[DAGCombiner] scalarize binop followed by extractelement

[spatel]

Summary

As noted in PR39973 and D55558:
https://bugs.llvm.org/show_bug.cgi?id=39973
...this is a partial implementation of a fold that we do as an IR canonicalization in instcombine:

// extelt (binop X, Y), Index --> binop (extelt X, Index), (extelt Y, Index)

We want to have this in the DAG too because as we can see in some of the test diffs (reductions), the pattern may not be visible in IR.
Given that this is already an IR canonicalization, any backend that would prefer a vector op over a scalar op is expected to already have the reverse transform in DAG lowering (not sure if that's a realistic expectation though).

There's an ARM test diff that shows that we also have a reverse transform in IR in CGP. I'm not sure what to make of that. Isn't the updated asm better than the existing code for that test?

Note that I changed some existing regression tests leading up to this patch trying to preserve their intent. Let me know if we need to do any more of that for the remaining diffs here. These were the preliminary commits (and so those tests are not affected by this patch):
rL349160
rL349163
rL349164
rL349166
rL349176
rL349177

[RKSimon]

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

15578 ↗

(On Diff #178281)

I wonder whether we need a oneuse check on the constant as well? Some of the cases below suggest that a vector constant load might be happening whatever.

test/CodeGen/SystemZ/knownbits.ll

51 ↗

(On Diff #178281)

This will need rebasing after rL349264

test/CodeGen/X86/and-load-fold.ll

11 ↗

(On Diff #178281)

X86ISelLowering's combineExtractWithShuffle is supposed to handle this.....

[spatel]

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

15578 ↗

(On Diff #178281)

If we put a one-use check on the constant, the effect will be to remove all of the horizontal-reduce-* and vector-reduce-* diffs from this patch.

I thought those were good changes though - better to get the data out of the vector unit as soon as possible since we know we're going to end up in a scalar?

[spatel]

Patch updated:
No code changes, but corrected code comment and rebased tests.

[uweigand]

Just looking at the SystemZ test cases:

• The knownbits.ll case shows what might be an unintended consequence. We have a vector "not" followed by a vector "and", which are merged into a single vector "and complement" operation, and this is then followed by an element extraction. Your patch moves the "and" to the scalar side, but leaves the "not" a vector operation. This means that we can now no longer combine the two into an "and complement". Would it maybe be better to do the same transformation also for unary operations like "not"?

• The vec-trunc-to-i1.ll case is supposed to verify that a particular vector operation doesn't crash (see PR 32275). With your patch, we no longer have that vector operation. While this leads to better code and thus is certainly a useful transformation, it means that we should probably modify the test case so it again tests the original problem. @jonpa, can you have a look?

[spatel]

In D55722#1332736, @uweigand wrote:

Just looking at the SystemZ test cases:

• The knownbits.ll case shows what might be an unintended consequence. We have a vector "not" followed by a vector "and", which are merged into a single vector "and complement" operation, and this is then followed by an element extraction. Your patch moves the "and" to the scalar side, but leaves the "not" a vector operation. This means that we can now no longer combine the two into an "and complement". Would it maybe be better to do the same transformation also for unary operations like "not"?

Thanks for explaining that output! Yes, I think we do want to extend this for unary ops like ISD::FNEG, but there's is no generic unary ISD::NOT node. We just do 'xor X, -1'. The same code example compiled for x86 behaves like we would expect - both logic ops are scalarized. So I looked at the debug output for this test with SystemZ target:

    t15: v4i32 = setcc t2, t4, setne:ch
  t17: i32 = extract_vector_elt t15, Constant:i32<3>
t18: i32 = and t17, Constant:i32<1>

And then as legalization progresses:

        t21: v16i8 = SystemZISD::BYTE_MASK Constant:i32<65535>
      t22: v4i32 = bitcast t21
    t23: v4i32 = xor t19, t22
  t17: i32 = extract_vector_elt t23, Constant:i32<3>
t18: i32 = and t17, Constant:i32<1>

So we do not recognize the t23 xor node as a 'not' because the constant value is hidden behind a bitcast of a target-specific constant. We can't handle that case here in generic combining without some kind of callback to know what to do with that constant node.

This does make me worry that I need to add an OpaqueConstant check into this somehow. Assuming that it is possible to have a build_vector that contains opaque constants?

[uweigand]

Ah, right, I missed that.

Now I'm wondering if we maybe introduce the BYTE_MASK too early. It looks like X86 (and other targets) leave an all-ones constant vector as a BUILD_VECTOR until the very end of instruction selection; maybe we should do the same on SystemZ to allow common optimizations to still recognize the value ...

[spatel]

Patch updated:
Avoid transforming with integers if the build vector contains any opaque constants. I'm not sure how that can happen, so there are no test diffs here vs. the last rev. There's apparently no opaque constant equivalent for FP constants.

[jonpa]

The vec-trunc-to-i1.ll case is supposed to verify that a particular vector operation doesn't crash (see PR 32275). With your patch, we no longer have that vector operation. While this leads to better code and thus is certainly a useful transformation, it means that we should probably modify the test case so it again tests the original problem. @jonpa, can you have a look?

I tried reverting my patch for the DAG type legalizer, and found that vec-trunc-to-i1.ll still fails with this new patch applied. It fails during type legalization, while this patch at least in this case gets enabled only after type legalization. So I think the test still serves its purpose.

[uweigand]

In D55722#1334123, @jonpa wrote:

I tried reverting my patch for the DAG type legalizer, and found that vec-trunc-to-i1.ll still fails with this new patch applied. It fails during type legalization, while this patch at least in this case gets enabled only after type legalization. So I think the test still serves its purpose.

Ah, good. Thanks for checking!

[spatel]

@efriedma / @dmgreen (or anyone else with ARM experience) - any thoughts about the ARM diffs?

[efriedma]

test/CodeGen/ARM/vector-promotion.ll

16 ↗

(On Diff #178478)

This is destroying the whole point of the test: we want to avoid the expensive float->int register transfer.

87 ↗

(On Diff #178478)

Same here; this is destroying the point of the test.

394 ↗

(On Diff #178478)

This transform is okay, but I'd like to see a new test replacing the load with an intrinsic that produces a vector, to preserve the original intent.

[spatel]

test/CodeGen/ARM/vector-promotion.ll

16 ↗

(On Diff #178478)

So this raises what appears to be a show-stopper for this patch. We have this reverse transform in IR, so we'd have to redo that as a late DAG transform (we'd have 2 reversals to get back to the code we started with). I don't think that I can cleanly use the TLI hook that was added for the CGP transform because it takes IR type params.

But this ties into something I noticed in InstCombine/DAGCombine. We have a DAG function called scalarizeExtractedVectorLoad() (except there seems to be a bug in how it is called - it's not used with a constant extract index).

I was wondering why we don't do that as an IR canonicalization. That would appear to make both this test and the next diff moot - we would remove all vector ops right from the start. Do you see a problem with that?

To make that clearer, I think the ideal ARM asm for this test has no vector ops:
ldr r0, [r0, #4]
orr r0, r0, #1
str r0, [r1]
bx lr

[efriedma]

test/CodeGen/ARM/vector-promotion.ll

16 ↗

(On Diff #178478)

I guess we could canonicalize load+op+store patterns in IR? It might have unexpected results in some cases, though... e.g. we currently emit different code for add i64 vs. add <1 x i64>, and the clang NEON intrinsics depend on that to some extent. Maybe we could discuss it.

I think that's really besides the point here, though... the "load" here is really just supposed to be a placeholder for "some operation that produces a vector".

[spatel]

Patch updated:
Avoid controversy by adding a default-off TLI hook for this transform. Enable it for x86, so those diffs are similar to the previous rev of the patch. No ARM changes now, and also no SystemZ changes (probably better to fix the BYTE_MASK transform first?).

[RKSimon]

LGTM - thanks

[andreadb]

The X86 changes look good to me.

llvm/trunk/include/llvm/CodeGen/TargetLowering.h

public:
   /// from this source type with this index. This is needed because
   /// EXTRACT_SUBVECTOR usually has custom lowering that depends on the index of
   /// the first element, and only the target knows which lowering is cheap.
   virtual bool isExtractSubvectorCheap(EVT ResVT, EVT SrcVT,
                                        unsigned Index) const {
     return false;
   }
 
+  /// Try to convert an extract element of a vector binary operation into an
+  /// extract element followed by a scalar operation.
+  virtual bool shouldScalarizeBinop(SDValue VecOp) const {
+    return false;
+  }
+
   // Return true if it is profitable to use a scalar input to a BUILD_VECTOR
   // even if the vector itself has multiple uses.
   virtual bool aggressivelyPreferBuildVectorSources(EVT VecVT) const {
     return false;
   }
 
   // Return true if CodeGenPrepare should consider splitting large offset of a
   // GEP to make the GEP fit into the addressing mode and can be sunk into the

llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

if (!Const || Const->getAPIntValue().getBitWidth() != BitWidth ||
         (Const->isOpaque() && NoOpaques))
       return false;
   }
   return true;
 }
 
 // Determines if a BUILD_VECTOR is composed of all-constants possibly mixed with
 // undef's.
-static bool isAnyConstantBuildVector(const SDNode *N) {
-  return ISD::isBuildVectorOfConstantSDNodes(N) ||
-         ISD::isBuildVectorOfConstantFPSDNodes(N);
+static bool isAnyConstantBuildVector(SDValue V, bool NoOpaques = false) {
+  if (V.getOpcode() != ISD::BUILD_VECTOR)
+    return false;
+  return isConstantOrConstantVector(V, NoOpaques) ||
+         ISD::isBuildVectorOfConstantFPSDNodes(V.getNode());
 }
 
 SDValue DAGCombiner::ReassociateOps(unsigned Opc, const SDLoc &DL, SDValue N0,
                                     SDValue N1, SDNodeFlags Flags) {
   // Don't reassociate reductions.
   if (Flags.hasVectorReduction())
     return SDValue();
 
SDValue DAGCombiner::scalarizeExtractedVectorLoad(SDNode *EVE, EVT InVecVT,
   AddToWorklist(Load.getNode());
   AddUsersToWorklist(Load.getNode()); // Add users too
   // Make sure to revisit this node to clean it up; it will usually be dead.
   AddToWorklist(EVE);
   ++OpsNarrowed;
   return SDValue(EVE, 0);
 }
 
+/// Transform a vector binary operation into a scalar binary operation by moving
+/// the math/logic after an extract element of a vector.
+static SDValue scalarizeExtractedBinop(SDNode *ExtElt, SelectionDAG &DAG,
+                                       bool LegalOperations) {
+  SDValue Vec = ExtElt->getOperand(0);
+  SDValue Index = ExtElt->getOperand(1);
+  auto *IndexC = dyn_cast<ConstantSDNode>(Index);
+  if (!IndexC || !ISD::isBinaryOp(Vec.getNode()) || !Vec.hasOneUse())
+    return SDValue();
+
+  // Targets may want to avoid this to prevent an expensive register transfer.
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  if (!TLI.shouldScalarizeBinop(Vec))
+    return SDValue();
+
+  // Extracting an element of a vector constant is constant-folded, so this
+  // transform is just replacing a vector op with a scalar op while moving the
+  // extract.
+  SDValue Op0 = Vec.getOperand(0);
+  SDValue Op1 = Vec.getOperand(1);
+  if (isAnyConstantBuildVector(Op0, true) ||
+      isAnyConstantBuildVector(Op1, true)) {
+    // extractelt (binop X, C), IndexC --> binop (extractelt X, IndexC), C'
+    // extractelt (binop C, X), IndexC --> binop C', (extractelt X, IndexC)
+    SDLoc DL(ExtElt);
+    EVT VT = ExtElt->getValueType(0);
+    SDValue Ext0 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, VT, Op0, Index);
+    SDValue Ext1 = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, VT, Op1, Index);
+    return DAG.getNode(Vec.getOpcode(), DL, VT, Ext0, Ext1);
+  }
+
+  return SDValue();
+}
+
 SDValue DAGCombiner::visitEXTRACT_VECTOR_ELT(SDNode *N) {
   SDValue VecOp = N->getOperand(0);
   SDValue Index = N->getOperand(1);
   EVT ScalarVT = N->getValueType(0);
   EVT VecVT = VecOp.getValueType();
   if (VecOp.isUndef())
     return DAG.getUNDEF(ScalarVT);
 
if (LegalTypes && BCSrc.getValueType().isInteger() &&
       if (ExtractIndex == BCTruncElt && XBitWidth > VecEltBitWidth) {
         assert(XBitWidth % VecEltBitWidth == 0 &&
                "Scalar bitwidth must be a multiple of vector element bitwidth");
         return DAG.getAnyExtOrTrunc(X, DL, ScalarVT);
       }
     }
   }
 
