(<4 x i32> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v4i32u':
-; SSE2: Found an estimated cost of 12 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 26 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 18 for instruction:   %ctlz
 ; AVX: Found an estimated cost of 18 for instruction:   %ctlz
   %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 1)
   ret <4 x i32> %ctlz
 }
 
 define <8 x i32> @var_ctlz_v8i32(<8 x i32> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v8i32':
-; SSE2: Found an estimated cost of 24 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 52 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 36 for instruction:   %ctlz
 ; AVX1: Found an estimated cost of 36 for instruction:   %ctlz
 ; AVX2: Found an estimated cost of 18 for instruction:   %ctlz
   %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 0)
   ret <8 x i32> %ctlz
 }
 
 define <8 x i32> @var_ctlz_v8i32u(<8 x i32> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v8i32u':
-; SSE2: Found an estimated cost of 24 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 52 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 36 for instruction:   %ctlz
 ; AVX1: Found an estimated cost of 36 for instruction:   %ctlz
 ; AVX2: Found an estimated cost of 18 for instruction:   %ctlz
   %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 1)
   ret <8 x i32> %ctlz
 }
 
 define <8 x i16> @var_ctlz_v8i16(<8 x i16> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v8i16':
-; SSE2: Found an estimated cost of 24 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 20 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 14 for instruction:   %ctlz
 ; AVX: Found an estimated cost of 14 for instruction:   %ctlz
   %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 0)
   ret <8 x i16> %ctlz
 }
 
 define <8 x i16> @var_ctlz_v8i16u(<8 x i16> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v8i16u':
-; SSE2: Found an estimated cost of 24 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 20 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 14 for instruction:   %ctlz
 ; AVX: Found an estimated cost of 14 for instruction:   %ctlz
   %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 1)
   ret <8 x i16> %ctlz
 }
 
 define <16 x i16> @var_ctlz_v16i16(<16 x i16> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v16i16':
-; SSE2: Found an estimated cost of 48 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 40 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 28 for instruction:   %ctlz
 ; AVX1: Found an estimated cost of 28 for instruction:   %ctlz
 ; AVX2: Found an estimated cost of 14 for instruction:   %ctlz
   %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 0)
   ret <16 x i16> %ctlz
 }
 
 define <16 x i16> @var_ctlz_v16i16u(<16 x i16> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v16i16u':
-; SSE2: Found an estimated cost of 48 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 40 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 28 for instruction:   %ctlz
 ; AVX1: Found an estimated cost of 28 for instruction:   %ctlz
 ; AVX2: Found an estimated cost of 14 for instruction:   %ctlz
   %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 1)
   ret <16 x i16> %ctlz
 }
 
 define <16 x i8> @var_ctlz_v16i8(<16 x i8> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v16i8':
-; SSE2: Found an estimated cost of 48 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 17 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 9 for instruction:   %ctlz
 ; AVX: Found an estimated cost of 9 for instruction:   %ctlz
   %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 0)
   ret <16 x i8> %ctlz
 }
 
 define <16 x i8> @var_ctlz_v16i8u(<16 x i8> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v16i8u':
-; SSE2: Found an estimated cost of 48 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 17 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 9 for instruction:   %ctlz
 ; AVX: Found an estimated cost of 9 for instruction:   %ctlz
   %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 1)
   ret <16 x i8> %ctlz
 }
 
 define <32 x i8> @var_ctlz_v32i8(<32 x i8> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v32i8':
-; SSE2: Found an estimated cost of 96 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 34 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 18 for instruction:   %ctlz
 ; AVX1: Found an estimated cost of 18 for instruction:   %ctlz
 ; AVX2: Found an estimated cost of 9 for instruction:   %ctlz
   %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 0)
   ret <32 x i8> %ctlz
 }
 
 define <32 x i8> @var_ctlz_v32i8u(<32 x i8> %a) {
 ; CHECK: 'Cost Model Analysis' for function 'var_ctlz_v32i8u':
-; SSE2: Found an estimated cost of 96 for instruction:   %ctlz
+; SSE2: Found an estimated cost of 34 for instruction:   %ctlz
 ; SSE42: Found an estimated cost of 18 for instruction:   %ctlz
 ; AVX1: Found an estimated cost of 18 for instruction:   %ctlz
 ; AVX2: Found an estimated cost of 9 for instruction:   %ctlz
   %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 1)
   ret <32 x i8> %ctlz
 }
 
 ; Verify the cost of scalar trailing zero count instructions.

test/CodeGen/X86/vec_ctbits.ll

 ; CHECK-NEXT:    retq
   %c = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
   ret <2 x i64> %c
 
 }
 define <2 x i64> @foolz(<2 x i64> %a) nounwind {
 ; CHECK-LABEL: foolz:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    movd %xmm0, %rax
-; CHECK-NEXT:    bsrq %rax, %rax
-; CHECK-NEXT:    xorq $63, %rax
-; CHECK-NEXT:    movd %rax, %xmm1
-; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-NEXT:    movd %xmm0, %rax
-; CHECK-NEXT:    bsrq %rax, %rax
-; CHECK-NEXT:    xorq $63, %rax
-; CHECK-NEXT:    movd %rax, %xmm0
-; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    psrlq $1, %xmm1
+; CHECK-NEXT:    por %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    psrlq $2, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    psrlq $4, %xmm1
+; CHECK-NEXT:    por %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    psrlq $8, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    psrlq $16, %xmm1
+; CHECK-NEXT:    por %xmm0, %xmm1
 ; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    psrlq $32, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
+; CHECK-NEXT:    pcmpeqd %xmm1, %xmm1
+; CHECK-NEXT:    pxor %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    psrlq $1, %xmm0
+; CHECK-NEXT:    pand {{.*}}(%rip), %xmm0
+; CHECK-NEXT:    psubq %xmm0, %xmm1
+; CHECK-NEXT:    movdqa {{.*#+}} xmm0 = [3689348814741910323,3689348814741910323]
+; CHECK-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-NEXT:    pand %xmm0, %xmm2
+; CHECK-NEXT:    psrlq $2, %xmm1
+; CHECK-NEXT:    pand %xmm0, %xmm1
+; CHECK-NEXT:    paddq %xmm2, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-NEXT:    psrlq $4, %xmm2
+; CHECK-NEXT:    paddq %xmm1, %xmm2
+; CHECK-NEXT:    pand {{.*}}(%rip), %xmm2
+; CHECK-NEXT:    pxor %xmm0, %xmm0
+; CHECK-NEXT:    psadbw %xmm2, %xmm0
 ; CHECK-NEXT:    retq
   %c = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
   ret <2 x i64> %c
 
 }
 
 define <2 x i64> @foopop(<2 x i64> %a) nounwind {
 ; CHECK-LABEL: foopop:
; CHECK-NEXT:    retq
   %c = call <2 x i32> @llvm.cttz.v2i32(<2 x i32> %a, i1 false)
   ret <2 x i32> %c
 
 }
 define <2 x i32> @promlz(<2 x i32> %a) nounwind {
 ; CHECK-LABEL: promlz:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    pand {{.*}}(%rip), %xmm0
-; CHECK-NEXT:    movd %xmm0, %rax
-; CHECK-NEXT:    bsrq %rax, %rax
-; CHECK-NEXT:    movl $127, %ecx
-; CHECK-NEXT:    cmoveq %rcx, %rax
-; CHECK-NEXT:    xorq $63, %rax
-; CHECK-NEXT:    movd %rax, %xmm1
-; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; CHECK-NEXT:    movd %xmm0, %rax
-; CHECK-NEXT:    bsrq %rax, %rax
-; CHECK-NEXT:    cmoveq %rcx, %rax
-; CHECK-NEXT:    xorq $63, %rax
-; CHECK-NEXT:    movd %rax, %xmm0
-; CHECK-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
-; CHECK-NEXT:    psubq {{.*}}(%rip), %xmm1
-; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    pxor %xmm1, %xmm1
+; CHECK-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-NEXT:    psrlq $1, %xmm2
+; CHECK-NEXT:    por %xmm0, %xmm2
+; CHECK-NEXT:    movdqa %xmm2, %xmm0
+; CHECK-NEXT:    psrlq $2, %xmm0
+; CHECK-NEXT:    por %xmm2, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-NEXT:    psrlq $4, %xmm2
+; CHECK-NEXT:    por %xmm0, %xmm2
+; CHECK-NEXT:    movdqa %xmm2, %xmm0
+; CHECK-NEXT:    psrlq $8, %xmm0
+; CHECK-NEXT:    por %xmm2, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm2
+; CHECK-NEXT:    psrlq $16, %xmm2
+; CHECK-NEXT:    por %xmm0, %xmm2
+; CHECK-NEXT:    movdqa %xmm2, %xmm0
+; CHECK-NEXT:    psrlq $32, %xmm0
+; CHECK-NEXT:    por %xmm2, %xmm0
+; CHECK-NEXT:    pcmpeqd %xmm2, %xmm2
+; CHECK-NEXT:    pxor %xmm0, %xmm2
+; CHECK-NEXT:    movdqa %xmm2, %xmm0
+; CHECK-NEXT:    psrlq $1, %xmm0
+; CHECK-NEXT:    pand {{.*}}(%rip), %xmm0
+; CHECK-NEXT:    psubq %xmm0, %xmm2
+; CHECK-NEXT:    movdqa {{.*#+}} xmm0 = [3689348814741910323,3689348814741910323]
+; CHECK-NEXT:    movdqa %xmm2, %xmm3
+; CHECK-NEXT:    pand %xmm0, %xmm3
+; CHECK-NEXT:    psrlq $2, %xmm2
+; CHECK-NEXT:    pand %xmm0, %xmm2
+; CHECK-NEXT:    paddq %xmm3, %xmm2
+; CHECK-NEXT:    movdqa %xmm2, %xmm0
+; CHECK-NEXT:    psrlq $4, %xmm0
+; CHECK-NEXT:    paddq %xmm2, %xmm0
+; CHECK-NEXT:    pand {{.*}}(%rip), %xmm0
+; CHECK-NEXT:    psadbw %xmm1, %xmm0
+; CHECK-NEXT:    psubq {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    retq
   %c = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %a, i1 false)
   ret <2 x i32> %c
 
 }
 
 define <2 x i32> @prompop(<2 x i32> %a) nounwind {
 ; CHECK-LABEL: prompop:

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

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse3 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE3
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSSE3
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512cd -mattr=+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512VLCD
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512cd | FileCheck %s --check-prefix=ALL --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=+sse4.1 | FileCheck %s --check-prefix=ALL --check-prefix=X32-SSE --check-prefix=X32-SSE41
 
 define <2 x i64> @testv2i64(<2 x i64> %in) nounwind {
 ; SSE2-LABEL: testv2i64:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    movd %xmm0, %rax
-; SSE2-NEXT:    bsrq %rax, %rax
-; SSE2-NEXT:    movl $127, %ecx
-; SSE2-NEXT:    cmoveq %rcx, %rax
-; SSE2-NEXT:    xorq $63, %rax
-; SSE2-NEXT:    movd %rax, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE2-NEXT:    movd %xmm0, %rax
-; SSE2-NEXT:    bsrq %rax, %rax
-; SSE2-NEXT:    cmoveq %rcx, %rax
-; SSE2-NEXT:    xorq $63, %rax
-; SSE2-NEXT:    movd %rax, %xmm0
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlq $1, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlq $2, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlq $4, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlq $8, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlq $16, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlq $32, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT:    pxor %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlq $1, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    psubq %xmm0, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [3689348814741910323,3689348814741910323]
+; SSE2-NEXT:    movdqa %xmm1, %xmm2
+; SSE2-NEXT:    pand %xmm0, %xmm2
+; SSE2-NEXT:    psrlq $2, %xmm1
+; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    paddq %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm2
+; SSE2-NEXT:    psrlq $4, %xmm2
+; SSE2-NEXT:    paddq %xmm1, %xmm2
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    pxor %xmm0, %xmm0
+; SSE2-NEXT:    psadbw %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: testv2i64:
 ; SSE3:       # BB#0:
-; SSE3-NEXT:    movd %xmm0, %rax
-; SSE3-NEXT:    bsrq %rax, %rax
-; SSE3-NEXT:    movl $127, %ecx
-; SSE3-NEXT:    cmoveq %rcx, %rax
-; SSE3-NEXT:    xorq $63, %rax
-; SSE3-NEXT:    movd %rax, %xmm1
-; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE3-NEXT:    movd %xmm0, %rax
-; SSE3-NEXT:    bsrq %rax, %rax
-; SSE3-NEXT:    cmoveq %rcx, %rax
-; SSE3-NEXT:    xorq $63, %rax
-; SSE3-NEXT:    movd %rax, %xmm0
-; SSE3-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlq $1, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlq $2, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlq $4, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
 ; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlq $8, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlq $16, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlq $32, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT:    pxor %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlq $1, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    psubq %xmm0, %xmm1
+; SSE3-NEXT:    movdqa {{.*#+}} xmm0 = [3689348814741910323,3689348814741910323]
+; SSE3-NEXT:    movdqa %xmm1, %xmm2
+; SSE3-NEXT:    pand %xmm0, %xmm2
+; SSE3-NEXT:    psrlq $2, %xmm1
+; SSE3-NEXT:    pand %xmm0, %xmm1
+; SSE3-NEXT:    paddq %xmm2, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm2
+; SSE3-NEXT:    psrlq $4, %xmm2
+; SSE3-NEXT:    paddq %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]
; X32-SSE-NEXT:    retl
 