+  if (SDValue BO = scalarizeExtractedBinop(N, DAG, LegalOperations))
+    return BO;
+
   // Transform: (EXTRACT_VECTOR_ELT( VECTOR_SHUFFLE )) -> EXTRACT_VECTOR_ELT.
   // We only perform this optimization before the op legalization phase because
   // we may introduce new vector instructions which are not backed by TD
   // patterns. For example on AVX, extracting elements from a wide vector
   // without using extract_subvector. However, if we can find an underlying
   // scalar value, then we can always use that.
   if (IndexC && VecOp.getOpcode() == ISD::VECTOR_SHUFFLE) {
     auto *Shuf = cast<ShuffleVectorSDNode>(VecOp);
if (!N0->hasOneUse())
     return SDValue();
 
   // If only one of N1,N2 is constant, bail out if it is not ALL_ZEROS as
   // discussed above.
   if (!N1.isUndef()) {
     if (!N1->hasOneUse())
       return SDValue();
 
-    bool N0AnyConst = isAnyConstantBuildVector(N0.getNode());
-    bool N1AnyConst = isAnyConstantBuildVector(N1.getNode());
+    bool N0AnyConst = isAnyConstantBuildVector(N0);
+    bool N1AnyConst = isAnyConstantBuildVector(N1);
     if (N0AnyConst && !N1AnyConst && !ISD::isBuildVectorAllZeros(N0.getNode()))
       return SDValue();
     if (!N0AnyConst && N1AnyConst && !ISD::isBuildVectorAllZeros(N1.getNode()))
       return SDValue();
   }
 
   // If both inputs are splats of the same value then we can safely merge this
   // to a single BUILD_VECTOR with undef elements based on the shuffle mask.

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

public:
     bool shouldUseStrictFP_TO_INT(EVT FpVT, EVT IntVT,
                                   bool IsSigned) const override;
 
     /// Return true if EXTRACT_SUBVECTOR is cheap for this result type
     /// with this index.
     bool isExtractSubvectorCheap(EVT ResVT, EVT SrcVT,
                                  unsigned Index) const override;
 
+    /// Scalar ops always have equal or better analysis/performance/power than
+    /// the vector equivalent, so this always makes sense if the scalar op is
+    /// supported.
+    bool shouldScalarizeBinop(SDValue) const override;
+
     bool storeOfVectorConstantIsCheap(EVT MemVT, unsigned NumElem,
                                       unsigned AddrSpace) const override {
       // If we can replace more than 2 scalar stores, there will be a reduction
       // in instructions even after we add a vector constant load.
       return NumElem > 2;
     }
 
     bool isLoadBitCastBeneficial(EVT LoadVT, EVT BitcastVT) const override;

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

bool X86TargetLowering::isExtractSubvectorCheap(EVT ResVT, EVT SrcVT,
   // extract half of vector.
   if (ResVT.getVectorElementType() == MVT::i1)
     return Index == 0 || ((ResVT.getSizeInBits() == SrcVT.getSizeInBits()*2) &&
                           (Index == ResVT.getVectorNumElements()));
 
   return (Index % ResVT.getVectorNumElements()) == 0;
 }
 
+bool X86TargetLowering::shouldScalarizeBinop(SDValue VecOp) const {
+  // If the vector op is not supported, try to convert to scalar.
+  EVT VecVT = VecOp.getValueType();
+  if (!isOperationLegalOrCustomOrPromote(VecOp.getOpcode(), VecVT))
+    return true;
+
+  // If the vector op is supported, but the scalar op is not, the transform may
+  // not be worthwhile.
+  EVT ScalarVT = VecVT.getScalarType();
+  return isOperationLegalOrCustomOrPromote(VecOp.getOpcode(), ScalarVT);
+}
+
 bool X86TargetLowering::isCheapToSpeculateCttz() const {
   // Speculate cttz only if we can directly use TZCNT.
   return Subtarget.hasBMI();
 }
 
 bool X86TargetLowering::isCheapToSpeculateCtlz() const {
   // Speculate ctlz only if we can directly use LZCNT.
   return Subtarget.hasLZCNT();

llvm/trunk/test/CodeGen/X86/and-load-fold.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=x86_64-- -mcpu=generic < %s | FileCheck %s
 
 ; Verify that the DAGCombiner doesn't wrongly remove the 'and' from the dag.
 
 define i8 @foo(<4 x i8>* %V) {
 ; CHECK-LABEL: foo:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; CHECK-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
-; CHECK-NEXT:    pand {{.*}}(%rip), %xmm0
-; CHECK-NEXT:    pextrw $4, %xmm0, %eax
+; CHECK-NEXT:    pextrw $2, %xmm0, %eax
+; CHECK-NEXT:    andb $95, %al
 ; CHECK-NEXT:    # kill: def $al killed $al killed $eax
 ; CHECK-NEXT:    retq
   %Vp = bitcast <4 x i8>* %V to <3 x i8>*
   %V3i8 = load <3 x i8>, <3 x i8>* %Vp, align 4
   %t0 = and <3 x i8> %V3i8, <i8 undef, i8 undef, i8 95>
   %t1 = extractelement <3 x i8> %t0, i64 2
   ret i8 %t1
 }
 

llvm/trunk/test/CodeGen/X86/extract-fp.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+sse2  | FileCheck %s
 
 define float @ext_fadd_v4f32(<4 x float> %x) {
 ; CHECK-LABEL: ext_fadd_v4f32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    addps {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
+; CHECK-NEXT:    addss {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    retq
   %bo = fadd <4 x float> %x, <float 1.0, float 2.0, float 3.0, float 42.0>
   %ext = extractelement <4 x float> %bo, i32 2
   ret float %ext
 }
 
 define float @ext_fsub_v4f32(<4 x float> %x) {
 ; CHECK-LABEL: ext_fsub_v4f32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    movaps {{.*#+}} xmm1 = <u,2.0E+0,u,u>
-; CHECK-NEXT:    subps %xmm0, %xmm1
-; CHECK-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1,2,3]
+; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    subss %xmm0, %xmm1
 ; CHECK-NEXT:    movaps %xmm1, %xmm0
 ; CHECK-NEXT:    retq
   %bo = fsub <4 x float> <float 1.0, float 2.0, float 3.0, float 42.0>, %x
   %ext = extractelement <4 x float> %bo, i32 1
   ret float %ext
 }
 
 define float @ext_fmul_v4f32(<4 x float> %x) {
 ; CHECK-LABEL: ext_fmul_v4f32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    mulps {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,1,2,3]
+; CHECK-NEXT:    mulss {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    retq
   %bo = fmul <4 x float> %x, <float 1.0, float 2.0, float 3.0, float 42.0>
   %ext = extractelement <4 x float> %bo, i32 3
   ret float %ext
 }
 
+; TODO: X / 1.0 --> X
+
 define float @ext_fdiv_v4f32(<4 x float> %x) {
 ; CHECK-LABEL: ext_fdiv_v4f32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; CHECK-NEXT:    divps %xmm1, %xmm0
+; CHECK-NEXT:    divss {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    retq
   %bo = fdiv <4 x float> %x, <float 1.0, float 2.0, float 3.0, float 42.0>
   %ext = extractelement <4 x float> %bo, i32 0
   ret float %ext
 }
 
 define float @ext_fdiv_v4f32_constant_op0(<4 x float> %x) {
 ; CHECK-LABEL: ext_fdiv_v4f32_constant_op0:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    movaps {{.*#+}} xmm1 = <u,2.0E+0,u,u>
-; CHECK-NEXT:    divps %xmm0, %xmm1
-; CHECK-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1,2,3]
+; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; CHECK-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    divss %xmm0, %xmm1
 ; CHECK-NEXT:    movaps %xmm1, %xmm0
 ; CHECK-NEXT:    retq
   %bo = fdiv <4 x float> <float 1.0, float 2.0, float 3.0, float 42.0>, %x
   %ext = extractelement <4 x float> %bo, i32 1
   ret float %ext
 }
 
 define float @ext_frem_v4f32(<4 x float> %x) {

llvm/trunk/test/CodeGen/X86/horizontal-reduce-smax.ll

 ; X86-SSE2-NEXT:    psrld $16, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v8i16:
 ; X86-SSE42:       ## %bb.0:
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI2_0, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v8i16:
 ; X86-AVX:       ## %bb.0:
-; X86-AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpxor LCPI2_0, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v8i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v8i16:
 ; X64-SSE42:       ## %bb.0:
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v8i16:
 ; X64-AVX:       ## %bb.0:
-; X64-AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X64-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vmovd %xmm0, %eax
+; X64-AVX-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X64-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <8 x i16> %a0, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <8 x i16> %a0, %1
   %3  = select <8 x i1> %2, <8 x i16> %a0, <8 x i16> %1
   %4  = shufflevector <8 x i16> %3, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp sgt <8 x i16> %3, %4
   %6  = select <8 x i1> %5, <8 x i16> %3, <8 x i16> %4

 ; X86-SSE2-NEXT:    pandn %xmm0, %xmm1
 ; X86-SSE2-NEXT:    por %xmm2, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i8:
 ; X86-SSE42:       ## %bb.0:
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X86-SSE42-NEXT:    psrlw $8, %xmm2
-; X86-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X86-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI3_0, %xmm0
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    xorb $127, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v16i8:
 ; X86-AVX:       ## %bb.0:
-; X86-AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpxor LCPI3_0, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    xorb $127, %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm2
 ; X64-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2

 ; X64-SSE2-NEXT:    pandn %xmm0, %xmm1
 ; X64-SSE2-NEXT:    por %xmm2, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i8:
 ; X64-SSE42:       ## %bb.0:
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X64-SSE42-NEXT:    psrlw $8, %xmm2
-; X64-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X64-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    xorb $127, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v16i8:
 ; X64-AVX:       ## %bb.0:
-; X64-AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X64-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    xorb $127, %al
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <16 x i8> %a0, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <16 x i8> %a0, %1
   %3  = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %1
   %4  = shufflevector <16 x i8> %3, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp sgt <16 x i8> %3, %4
   %6  = select <16 x i1> %5, <16 x i8> %3, <16 x i8> %4

 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i16:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxsw %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI6_0, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v16i16:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X86-AVX1-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpxor LCPI6_0, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vmovd %xmm0, %eax
+; X86-AVX1-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X86-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v16i16:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpxor LCPI6_0, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vmovd %xmm0, %eax
+; X86-AVX2-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X86-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i16:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxsw %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v16i16:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X64-AVX1-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovd %xmm0, %eax
+; X64-AVX1-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X64-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v16i16:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vmovd %xmm0, %eax
+; X64-AVX2-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X64-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v16i16:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vmovd %xmm0, %eax
+; X64-AVX512-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X64-AVX512-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <16 x i16> %a0, <16 x i16> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <16 x i16> %a0, %1
   %3  = select <16 x i1> %2, <16 x i16> %a0, <16 x i16> %1
   %4  = shufflevector <16 x i16> %3, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp sgt <16 x i16> %3, %4

 ; X86-SSE2-NEXT:    por %xmm1, %xmm2
 ; X86-SSE2-NEXT:    movd %xmm2, %eax
 ; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxsb %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X86-SSE42-NEXT:    psrlw $8, %xmm2
-; X86-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X86-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI7_0, %xmm0
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    xorb $127, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v32i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X86-AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpxor LCPI7_0, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    xorb $127, %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v32i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpxor LCPI7_0, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    xorb $127, %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm2
 ; X64-SSE2-NEXT:    pcmpgtb %xmm1, %xmm2

 ; X64-SSE2-NEXT:    por %xmm1, %xmm2
 ; X64-SSE2-NEXT:    movd %xmm2, %eax
 ; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxsb %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X64-SSE42-NEXT:    psrlw $8, %xmm2
-; X64-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X64-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    xorb $127, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X64-AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    xorb $127, %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    xorb $127, %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    xorb $127, %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp sgt <32 x i8> %3, %4

 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i16:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxsw %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pmaxsw %xmm2, %xmm0
 ; X86-SSE42-NEXT:    pmaxsw %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI10_0, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v32i16:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X86-AVX1-NEXT:    vpmaxsw %xmm2, %xmm3, %xmm2
 ; X86-AVX1-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpmaxsw %xmm2, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpxor LCPI10_0, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vmovd %xmm0, %eax
+; X86-AVX1-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X86-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v32i16:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpxor LCPI10_0, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vmovd %xmm0, %eax
+; X86-AVX2-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X86-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pmaxsw %xmm3, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm2, %xmm0

 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i16:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxsw %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pmaxsw %xmm2, %xmm0
 ; X64-SSE42-NEXT:    pmaxsw %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i16:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X64-AVX1-NEXT:    vpmaxsw %xmm2, %xmm3, %xmm2
 ; X64-AVX1-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpmaxsw %xmm2, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovd %xmm0, %eax
+; X64-AVX1-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X64-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i16:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vmovd %xmm0, %eax
+; X64-AVX2-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X64-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i16:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; X64-AVX512-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vmovd %xmm0, %eax
+; X64-AVX512-NEXT:    xorl $32767, %eax ## imm = 0x7FFF
 ; X64-AVX512-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i16> %a0, <32 x i16> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <32 x i16> %a0, %1
   %3  = select <32 x i1> %2, <32 x i16> %a0, <32 x i16> %1
   %4  = shufflevector <32 x i16> %3, <32 x i16> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp sgt <32 x i16> %3, %4

 ; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v64i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxsb %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pmaxsb %xmm2, %xmm0
 ; X86-SSE42-NEXT:    pmaxsb %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X86-SSE42-NEXT:    psrlw $8, %xmm2
-; X86-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X86-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI11_0, %xmm0
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    xorb $127, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v64i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X86-AVX1-NEXT:    vpmaxsb %xmm2, %xmm3, %xmm2
 ; X86-AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpmaxsb %xmm2, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpxor LCPI11_0, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    xorb $127, %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v64i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpxor LCPI11_0, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    xorb $127, %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm4
 ; X64-SSE2-NEXT:    pcmpgtb %xmm3, %xmm4

 ; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v64i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxsb %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pmaxsb %xmm2, %xmm0
 ; X64-SSE42-NEXT:    pmaxsb %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X64-SSE42-NEXT:    psrlw $8, %xmm2
-; X64-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X64-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    xorb $127, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v64i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X64-AVX1-NEXT:    vpmaxsb %xmm2, %xmm3, %xmm2
 ; X64-AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpmaxsb %xmm2, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    xorb $127, %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v64i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    xorb $127, %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v64i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; X64-AVX512-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    xorb $127, %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp sgt <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp sgt <64 x i8> %3, %4

llvm/trunk/test/CodeGen/X86/horizontal-reduce-smin.ll

 ; X86-SSE2-NEXT:    psrld $16, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v8i16:
 ; X86-SSE42:       ## %bb.0:
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI2_0, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v8i16:
 ; X86-AVX:       ## %bb.0:
-; X86-AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpxor LCPI2_0, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v8i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v8i16:
 ; X64-SSE42:       ## %bb.0:
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v8i16:
 ; X64-AVX:       ## %bb.0:
-; X64-AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vmovd %xmm0, %eax
+; X64-AVX-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <8 x i16> %a0, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <8 x i16> %a0, %1
   %3  = select <8 x i1> %2, <8 x i16> %a0, <8 x i16> %1
   %4  = shufflevector <8 x i16> %3, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp slt <8 x i16> %3, %4
   %6  = select <8 x i1> %5, <8 x i16> %3, <8 x i16> %4

 ; X86-SSE2-NEXT:    pandn %xmm0, %xmm1
 ; X86-SSE2-NEXT:    por %xmm2, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i8:
 ; X86-SSE42:       ## %bb.0:
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X86-SSE42-NEXT:    psrlw $8, %xmm2
-; X86-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X86-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI3_0, %xmm0
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    xorb $-128, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v16i8:
 ; X86-AVX:       ## %bb.0:
-; X86-AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpxor LCPI3_0, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    xorb $-128, %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm2
 ; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm2

 ; X64-SSE2-NEXT:    pandn %xmm0, %xmm1
 ; X64-SSE2-NEXT:    por %xmm2, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i8:
 ; X64-SSE42:       ## %bb.0:
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X64-SSE42-NEXT:    psrlw $8, %xmm2
-; X64-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X64-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    xorb $-128, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX-LABEL: test_reduce_v16i8:
 ; X64-AVX:       ## %bb.0:
-; X64-AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X64-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX-NEXT:    xorb $-128, %al
 ; X64-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX-NEXT:    retq
   %1  = shufflevector <16 x i8> %a0, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <16 x i8> %a0, %1
   %3  = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %1
   %4  = shufflevector <16 x i8> %3, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp slt <16 x i8> %3, %4
   %6  = select <16 x i1> %5, <16 x i8> %3, <16 x i8> %4

 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i16:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pminsw %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI6_0, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v16i16:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X86-AVX1-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpxor LCPI6_0, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vmovd %xmm0, %eax
+; X86-AVX1-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v16i16:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpxor LCPI6_0, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vmovd %xmm0, %eax
+; X86-AVX2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i16:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pminsw %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v16i16:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X64-AVX1-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovd %xmm0, %eax
+; X64-AVX1-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v16i16:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vmovd %xmm0, %eax
+; X64-AVX2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v16i16:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vmovd %xmm0, %eax
+; X64-AVX512-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-AVX512-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <16 x i16> %a0, <16 x i16> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <16 x i16> %a0, %1
   %3  = select <16 x i1> %2, <16 x i16> %a0, <16 x i16> %1
   %4  = shufflevector <16 x i16> %3, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp slt <16 x i16> %3, %4

 ; X86-SSE2-NEXT:    por %xmm1, %xmm2
 ; X86-SSE2-NEXT:    movd %xmm2, %eax
 ; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pminsb %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X86-SSE42-NEXT:    psrlw $8, %xmm2
-; X86-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X86-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI7_0, %xmm0
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    xorb $-128, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v32i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X86-AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpxor LCPI7_0, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    xorb $-128, %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v32i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpxor LCPI7_0, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    xorb $-128, %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm2
 ; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm2

 ; X64-SSE2-NEXT:    por %xmm1, %xmm2
 ; X64-SSE2-NEXT:    movd %xmm2, %eax
 ; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pminsb %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X64-SSE42-NEXT:    psrlw $8, %xmm2
-; X64-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X64-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    xorb $-128, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X64-AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    xorb $-128, %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    xorb $-128, %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    xorb $-128, %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp slt <32 x i8> %3, %4

 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i16:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pminsw %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pminsw %xmm2, %xmm0
 ; X86-SSE42-NEXT:    pminsw %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI10_0, %xmm0
 ; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v32i16:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X86-AVX1-NEXT:    vpminsw %xmm2, %xmm3, %xmm2
 ; X86-AVX1-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpminsw %xmm2, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpxor LCPI10_0, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vmovd %xmm0, %eax
+; X86-AVX1-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v32i16:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpxor LCPI10_0, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vmovd %xmm0, %eax
+; X86-AVX2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pminsw %xmm3, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm2, %xmm0

 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i16:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pminsw %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pminsw %xmm2, %xmm0
 ; X64-SSE42-NEXT:    pminsw %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i16:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X64-AVX1-NEXT:    vpminsw %xmm2, %xmm3, %xmm2
 ; X64-AVX1-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpminsw %xmm2, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovd %xmm0, %eax
+; X64-AVX1-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i16:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vmovd %xmm0, %eax
+; X64-AVX2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i16:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; X64-AVX512-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vmovd %xmm0, %eax
+; X64-AVX512-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-AVX512-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i16> %a0, <32 x i16> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <32 x i16> %a0, %1
   %3  = select <32 x i1> %2, <32 x i16> %a0, <32 x i16> %1
   %4  = shufflevector <32 x i16> %3, <32 x i16> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp slt <32 x i16> %3, %4

 ; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v64i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pminsb %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pminsb %xmm2, %xmm0
 ; X86-SSE42-NEXT:    pminsb %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X86-SSE42-NEXT:    psrlw $8, %xmm2
-; X86-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X86-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor LCPI11_0, %xmm0
+; X86-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X86-SSE42-NEXT:    psrlw $8, %xmm1
+; X86-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    xorb $-128, %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v64i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X86-AVX1-NEXT:    vpminsb %xmm2, %xmm3, %xmm2
 ; X86-AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpminsb %xmm2, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpxor LCPI11_0, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    xorb $-128, %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v64i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpxor LCPI11_0, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    xorb $-128, %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa %xmm2, %xmm4
 ; X64-SSE2-NEXT:    pcmpgtb %xmm0, %xmm4

 ; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v64i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pminsb %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pminsb %xmm2, %xmm0
 ; X64-SSE42-NEXT:    pminsb %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X64-SSE42-NEXT:    psrlw $8, %xmm2
-; X64-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X64-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor {{.*}}(%rip), %xmm0
+; X64-SSE42-NEXT:    movdqa %xmm0, %xmm1
+; X64-SSE42-NEXT:    psrlw $8, %xmm1
+; X64-SSE42-NEXT:    pminub %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    xorb $-128, %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v64i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X64-AVX1-NEXT:    vpminsb %xmm2, %xmm3, %xmm2
 ; X64-AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpminsb %xmm2, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    xorb $-128, %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v64i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    xorb $-128, %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v64i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; X64-AVX512-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    xorb $-128, %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp slt <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp slt <64 x i8> %3, %4

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

 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v8i16:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    notl %eax
 ; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v8i16:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vmovd %xmm0, %eax
+; X86-AVX-NEXT:    notl %eax
 ; X86-AVX-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v8i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v8i16:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    notl %eax
 ; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v8i16:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovd %xmm0, %eax
+; X64-AVX1-NEXT:    notl %eax
 ; X64-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v8i16:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vmovd %xmm0, %eax
+; X64-AVX2-NEXT:    notl %eax
 ; X64-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v8i16:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vmovd %xmm0, %eax
+; X64-AVX512-NEXT:    notl %eax
 ; X64-AVX512-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <8 x i16> %a0, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <8 x i16> %a0, %1
   %3  = select <8 x i1> %2, <8 x i16> %a0, <8 x i16> %1
   %4  = shufflevector <8 x i16> %3, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ugt <8 x i16> %3, %4
   %6  = select <8 x i1> %5, <8 x i16> %3, <8 x i16> %4