   %out = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %in, i1 0)
   ret <2 x i64> %out
 }
 
 define <2 x i64> @testv2i64u(<2 x i64> %in) nounwind {
 ; SSE2-LABEL: testv2i64u:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    movd %xmm0, %rax
-; SSE2-NEXT:    bsrq %rax, %rax
-; SSE2-NEXT:    xorq $63, %rax
-; SSE2-NEXT:    movd %rax, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE2-NEXT:    movd %xmm0, %rax
-; SSE2-NEXT:    bsrq %rax, %rax
-; SSE2-NEXT:    xorq $63, %rax
-; SSE2-NEXT:    movd %rax, %xmm0
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlq $1, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlq $2, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlq $4, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlq $8, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlq $16, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlq $32, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT:    pxor %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlq $1, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    psubq %xmm0, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [3689348814741910323,3689348814741910323]
+; SSE2-NEXT:    movdqa %xmm1, %xmm2
+; SSE2-NEXT:    pand %xmm0, %xmm2
+; SSE2-NEXT:    psrlq $2, %xmm1
+; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    paddq %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm2
+; SSE2-NEXT:    psrlq $4, %xmm2
+; SSE2-NEXT:    paddq %xmm1, %xmm2
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    pxor %xmm0, %xmm0
+; SSE2-NEXT:    psadbw %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: testv2i64u:
 ; SSE3:       # BB#0:
-; SSE3-NEXT:    movd %xmm0, %rax
-; SSE3-NEXT:    bsrq %rax, %rax
-; SSE3-NEXT:    xorq $63, %rax
-; SSE3-NEXT:    movd %rax, %xmm1
-; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE3-NEXT:    movd %xmm0, %rax
-; SSE3-NEXT:    bsrq %rax, %rax
-; SSE3-NEXT:    xorq $63, %rax
-; SSE3-NEXT:    movd %rax, %xmm0
-; SSE3-NEXT:    punpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm0[0]
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlq $1, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlq $2, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlq $4, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlq $8, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlq $16, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlq $32, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT:    pxor %xmm0, %xmm1
 ; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlq $1, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    psubq %xmm0, %xmm1
+; SSE3-NEXT:    movdqa {{.*#+}} xmm0 = [3689348814741910323,3689348814741910323]
+; SSE3-NEXT:    movdqa %xmm1, %xmm2
+; SSE3-NEXT:    pand %xmm0, %xmm2
+; SSE3-NEXT:    psrlq $2, %xmm1
+; SSE3-NEXT:    pand %xmm0, %xmm1
+; SSE3-NEXT:    paddq %xmm2, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm2
+; SSE3-NEXT:    psrlq $4, %xmm2
+; SSE3-NEXT:    paddq %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]
; X32-SSE-NEXT:    retl
 
   %out = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %in, i1 -1)
   ret <2 x i64> %out
 }
 
 define <4 x i32> @testv4i32(<4 x i32> %in) nounwind {
 ; SSE2-LABEL: testv4i32:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    movl $63, %ecx
-; SSE2-NEXT:    cmovel %ecx, %eax
-; SSE2-NEXT:    xorl $31, %eax
-; SSE2-NEXT:    movd %eax, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
-; SSE2-NEXT:    movd %xmm2, %eax
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    cmovel %ecx, %eax
-; SSE2-NEXT:    xorl $31, %eax
-; SSE2-NEXT:    movd %eax, %xmm2
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    cmovel %ecx, %eax
-; SSE2-NEXT:    xorl $31, %eax
-; SSE2-NEXT:    movd %eax, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    cmovel %ecx, %eax
-; SSE2-NEXT:    xorl $31, %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrld $1, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrld $2, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrld $4, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrld $8, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrld $16, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    pcmpeqd %xmm2, %xmm2
+; SSE2-NEXT:    pxor %xmm1, %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    psrld $1, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    psubd %xmm0, %xmm2
+; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [858993459,858993459,858993459,858993459]
+; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    psrld $2, %xmm2
+; SSE2-NEXT:    pand %xmm0, %xmm2
+; SSE2-NEXT:    paddd %xmm1, %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    psrld $4, %xmm0
+; SSE2-NEXT:    paddd %xmm2, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; SSE2-NEXT:    psadbw %xmm1, %xmm2
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    psadbw %xmm1, %xmm0
+; SSE2-NEXT:    packuswb %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: testv4i32:
 ; SSE3:       # BB#0:
-; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; SSE3-NEXT:    movd %xmm1, %eax
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    movl $63, %ecx
-; SSE3-NEXT:    cmovel %ecx, %eax
-; SSE3-NEXT:    xorl $31, %eax
-; SSE3-NEXT:    movd %eax, %xmm1
-; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
-; SSE3-NEXT:    movd %xmm2, %eax
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    cmovel %ecx, %eax
-; SSE3-NEXT:    xorl $31, %eax
-; SSE3-NEXT:    movd %eax, %xmm2
-; SSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE3-NEXT:    movd %xmm0, %eax
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    cmovel %ecx, %eax
-; SSE3-NEXT:    xorl $31, %eax
-; SSE3-NEXT:    movd %eax, %xmm1
-; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE3-NEXT:    movd %xmm0, %eax
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    cmovel %ecx, %eax
-; SSE3-NEXT:    xorl $31, %eax
-; SSE3-NEXT:    movd %eax, %xmm0
-; SSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrld $1, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrld $2, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrld $4, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
 ; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrld $8, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrld $16, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    pcmpeqd %xmm2, %xmm2
+; SSE3-NEXT:    pxor %xmm1, %xmm2
+; SSE3-NEXT:    movdqa %xmm2, %xmm0
+; SSE3-NEXT:    psrld $1, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    psubd %xmm0, %xmm2
+; SSE3-NEXT:    movdqa {{.*#+}} xmm0 = [858993459,858993459,858993459,858993459]
+; SSE3-NEXT:    movdqa %xmm2, %xmm1
+; SSE3-NEXT:    pand %xmm0, %xmm1
+; SSE3-NEXT:    psrld $2, %xmm2
+; SSE3-NEXT:    pand %xmm0, %xmm2
+; SSE3-NEXT:    paddd %xmm1, %xmm2
+; SSE3-NEXT:    movdqa %xmm2, %xmm0
+; SSE3-NEXT:    psrld $4, %xmm0
+; SSE3-NEXT:    paddd %xmm2, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    pxor %xmm1, %xmm1
+; SSE3-NEXT:    movdqa %xmm0, %xmm2
+; SSE3-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; 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]
; X32-SSE-NEXT:    retl
 
   %out = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %in, i1 0)
   ret <4 x i32> %out
 }
 
 define <4 x i32> @testv4i32u(<4 x i32> %in) nounwind {
 ; SSE2-LABEL: testv4i32u:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; SSE2-NEXT:    movd %xmm1, %eax
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    xorl $31, %eax
-; SSE2-NEXT:    movd %eax, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
-; SSE2-NEXT:    movd %xmm2, %eax
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    xorl $31, %eax
-; SSE2-NEXT:    movd %eax, %xmm2
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    xorl $31, %eax
-; SSE2-NEXT:    movd %eax, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    xorl $31, %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrld $1, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrld $2, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrld $4, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrld $8, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrld $16, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    pcmpeqd %xmm2, %xmm2
+; SSE2-NEXT:    pxor %xmm1, %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    psrld $1, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    psubd %xmm0, %xmm2
+; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [858993459,858993459,858993459,858993459]
+; SSE2-NEXT:    movdqa %xmm2, %xmm1
+; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    psrld $2, %xmm2
+; SSE2-NEXT:    pand %xmm0, %xmm2
+; SSE2-NEXT:    paddd %xmm1, %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    psrld $4, %xmm0
+; SSE2-NEXT:    paddd %xmm2, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; SSE2-NEXT:    psadbw %xmm1, %xmm2
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    psadbw %xmm1, %xmm0
+; SSE2-NEXT:    packuswb %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: testv4i32u:
 ; SSE3:       # BB#0:
-; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[3,1,2,3]
-; SSE3-NEXT:    movd %xmm1, %eax
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    xorl $31, %eax
-; SSE3-NEXT:    movd %eax, %xmm1
-; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
-; SSE3-NEXT:    movd %xmm2, %eax
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    xorl $31, %eax
-; SSE3-NEXT:    movd %eax, %xmm2
-; SSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE3-NEXT:    movd %xmm0, %eax
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    xorl $31, %eax
-; SSE3-NEXT:    movd %eax, %xmm1
-; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE3-NEXT:    movd %xmm0, %eax
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    xorl $31, %eax
-; SSE3-NEXT:    movd %eax, %xmm0
-; SSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrld $1, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrld $2, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrld $4, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
 ; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrld $8, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrld $16, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    pcmpeqd %xmm2, %xmm2
+; SSE3-NEXT:    pxor %xmm1, %xmm2
+; SSE3-NEXT:    movdqa %xmm2, %xmm0
+; SSE3-NEXT:    psrld $1, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    psubd %xmm0, %xmm2
+; SSE3-NEXT:    movdqa {{.*#+}} xmm0 = [858993459,858993459,858993459,858993459]
+; SSE3-NEXT:    movdqa %xmm2, %xmm1
+; SSE3-NEXT:    pand %xmm0, %xmm1
+; SSE3-NEXT:    psrld $2, %xmm2
+; SSE3-NEXT:    pand %xmm0, %xmm2
+; SSE3-NEXT:    paddd %xmm1, %xmm2
+; SSE3-NEXT:    movdqa %xmm2, %xmm0
+; SSE3-NEXT:    psrld $4, %xmm0
+; SSE3-NEXT:    paddd %xmm2, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    pxor %xmm1, %xmm1
+; SSE3-NEXT:    movdqa %xmm0, %xmm2
+; SSE3-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; 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]
; X32-SSE-NEXT:    retl
 
   %out = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %in, i1 -1)
   ret <4 x i32> %out
 }
 
 define <8 x i16> @testv8i16(<8 x i16> %in) nounwind {
 ; SSE2-LABEL: testv8i16:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pextrw $7, %xmm0, %eax
-; SSE2-NEXT:    bsrw %ax, %cx
-; SSE2-NEXT:    movw $31, %ax
-; SSE2-NEXT:    cmovew %ax, %cx
-; SSE2-NEXT:    xorl $15, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm1
-; SSE2-NEXT:    pextrw $3, %xmm0, %ecx
-; SSE2-NEXT:    bsrw %cx, %cx
-; SSE2-NEXT:    cmovew %ax, %cx
-; SSE2-NEXT:    xorl $15, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSE2-NEXT:    pextrw $5, %xmm0, %ecx
-; SSE2-NEXT:    bsrw %cx, %cx
-; SSE2-NEXT:    cmovew %ax, %cx
-; SSE2-NEXT:    xorl $15, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm3
-; SSE2-NEXT:    pextrw $1, %xmm0, %ecx
-; SSE2-NEXT:    bsrw %cx, %cx
-; SSE2-NEXT:    cmovew %ax, %cx
-; SSE2-NEXT:    xorl $15, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm1
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; SSE2-NEXT:    pextrw $6, %xmm0, %ecx
-; SSE2-NEXT:    bsrw %cx, %cx
-; SSE2-NEXT:    cmovew %ax, %cx
-; SSE2-NEXT:    xorl $15, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm2
-; SSE2-NEXT:    pextrw $2, %xmm0, %ecx
-; SSE2-NEXT:    bsrw %cx, %cx
-; SSE2-NEXT:    cmovew %ax, %cx
-; SSE2-NEXT:    xorl $15, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm3
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; SSE2-NEXT:    pextrw $4, %xmm0, %ecx
-; SSE2-NEXT:    bsrw %cx, %cx
-; SSE2-NEXT:    cmovew %ax, %cx
-; SSE2-NEXT:    xorl $15, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm2
-; SSE2-NEXT:    movd %xmm0, %ecx
-; SSE2-NEXT:    bsrw %cx, %cx
-; SSE2-NEXT:    cmovew %ax, %cx
-; SSE2-NEXT:    xorl $15, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm0
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlw $2, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlw $4, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlw $8, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT:    pxor %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    psubw %xmm0, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [13107,13107,13107,13107,13107,13107,13107,13107]
+; SSE2-NEXT:    movdqa %xmm1, %xmm2
+; SSE2-NEXT:    pand %xmm0, %xmm2
+; SSE2-NEXT:    psrlw $2, %xmm1
+; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    paddw %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm2
+; SSE2-NEXT:    psrlw $4, %xmm2
+; SSE2-NEXT:    paddw %xmm1, %xmm2
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    psllw $8, %xmm0
+; SSE2-NEXT:    paddb %xmm2, %xmm0
+; SSE2-NEXT:    psrlw $8, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: testv8i16:
 ; SSE3:       # BB#0:
-; SSE3-NEXT:    pextrw $7, %xmm0, %eax
-; SSE3-NEXT:    bsrw %ax, %cx
-; SSE3-NEXT:    movw $31, %ax
-; SSE3-NEXT:    cmovew %ax, %cx
-; SSE3-NEXT:    xorl $15, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm1
-; SSE3-NEXT:    pextrw $3, %xmm0, %ecx
-; SSE3-NEXT:    bsrw %cx, %cx
-; SSE3-NEXT:    cmovew %ax, %cx
-; SSE3-NEXT:    xorl $15, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm2
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSE3-NEXT:    pextrw $5, %xmm0, %ecx
-; SSE3-NEXT:    bsrw %cx, %cx
-; SSE3-NEXT:    cmovew %ax, %cx
-; SSE3-NEXT:    xorl $15, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm3
-; SSE3-NEXT:    pextrw $1, %xmm0, %ecx
-; SSE3-NEXT:    bsrw %cx, %cx
-; SSE3-NEXT:    cmovew %ax, %cx
-; SSE3-NEXT:    xorl $15, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm1
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; SSE3-NEXT:    pextrw $6, %xmm0, %ecx
-; SSE3-NEXT:    bsrw %cx, %cx
-; SSE3-NEXT:    cmovew %ax, %cx
-; SSE3-NEXT:    xorl $15, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm2
-; SSE3-NEXT:    pextrw $2, %xmm0, %ecx
-; SSE3-NEXT:    bsrw %cx, %cx
-; SSE3-NEXT:    cmovew %ax, %cx
-; SSE3-NEXT:    xorl $15, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm3
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; SSE3-NEXT:    pextrw $4, %xmm0, %ecx
-; SSE3-NEXT:    bsrw %cx, %cx
-; SSE3-NEXT:    cmovew %ax, %cx
-; SSE3-NEXT:    xorl $15, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm2
-; SSE3-NEXT:    movd %xmm0, %ecx
-; SSE3-NEXT:    bsrw %cx, %cx
-; SSE3-NEXT:    cmovew %ax, %cx
-; SSE3-NEXT:    xorl $15, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm0
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlw $1, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlw $2, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlw $4, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlw $8, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT:    pxor %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlw $1, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    psubw %xmm0, %xmm1
+; SSE3-NEXT:    movdqa {{.*#+}} xmm0 = [13107,13107,13107,13107,13107,13107,13107,13107]
+; SSE3-NEXT:    movdqa %xmm1, %xmm2
+; SSE3-NEXT:    pand %xmm0, %xmm2
+; SSE3-NEXT:    psrlw $2, %xmm1
+; SSE3-NEXT:    pand %xmm0, %xmm1
+; SSE3-NEXT:    paddw %xmm2, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm2
+; SSE3-NEXT:    psrlw $4, %xmm2
+; SSE3-NEXT:    paddw %xmm1, %xmm2
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm2
+; 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]

 ; X32-SSE-NEXT:    retl
   %out = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %in, i1 0)
   ret <8 x i16> %out
 }
 
 define <8 x i16> @testv8i16u(<8 x i16> %in) nounwind {
 ; SSE2-LABEL: testv8i16u:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pextrw $7, %xmm0, %eax
-; SSE2-NEXT:    bsrw %ax, %ax
-; SSE2-NEXT:    xorl $15, %eax
-; SSE2-NEXT:    movd %eax, %xmm1
-; SSE2-NEXT:    pextrw $3, %xmm0, %eax
-; SSE2-NEXT:    bsrw %ax, %ax
-; SSE2-NEXT:    xorl $15, %eax
-; SSE2-NEXT:    movd %eax, %xmm2
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSE2-NEXT:    pextrw $5, %xmm0, %eax
-; SSE2-NEXT:    bsrw %ax, %ax
-; SSE2-NEXT:    xorl $15, %eax
-; SSE2-NEXT:    movd %eax, %xmm3
-; SSE2-NEXT:    pextrw $1, %xmm0, %eax
-; SSE2-NEXT:    bsrw %ax, %ax
-; SSE2-NEXT:    xorl $15, %eax
-; SSE2-NEXT:    movd %eax, %xmm1
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; SSE2-NEXT:    pextrw $6, %xmm0, %eax
-; SSE2-NEXT:    bsrw %ax, %ax
-; SSE2-NEXT:    xorl $15, %eax
-; SSE2-NEXT:    movd %eax, %xmm2
-; SSE2-NEXT:    pextrw $2, %xmm0, %eax
-; SSE2-NEXT:    bsrw %ax, %ax
-; SSE2-NEXT:    xorl $15, %eax
-; SSE2-NEXT:    movd %eax, %xmm3
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; SSE2-NEXT:    pextrw $4, %xmm0, %eax
-; SSE2-NEXT:    bsrw %ax, %ax
-; SSE2-NEXT:    xorl $15, %eax
-; SSE2-NEXT:    movd %eax, %xmm2
-; SSE2-NEXT:    movd %xmm0, %eax
-; SSE2-NEXT:    bsrw %ax, %ax
-; SSE2-NEXT:    xorl $15, %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlw $2, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlw $4, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlw $8, %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE2-NEXT:    pxor %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    psubw %xmm0, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [13107,13107,13107,13107,13107,13107,13107,13107]
+; SSE2-NEXT:    movdqa %xmm1, %xmm2
+; SSE2-NEXT:    pand %xmm0, %xmm2
+; SSE2-NEXT:    psrlw $2, %xmm1
+; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    paddw %xmm2, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm2
+; SSE2-NEXT:    psrlw $4, %xmm2
+; SSE2-NEXT:    paddw %xmm1, %xmm2
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    psllw $8, %xmm0
+; SSE2-NEXT:    paddb %xmm2, %xmm0
+; SSE2-NEXT:    psrlw $8, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: testv8i16u:
 ; SSE3:       # BB#0:
-; SSE3-NEXT:    pextrw $7, %xmm0, %eax
-; SSE3-NEXT:    bsrw %ax, %ax
-; SSE3-NEXT:    xorl $15, %eax
-; SSE3-NEXT:    movd %eax, %xmm1
-; SSE3-NEXT:    pextrw $3, %xmm0, %eax
-; SSE3-NEXT:    bsrw %ax, %ax
-; SSE3-NEXT:    xorl $15, %eax
-; SSE3-NEXT:    movd %eax, %xmm2
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSE3-NEXT:    pextrw $5, %xmm0, %eax
-; SSE3-NEXT:    bsrw %ax, %ax
-; SSE3-NEXT:    xorl $15, %eax
-; SSE3-NEXT:    movd %eax, %xmm3
-; SSE3-NEXT:    pextrw $1, %xmm0, %eax
-; SSE3-NEXT:    bsrw %ax, %ax
-; SSE3-NEXT:    xorl $15, %eax
-; SSE3-NEXT:    movd %eax, %xmm1
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; SSE3-NEXT:    pextrw $6, %xmm0, %eax
-; SSE3-NEXT:    bsrw %ax, %ax
-; SSE3-NEXT:    xorl $15, %eax
-; SSE3-NEXT:    movd %eax, %xmm2
-; SSE3-NEXT:    pextrw $2, %xmm0, %eax
-; SSE3-NEXT:    bsrw %ax, %ax
-; SSE3-NEXT:    xorl $15, %eax
-; SSE3-NEXT:    movd %eax, %xmm3
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; SSE3-NEXT:    pextrw $4, %xmm0, %eax
-; SSE3-NEXT:    bsrw %ax, %ax
-; SSE3-NEXT:    xorl $15, %eax
-; SSE3-NEXT:    movd %eax, %xmm2
-; SSE3-NEXT:    movd %xmm0, %eax
-; SSE3-NEXT:    bsrw %ax, %ax
-; SSE3-NEXT:    xorl $15, %eax
-; SSE3-NEXT:    movd %eax, %xmm0
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
-; SSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlw $1, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlw $2, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlw $4, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlw $8, %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    pcmpeqd %xmm1, %xmm1
+; SSE3-NEXT:    pxor %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlw $1, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    psubw %xmm0, %xmm1
+; SSE3-NEXT:    movdqa {{.*#+}} xmm0 = [13107,13107,13107,13107,13107,13107,13107,13107]
+; SSE3-NEXT:    movdqa %xmm1, %xmm2
+; SSE3-NEXT:    pand %xmm0, %xmm2
+; SSE3-NEXT:    psrlw $2, %xmm1
+; SSE3-NEXT:    pand %xmm0, %xmm1
+; SSE3-NEXT:    paddw %xmm2, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm2
+; SSE3-NEXT:    psrlw $4, %xmm2
+; SSE3-NEXT:    paddw %xmm1, %xmm2
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm2
+; 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]