 ; X86-SSE2-NEXT:    pmaxub %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movd %xmm0, %eax
 ; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X86-SSE42-NEXT:    psrlw $8, %xmm2
-; X86-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X86-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE42-NEXT:    psrlw $8, %xmm0
+; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
 ; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    notb %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX-LABEL: test_reduce_v16i8:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX-NEXT:    notb %al
 ; X86-AVX-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; X64-SSE2-NEXT:    pmaxub %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; X64-SSE2-NEXT:    pmaxub %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
 ; X64-SSE2-NEXT:    pmaxub %xmm0, %xmm1
 ; X64-SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; X64-SSE2-NEXT:    psrlw $8, %xmm0
 ; X64-SSE2-NEXT:    pmaxub %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movd %xmm0, %eax
 ; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X64-SSE42-NEXT:    psrlw $8, %xmm2
-; X64-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X64-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE42-NEXT:    psrlw $8, %xmm0
+; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
 ; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    notb %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v16i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    notb %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v16i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    notb %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v16i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    notb %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <16 x i8> %a0, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <16 x i8> %a0, %1
   %3  = select <16 x i1> %2, <16 x i8> %a0, <16 x i8> %1
   %4  = shufflevector <16 x i8> %3, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ugt <16 x i8> %3, %4
   %6  = select <16 x i1> %5, <16 x i8> %3, <16 x i8> %4

 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v16i16:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    notl %eax
 ; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v16i16:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X86-AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vmovd %xmm0, %eax
+; X86-AVX1-NEXT:    notl %eax
 ; X86-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v16i16:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vmovd %xmm0, %eax
+; X86-AVX2-NEXT:    notl %eax
 ; X86-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v16i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1

 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm0
 ; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm2, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v16i16:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    notl %eax
 ; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v16i16:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X64-AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovd %xmm0, %eax
+; X64-AVX1-NEXT:    notl %eax
 ; X64-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v16i16:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vmovd %xmm0, %eax
+; X64-AVX2-NEXT:    notl %eax
 ; X64-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v16i16:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vmovd %xmm0, %eax
+; X64-AVX512-NEXT:    notl %eax
 ; X64-AVX512-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <16 x i16> %a0, <16 x i16> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <16 x i16> %a0, %1
   %3  = select <16 x i1> %2, <16 x i16> %a0, <16 x i16> %1
   %4  = shufflevector <16 x i16> %3, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ugt <16 x i16> %3, %4

 ; X86-SSE2-NEXT:    movd %xmm0, %eax
 ; X86-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X86-SSE42-NEXT:    psrlw $8, %xmm2
-; X86-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X86-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE42-NEXT:    psrlw $8, %xmm0
+; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
 ; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    notb %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v32i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X86-AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    notb %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v32i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    notb %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pmaxub %xmm1, %xmm0
 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]

 ; X64-SSE2-NEXT:    movd %xmm0, %eax
 ; X64-SSE2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X64-SSE42-NEXT:    psrlw $8, %xmm2
-; X64-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X64-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE42-NEXT:    psrlw $8, %xmm0
+; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
 ; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    notb %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X64-AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    notb %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    notb %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    notb %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i8> %a0, <32 x i8> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <32 x i8> %a0, %1
   %3  = select <32 x i1> %2, <32 x i8> %a0, <32 x i8> %1
   %4  = shufflevector <32 x i8> %3, <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ugt <32 x i8> %3, %4

 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X86-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i16:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxuw %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pmaxuw %xmm2, %xmm0
 ; X86-SSE42-NEXT:    pmaxuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X86-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X86-SSE42-NEXT:    movd %xmm0, %eax
+; X86-SSE42-NEXT:    notl %eax
 ; X86-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v32i16:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X86-AVX1-NEXT:    vpmaxuw %xmm2, %xmm3, %xmm2
 ; X86-AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpmaxuw %xmm2, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vmovd %xmm0, %eax
+; X86-AVX1-NEXT:    notl %eax
 ; X86-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v32i16:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vmovd %xmm0, %eax
+; X86-AVX2-NEXT:    notl %eax
 ; X86-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v32i16:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm3

 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X64-SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i16:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxuw %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pmaxuw %xmm2, %xmm0
 ; X64-SSE42-NEXT:    pmaxuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X64-SSE42-NEXT:    phminposuw %xmm1, %xmm0
 ; X64-SSE42-NEXT:    movd %xmm0, %eax
+; X64-SSE42-NEXT:    notl %eax
 ; X64-SSE42-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v32i16:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X64-AVX1-NEXT:    vpmaxuw %xmm2, %xmm3, %xmm2
 ; X64-AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpmaxuw %xmm2, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovd %xmm0, %eax
+; X64-AVX1-NEXT:    notl %eax
 ; X64-AVX1-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v32i16:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vmovd %xmm0, %eax
+; X64-AVX2-NEXT:    notl %eax
 ; X64-AVX2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v32i16:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; X64-AVX512-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vmovd %xmm0, %eax
+; X64-AVX512-NEXT:    notl %eax
 ; X64-AVX512-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <32 x i16> %a0, <32 x i16> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <32 x i16> %a0, %1
   %3  = select <32 x i1> %2, <32 x i16> %a0, <32 x i16> %1
   %4  = shufflevector <32 x i16> %3, <32 x i16> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ugt <32 x i16> %3, %4

 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v64i8:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pmaxub %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pmaxub %xmm2, %xmm0
 ; X86-SSE42-NEXT:    pmaxub %xmm1, %xmm0
 ; X86-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X86-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X86-SSE42-NEXT:    psrlw $8, %xmm2
-; X86-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X86-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X86-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X86-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X86-SSE42-NEXT:    movdqa %xmm1, %xmm0
+; X86-SSE42-NEXT:    psrlw $8, %xmm0
+; X86-SSE42-NEXT:    pminub %xmm1, %xmm0
+; X86-SSE42-NEXT:    phminposuw %xmm0, %xmm0
 ; X86-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X86-SSE42-NEXT:    notb %al
 ; X86-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-SSE42-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: test_reduce_v64i8:
 ; X86-AVX1:       ## %bb.0:
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X86-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X86-AVX1-NEXT:    vpmaxub %xmm2, %xmm3, %xmm2
 ; X86-AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpmaxub %xmm2, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX1-NEXT:    notb %al
 ; X86-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl
 ;
 ; X86-AVX2-LABEL: test_reduce_v64i8:
 ; X86-AVX2:       ## %bb.0:
 ; X86-AVX2-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; X86-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X86-AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X86-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X86-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X86-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X86-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X86-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X86-AVX2-NEXT:    notb %al
 ; X86-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE2-LABEL: test_reduce_v64i8:
 ; X64-SSE2:       ## %bb.0:
 ; X64-SSE2-NEXT:    pmaxub %xmm3, %xmm1
 ; X64-SSE2-NEXT:    pmaxub %xmm2, %xmm0

 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v64i8:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pmaxub %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pmaxub %xmm2, %xmm0
 ; X64-SSE42-NEXT:    pmaxub %xmm1, %xmm0
 ; X64-SSE42-NEXT:    pcmpeqd %xmm1, %xmm1
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
-; X64-SSE42-NEXT:    movdqa %xmm0, %xmm2
-; X64-SSE42-NEXT:    psrlw $8, %xmm2
-; X64-SSE42-NEXT:    pminub %xmm0, %xmm2
-; X64-SSE42-NEXT:    phminposuw %xmm2, %xmm0
-; X64-SSE42-NEXT:    pxor %xmm1, %xmm0
+; X64-SSE42-NEXT:    pxor %xmm0, %xmm1
+; X64-SSE42-NEXT:    movdqa %xmm1, %xmm0
+; X64-SSE42-NEXT:    psrlw $8, %xmm0
+; X64-SSE42-NEXT:    pminub %xmm1, %xmm0
+; X64-SSE42-NEXT:    phminposuw %xmm0, %xmm0
 ; X64-SSE42-NEXT:    pextrb $0, %xmm0, %eax
+; X64-SSE42-NEXT:    notb %al
 ; X64-SSE42-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-SSE42-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: test_reduce_v64i8:
 ; X64-AVX1:       ## %bb.0:
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X64-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; X64-AVX1-NEXT:    vpmaxub %xmm2, %xmm3, %xmm2
 ; X64-AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpmaxub %xmm2, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX1-NEXT:    notb %al
 ; X64-AVX1-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: test_reduce_v64i8:
 ; X64-AVX2:       ## %bb.0:
 ; X64-AVX2-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; X64-AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; X64-AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; X64-AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; X64-AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; X64-AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX2-NEXT:    notb %al
 ; X64-AVX2-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
 ;
 ; X64-AVX512-LABEL: test_reduce_v64i8:
 ; X64-AVX512:       ## %bb.0:
 ; X64-AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; X64-AVX512-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; X64-AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; X64-AVX512-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; X64-AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; X64-AVX512-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; X64-AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; X64-AVX512-NEXT:    notb %al
 ; X64-AVX512-NEXT:    ## kill: def $al killed $al killed $eax
 ; X64-AVX512-NEXT:    vzeroupper
 ; X64-AVX512-NEXT:    retq
   %1  = shufflevector <64 x i8> %a0, <64 x i8> undef, <64 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2  = icmp ugt <64 x i8> %a0, %1
   %3  = select <64 x i1> %2, <64 x i8> %a0, <64 x i8> %1
   %4  = shufflevector <64 x i8> %3, <64 x i8> undef, <64 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %5  = icmp ugt <64 x i8> %3, %4

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

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

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

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

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

 ; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X86-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X86-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    psrld $16, %xmm1
 ; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    pminsw %xmm0, %xmm1
-; X86-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X86-SSE2-NEXT:    movd %xmm1, %eax
+; X86-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X86-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X86-SSE2-NEXT:    retl
 ;
 ; X86-SSE42-LABEL: test_reduce_v32i16:
 ; X86-SSE42:       ## %bb.0:
 ; X86-SSE42-NEXT:    pminuw %xmm3, %xmm1
 ; X86-SSE42-NEXT:    pminuw %xmm2, %xmm0
 ; X86-SSE42-NEXT:    pminuw %xmm1, %xmm0

 ; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm0
 ; X64-SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; X64-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    psrld $16, %xmm1
 ; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    pminsw %xmm0, %xmm1
-; X64-SSE2-NEXT:    pxor %xmm4, %xmm1
 ; X64-SSE2-NEXT:    movd %xmm1, %eax
+; X64-SSE2-NEXT:    xorl $32768, %eax ## imm = 0x8000
 ; X64-SSE2-NEXT:    ## kill: def $ax killed $ax killed $eax
 ; X64-SSE2-NEXT:    retq
 ;
 ; X64-SSE42-LABEL: test_reduce_v32i16:
 ; X64-SSE42:       ## %bb.0:
 ; X64-SSE42-NEXT:    pminuw %xmm3, %xmm1
 ; X64-SSE42-NEXT:    pminuw %xmm2, %xmm0
 ; X64-SSE42-NEXT:    pminuw %xmm1, %xmm0

llvm/trunk/test/CodeGen/X86/known-bits-vector.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X32
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X64
 
 define i32 @knownbits_mask_extract_sext(<8 x i16> %a0) nounwind {
 ; X32-LABEL: knownbits_mask_extract_sext:
 ; X32:       # %bb.0:
-; X32-NEXT:    movl $15, %eax
-; X32-NEXT:    vmovd %eax, %xmm1
-; X32-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; X32-NEXT:    vpextrw $0, %xmm0, %eax
+; X32-NEXT:    vmovd %xmm0, %eax
+; X32-NEXT:    andl $15, %eax
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: knownbits_mask_extract_sext:
 ; X64:       # %bb.0:
-; X64-NEXT:    movl $15, %eax
-; X64-NEXT:    vmovd %eax, %xmm1
-; X64-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vpextrw $0, %xmm0, %eax
+; X64-NEXT:    vmovd %xmm0, %eax
+; X64-NEXT:    andl $15, %eax
 ; X64-NEXT:    retq
   %1 = and <8 x i16> %a0, <i16 15, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1, i16 -1>
   %2 = extractelement <8 x i16> %1, i32 0
   %3 = sext i16 %2 to i32
   ret i32 %3
 }
 
 define float @knownbits_mask_extract_uitofp(<2 x i64> %a0) nounwind {
 ; X32-LABEL: knownbits_mask_extract_uitofp:
 ; X32:       # %bb.0:
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
 ; X32-NEXT:    vmovd %xmm0, %eax
 ; X32-NEXT:    vcvtsi2ssl %eax, %xmm1, %xmm0
 ; X32-NEXT:    vmovss %xmm0, (%esp)
 ; X32-NEXT:    flds (%esp)
 ; X32-NEXT:    popl %eax
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: knownbits_mask_extract_uitofp:
 ; X64:       # %bb.0:
-; X64-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
 ; X64-NEXT:    vmovq %xmm0, %rax
+; X64-NEXT:    movzwl %ax, %eax
 ; X64-NEXT:    vcvtsi2ssl %eax, %xmm1, %xmm0
 ; X64-NEXT:    retq
   %1 = and <2 x i64> %a0, <i64 65535, i64 -1>
   %2 = extractelement <2 x i64> %1, i32 0
   %3 = uitofp i64 %2 to float
   ret float %3
 }
 

llvm/trunk/test/CodeGen/X86/known-signbits-vector.ll

 ; X32-NEXT:    vcvtsi2ssl %eax, %xmm1, %xmm0
 ; X32-NEXT:    vmovss %xmm0, (%esp)
 ; X32-NEXT:    flds (%esp)
 ; X32-NEXT:    popl %eax
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: signbits_ashr_extract_sitofp_0:
 ; X64:       # %bb.0:
-; X64-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
+; X64-NEXT:    shrq $32, %rax
 ; X64-NEXT:    vcvtsi2ssl %eax, %xmm1, %xmm0
 ; X64-NEXT:    retq
   %1 = ashr <2 x i64> %a0, <i64 32, i64 32>
   %2 = extractelement <2 x i64> %1, i32 0
   %3 = sitofp i64 %2 to float
   ret float %3
 }
 

 ; X32-NEXT:    vcvtsi2ssl %eax, %xmm2, %xmm0
 ; X32-NEXT:    vmovss %xmm0, (%esp)
 ; X32-NEXT:    flds (%esp)
 ; X32-NEXT:    popl %eax
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: signbits_ashr_extract_sitofp_1:
 ; X64:       # %bb.0:
-; X64-NEXT:    vpsrlq $32, %xmm0, %xmm0
-; X64-NEXT:    vmovdqa {{.*#+}} xmm1 = [2147483648,1]
-; X64-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vpsubq %xmm1, %xmm0, %xmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
-; X64-NEXT:    vcvtsi2ssl %eax, %xmm2, %xmm0
+; X64-NEXT:    shrq $32, %rax
+; X64-NEXT:    vcvtsi2ssl %eax, %xmm1, %xmm0
 ; X64-NEXT:    retq
   %1 = ashr <2 x i64> %a0, <i64 32, i64 63>
   %2 = extractelement <2 x i64> %1, i32 0
   %3 = sitofp i64 %2 to float
   ret float %3
 }
 
 define float @signbits_ashr_shl_extract_sitofp(<2 x i64> %a0) nounwind {

 ; X32-NEXT:    vcvtsi2ssl %eax, %xmm2, %xmm0
 ; X32-NEXT:    vmovss %xmm0, (%esp)
 ; X32-NEXT:    flds (%esp)
 ; X32-NEXT:    popl %eax
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: signbits_ashr_shl_extract_sitofp:
 ; X64:       # %bb.0:
-; X64-NEXT:    vpsrlq $61, %xmm0, %xmm0
-; X64-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,8]
-; X64-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vpsubq %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vpsllq $20, %xmm0, %xmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
-; X64-NEXT:    vcvtsi2ssl %eax, %xmm2, %xmm0
+; X64-NEXT:    sarq $61, %rax
+; X64-NEXT:    shll $20, %eax
+; X64-NEXT:    vcvtsi2ssl %eax, %xmm1, %xmm0
 ; X64-NEXT:    retq
   %1 = ashr <2 x i64> %a0, <i64 61, i64 60>
   %2 = shl <2 x i64> %1, <i64 20, i64 16>
   %3 = extractelement <2 x i64> %2, i32 0
   %4 = sitofp i64 %3 to float
   ret float %4
 }
 

 ; X32-NEXT:    vmovss %xmm0, (%esp)
 ; X32-NEXT:    flds (%esp)
 ; X32-NEXT:    popl %eax
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: signbits_ashr_insert_ashr_extract_sitofp:
 ; X64:       # %bb.0:
 ; X64-NEXT:    sarq $30, %rdi
-; X64-NEXT:    vmovq %rdi, %xmm0
-; X64-NEXT:    vpsrlq $3, %xmm0, %xmm0
-; X64-NEXT:    vmovq %xmm0, %rax
-; X64-NEXT:    vcvtsi2ssl %eax, %xmm1, %xmm0
+; X64-NEXT:    shrq $3, %rdi
+; X64-NEXT:    vcvtsi2ssl %edi, %xmm0, %xmm0
 ; X64-NEXT:    retq
   %1 = ashr i64 %a0, 30
   %2 = insertelement <2 x i64> undef, i64 %1, i32 0
   %3 = insertelement <2 x i64> %2, i64 %a1, i32 1
   %4 = ashr <2 x i64> %3, <i64 3, i64 3>
   %5 = extractelement <2 x i64> %4, i32 0
   %6 = sitofp i64 %5 to float
   ret float %6

llvm/trunk/test/CodeGen/X86/pr30511.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-pc-linux-gnu | FileCheck %s
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-pc-linux-gnu"
 
 define i64 @PR30511(<2 x double> %a) {
 ; CHECK-LABEL: PR30511:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    addpd {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    cvtdq2pd %xmm0, %xmm0
-; CHECK-NEXT:    mulpd {{.*}}(%rip), %xmm0
+; CHECK-NEXT:    mulsd {{.*}}(%rip), %xmm0
 ; CHECK-NEXT:    movq %xmm0, %rax
 ; CHECK-NEXT:    retq
   %1 = fadd <2 x double> %a, <double 0x4338000000000000, double 0x4338000000000000>
   %2 = bitcast <2 x double> %1 to <2 x i64>
   %3 = trunc <2 x i64> %2 to <2 x i32>
   %4 = sitofp <2 x i32> %3 to <2 x double>
   %5 = fmul <2 x double> %4, <double 0x3E76800000000000, double 0x3E76800000000000>
   %6 = extractelement <2 x double> %5, i32 0
   %7 = bitcast double %6 to i64
   ret i64 %7
 }
 

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

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+sse2 < %s | FileCheck %s --check-prefixes=CHECK,SSE2
 ; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 < %s | FileCheck %s --check-prefixes=CHECK,SSE41
 
 define i32 @test_eq_1(<4 x i32> %A, <4 x i32> %B) {
 ; SSE2-LABEL: test_eq_1:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pcmpgtd %xmm0, %xmm1
-; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    notl %eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_eq_1:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpgtd %xmm0, %xmm1
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    pextrd $1, %xmm0, %eax
+; SSE41-NEXT:    pextrd $1, %xmm1, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    retq
   %cmp = icmp slt <4 x i32> %A, %B
   %sext = sext <4 x i1> %cmp to <4 x i32>
   %cmp1 = icmp eq <4 x i32> %sext, zeroinitializer
   %t0 = extractelement <4 x i1> %cmp1, i32 1
   %t1 = sext i1 %t0 to i32
   ret i32 %t1
 }
; CHECK-NEXT:    retq
   %t1 = sext i1 %t0 to i32
   ret i32 %t1
 }
 
 define i32 @test_ge_1(<4 x i32> %A, <4 x i32> %B) {
 ; SSE2-LABEL: test_ge_1:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pcmpgtd %xmm0, %xmm1
-; SSE2-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    notl %eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_ge_1:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpgtd %xmm0, %xmm1
-; SSE41-NEXT:    pcmpeqd %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    pextrd $1, %xmm0, %eax
+; SSE41-NEXT:    pextrd $1, %xmm1, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    retq
   %cmp = icmp slt <4 x i32> %A, %B
   %sext = sext <4 x i1> %cmp to <4 x i32>
   %cmp1 = icmp sge <4 x i32> %sext, zeroinitializer
   %t0 = extractelement <4 x i1> %cmp1, i32 1
   %t1 = sext i1 %t0 to i32
   ret i32 %t1
 }
; CHECK-NEXT:    retq
   %t1 = sext i1 %t0 to i32
   ret i32 %t1
 }
 
 define i32 @test_eq_2(<4 x i32> %A, <4 x i32> %B) {
 ; SSE2-LABEL: test_eq_2:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pcmpgtd %xmm1, %xmm0
-; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    notl %eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_eq_2:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpgtd %xmm1, %xmm0
-; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    pextrd $1, %xmm1, %eax
+; SSE41-NEXT:    pextrd $1, %xmm0, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    retq
   %cmp = icmp slt <4 x i32> %B, %A
   %sext = sext <4 x i1> %cmp to <4 x i32>
   %cmp1 = icmp eq <4 x i32> %sext, zeroinitializer
   %t0 = extractelement <4 x i1> %cmp1, i32 1
   %t1 = sext i1 %t0 to i32
   ret i32 %t1
 }
; SSE41-NEXT:    retq
   %t1 = sext i1 %t0 to i32
   ret i32 %t1
 }
 
 define i32 @test_le_2(<4 x i32> %A, <4 x i32> %B) {
 ; SSE2-LABEL: test_le_2:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pcmpgtd %xmm1, %xmm0
-; SSE2-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE2-NEXT:    pxor %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    notl %eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_le_2:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpgtd %xmm1, %xmm0
-; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm0, %xmm1
-; SSE41-NEXT:    pextrd $1, %xmm1, %eax
+; SSE41-NEXT:    pextrd $1, %xmm0, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    retq
   %cmp = icmp slt <4 x i32> %B, %A
   %sext = sext <4 x i1> %cmp to <4 x i32>
   %cmp1 = icmp sle <4 x i32> zeroinitializer, %sext
   %t0 = extractelement <4 x i1> %cmp1, i32 1
   %t1 = sext i1 %t0 to i32
   ret i32 %t1
 }