 ; X32-SSE-NEXT:    retl
   %out = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %in, i1 -1)
   ret <8 x i16> %out
 }
 
 define <16 x i8> @testv16i8(<16 x i8> %in) nounwind {
 ; SSE2-LABEL: testv16i8:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pushq %rbp
-; SSE2-NEXT:    pushq %rbx
-; SSE2-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
-; SSE2-NEXT:    bsrl %eax, %ecx
-; SSE2-NEXT:    movl $15, %eax
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm0
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ebx
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edi
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r9d
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edx
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r11d
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %esi
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r8d
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE2-NEXT:    bsrl %ecx, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm1
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE2-NEXT:    bsrl %edx, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm2
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edx
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r10d
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ebp
-; SSE2-NEXT:    bsrl %ebp, %ebp
-; SSE2-NEXT:    cmovel %eax, %ebp
-; SSE2-NEXT:    xorl $7, %ebp
-; SSE2-NEXT:    movd %ebp, %xmm0
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE2-NEXT:    bsrl %edi, %edi
-; SSE2-NEXT:    cmovel %eax, %edi
-; SSE2-NEXT:    xorl $7, %edi
-; SSE2-NEXT:    movd %edi, %xmm1
-; SSE2-NEXT:    bsrl %ecx, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
-; SSE2-NEXT:    bsrl %esi, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm3
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %esi
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE2-NEXT:    bsrl %ecx, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm1
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE2-NEXT:    bsrl %ebx, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm0
-; SSE2-NEXT:    bsrl %edx, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm3
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE2-NEXT:    bsrl %r11d, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm0
-; SSE2-NEXT:    bsrl %esi, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3],xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; SSE2-NEXT:    bsrl %r9d, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm0
-; SSE2-NEXT:    bsrl %r10d, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm3
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE2-NEXT:    bsrl %r8d, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm4
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE2-NEXT:    bsrl %ecx, %ecx
-; SSE2-NEXT:    cmovel %eax, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm0
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE2-NEXT:    popq %rbx
-; SSE2-NEXT:    popq %rbp
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlw $2, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlw $4, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE2-NEXT:    pand %xmm2, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    pcmpeqd %xmm3, %xmm3
+; SSE2-NEXT:    pxor %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm0
+; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    psubb %xmm0, %xmm3
+; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE2-NEXT:    movdqa %xmm3, %xmm1
+; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    psrlw $2, %xmm3
+; SSE2-NEXT:    pand %xmm0, %xmm3
+; SSE2-NEXT:    paddb %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm0
+; SSE2-NEXT:    psrlw $4, %xmm0
+; SSE2-NEXT:    paddb %xmm3, %xmm0
+; SSE2-NEXT:    pand %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: testv16i8:
 ; SSE3:       # BB#0:
-; SSE3-NEXT:    pushq %rbp
-; SSE3-NEXT:    pushq %rbx
-; SSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
-; SSE3-NEXT:    bsrl %eax, %ecx
-; SSE3-NEXT:    movl $15, %eax
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm0
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ebx
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edi
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r9d
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edx
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r11d
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %esi
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r8d
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE3-NEXT:    bsrl %ecx, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm1
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE3-NEXT:    bsrl %edx, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm2
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edx
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r10d
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ebp
-; SSE3-NEXT:    bsrl %ebp, %ebp
-; SSE3-NEXT:    cmovel %eax, %ebp
-; SSE3-NEXT:    xorl $7, %ebp
-; SSE3-NEXT:    movd %ebp, %xmm0
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE3-NEXT:    bsrl %edi, %edi
-; SSE3-NEXT:    cmovel %eax, %edi
-; SSE3-NEXT:    xorl $7, %edi
-; SSE3-NEXT:    movd %edi, %xmm1
-; SSE3-NEXT:    bsrl %ecx, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm2
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
-; SSE3-NEXT:    bsrl %esi, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm3
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %esi
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE3-NEXT:    bsrl %ecx, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm1
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE3-NEXT:    bsrl %ebx, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm0
-; SSE3-NEXT:    bsrl %edx, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm3
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE3-NEXT:    bsrl %r11d, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm0
-; SSE3-NEXT:    bsrl %esi, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm2
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3],xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; SSE3-NEXT:    bsrl %r9d, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm0
-; SSE3-NEXT:    bsrl %r10d, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm3
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE3-NEXT:    bsrl %r8d, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm4
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE3-NEXT:    bsrl %ecx, %ecx
-; SSE3-NEXT:    cmovel %eax, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm0
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE3-NEXT:    popq %rbx
-; SSE3-NEXT:    popq %rbp
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlw $1, %xmm1
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlw $2, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlw $4, %xmm1
+; SSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE3-NEXT:    pand %xmm2, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    pcmpeqd %xmm3, %xmm3
+; SSE3-NEXT:    pxor %xmm1, %xmm3
+; SSE3-NEXT:    movdqa %xmm3, %xmm0
+; SSE3-NEXT:    psrlw $1, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    psubb %xmm0, %xmm3
+; SSE3-NEXT:    movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE3-NEXT:    movdqa %xmm3, %xmm1
+; SSE3-NEXT:    pand %xmm0, %xmm1
+; SSE3-NEXT:    psrlw $2, %xmm3
+; SSE3-NEXT:    pand %xmm0, %xmm3
+; SSE3-NEXT:    paddb %xmm1, %xmm3
+; 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]

 ; 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:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pushq %rbx
-; SSE2-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    xorl $7, %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edi
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edx
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r9d
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r10d
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r8d
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %esi
-; SSE2-NEXT:    bsrl %esi, %esi
-; SSE2-NEXT:    xorl $7, %esi
-; SSE2-NEXT:    movd %esi, %xmm1
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    xorl $7, %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %esi
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r11d
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ebx
-; SSE2-NEXT:    bsrl %ebx, %ebx
-; SSE2-NEXT:    xorl $7, %ebx
-; SSE2-NEXT:    movd %ebx, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
-; SSE2-NEXT:    bsrl %edx, %edx
-; SSE2-NEXT:    xorl $7, %edx
-; SSE2-NEXT:    movd %edx, %xmm0
-; SSE2-NEXT:    bsrl %esi, %edx
-; SSE2-NEXT:    xorl $7, %edx
-; SSE2-NEXT:    movd %edx, %xmm3
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE2-NEXT:    bsrl %ecx, %ecx
-; SSE2-NEXT:    xorl $7, %ecx
-; SSE2-NEXT:    movd %ecx, %xmm0
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edx
-; SSE2-NEXT:    bsrl %edx, %edx
-; SSE2-NEXT:    xorl $7, %edx
-; SSE2-NEXT:    movd %edx, %xmm1
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
-; SSE2-NEXT:    bsrl %edi, %edx
-; SSE2-NEXT:    xorl $7, %edx
-; SSE2-NEXT:    movd %edx, %xmm0
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    xorl $7, %eax
-; SSE2-NEXT:    movd %eax, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
-; SSE2-NEXT:    bsrl %r10d, %eax
-; SSE2-NEXT:    xorl $7, %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    bsrl %ecx, %eax
-; SSE2-NEXT:    xorl $7, %eax
-; SSE2-NEXT:    movd %eax, %xmm3
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; SSE2-NEXT:    bsrl %r9d, %eax
-; SSE2-NEXT:    xorl $7, %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    bsrl %r11d, %eax
-; SSE2-NEXT:    xorl $7, %eax
-; SSE2-NEXT:    movd %eax, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
-; SSE2-NEXT:    bsrl %r8d, %eax
-; SSE2-NEXT:    xorl $7, %eax
-; SSE2-NEXT:    movd %eax, %xmm4
-; SSE2-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
-; SSE2-NEXT:    bsrl %eax, %eax
-; SSE2-NEXT:    xorl $7, %eax
-; SSE2-NEXT:    movd %eax, %xmm0
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE2-NEXT:    popq %rbx
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlw $1, %xmm1
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    psrlw $2, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrlw $4, %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE2-NEXT:    pand %xmm2, %xmm1
+; SSE2-NEXT:    por %xmm0, %xmm1
+; SSE2-NEXT:    pcmpeqd %xmm3, %xmm3
+; SSE2-NEXT:    pxor %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm0
+; SSE2-NEXT:    psrlw $1, %xmm0
+; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    psubb %xmm0, %xmm3
+; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE2-NEXT:    movdqa %xmm3, %xmm1
+; SSE2-NEXT:    pand %xmm0, %xmm1
+; SSE2-NEXT:    psrlw $2, %xmm3
+; SSE2-NEXT:    pand %xmm0, %xmm3
+; SSE2-NEXT:    paddb %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm3, %xmm0
+; SSE2-NEXT:    psrlw $4, %xmm0
+; SSE2-NEXT:    paddb %xmm3, %xmm0
+; SSE2-NEXT:    pand %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: testv16i8u:
 ; SSE3:       # BB#0:
-; SSE3-NEXT:    pushq %rbx
-; SSE3-NEXT:    movaps %xmm0, -{{[0-9]+}}(%rsp)
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    xorl $7, %eax
-; SSE3-NEXT:    movd %eax, %xmm0
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edi
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edx
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r9d
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r10d
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r8d
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %esi
-; SSE3-NEXT:    bsrl %esi, %esi
-; SSE3-NEXT:    xorl $7, %esi
-; SSE3-NEXT:    movd %esi, %xmm1
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    xorl $7, %eax
-; SSE3-NEXT:    movd %eax, %xmm0
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %esi
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %r11d
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ebx
-; SSE3-NEXT:    bsrl %ebx, %ebx
-; SSE3-NEXT:    xorl $7, %ebx
-; SSE3-NEXT:    movd %ebx, %xmm2
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
-; SSE3-NEXT:    bsrl %edx, %edx
-; SSE3-NEXT:    xorl $7, %edx
-; SSE3-NEXT:    movd %edx, %xmm0
-; SSE3-NEXT:    bsrl %esi, %edx
-; SSE3-NEXT:    xorl $7, %edx
-; SSE3-NEXT:    movd %edx, %xmm3
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE3-NEXT:    bsrl %ecx, %ecx
-; SSE3-NEXT:    xorl $7, %ecx
-; SSE3-NEXT:    movd %ecx, %xmm0
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %ecx
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %edx
-; SSE3-NEXT:    bsrl %edx, %edx
-; SSE3-NEXT:    xorl $7, %edx
-; SSE3-NEXT:    movd %edx, %xmm1
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
-; SSE3-NEXT:    bsrl %edi, %edx
-; SSE3-NEXT:    xorl $7, %edx
-; SSE3-NEXT:    movd %edx, %xmm0
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    xorl $7, %eax
-; SSE3-NEXT:    movd %eax, %xmm2
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
-; SSE3-NEXT:    bsrl %r10d, %eax
-; SSE3-NEXT:    xorl $7, %eax
-; SSE3-NEXT:    movd %eax, %xmm0
-; SSE3-NEXT:    bsrl %ecx, %eax
-; SSE3-NEXT:    xorl $7, %eax
-; SSE3-NEXT:    movd %eax, %xmm3
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; SSE3-NEXT:    bsrl %r9d, %eax
-; SSE3-NEXT:    xorl $7, %eax
-; SSE3-NEXT:    movd %eax, %xmm0
-; SSE3-NEXT:    bsrl %r11d, %eax
-; SSE3-NEXT:    xorl $7, %eax
-; SSE3-NEXT:    movd %eax, %xmm2
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
-; SSE3-NEXT:    bsrl %r8d, %eax
-; SSE3-NEXT:    xorl $7, %eax
-; SSE3-NEXT:    movd %eax, %xmm4
-; SSE3-NEXT:    movzbl -{{[0-9]+}}(%rsp), %eax
-; SSE3-NEXT:    bsrl %eax, %eax
-; SSE3-NEXT:    xorl $7, %eax
-; SSE3-NEXT:    movd %eax, %xmm0
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
-; SSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE3-NEXT:    popq %rbx
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlw $1, %xmm1
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSE3-NEXT:    psrlw $2, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    por %xmm1, %xmm0
+; SSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSE3-NEXT:    psrlw $4, %xmm1
+; SSE3-NEXT:    movdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; SSE3-NEXT:    pand %xmm2, %xmm1
+; SSE3-NEXT:    por %xmm0, %xmm1
+; SSE3-NEXT:    pcmpeqd %xmm3, %xmm3
+; SSE3-NEXT:    pxor %xmm1, %xmm3
+; SSE3-NEXT:    movdqa %xmm3, %xmm0
+; SSE3-NEXT:    psrlw $1, %xmm0
+; SSE3-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE3-NEXT:    psubb %xmm0, %xmm3
+; SSE3-NEXT:    movdqa {{.*#+}} xmm0 = [51,51,51,51,51,51,51,51,51,51,51,51,51,51,51,51]
+; SSE3-NEXT:    movdqa %xmm3, %xmm1
+; SSE3-NEXT:    pand %xmm0, %xmm1
+; SSE3-NEXT:    psrlw $2, %xmm3
+; SSE3-NEXT:    pand %xmm0, %xmm3
+; SSE3-NEXT:    paddb %xmm1, %xmm3
+; 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]

test/Transforms/SLPVectorizer/X86/ctlz.ll

;
   store i32 %ctlz4, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 4), align 2
   store i32 %ctlz5, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 5), align 2
   store i32 %ctlz6, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 6), align 2
   store i32 %ctlz7, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 7), align 2
   ret void
 }
 
 define void @ctlz_8i16() #0 {
-; SSE2-LABEL: @ctlz_8i16(
-; SSE2-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2
-; SSE2-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2
-; SSE2-NEXT:    [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2
-; SSE2-NEXT:    [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2
-; SSE2-NEXT:    [[LD4:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2
-; SSE2-NEXT:    [[LD5:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2
-; SSE2-NEXT:    [[LD6:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 6), align 2
-; SSE2-NEXT:    [[LD7:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 7), align 2
-; SSE2-NEXT:    [[CTLZ0:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD0]], i1 false)
-; SSE2-NEXT:    [[CTLZ1:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD1]], i1 false)
-; SSE2-NEXT:    [[CTLZ2:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD2]], i1 false)
-; SSE2-NEXT:    [[CTLZ3:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD3]], i1 false)
-; SSE2-NEXT:    [[CTLZ4:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD4]], i1 false)
-; SSE2-NEXT:    [[CTLZ5:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD5]], i1 false)
-; SSE2-NEXT:    [[CTLZ6:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD6]], i1 false)
-; SSE2-NEXT:    [[CTLZ7:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD7]], i1 false)
-; SSE2-NEXT:    store i16 [[CTLZ0]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 0), align 2
-; SSE2-NEXT:    store i16 [[CTLZ1]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 1), align 2
-; SSE2-NEXT:    store i16 [[CTLZ2]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 2), align 2
-; SSE2-NEXT:    store i16 [[CTLZ3]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 3), align 2
-; SSE2-NEXT:    store i16 [[CTLZ4]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 4), align 2
-; SSE2-NEXT:    store i16 [[CTLZ5]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 5), align 2
-; SSE2-NEXT:    store i16 [[CTLZ6]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 6), align 2
-; SSE2-NEXT:    store i16 [[CTLZ7]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 7), align 2
-; SSE2-NEXT:    ret void
-;
-; SSE42-LABEL: @ctlz_8i16(
-; SSE42-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
-; SSE42-NEXT:    [[TMP2:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP1]], i1 false)
-; SSE42-NEXT:    store <8 x i16> [[TMP2]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
-; SSE42-NEXT:    ret void
-;
-; AVX-LABEL: @ctlz_8i16(
-; AVX-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
-; AVX-NEXT:    [[TMP2:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP1]], i1 false)
-; AVX-NEXT:    store <8 x i16> [[TMP2]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
-; AVX-NEXT:    ret void
+; CHECK-LABEL: @ctlz_8i16(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
+; CHECK-NEXT:    [[TMP2:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP1]], i1 false)
+; CHECK-NEXT:    store <8 x i16> [[TMP2]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
+; CHECK-NEXT:    ret void
 ;
   %ld0 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2
   %ld1 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2
   %ld2 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2
   %ld3 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2
   %ld4 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2
   %ld5 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2
   %ld6 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 6), align 2
;
   store i16 %ctlz4, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 4), align 2
   store i16 %ctlz5, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 5), align 2
   store i16 %ctlz6, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 6), align 2
   store i16 %ctlz7, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 7), align 2
   ret void
 }
 
 define void @ctlz_16i16() #0 {
-; SSE2-LABEL: @ctlz_16i16(
-; SSE2-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2
-; SSE2-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2
-; SSE2-NEXT:    [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2
-; SSE2-NEXT:    [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2
-; SSE2-NEXT:    [[LD4:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2
-; SSE2-NEXT:    [[LD5:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2
-; SSE2-NEXT:    [[LD6:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 6), align 2
-; SSE2-NEXT:    [[LD7:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 7), align 2
-; SSE2-NEXT:    [[LD8:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 8), align 2
-; SSE2-NEXT:    [[LD9:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 9), align 2
-; SSE2-NEXT:    [[LD10:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 10), align 2
-; SSE2-NEXT:    [[LD11:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 11), align 2
-; SSE2-NEXT:    [[LD12:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 12), align 2
-; SSE2-NEXT:    [[LD13:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 13), align 2
-; SSE2-NEXT:    [[LD14:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 14), align 2
-; SSE2-NEXT:    [[LD15:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 15), align 2
-; SSE2-NEXT:    [[CTLZ0:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD0]], i1 false)
-; SSE2-NEXT:    [[CTLZ1:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD1]], i1 false)
-; SSE2-NEXT:    [[CTLZ2:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD2]], i1 false)
-; SSE2-NEXT:    [[CTLZ3:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD3]], i1 false)
-; SSE2-NEXT:    [[CTLZ4:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD4]], i1 false)
-; SSE2-NEXT:    [[CTLZ5:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD5]], i1 false)
-; SSE2-NEXT:    [[CTLZ6:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD6]], i1 false)
-; SSE2-NEXT:    [[CTLZ7:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD7]], i1 false)
-; SSE2-NEXT:    [[CTLZ8:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD8]], i1 false)
-; SSE2-NEXT:    [[CTLZ9:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD9]], i1 false)
-; SSE2-NEXT:    [[CTLZ10:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD10]], i1 false)
-; SSE2-NEXT:    [[CTLZ11:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD11]], i1 false)
-; SSE2-NEXT:    [[CTLZ12:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD12]], i1 false)
-; SSE2-NEXT:    [[CTLZ13:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD13]], i1 false)
-; SSE2-NEXT:    [[CTLZ14:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD14]], i1 false)
-; SSE2-NEXT:    [[CTLZ15:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD15]], i1 false)
-; SSE2-NEXT:    store i16 [[CTLZ0]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 0), align 2
-; SSE2-NEXT:    store i16 [[CTLZ1]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 1), align 2
-; SSE2-NEXT:    store i16 [[CTLZ2]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 2), align 2
-; SSE2-NEXT:    store i16 [[CTLZ3]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 3), align 2
-; SSE2-NEXT:    store i16 [[CTLZ4]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 4), align 2
-; SSE2-NEXT:    store i16 [[CTLZ5]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 5), align 2
-; SSE2-NEXT:    store i16 [[CTLZ6]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 6), align 2
-; SSE2-NEXT:    store i16 [[CTLZ7]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 7), align 2
-; SSE2-NEXT:    store i16 [[CTLZ8]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 8), align 2
-; SSE2-NEXT:    store i16 [[CTLZ9]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 9), align 2
-; SSE2-NEXT:    store i16 [[CTLZ10]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 10), align 2
-; SSE2-NEXT:    store i16 [[CTLZ11]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 11), align 2
-; SSE2-NEXT:    store i16 [[CTLZ12]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 12), align 2
-; SSE2-NEXT:    store i16 [[CTLZ13]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 13), align 2
-; SSE2-NEXT:    store i16 [[CTLZ14]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 14), align 2
-; SSE2-NEXT:    store i16 [[CTLZ15]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 15), align 2
-; SSE2-NEXT:    ret void
-;
-; SSE42-LABEL: @ctlz_16i16(
-; SSE42-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
-; SSE42-NEXT:    [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 8) to <8 x i16>*), align 2
-; SSE42-NEXT:    [[TMP3:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP1]], i1 false)
-; SSE42-NEXT:    [[TMP4:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP2]], i1 false)
-; SSE42-NEXT:    store <8 x i16> [[TMP3]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
-; SSE42-NEXT:    store <8 x i16> [[TMP4]], <8 x i16>* bitcast (i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 8) to <8 x i16>*), align 2
-; SSE42-NEXT:    ret void
+; SSE-LABEL: @ctlz_16i16(
+; SSE-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
+; SSE-NEXT:    [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 8) to <8 x i16>*), align 2
+; SSE-NEXT:    [[TMP3:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP1]], i1 false)
+; SSE-NEXT:    [[TMP4:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP2]], i1 false)
+; SSE-NEXT:    store <8 x i16> [[TMP3]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
+; SSE-NEXT:    store <8 x i16> [[TMP4]], <8 x i16>* bitcast (i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 8) to <8 x i16>*), align 2
+; SSE-NEXT:    ret void
 ;
 ; AVX-LABEL: @ctlz_16i16(
 ; AVX-NEXT:    [[TMP1:%.*]] = load <16 x i16>, <16 x i16>* bitcast ([16 x i16]* @src16 to <16 x i16>*), align 2
 ; AVX-NEXT:    [[TMP2:%.*]] = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> [[TMP1]], i1 false)
 ; AVX-NEXT:    store <16 x i16> [[TMP2]], <16 x i16>* bitcast ([16 x i16]* @dst16 to <16 x i16>*), align 2
 ; AVX-NEXT:    ret void
 ;
   %ld0  = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64  0), align 2
;
   store i16 %ctlz12, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 12), align 2
   store i16 %ctlz13, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 13), align 2
   store i16 %ctlz14, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 14), align 2
   store i16 %ctlz15, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 15), align 2
   ret void
 }
 
 define void @ctlz_16i8() #0 {
-; SSE2-LABEL: @ctlz_16i8(
-; SSE2-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 0), align 1
-; SSE2-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 1), align 1
-; SSE2-NEXT:    [[LD2:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 2), align 1
-; SSE2-NEXT:    [[LD3:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 3), align 1
-; SSE2-NEXT:    [[LD4:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 4), align 1
-; SSE2-NEXT:    [[LD5:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 5), align 1
-; SSE2-NEXT:    [[LD6:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 6), align 1
-; SSE2-NEXT:    [[LD7:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 7), align 1
-; SSE2-NEXT:    [[LD8:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 8), align 1
-; SSE2-NEXT:    [[LD9:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 9), align 1
-; SSE2-NEXT:    [[LD10:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 10), align 1
-; SSE2-NEXT:    [[LD11:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 11), align 1
-; SSE2-NEXT:    [[LD12:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 12), align 1
-; SSE2-NEXT:    [[LD13:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 13), align 1
-; SSE2-NEXT:    [[LD14:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 14), align 1
-; SSE2-NEXT:    [[LD15:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 15), align 1
-; SSE2-NEXT:    [[CTLZ0:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD0]], i1 false)
-; SSE2-NEXT:    [[CTLZ1:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD1]], i1 false)
-; SSE2-NEXT:    [[CTLZ2:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD2]], i1 false)
-; SSE2-NEXT:    [[CTLZ3:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD3]], i1 false)
-; SSE2-NEXT:    [[CTLZ4:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD4]], i1 false)
-; SSE2-NEXT:    [[CTLZ5:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD5]], i1 false)
-; SSE2-NEXT:    [[CTLZ6:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD6]], i1 false)
-; SSE2-NEXT:    [[CTLZ7:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD7]], i1 false)
-; SSE2-NEXT:    [[CTLZ8:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD8]], i1 false)
-; SSE2-NEXT:    [[CTLZ9:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD9]], i1 false)
-; SSE2-NEXT:    [[CTLZ10:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD10]], i1 false)
-; SSE2-NEXT:    [[CTLZ11:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD11]], i1 false)
-; SSE2-NEXT:    [[CTLZ12:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD12]], i1 false)
-; SSE2-NEXT:    [[CTLZ13:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD13]], i1 false)
-; SSE2-NEXT:    [[CTLZ14:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD14]], i1 false)
-; SSE2-NEXT:    [[CTLZ15:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD15]], i1 false)
-; SSE2-NEXT:    store i8 [[CTLZ0]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 0), align 1
-; SSE2-NEXT:    store i8 [[CTLZ1]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 1), align 1
-; SSE2-NEXT:    store i8 [[CTLZ2]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 2), align 1
-; SSE2-NEXT:    store i8 [[CTLZ3]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 3), align 1
-; SSE2-NEXT:    store i8 [[CTLZ4]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 4), align 1
-; SSE2-NEXT:    store i8 [[CTLZ5]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 5), align 1
-; SSE2-NEXT:    store i8 [[CTLZ6]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 6), align 1
-; SSE2-NEXT:    store i8 [[CTLZ7]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 7), align 1
-; SSE2-NEXT:    store i8 [[CTLZ8]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 8), align 1
-; SSE2-NEXT:    store i8 [[CTLZ9]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 9), align 1
-; SSE2-NEXT:    store i8 [[CTLZ10]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 10), align 1
-; SSE2-NEXT:    store i8 [[CTLZ11]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 11), align 1
-; SSE2-NEXT:    store i8 [[CTLZ12]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 12), align 1
-; SSE2-NEXT:    store i8 [[CTLZ13]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 13), align 1
-; SSE2-NEXT:    store i8 [[CTLZ14]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 14), align 1
-; SSE2-NEXT:    store i8 [[CTLZ15]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 15), align 1
-; SSE2-NEXT:    ret void
-;
-; SSE42-LABEL: @ctlz_16i8(
-; SSE42-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
-; SSE42-NEXT:    [[TMP2:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 false)
-; SSE42-NEXT:    store <16 x i8> [[TMP2]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
-; SSE42-NEXT:    ret void
-;
-; AVX-LABEL: @ctlz_16i8(
-; AVX-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
-; AVX-NEXT:    [[TMP2:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 false)
-; AVX-NEXT:    store <16 x i8> [[TMP2]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
-; AVX-NEXT:    ret void
+; CHECK-LABEL: @ctlz_16i8(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
+; CHECK-NEXT:    [[TMP2:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 false)
+; CHECK-NEXT:    store <16 x i8> [[TMP2]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
+; CHECK-NEXT:    ret void
 ;
   %ld0  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  0), align 1
   %ld1  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  1), align 1
   %ld2  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  2), align 1
   %ld3  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  3), align 1
   %ld4  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  4), align 1
   %ld5  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  5), align 1
   %ld6  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  6), align 1
;
   store i8 %ctlz12, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 12), align 1
   store i8 %ctlz13, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 13), align 1
   store i8 %ctlz14, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 14), align 1
   store i8 %ctlz15, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 15), align 1
   ret void
 }
 