llvm/trunk/test/CodeGen/X86/shrink_vmul-widen.ll

 define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; X86-SSE-LABEL: PR34947:
 ; X86-SSE:       # %bb.0:
 ; X86-SSE-NEXT:    pushl %esi
 ; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SSE-NEXT:    movdqa (%eax), %xmm5
 ; X86-SSE-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    movdqa (%ecx), %xmm3
+; X86-SSE-NEXT:    movdqa (%ecx), %xmm2
 ; X86-SSE-NEXT:    movdqa 16(%ecx), %xmm6
 ; X86-SSE-NEXT:    pxor %xmm0, %xmm0
 ; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
 ; X86-SSE-NEXT:    movdqa %xmm5, %xmm4
 ; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
 ; X86-SSE-NEXT:    punpckhwd {{.*#+}} xmm5 = xmm5[4],xmm0[4],xmm5[5],xmm0[5],xmm5[6],xmm0[6],xmm5[7],xmm0[7]
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[3,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm0, %eax
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm6[3,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm0, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    movd %edx, %xmm0
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[2,3,0,1]
-; X86-SSE-NEXT:    movd %xmm2, %eax
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm6[2,3,0,1]
-; X86-SSE-NEXT:    movd %xmm2, %esi
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[2,3,0,1]
+; X86-SSE-NEXT:    movd %xmm3, %eax
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm6[2,3,0,1]
+; X86-SSE-NEXT:    movd %xmm3, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    movd %edx, %xmm7
 ; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm0[0],xmm7[1],xmm0[1]
 ; X86-SSE-NEXT:    movd %xmm5, %eax
 ; X86-SSE-NEXT:    movd %xmm6, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
-; X86-SSE-NEXT:    movd %edx, %xmm2
+; X86-SSE-NEXT:    movd %edx, %xmm3
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[1,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm5, %eax
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm6[1,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm5, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    movd %edx, %xmm5
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1]
-; X86-SSE-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm7[0]
+; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
+; X86-SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm7[0]
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm4[3,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm6, %eax
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[3,1,2,3]
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm2[3,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm6, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    movd %edx, %xmm6
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm4[2,3,0,1]
 ; X86-SSE-NEXT:    movd %xmm7, %eax
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm3[2,3,0,1]
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm2[2,3,0,1]
 ; X86-SSE-NEXT:    movd %xmm7, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    movd %edx, %xmm7
 ; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm6[0],xmm7[1],xmm6[1]
 ; X86-SSE-NEXT:    movd %xmm4, %eax
-; X86-SSE-NEXT:    movd %xmm3, %esi
+; X86-SSE-NEXT:    movd %xmm2, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[1,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm4, %eax
 ; X86-SSE-NEXT:    movd %edx, %xmm4
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,2,3]
-; X86-SSE-NEXT:    movd %xmm3, %esi
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,2,3]
+; X86-SSE-NEXT:    movd %xmm2, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
-; X86-SSE-NEXT:    movd %edx, %xmm3
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; X86-SSE-NEXT:    movd %edx, %xmm2
+; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
 ; X86-SSE-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm7[0]
 ; X86-SSE-NEXT:    movd %xmm1, %eax
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,0],xmm6[0,0]
+; X86-SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,0],xmm6[0,0]
 ; X86-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [8199,8199,8199,8199]
 ; X86-SSE-NEXT:    pmuludq %xmm1, %xmm4
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; X86-SSE-NEXT:    pmuludq %xmm1, %xmm3
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; X86-SSE-NEXT:    pmuludq %xmm1, %xmm2
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
 ; X86-SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,0],xmm0[0,0]
+; X86-SSE-NEXT:    pmuludq %xmm1, %xmm3
+; X86-SSE-NEXT:    pmuludq %xmm1, %xmm5
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm5[0,2,2,3]
+; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl 32(%ecx)
-; X86-SSE-NEXT:    pmuludq %xmm1, %xmm2
-; X86-SSE-NEXT:    pmuludq %xmm1, %xmm5
-; X86-SSE-NEXT:    movd %edx, %xmm0
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[0,2,2,3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; X86-SSE-NEXT:    movl $8199, %eax # imm = 0x2007
-; X86-SSE-NEXT:    movd %eax, %xmm2
-; X86-SSE-NEXT:    pmuludq %xmm0, %xmm2
-; X86-SSE-NEXT:    movd %xmm2, (%eax)
-; X86-SSE-NEXT:    movdqa %xmm1, (%eax)
+; X86-SSE-NEXT:    movdqa %xmm0, (%eax)
 ; X86-SSE-NEXT:    movdqa %xmm4, (%eax)
+; X86-SSE-NEXT:    imull $8199, %edx, %eax # imm = 0x2007
+; X86-SSE-NEXT:    movl %eax, (%eax)
 ; X86-SSE-NEXT:    popl %esi
 ; X86-SSE-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: PR34947:
 ; X86-AVX1:       # %bb.0:
 ; X86-AVX1-NEXT:    pushl %ebp
 ; X86-AVX1-NEXT:    pushl %ebx
 ; X86-AVX1-NEXT:    pushl %edi

 ; X86-AVX1-NEXT:    vmovd %edx, %xmm0
 ; X86-AVX1-NEXT:    vpinsrd $1, %edi, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpinsrd $2, %esi, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpinsrd $3, %ebx, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vmovd %ebp, %xmm1
 ; X86-AVX1-NEXT:    vpinsrd $1, (%esp), %xmm1, %xmm1 # 4-byte Folded Reload
 ; X86-AVX1-NEXT:    vpinsrd $2, {{[-0-9]+}}(%e{{[sb]}}p), %xmm1, %xmm1 # 4-byte Folded Reload
 ; X86-AVX1-NEXT:    vpinsrd $3, {{[-0-9]+}}(%e{{[sb]}}p), %xmm1, %xmm1 # 4-byte Folded Reload
-; X86-AVX1-NEXT:    vmovd {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 4-byte Folded Reload
-; X86-AVX1-NEXT:    # xmm2 = mem[0],zero,zero,zero
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [8199,8199,8199,8199]
-; X86-AVX1-NEXT:    vpmulld %xmm3, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
+; X86-AVX1-NEXT:    imull $8199, {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X86-AVX1-NEXT:    # imm = 0x2007
+; X86-AVX1-NEXT:    movl %eax, (%eax)
+; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [8199,8199,8199,8199]
+; X86-AVX1-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
-; X86-AVX1-NEXT:    vpmulld %xmm3, %xmm2, %xmm1
-; X86-AVX1-NEXT:    vmovd %xmm1, (%eax)
 ; X86-AVX1-NEXT:    vmovaps %ymm0, (%eax)
 ; X86-AVX1-NEXT:    addl $16, %esp
 ; X86-AVX1-NEXT:    popl %esi
 ; X86-AVX1-NEXT:    popl %edi
 ; X86-AVX1-NEXT:    popl %ebx
 ; X86-AVX1-NEXT:    popl %ebp
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl

 ; X86-AVX2-NEXT:    vpextrd $3, %xmm1, %eax
 ; X86-AVX2-NEXT:    xorl %edx, %edx
 ; X86-AVX2-NEXT:    divl %ecx
 ; X86-AVX2-NEXT:    vpinsrd $3, %edx, %xmm4, %xmm1
 ; X86-AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
 ; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    xorl %edx, %edx
 ; X86-AVX2-NEXT:    divl 32(%esi)
-; X86-AVX2-NEXT:    vmovd %edx, %xmm0
-; X86-AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [8199,8199,8199,8199,8199,8199,8199,8199]
-; X86-AVX2-NEXT:    vpmulld %ymm2, %ymm1, %ymm1
-; X86-AVX2-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vmovd %xmm0, (%eax)
-; X86-AVX2-NEXT:    vmovdqa %ymm1, (%eax)
+; X86-AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm0 = [8199,8199,8199,8199,8199,8199,8199,8199]
+; X86-AVX2-NEXT:    vpmulld %ymm0, %ymm1, %ymm0
+; X86-AVX2-NEXT:    imull $8199, %edx, %eax # imm = 0x2007
+; X86-AVX2-NEXT:    movl %eax, (%eax)
+; X86-AVX2-NEXT:    vmovdqa %ymm0, (%eax)
 ; X86-AVX2-NEXT:    popl %esi
 ; X86-AVX2-NEXT:    popl %edi
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE-LABEL: PR34947:
 ; X64-SSE:       # %bb.0:
 ; X64-SSE-NEXT:    movdqa (%rdi), %xmm5
 ; X64-SSE-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
 ; X64-SSE-NEXT:    movdqa (%rsi), %xmm2
 ; X64-SSE-NEXT:    movdqa 16(%rsi), %xmm6
 ; X64-SSE-NEXT:    pxor %xmm0, %xmm0
 ; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X64-SSE-NEXT:    movdqa %xmm5, %xmm4
-; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
+; X64-SSE-NEXT:    movdqa %xmm5, %xmm3
+; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3]
 ; X64-SSE-NEXT:    punpckhwd {{.*#+}} xmm5 = xmm5[4],xmm0[4],xmm5[5],xmm0[5],xmm5[6],xmm0[6],xmm5[7],xmm0[7]
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[3,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm0, %eax
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm6[3,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm0, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm8
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[2,3,0,1]
-; X64-SSE-NEXT:    movd %xmm3, %eax
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm6[2,3,0,1]
-; X64-SSE-NEXT:    movd %xmm3, %ecx
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm5[2,3,0,1]
+; X64-SSE-NEXT:    movd %xmm4, %eax
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm6[2,3,0,1]
+; X64-SSE-NEXT:    movd %xmm4, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm7
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm8[0],xmm7[1],xmm8[1]
 ; X64-SSE-NEXT:    movd %xmm5, %eax
 ; X64-SSE-NEXT:    movd %xmm6, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
-; X64-SSE-NEXT:    movd %edx, %xmm3
+; X64-SSE-NEXT:    movd %edx, %xmm4
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[1,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm5, %eax
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm6[1,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm5, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm5
-; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
-; X64-SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm7[0]
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm4[3,1,2,3]
+; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
+; X64-SSE-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm7[0]
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[3,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm6, %eax
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm2[3,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm6, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm6
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm4[2,3,0,1]
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm3[2,3,0,1]
 ; X64-SSE-NEXT:    movd %xmm7, %eax
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm2[2,3,0,1]
 ; X64-SSE-NEXT:    movd %xmm7, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm7
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm6[0],xmm7[1],xmm6[1]
-; X64-SSE-NEXT:    movd %xmm4, %eax
+; X64-SSE-NEXT:    movd %xmm3, %eax
 ; X64-SSE-NEXT:    movd %xmm2, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm0
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[1,1,2,3]
-; X64-SSE-NEXT:    movd %xmm4, %eax
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,2,3]
+; X64-SSE-NEXT:    movd %xmm3, %eax
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm2, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm2
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; X64-SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm7[0]
 ; X64-SSE-NEXT:    movd %xmm1, %eax
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl 32(%rsi)
-; X64-SSE-NEXT:    movd %edx, %xmm1
-; X64-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [8199,8199,8199,8199]
-; X64-SSE-NEXT:    pmuludq %xmm4, %xmm0
+; X64-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [8199,8199,8199,8199]
+; X64-SSE-NEXT:    pmuludq %xmm1, %xmm0
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; X64-SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,0],xmm6[0,0]
-; X64-SSE-NEXT:    pmuludq %xmm4, %xmm2
+; X64-SSE-NEXT:    pmuludq %xmm1, %xmm2
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X64-SSE-NEXT:    pmuludq %xmm4, %xmm3
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
+; X64-SSE-NEXT:    pmuludq %xmm1, %xmm4
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[0,2,2,3]
 ; X64-SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,0],xmm8[0,0]
-; X64-SSE-NEXT:    pmuludq %xmm4, %xmm5
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[0,2,2,3]
-; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
-; X64-SSE-NEXT:    movl $8199, %eax # imm = 0x2007
-; X64-SSE-NEXT:    movd %eax, %xmm3
-; X64-SSE-NEXT:    pmuludq %xmm1, %xmm3
-; X64-SSE-NEXT:    movd %xmm3, (%rax)
+; X64-SSE-NEXT:    pmuludq %xmm1, %xmm5
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm5[0,2,2,3]
+; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; X64-SSE-NEXT:    imull $8199, %edx, %eax # imm = 0x2007
+; X64-SSE-NEXT:    movl %eax, (%rax)
 ; X64-SSE-NEXT:    movdqa %xmm2, (%rax)
 ; X64-SSE-NEXT:    movdqa %xmm0, (%rax)
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: PR34947:
 ; X64-AVX1:       # %bb.0:
 ; X64-AVX1-NEXT:    pushq %rbp
 ; X64-AVX1-NEXT:    pushq %rbx