 define void @ctlz_32i8() #0 {
-; SSE2-LABEL: @ctlz_32i8(
-; SSE2-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 0), align 1
-; SSE2-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 1), align 1
-; SSE2-NEXT:    [[LD2:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 2), align 1
-; SSE2-NEXT:    [[LD3:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 3), align 1
-; SSE2-NEXT:    [[LD4:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 4), align 1
-; SSE2-NEXT:    [[LD5:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 5), align 1
-; SSE2-NEXT:    [[LD6:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 6), align 1
-; SSE2-NEXT:    [[LD7:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 7), align 1
-; SSE2-NEXT:    [[LD8:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 8), align 1
-; SSE2-NEXT:    [[LD9:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 9), align 1
-; SSE2-NEXT:    [[LD10:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 10), align 1
-; SSE2-NEXT:    [[LD11:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 11), align 1
-; SSE2-NEXT:    [[LD12:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 12), align 1
-; SSE2-NEXT:    [[LD13:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 13), align 1
-; SSE2-NEXT:    [[LD14:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 14), align 1
-; SSE2-NEXT:    [[LD15:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 15), align 1
-; SSE2-NEXT:    [[LD16:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 16), align 1
-; SSE2-NEXT:    [[LD17:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 17), align 1
-; SSE2-NEXT:    [[LD18:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 18), align 1
-; SSE2-NEXT:    [[LD19:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 19), align 1
-; SSE2-NEXT:    [[LD20:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 20), align 1
-; SSE2-NEXT:    [[LD21:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 21), align 1
-; SSE2-NEXT:    [[LD22:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 22), align 1
-; SSE2-NEXT:    [[LD23:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 23), align 1
-; SSE2-NEXT:    [[LD24:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 24), align 1
-; SSE2-NEXT:    [[LD25:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 25), align 1
-; SSE2-NEXT:    [[LD26:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 26), align 1
-; SSE2-NEXT:    [[LD27:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 27), align 1
-; SSE2-NEXT:    [[LD28:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 28), align 1
-; SSE2-NEXT:    [[LD29:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 29), align 1
-; SSE2-NEXT:    [[LD30:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 30), align 1
-; SSE2-NEXT:    [[LD31:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 31), align 1
-; SSE2-NEXT:    [[CTLZ0:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD0]], i1 false)
-; SSE2-NEXT:    [[CTLZ1:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD1]], i1 false)
-; SSE2-NEXT:    [[CTLZ2:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD2]], i1 false)
-; SSE2-NEXT:    [[CTLZ3:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD3]], i1 false)
-; SSE2-NEXT:    [[CTLZ4:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD4]], i1 false)
-; SSE2-NEXT:    [[CTLZ5:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD5]], i1 false)
-; SSE2-NEXT:    [[CTLZ6:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD6]], i1 false)
-; SSE2-NEXT:    [[CTLZ7:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD7]], i1 false)
-; SSE2-NEXT:    [[CTLZ8:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD8]], i1 false)
-; SSE2-NEXT:    [[CTLZ9:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD9]], i1 false)
-; SSE2-NEXT:    [[CTLZ10:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD10]], i1 false)
-; SSE2-NEXT:    [[CTLZ11:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD11]], i1 false)
-; SSE2-NEXT:    [[CTLZ12:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD12]], i1 false)
-; SSE2-NEXT:    [[CTLZ13:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD13]], i1 false)
-; SSE2-NEXT:    [[CTLZ14:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD14]], i1 false)
-; SSE2-NEXT:    [[CTLZ15:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD15]], i1 false)
-; SSE2-NEXT:    [[CTLZ16:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD16]], i1 false)
-; SSE2-NEXT:    [[CTLZ17:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD17]], i1 false)
-; SSE2-NEXT:    [[CTLZ18:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD18]], i1 false)
-; SSE2-NEXT:    [[CTLZ19:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD19]], i1 false)
-; SSE2-NEXT:    [[CTLZ20:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD20]], i1 false)
-; SSE2-NEXT:    [[CTLZ21:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD21]], i1 false)
-; SSE2-NEXT:    [[CTLZ22:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD22]], i1 false)
-; SSE2-NEXT:    [[CTLZ23:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD23]], i1 false)
-; SSE2-NEXT:    [[CTLZ24:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD24]], i1 false)
-; SSE2-NEXT:    [[CTLZ25:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD25]], i1 false)
-; SSE2-NEXT:    [[CTLZ26:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD26]], i1 false)
-; SSE2-NEXT:    [[CTLZ27:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD27]], i1 false)
-; SSE2-NEXT:    [[CTLZ28:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD28]], i1 false)
-; SSE2-NEXT:    [[CTLZ29:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD29]], i1 false)
-; SSE2-NEXT:    [[CTLZ30:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD30]], i1 false)
-; SSE2-NEXT:    [[CTLZ31:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD31]], i1 false)
-; SSE2-NEXT:    store i8 [[CTLZ0]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 0), align 1
-; SSE2-NEXT:    store i8 [[CTLZ1]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 1), align 1
-; SSE2-NEXT:    store i8 [[CTLZ2]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 2), align 1
-; SSE2-NEXT:    store i8 [[CTLZ3]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 3), align 1
-; SSE2-NEXT:    store i8 [[CTLZ4]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 4), align 1
-; SSE2-NEXT:    store i8 [[CTLZ5]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 5), align 1
-; SSE2-NEXT:    store i8 [[CTLZ6]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 6), align 1
-; SSE2-NEXT:    store i8 [[CTLZ7]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 7), align 1
-; SSE2-NEXT:    store i8 [[CTLZ8]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 8), align 1
-; SSE2-NEXT:    store i8 [[CTLZ9]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 9), align 1
-; SSE2-NEXT:    store i8 [[CTLZ10]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 10), align 1
-; SSE2-NEXT:    store i8 [[CTLZ11]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 11), align 1
-; SSE2-NEXT:    store i8 [[CTLZ12]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 12), align 1
-; SSE2-NEXT:    store i8 [[CTLZ13]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 13), align 1
-; SSE2-NEXT:    store i8 [[CTLZ14]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 14), align 1
-; SSE2-NEXT:    store i8 [[CTLZ15]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 15), align 1
-; SSE2-NEXT:    store i8 [[CTLZ16]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 16), align 1
-; SSE2-NEXT:    store i8 [[CTLZ17]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 17), align 1
-; SSE2-NEXT:    store i8 [[CTLZ18]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 18), align 1
-; SSE2-NEXT:    store i8 [[CTLZ19]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 19), align 1
-; SSE2-NEXT:    store i8 [[CTLZ20]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 20), align 1
-; SSE2-NEXT:    store i8 [[CTLZ21]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 21), align 1
-; SSE2-NEXT:    store i8 [[CTLZ22]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 22), align 1
-; SSE2-NEXT:    store i8 [[CTLZ23]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 23), align 1
-; SSE2-NEXT:    store i8 [[CTLZ24]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 24), align 1
-; SSE2-NEXT:    store i8 [[CTLZ25]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 25), align 1
-; SSE2-NEXT:    store i8 [[CTLZ26]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 26), align 1
-; SSE2-NEXT:    store i8 [[CTLZ27]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 27), align 1
-; SSE2-NEXT:    store i8 [[CTLZ28]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 28), align 1
-; SSE2-NEXT:    store i8 [[CTLZ29]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 29), align 1
-; SSE2-NEXT:    store i8 [[CTLZ30]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 30), align 1
-; SSE2-NEXT:    store i8 [[CTLZ31]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 31), align 1
-; SSE2-NEXT:    ret void
-;
-; SSE42-LABEL: @ctlz_32i8(
-; SSE42-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
-; SSE42-NEXT:    [[TMP2:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 16) to <16 x i8>*), align 1
-; SSE42-NEXT:    [[TMP3:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 false)
-; SSE42-NEXT:    [[TMP4:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP2]], i1 false)
-; SSE42-NEXT:    store <16 x i8> [[TMP3]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
-; SSE42-NEXT:    store <16 x i8> [[TMP4]], <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 16) to <16 x i8>*), align 1
-; SSE42-NEXT:    ret void
-;
-; AVX-LABEL: @ctlz_32i8(
-; AVX-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
-; AVX-NEXT:    [[TMP2:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 16) to <16 x i8>*), align 1
-; AVX-NEXT:    [[TMP3:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 false)
-; AVX-NEXT:    [[TMP4:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP2]], i1 false)
-; AVX-NEXT:    store <16 x i8> [[TMP3]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
-; AVX-NEXT:    store <16 x i8> [[TMP4]], <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 16) to <16 x i8>*), align 1
-; AVX-NEXT:    ret void
+; CHECK-LABEL: @ctlz_32i8(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 16) to <16 x i8>*), align 1
+; CHECK-NEXT:    [[TMP3:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 false)
+; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP2]], i1 false)
+; CHECK-NEXT:    store <16 x i8> [[TMP3]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
+; CHECK-NEXT:    store <16 x i8> [[TMP4]], <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 16) to <16 x i8>*), align 1
+; CHECK-NEXT:    ret void
 ;
   %ld0  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  0), align 1
   %ld1  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  1), align 1
   %ld2  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  2), align 1
   %ld3  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  3), align 1
   %ld4  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  4), align 1
   %ld5  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  5), align 1
   %ld6  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  6), align 1
;
   store i32 %ctlz4, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 4), align 2
   store i32 %ctlz5, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 5), align 2
   store i32 %ctlz6, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 6), align 2
   store i32 %ctlz7, i32* getelementptr inbounds ([8 x i32], [8 x i32]* @dst32, i32 0, i64 7), align 2
   ret void
 }
 
 define void @ctlz_undef_8i16() #0 {
-; SSE2-LABEL: @ctlz_undef_8i16(
-; SSE2-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2
-; SSE2-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2
-; SSE2-NEXT:    [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2
-; SSE2-NEXT:    [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2
-; SSE2-NEXT:    [[LD4:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2
-; SSE2-NEXT:    [[LD5:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2
-; SSE2-NEXT:    [[LD6:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 6), align 2
-; SSE2-NEXT:    [[LD7:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 7), align 2
-; SSE2-NEXT:    [[CTLZ0:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD0]], i1 true)
-; SSE2-NEXT:    [[CTLZ1:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD1]], i1 true)
-; SSE2-NEXT:    [[CTLZ2:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD2]], i1 true)
-; SSE2-NEXT:    [[CTLZ3:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD3]], i1 true)
-; SSE2-NEXT:    [[CTLZ4:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD4]], i1 true)
-; SSE2-NEXT:    [[CTLZ5:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD5]], i1 true)
-; SSE2-NEXT:    [[CTLZ6:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD6]], i1 true)
-; SSE2-NEXT:    [[CTLZ7:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD7]], i1 true)
-; SSE2-NEXT:    store i16 [[CTLZ0]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 0), align 2
-; SSE2-NEXT:    store i16 [[CTLZ1]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 1), align 2
-; SSE2-NEXT:    store i16 [[CTLZ2]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 2), align 2
-; SSE2-NEXT:    store i16 [[CTLZ3]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 3), align 2
-; SSE2-NEXT:    store i16 [[CTLZ4]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 4), align 2
-; SSE2-NEXT:    store i16 [[CTLZ5]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 5), align 2
-; SSE2-NEXT:    store i16 [[CTLZ6]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 6), align 2
-; SSE2-NEXT:    store i16 [[CTLZ7]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 7), align 2
-; SSE2-NEXT:    ret void
-;
-; SSE42-LABEL: @ctlz_undef_8i16(
-; SSE42-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
-; SSE42-NEXT:    [[TMP2:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP1]], i1 true)
-; SSE42-NEXT:    store <8 x i16> [[TMP2]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
-; SSE42-NEXT:    ret void
-;
-; AVX-LABEL: @ctlz_undef_8i16(
-; AVX-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
-; AVX-NEXT:    [[TMP2:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP1]], i1 true)
-; AVX-NEXT:    store <8 x i16> [[TMP2]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
-; AVX-NEXT:    ret void
+; CHECK-LABEL: @ctlz_undef_8i16(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
+; CHECK-NEXT:    [[TMP2:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP1]], i1 true)
+; CHECK-NEXT:    store <8 x i16> [[TMP2]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
+; CHECK-NEXT:    ret void
 ;
   %ld0 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2
   %ld1 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2
   %ld2 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2
   %ld3 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2
   %ld4 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2
   %ld5 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2
   %ld6 = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 6), align 2
;
   store i16 %ctlz4, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 4), align 2
   store i16 %ctlz5, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 5), align 2
   store i16 %ctlz6, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 6), align 2
   store i16 %ctlz7, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 7), align 2
   ret void
 }
 
 define void @ctlz_undef_16i16() #0 {
-; SSE2-LABEL: @ctlz_undef_16i16(
-; SSE2-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 0), align 2
-; SSE2-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 1), align 2
-; SSE2-NEXT:    [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 2), align 2
-; SSE2-NEXT:    [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 3), align 2
-; SSE2-NEXT:    [[LD4:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 4), align 2
-; SSE2-NEXT:    [[LD5:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 5), align 2
-; SSE2-NEXT:    [[LD6:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 6), align 2
-; SSE2-NEXT:    [[LD7:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 7), align 2
-; SSE2-NEXT:    [[LD8:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 8), align 2
-; SSE2-NEXT:    [[LD9:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 9), align 2
-; SSE2-NEXT:    [[LD10:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 10), align 2
-; SSE2-NEXT:    [[LD11:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 11), align 2
-; SSE2-NEXT:    [[LD12:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 12), align 2
-; SSE2-NEXT:    [[LD13:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 13), align 2
-; SSE2-NEXT:    [[LD14:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 14), align 2
-; SSE2-NEXT:    [[LD15:%.*]] = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 15), align 2
-; SSE2-NEXT:    [[CTLZ0:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD0]], i1 true)
-; SSE2-NEXT:    [[CTLZ1:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD1]], i1 true)
-; SSE2-NEXT:    [[CTLZ2:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD2]], i1 true)
-; SSE2-NEXT:    [[CTLZ3:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD3]], i1 true)
-; SSE2-NEXT:    [[CTLZ4:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD4]], i1 true)
-; SSE2-NEXT:    [[CTLZ5:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD5]], i1 true)
-; SSE2-NEXT:    [[CTLZ6:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD6]], i1 true)
-; SSE2-NEXT:    [[CTLZ7:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD7]], i1 true)
-; SSE2-NEXT:    [[CTLZ8:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD8]], i1 true)
-; SSE2-NEXT:    [[CTLZ9:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD9]], i1 true)
-; SSE2-NEXT:    [[CTLZ10:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD10]], i1 true)
-; SSE2-NEXT:    [[CTLZ11:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD11]], i1 true)
-; SSE2-NEXT:    [[CTLZ12:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD12]], i1 true)
-; SSE2-NEXT:    [[CTLZ13:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD13]], i1 true)
-; SSE2-NEXT:    [[CTLZ14:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD14]], i1 true)
-; SSE2-NEXT:    [[CTLZ15:%.*]] = call i16 @llvm.ctlz.i16(i16 [[LD15]], i1 true)
-; SSE2-NEXT:    store i16 [[CTLZ0]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 0), align 2
-; SSE2-NEXT:    store i16 [[CTLZ1]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 1), align 2
-; SSE2-NEXT:    store i16 [[CTLZ2]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 2), align 2
-; SSE2-NEXT:    store i16 [[CTLZ3]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 3), align 2
-; SSE2-NEXT:    store i16 [[CTLZ4]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 4), align 2
-; SSE2-NEXT:    store i16 [[CTLZ5]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 5), align 2
-; SSE2-NEXT:    store i16 [[CTLZ6]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 6), align 2
-; SSE2-NEXT:    store i16 [[CTLZ7]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 7), align 2
-; SSE2-NEXT:    store i16 [[CTLZ8]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 8), align 2
-; SSE2-NEXT:    store i16 [[CTLZ9]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 9), align 2
-; SSE2-NEXT:    store i16 [[CTLZ10]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 10), align 2
-; SSE2-NEXT:    store i16 [[CTLZ11]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 11), align 2
-; SSE2-NEXT:    store i16 [[CTLZ12]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 12), align 2
-; SSE2-NEXT:    store i16 [[CTLZ13]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 13), align 2
-; SSE2-NEXT:    store i16 [[CTLZ14]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 14), align 2
-; SSE2-NEXT:    store i16 [[CTLZ15]], i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 15), align 2
-; SSE2-NEXT:    ret void
-;
-; SSE42-LABEL: @ctlz_undef_16i16(
-; SSE42-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
-; SSE42-NEXT:    [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 8) to <8 x i16>*), align 2
-; SSE42-NEXT:    [[TMP3:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP1]], i1 true)
-; SSE42-NEXT:    [[TMP4:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP2]], i1 true)
-; SSE42-NEXT:    store <8 x i16> [[TMP3]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
-; SSE42-NEXT:    store <8 x i16> [[TMP4]], <8 x i16>* bitcast (i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 8) to <8 x i16>*), align 2
-; SSE42-NEXT:    ret void
+; SSE-LABEL: @ctlz_undef_16i16(
+; SSE-NEXT:    [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* bitcast ([16 x i16]* @src16 to <8 x i16>*), align 2
+; SSE-NEXT:    [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* bitcast (i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64 8) to <8 x i16>*), align 2
+; SSE-NEXT:    [[TMP3:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP1]], i1 true)
+; SSE-NEXT:    [[TMP4:%.*]] = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> [[TMP2]], i1 true)
+; SSE-NEXT:    store <8 x i16> [[TMP3]], <8 x i16>* bitcast ([16 x i16]* @dst16 to <8 x i16>*), align 2
+; SSE-NEXT:    store <8 x i16> [[TMP4]], <8 x i16>* bitcast (i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 8) to <8 x i16>*), align 2
+; SSE-NEXT:    ret void
 ;
 ; AVX-LABEL: @ctlz_undef_16i16(
 ; AVX-NEXT:    [[TMP1:%.*]] = load <16 x i16>, <16 x i16>* bitcast ([16 x i16]* @src16 to <16 x i16>*), align 2
 ; AVX-NEXT:    [[TMP2:%.*]] = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> [[TMP1]], i1 true)
 ; AVX-NEXT:    store <16 x i16> [[TMP2]], <16 x i16>* bitcast ([16 x i16]* @dst16 to <16 x i16>*), align 2
 ; AVX-NEXT:    ret void
 ;
   %ld0  = load i16, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @src16, i16 0, i64  0), align 2
;
   store i16 %ctlz12, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 12), align 2
   store i16 %ctlz13, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 13), align 2
   store i16 %ctlz14, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 14), align 2
   store i16 %ctlz15, i16* getelementptr inbounds ([16 x i16], [16 x i16]* @dst16, i16 0, i64 15), align 2
   ret void
 }
 
 define void @ctlz_undef_16i8() #0 {
-; SSE2-LABEL: @ctlz_undef_16i8(
-; SSE2-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 0), align 1
-; SSE2-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 1), align 1
-; SSE2-NEXT:    [[LD2:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 2), align 1
-; SSE2-NEXT:    [[LD3:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 3), align 1
-; SSE2-NEXT:    [[LD4:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 4), align 1
-; SSE2-NEXT:    [[LD5:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 5), align 1
-; SSE2-NEXT:    [[LD6:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 6), align 1
-; SSE2-NEXT:    [[LD7:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 7), align 1
-; SSE2-NEXT:    [[LD8:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 8), align 1
-; SSE2-NEXT:    [[LD9:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 9), align 1
-; SSE2-NEXT:    [[LD10:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 10), align 1
-; SSE2-NEXT:    [[LD11:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 11), align 1
-; SSE2-NEXT:    [[LD12:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 12), align 1
-; SSE2-NEXT:    [[LD13:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 13), align 1
-; SSE2-NEXT:    [[LD14:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 14), align 1
-; SSE2-NEXT:    [[LD15:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 15), align 1
-; SSE2-NEXT:    [[CTLZ0:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD0]], i1 true)
-; SSE2-NEXT:    [[CTLZ1:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD1]], i1 true)
-; SSE2-NEXT:    [[CTLZ2:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD2]], i1 true)
-; SSE2-NEXT:    [[CTLZ3:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD3]], i1 true)
-; SSE2-NEXT:    [[CTLZ4:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD4]], i1 true)
-; SSE2-NEXT:    [[CTLZ5:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD5]], i1 true)
-; SSE2-NEXT:    [[CTLZ6:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD6]], i1 true)
-; SSE2-NEXT:    [[CTLZ7:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD7]], i1 true)
-; SSE2-NEXT:    [[CTLZ8:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD8]], i1 true)
-; SSE2-NEXT:    [[CTLZ9:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD9]], i1 true)
-; SSE2-NEXT:    [[CTLZ10:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD10]], i1 true)
-; SSE2-NEXT:    [[CTLZ11:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD11]], i1 true)
-; SSE2-NEXT:    [[CTLZ12:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD12]], i1 true)
-; SSE2-NEXT:    [[CTLZ13:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD13]], i1 true)
-; SSE2-NEXT:    [[CTLZ14:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD14]], i1 true)
-; SSE2-NEXT:    [[CTLZ15:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD15]], i1 true)
-; SSE2-NEXT:    store i8 [[CTLZ0]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 0), align 1
-; SSE2-NEXT:    store i8 [[CTLZ1]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 1), align 1
-; SSE2-NEXT:    store i8 [[CTLZ2]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 2), align 1
-; SSE2-NEXT:    store i8 [[CTLZ3]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 3), align 1
-; SSE2-NEXT:    store i8 [[CTLZ4]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 4), align 1
-; SSE2-NEXT:    store i8 [[CTLZ5]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 5), align 1
-; SSE2-NEXT:    store i8 [[CTLZ6]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 6), align 1
-; SSE2-NEXT:    store i8 [[CTLZ7]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 7), align 1
-; SSE2-NEXT:    store i8 [[CTLZ8]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 8), align 1
-; SSE2-NEXT:    store i8 [[CTLZ9]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 9), align 1
-; SSE2-NEXT:    store i8 [[CTLZ10]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 10), align 1
-; SSE2-NEXT:    store i8 [[CTLZ11]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 11), align 1
-; SSE2-NEXT:    store i8 [[CTLZ12]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 12), align 1
-; SSE2-NEXT:    store i8 [[CTLZ13]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 13), align 1
-; SSE2-NEXT:    store i8 [[CTLZ14]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 14), align 1
-; SSE2-NEXT:    store i8 [[CTLZ15]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 15), align 1
-; SSE2-NEXT:    ret void
-;
-; SSE42-LABEL: @ctlz_undef_16i8(
-; SSE42-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
-; SSE42-NEXT:    [[TMP2:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 true)
-; SSE42-NEXT:    store <16 x i8> [[TMP2]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
-; SSE42-NEXT:    ret void
-;
-; AVX-LABEL: @ctlz_undef_16i8(
-; AVX-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
-; AVX-NEXT:    [[TMP2:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 true)
-; AVX-NEXT:    store <16 x i8> [[TMP2]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
-; AVX-NEXT:    ret void
+; CHECK-LABEL: @ctlz_undef_16i8(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
+; CHECK-NEXT:    [[TMP2:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 true)
+; CHECK-NEXT:    store <16 x i8> [[TMP2]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
+; CHECK-NEXT:    ret void
 ;
   %ld0  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  0), align 1
   %ld1  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  1), align 1
   %ld2  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  2), align 1
   %ld3  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  3), align 1
   %ld4  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  4), align 1
   %ld5  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  5), align 1
   %ld6  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  6), align 1
;
   store i8 %ctlz12, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 12), align 1
   store i8 %ctlz13, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 13), align 1
   store i8 %ctlz14, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 14), align 1
   store i8 %ctlz15, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 15), align 1
   ret void
 }
 