 ; X64-AVX1-NEXT:    vpinsrd $2, %ecx, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpinsrd $3, %edi, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [8199,8199,8199,8199]
 ; X64-AVX1-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovd %esi, %xmm2
 ; X64-AVX1-NEXT:    vpinsrd $1, %r11d, %xmm2, %xmm2
 ; X64-AVX1-NEXT:    vpinsrd $2, %r10d, %xmm2, %xmm2
 ; X64-AVX1-NEXT:    vpinsrd $3, %r9d, %xmm2, %xmm2
-; X64-AVX1-NEXT:    vpmulld %xmm1, %xmm2, %xmm2
-; X64-AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
-; X64-AVX1-NEXT:    vmovd %r8d, %xmm2
 ; X64-AVX1-NEXT:    vpmulld %xmm1, %xmm2, %xmm1
-; X64-AVX1-NEXT:    vmovd %xmm1, (%rax)
+; X64-AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-AVX1-NEXT:    imull $8199, %r8d, %eax # imm = 0x2007
+; X64-AVX1-NEXT:    movl %eax, (%rax)
 ; X64-AVX1-NEXT:    vmovaps %ymm0, (%rax)
 ; X64-AVX1-NEXT:    popq %rbx
 ; X64-AVX1-NEXT:    popq %rbp
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: PR34947:
 ; X64-AVX2:       # %bb.0:

 ; X64-AVX2-NEXT:    vpextrd $3, %xmm1, %eax
 ; X64-AVX2-NEXT:    xorl %edx, %edx
 ; X64-AVX2-NEXT:    divl %ecx
 ; X64-AVX2-NEXT:    vpinsrd $3, %edx, %xmm4, %xmm1
 ; X64-AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
 ; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    xorl %edx, %edx
 ; X64-AVX2-NEXT:    divl 32(%rsi)
-; X64-AVX2-NEXT:    vmovd %edx, %xmm0
-; X64-AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [8199,8199,8199,8199,8199,8199,8199,8199]
-; X64-AVX2-NEXT:    vpmulld %ymm2, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vmovd %xmm0, (%rax)
-; X64-AVX2-NEXT:    vmovdqa %ymm1, (%rax)
+; X64-AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm0 = [8199,8199,8199,8199,8199,8199,8199,8199]
+; X64-AVX2-NEXT:    vpmulld %ymm0, %ymm1, %ymm0
+; X64-AVX2-NEXT:    imull $8199, %edx, %eax # imm = 0x2007
+; X64-AVX2-NEXT:    movl %eax, (%rax)
+; X64-AVX2-NEXT:    vmovdqa %ymm0, (%rax)
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
   %a0 = load <9 x i16>, <9 x i16>* %p0, align 64
   %a1 = load <9 x i32>, <9 x i32>* %p1, align 64
   %ext0 = zext <9 x i16> %a0 to <9 x i32>
   %rem = urem <9 x i32> %ext0, %a1
   %mul = mul <9 x i32> <i32 8199, i32 8199, i32 8199, i32 8199, i32 8199, i32 8199, i32 8199, i32 8199, i32 8199>, %rem
   store <9 x i32> %mul, <9 x i32>* undef, align 64
   ret void
 }

llvm/trunk/test/CodeGen/X86/shrink_vmul.ll

 define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; X86-SSE-LABEL: PR34947:
 ; X86-SSE:       # %bb.0:
 ; X86-SSE-NEXT:    pushl %esi
 ; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-SSE-NEXT:    movdqa (%eax), %xmm5
 ; X86-SSE-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; X86-SSE-NEXT:    movdqa (%ecx), %xmm3
+; X86-SSE-NEXT:    movdqa (%ecx), %xmm2
 ; X86-SSE-NEXT:    movdqa 16(%ecx), %xmm6
 ; X86-SSE-NEXT:    pxor %xmm0, %xmm0
 ; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
 ; X86-SSE-NEXT:    movdqa %xmm5, %xmm4
 ; X86-SSE-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
 ; X86-SSE-NEXT:    punpckhwd {{.*#+}} xmm5 = xmm5[4],xmm0[4],xmm5[5],xmm0[5],xmm5[6],xmm0[6],xmm5[7],xmm0[7]
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[3,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm0, %eax
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm6[3,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm0, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    movd %edx, %xmm0
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[2,3,0,1]
-; X86-SSE-NEXT:    movd %xmm2, %eax
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm6[2,3,0,1]
-; X86-SSE-NEXT:    movd %xmm2, %esi
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[2,3,0,1]
+; X86-SSE-NEXT:    movd %xmm3, %eax
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm6[2,3,0,1]
+; X86-SSE-NEXT:    movd %xmm3, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    movd %edx, %xmm7
 ; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm0[0],xmm7[1],xmm0[1]
 ; X86-SSE-NEXT:    movd %xmm5, %eax
 ; X86-SSE-NEXT:    movd %xmm6, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
-; X86-SSE-NEXT:    movd %edx, %xmm2
+; X86-SSE-NEXT:    movd %edx, %xmm3
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[1,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm5, %eax
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm6[1,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm5, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    movd %edx, %xmm5
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1]
-; X86-SSE-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm7[0]
+; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
+; X86-SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm7[0]
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm4[3,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm6, %eax
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[3,1,2,3]
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm2[3,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm6, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    movd %edx, %xmm6
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm4[2,3,0,1]
 ; X86-SSE-NEXT:    movd %xmm7, %eax
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm3[2,3,0,1]
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm2[2,3,0,1]
 ; X86-SSE-NEXT:    movd %xmm7, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    movd %edx, %xmm7
 ; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm6[0],xmm7[1],xmm6[1]
 ; X86-SSE-NEXT:    movd %xmm4, %eax
-; X86-SSE-NEXT:    movd %xmm3, %esi
+; X86-SSE-NEXT:    movd %xmm2, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[1,1,2,3]
 ; X86-SSE-NEXT:    movd %xmm4, %eax
 ; X86-SSE-NEXT:    movd %edx, %xmm4
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,2,3]
-; X86-SSE-NEXT:    movd %xmm3, %esi
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,2,3]
+; X86-SSE-NEXT:    movd %xmm2, %esi
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl %esi
-; X86-SSE-NEXT:    movd %edx, %xmm3
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; X86-SSE-NEXT:    movd %edx, %xmm2
+; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
 ; X86-SSE-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm7[0]
 ; X86-SSE-NEXT:    movd %xmm1, %eax
-; X86-SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,0],xmm6[0,0]
+; X86-SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,0],xmm6[0,0]
 ; X86-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [8199,8199,8199,8199]
 ; X86-SSE-NEXT:    pmuludq %xmm1, %xmm4
 ; X86-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; X86-SSE-NEXT:    pmuludq %xmm1, %xmm3
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; X86-SSE-NEXT:    pmuludq %xmm1, %xmm2
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
 ; X86-SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,0],xmm0[0,0]
+; X86-SSE-NEXT:    pmuludq %xmm1, %xmm3
+; X86-SSE-NEXT:    pmuludq %xmm1, %xmm5
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,2,2,3]
+; X86-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm5[0,2,2,3]
+; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; X86-SSE-NEXT:    xorl %edx, %edx
 ; X86-SSE-NEXT:    divl 32(%ecx)
-; X86-SSE-NEXT:    pmuludq %xmm1, %xmm2
-; X86-SSE-NEXT:    pmuludq %xmm1, %xmm5
-; X86-SSE-NEXT:    movd %edx, %xmm0
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
-; X86-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[0,2,2,3]
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; X86-SSE-NEXT:    movl $8199, %eax # imm = 0x2007
-; X86-SSE-NEXT:    movd %eax, %xmm2
-; X86-SSE-NEXT:    pmuludq %xmm0, %xmm2
-; X86-SSE-NEXT:    movd %xmm2, (%eax)
-; X86-SSE-NEXT:    movdqa %xmm1, (%eax)
+; X86-SSE-NEXT:    movdqa %xmm0, (%eax)
 ; X86-SSE-NEXT:    movdqa %xmm4, (%eax)
+; X86-SSE-NEXT:    imull $8199, %edx, %eax # imm = 0x2007
+; X86-SSE-NEXT:    movl %eax, (%eax)
 ; X86-SSE-NEXT:    popl %esi
 ; X86-SSE-NEXT:    retl
 ;
 ; X86-AVX1-LABEL: PR34947:
 ; X86-AVX1:       # %bb.0:
 ; X86-AVX1-NEXT:    pushl %ebp
 ; X86-AVX1-NEXT:    pushl %ebx
 ; X86-AVX1-NEXT:    pushl %edi

 ; X86-AVX1-NEXT:    vmovd %edx, %xmm0
 ; X86-AVX1-NEXT:    vpinsrd $1, %edi, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpinsrd $2, %esi, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vpinsrd $3, %ebx, %xmm0, %xmm0
 ; X86-AVX1-NEXT:    vmovd %ebp, %xmm1
 ; X86-AVX1-NEXT:    vpinsrd $1, (%esp), %xmm1, %xmm1 # 4-byte Folded Reload
 ; X86-AVX1-NEXT:    vpinsrd $2, {{[-0-9]+}}(%e{{[sb]}}p), %xmm1, %xmm1 # 4-byte Folded Reload
 ; X86-AVX1-NEXT:    vpinsrd $3, {{[-0-9]+}}(%e{{[sb]}}p), %xmm1, %xmm1 # 4-byte Folded Reload
-; X86-AVX1-NEXT:    vmovd {{[-0-9]+}}(%e{{[sb]}}p), %xmm2 # 4-byte Folded Reload
-; X86-AVX1-NEXT:    # xmm2 = mem[0],zero,zero,zero
-; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [8199,8199,8199,8199]
-; X86-AVX1-NEXT:    vpmulld %xmm3, %xmm0, %xmm0
-; X86-AVX1-NEXT:    vpmulld %xmm3, %xmm1, %xmm1
+; X86-AVX1-NEXT:    imull $8199, {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X86-AVX1-NEXT:    # imm = 0x2007
+; X86-AVX1-NEXT:    movl %eax, (%eax)
+; X86-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [8199,8199,8199,8199]
+; X86-AVX1-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
+; X86-AVX1-NEXT:    vpmulld %xmm2, %xmm1, %xmm1
 ; X86-AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
-; X86-AVX1-NEXT:    vpmulld %xmm3, %xmm2, %xmm1
-; X86-AVX1-NEXT:    vmovd %xmm1, (%eax)
 ; X86-AVX1-NEXT:    vmovaps %ymm0, (%eax)
 ; X86-AVX1-NEXT:    addl $16, %esp
 ; X86-AVX1-NEXT:    popl %esi
 ; X86-AVX1-NEXT:    popl %edi
 ; X86-AVX1-NEXT:    popl %ebx
 ; X86-AVX1-NEXT:    popl %ebp
 ; X86-AVX1-NEXT:    vzeroupper
 ; X86-AVX1-NEXT:    retl