 define void @ctlz_undef_32i8() #0 {
-; SSE2-LABEL: @ctlz_undef_32i8(
-; SSE2-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 0), align 1
-; SSE2-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 1), align 1
-; SSE2-NEXT:    [[LD2:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 2), align 1
-; SSE2-NEXT:    [[LD3:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 3), align 1
-; SSE2-NEXT:    [[LD4:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 4), align 1
-; SSE2-NEXT:    [[LD5:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 5), align 1
-; SSE2-NEXT:    [[LD6:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 6), align 1
-; SSE2-NEXT:    [[LD7:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 7), align 1
-; SSE2-NEXT:    [[LD8:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 8), align 1
-; SSE2-NEXT:    [[LD9:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 9), align 1
-; SSE2-NEXT:    [[LD10:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 10), align 1
-; SSE2-NEXT:    [[LD11:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 11), align 1
-; SSE2-NEXT:    [[LD12:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 12), align 1
-; SSE2-NEXT:    [[LD13:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 13), align 1
-; SSE2-NEXT:    [[LD14:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 14), align 1
-; SSE2-NEXT:    [[LD15:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 15), align 1
-; SSE2-NEXT:    [[LD16:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 16), align 1
-; SSE2-NEXT:    [[LD17:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 17), align 1
-; SSE2-NEXT:    [[LD18:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 18), align 1
-; SSE2-NEXT:    [[LD19:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 19), align 1
-; SSE2-NEXT:    [[LD20:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 20), align 1
-; SSE2-NEXT:    [[LD21:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 21), align 1
-; SSE2-NEXT:    [[LD22:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 22), align 1
-; SSE2-NEXT:    [[LD23:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 23), align 1
-; SSE2-NEXT:    [[LD24:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 24), align 1
-; SSE2-NEXT:    [[LD25:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 25), align 1
-; SSE2-NEXT:    [[LD26:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 26), align 1
-; SSE2-NEXT:    [[LD27:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 27), align 1
-; SSE2-NEXT:    [[LD28:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 28), align 1
-; SSE2-NEXT:    [[LD29:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 29), align 1
-; SSE2-NEXT:    [[LD30:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 30), align 1
-; SSE2-NEXT:    [[LD31:%.*]] = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 31), align 1
-; SSE2-NEXT:    [[CTLZ0:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD0]], i1 true)
-; SSE2-NEXT:    [[CTLZ1:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD1]], i1 true)
-; SSE2-NEXT:    [[CTLZ2:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD2]], i1 true)
-; SSE2-NEXT:    [[CTLZ3:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD3]], i1 true)
-; SSE2-NEXT:    [[CTLZ4:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD4]], i1 true)
-; SSE2-NEXT:    [[CTLZ5:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD5]], i1 true)
-; SSE2-NEXT:    [[CTLZ6:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD6]], i1 true)
-; SSE2-NEXT:    [[CTLZ7:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD7]], i1 true)
-; SSE2-NEXT:    [[CTLZ8:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD8]], i1 true)
-; SSE2-NEXT:    [[CTLZ9:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD9]], i1 true)
-; SSE2-NEXT:    [[CTLZ10:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD10]], i1 true)
-; SSE2-NEXT:    [[CTLZ11:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD11]], i1 true)
-; SSE2-NEXT:    [[CTLZ12:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD12]], i1 true)
-; SSE2-NEXT:    [[CTLZ13:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD13]], i1 true)
-; SSE2-NEXT:    [[CTLZ14:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD14]], i1 true)
-; SSE2-NEXT:    [[CTLZ15:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD15]], i1 true)
-; SSE2-NEXT:    [[CTLZ16:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD16]], i1 true)
-; SSE2-NEXT:    [[CTLZ17:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD17]], i1 true)
-; SSE2-NEXT:    [[CTLZ18:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD18]], i1 true)
-; SSE2-NEXT:    [[CTLZ19:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD19]], i1 true)
-; SSE2-NEXT:    [[CTLZ20:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD20]], i1 true)
-; SSE2-NEXT:    [[CTLZ21:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD21]], i1 true)
-; SSE2-NEXT:    [[CTLZ22:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD22]], i1 true)
-; SSE2-NEXT:    [[CTLZ23:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD23]], i1 true)
-; SSE2-NEXT:    [[CTLZ24:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD24]], i1 true)
-; SSE2-NEXT:    [[CTLZ25:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD25]], i1 true)
-; SSE2-NEXT:    [[CTLZ26:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD26]], i1 true)
-; SSE2-NEXT:    [[CTLZ27:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD27]], i1 true)
-; SSE2-NEXT:    [[CTLZ28:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD28]], i1 true)
-; SSE2-NEXT:    [[CTLZ29:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD29]], i1 true)
-; SSE2-NEXT:    [[CTLZ30:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD30]], i1 true)
-; SSE2-NEXT:    [[CTLZ31:%.*]] = call i8 @llvm.ctlz.i8(i8 [[LD31]], i1 true)
-; SSE2-NEXT:    store i8 [[CTLZ0]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 0), align 1
-; SSE2-NEXT:    store i8 [[CTLZ1]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 1), align 1
-; SSE2-NEXT:    store i8 [[CTLZ2]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 2), align 1
-; SSE2-NEXT:    store i8 [[CTLZ3]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 3), align 1
-; SSE2-NEXT:    store i8 [[CTLZ4]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 4), align 1
-; SSE2-NEXT:    store i8 [[CTLZ5]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 5), align 1
-; SSE2-NEXT:    store i8 [[CTLZ6]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 6), align 1
-; SSE2-NEXT:    store i8 [[CTLZ7]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 7), align 1
-; SSE2-NEXT:    store i8 [[CTLZ8]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 8), align 1
-; SSE2-NEXT:    store i8 [[CTLZ9]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 9), align 1
-; SSE2-NEXT:    store i8 [[CTLZ10]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 10), align 1
-; SSE2-NEXT:    store i8 [[CTLZ11]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 11), align 1
-; SSE2-NEXT:    store i8 [[CTLZ12]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 12), align 1
-; SSE2-NEXT:    store i8 [[CTLZ13]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 13), align 1
-; SSE2-NEXT:    store i8 [[CTLZ14]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 14), align 1
-; SSE2-NEXT:    store i8 [[CTLZ15]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 15), align 1
-; SSE2-NEXT:    store i8 [[CTLZ16]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 16), align 1
-; SSE2-NEXT:    store i8 [[CTLZ17]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 17), align 1
-; SSE2-NEXT:    store i8 [[CTLZ18]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 18), align 1
-; SSE2-NEXT:    store i8 [[CTLZ19]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 19), align 1
-; SSE2-NEXT:    store i8 [[CTLZ20]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 20), align 1
-; SSE2-NEXT:    store i8 [[CTLZ21]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 21), align 1
-; SSE2-NEXT:    store i8 [[CTLZ22]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 22), align 1
-; SSE2-NEXT:    store i8 [[CTLZ23]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 23), align 1
-; SSE2-NEXT:    store i8 [[CTLZ24]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 24), align 1
-; SSE2-NEXT:    store i8 [[CTLZ25]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 25), align 1
-; SSE2-NEXT:    store i8 [[CTLZ26]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 26), align 1
-; SSE2-NEXT:    store i8 [[CTLZ27]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 27), align 1
-; SSE2-NEXT:    store i8 [[CTLZ28]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 28), align 1
-; SSE2-NEXT:    store i8 [[CTLZ29]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 29), align 1
-; SSE2-NEXT:    store i8 [[CTLZ30]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 30), align 1
-; SSE2-NEXT:    store i8 [[CTLZ31]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 31), align 1
-; SSE2-NEXT:    ret void
-;
-; SSE42-LABEL: @ctlz_undef_32i8(
-; SSE42-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
-; SSE42-NEXT:    [[TMP2:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 16) to <16 x i8>*), align 1
-; SSE42-NEXT:    [[TMP3:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 true)
-; SSE42-NEXT:    [[TMP4:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP2]], i1 true)
-; SSE42-NEXT:    store <16 x i8> [[TMP3]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
-; SSE42-NEXT:    store <16 x i8> [[TMP4]], <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 16) to <16 x i8>*), align 1
-; SSE42-NEXT:    ret void
-;
-; AVX-LABEL: @ctlz_undef_32i8(
-; AVX-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
-; AVX-NEXT:    [[TMP2:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 16) to <16 x i8>*), align 1
-; AVX-NEXT:    [[TMP3:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 true)
-; AVX-NEXT:    [[TMP4:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP2]], i1 true)
-; AVX-NEXT:    store <16 x i8> [[TMP3]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
-; AVX-NEXT:    store <16 x i8> [[TMP4]], <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 16) to <16 x i8>*), align 1
-; AVX-NEXT:    ret void
+; CHECK-LABEL: @ctlz_undef_32i8(
+; CHECK-NEXT:    [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* bitcast ([32 x i8]* @src8 to <16 x i8>*), align 1
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i8>, <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64 16) to <16 x i8>*), align 1
+; CHECK-NEXT:    [[TMP3:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP1]], i1 true)
+; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> [[TMP2]], i1 true)
+; CHECK-NEXT:    store <16 x i8> [[TMP3]], <16 x i8>* bitcast ([32 x i8]* @dst8 to <16 x i8>*), align 1
+; CHECK-NEXT:    store <16 x i8> [[TMP4]], <16 x i8>* bitcast (i8* getelementptr inbounds ([32 x i8], [32 x i8]* @dst8, i8 0, i64 16) to <16 x i8>*), align 1
+; CHECK-NEXT:    ret void
 ;
   %ld0  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  0), align 1
   %ld1  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  1), align 1
   %ld2  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  2), align 1
   %ld3  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  3), align 1
   %ld4  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  4), align 1
   %ld5  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  5), align 1
   %ld6  = load i8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @src8, i8 0, i64  6), align 1