 ; X86-AVX2-NEXT:    vpextrd $3, %xmm1, %eax
 ; X86-AVX2-NEXT:    xorl %edx, %edx
 ; X86-AVX2-NEXT:    divl %ecx
 ; X86-AVX2-NEXT:    vpinsrd $3, %edx, %xmm4, %xmm1
 ; X86-AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
 ; X86-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X86-AVX2-NEXT:    xorl %edx, %edx
 ; X86-AVX2-NEXT:    divl 32(%esi)
-; X86-AVX2-NEXT:    vmovd %edx, %xmm0
-; X86-AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [8199,8199,8199,8199,8199,8199,8199,8199]
-; X86-AVX2-NEXT:    vpmulld %ymm2, %ymm1, %ymm1
-; X86-AVX2-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
-; X86-AVX2-NEXT:    vmovd %xmm0, (%eax)
-; X86-AVX2-NEXT:    vmovdqa %ymm1, (%eax)
+; X86-AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm0 = [8199,8199,8199,8199,8199,8199,8199,8199]
+; X86-AVX2-NEXT:    vpmulld %ymm0, %ymm1, %ymm0
+; X86-AVX2-NEXT:    imull $8199, %edx, %eax # imm = 0x2007
+; X86-AVX2-NEXT:    movl %eax, (%eax)
+; X86-AVX2-NEXT:    vmovdqa %ymm0, (%eax)
 ; X86-AVX2-NEXT:    popl %esi
 ; X86-AVX2-NEXT:    popl %edi
 ; X86-AVX2-NEXT:    vzeroupper
 ; X86-AVX2-NEXT:    retl
 ;
 ; X64-SSE-LABEL: PR34947:
 ; X64-SSE:       # %bb.0:
 ; X64-SSE-NEXT:    movdqa (%rdi), %xmm5
 ; X64-SSE-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
 ; X64-SSE-NEXT:    movdqa (%rsi), %xmm2
 ; X64-SSE-NEXT:    movdqa 16(%rsi), %xmm6
 ; X64-SSE-NEXT:    pxor %xmm0, %xmm0
 ; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; X64-SSE-NEXT:    movdqa %xmm5, %xmm4
-; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
+; X64-SSE-NEXT:    movdqa %xmm5, %xmm3
+; X64-SSE-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3]
 ; X64-SSE-NEXT:    punpckhwd {{.*#+}} xmm5 = xmm5[4],xmm0[4],xmm5[5],xmm0[5],xmm5[6],xmm0[6],xmm5[7],xmm0[7]
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[3,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm0, %eax
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm6[3,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm0, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm8
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[2,3,0,1]
-; X64-SSE-NEXT:    movd %xmm3, %eax
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm6[2,3,0,1]
-; X64-SSE-NEXT:    movd %xmm3, %ecx
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm5[2,3,0,1]
+; X64-SSE-NEXT:    movd %xmm4, %eax
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm6[2,3,0,1]
+; X64-SSE-NEXT:    movd %xmm4, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm7
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm8[0],xmm7[1],xmm8[1]
 ; X64-SSE-NEXT:    movd %xmm5, %eax
 ; X64-SSE-NEXT:    movd %xmm6, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
-; X64-SSE-NEXT:    movd %edx, %xmm3
+; X64-SSE-NEXT:    movd %edx, %xmm4
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[1,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm5, %eax
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm6[1,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm5, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm5
-; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1]
-; X64-SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm7[0]
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm4[3,1,2,3]
+; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
+; X64-SSE-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm7[0]
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[3,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm6, %eax
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm2[3,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm6, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm6
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm4[2,3,0,1]
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm3[2,3,0,1]
 ; X64-SSE-NEXT:    movd %xmm7, %eax
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm2[2,3,0,1]
 ; X64-SSE-NEXT:    movd %xmm7, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm7
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm7 = xmm7[0],xmm6[0],xmm7[1],xmm6[1]
-; X64-SSE-NEXT:    movd %xmm4, %eax
+; X64-SSE-NEXT:    movd %xmm3, %eax
 ; X64-SSE-NEXT:    movd %xmm2, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm0
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[1,1,2,3]
-; X64-SSE-NEXT:    movd %xmm4, %eax
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,2,3]
+; X64-SSE-NEXT:    movd %xmm3, %eax
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,2,3]
 ; X64-SSE-NEXT:    movd %xmm2, %ecx
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl %ecx
 ; X64-SSE-NEXT:    movd %edx, %xmm2
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; X64-SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm7[0]
 ; X64-SSE-NEXT:    movd %xmm1, %eax
 ; X64-SSE-NEXT:    xorl %edx, %edx
 ; X64-SSE-NEXT:    divl 32(%rsi)
-; X64-SSE-NEXT:    movd %edx, %xmm1
-; X64-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [8199,8199,8199,8199]
-; X64-SSE-NEXT:    pmuludq %xmm4, %xmm0
+; X64-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [8199,8199,8199,8199]
+; X64-SSE-NEXT:    pmuludq %xmm1, %xmm0
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; X64-SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,0],xmm6[0,0]
-; X64-SSE-NEXT:    pmuludq %xmm4, %xmm2
+; X64-SSE-NEXT:    pmuludq %xmm1, %xmm2
 ; X64-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
 ; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; X64-SSE-NEXT:    pmuludq %xmm4, %xmm3
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
+; X64-SSE-NEXT:    pmuludq %xmm1, %xmm4
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[0,2,2,3]
 ; X64-SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,0],xmm8[0,0]
-; X64-SSE-NEXT:    pmuludq %xmm4, %xmm5
-; X64-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[0,2,2,3]
-; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
-; X64-SSE-NEXT:    movl $8199, %eax # imm = 0x2007
-; X64-SSE-NEXT:    movd %eax, %xmm3
-; X64-SSE-NEXT:    pmuludq %xmm1, %xmm3
-; X64-SSE-NEXT:    movd %xmm3, (%rax)
+; X64-SSE-NEXT:    pmuludq %xmm1, %xmm5
+; X64-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm5[0,2,2,3]
+; X64-SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; X64-SSE-NEXT:    imull $8199, %edx, %eax # imm = 0x2007
+; X64-SSE-NEXT:    movl %eax, (%rax)
 ; X64-SSE-NEXT:    movdqa %xmm2, (%rax)
 ; X64-SSE-NEXT:    movdqa %xmm0, (%rax)
 ; X64-SSE-NEXT:    retq
 ;
 ; X64-AVX1-LABEL: PR34947:
 ; X64-AVX1:       # %bb.0:
 ; X64-AVX1-NEXT:    pushq %rbp
 ; X64-AVX1-NEXT:    pushq %rbx

 ; X64-AVX1-NEXT:    vpinsrd $2, %ecx, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vpinsrd $3, %edi, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [8199,8199,8199,8199]
 ; X64-AVX1-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
 ; X64-AVX1-NEXT:    vmovd %esi, %xmm2
 ; X64-AVX1-NEXT:    vpinsrd $1, %r11d, %xmm2, %xmm2
 ; X64-AVX1-NEXT:    vpinsrd $2, %r10d, %xmm2, %xmm2
 ; X64-AVX1-NEXT:    vpinsrd $3, %r9d, %xmm2, %xmm2
-; X64-AVX1-NEXT:    vpmulld %xmm1, %xmm2, %xmm2
-; X64-AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm2, %ymm0
-; X64-AVX1-NEXT:    vmovd %r8d, %xmm2
 ; X64-AVX1-NEXT:    vpmulld %xmm1, %xmm2, %xmm1
-; X64-AVX1-NEXT:    vmovd %xmm1, (%rax)
+; X64-AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; X64-AVX1-NEXT:    imull $8199, %r8d, %eax # imm = 0x2007
+; X64-AVX1-NEXT:    movl %eax, (%rax)
 ; X64-AVX1-NEXT:    vmovaps %ymm0, (%rax)
 ; X64-AVX1-NEXT:    popq %rbx
 ; X64-AVX1-NEXT:    popq %rbp
 ; X64-AVX1-NEXT:    vzeroupper
 ; X64-AVX1-NEXT:    retq
 ;
 ; X64-AVX2-LABEL: PR34947:
 ; X64-AVX2:       # %bb.0:

 ; X64-AVX2-NEXT:    vpextrd $3, %xmm1, %eax
 ; X64-AVX2-NEXT:    xorl %edx, %edx
 ; X64-AVX2-NEXT:    divl %ecx
 ; X64-AVX2-NEXT:    vpinsrd $3, %edx, %xmm4, %xmm1
 ; X64-AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm1
 ; X64-AVX2-NEXT:    vmovd %xmm0, %eax
 ; X64-AVX2-NEXT:    xorl %edx, %edx
 ; X64-AVX2-NEXT:    divl 32(%rsi)
-; X64-AVX2-NEXT:    vmovd %edx, %xmm0
-; X64-AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [8199,8199,8199,8199,8199,8199,8199,8199]
-; X64-AVX2-NEXT:    vpmulld %ymm2, %ymm1, %ymm1
-; X64-AVX2-NEXT:    vpmulld %xmm2, %xmm0, %xmm0
-; X64-AVX2-NEXT:    vmovd %xmm0, (%rax)
-; X64-AVX2-NEXT:    vmovdqa %ymm1, (%rax)
+; X64-AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm0 = [8199,8199,8199,8199,8199,8199,8199,8199]
+; X64-AVX2-NEXT:    vpmulld %ymm0, %ymm1, %ymm0
+; X64-AVX2-NEXT:    imull $8199, %edx, %eax # imm = 0x2007
+; X64-AVX2-NEXT:    movl %eax, (%rax)
+; X64-AVX2-NEXT:    vmovdqa %ymm0, (%rax)
 ; X64-AVX2-NEXT:    vzeroupper
 ; X64-AVX2-NEXT:    retq
   %a0 = load <9 x i16>, <9 x i16>* %p0, align 64
   %a1 = load <9 x i32>, <9 x i32>* %p1, align 64
   %ext0 = zext <9 x i16> %a0 to <9 x i32>
   %rem = urem <9 x i32> %ext0, %a1
   %mul = mul <9 x i32> <i32 8199, i32 8199, i32 8199, i32 8199, i32 8199, i32 8199, i32 8199, i32 8199, i32 8199>, %rem
   store <9 x i32> %mul, <9 x i32>* undef, align 64
   ret void
 }

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

 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    xorps %xmm4, %xmm4
 ; SSE2-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
 ; SSE2-NEXT:    movdqa %xmm0, %xmm5
 ; SSE2-NEXT:    pslld %xmm4, %xmm5
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32,32,32,32]
-; SSE2-NEXT:    psubd %xmm2, %xmm4
-; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
-; SSE2-NEXT:    movss {{.*#+}} xmm3 = xmm4[0],xmm3[1,2,3]
-; SSE2-NEXT:    psrld %xmm3, %xmm1
+; SSE2-NEXT:    movd %xmm2, %eax
+; SSE2-NEXT:    movl $32, %ecx
+; SSE2-NEXT:    subl %eax, %ecx
+; SSE2-NEXT:    movd %ecx, %xmm4
+; SSE2-NEXT:    psrld %xmm4, %xmm1
 ; SSE2-NEXT:    por %xmm5, %xmm1
+; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
 ; SSE2-NEXT:    pand %xmm2, %xmm0
 ; SSE2-NEXT:    pandn %xmm1, %xmm2
 ; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v4i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3

 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
 ; X32-SSE-NEXT:    pxor %xmm3, %xmm3
 ; X32-SSE-NEXT:    xorps %xmm4, %xmm4
 ; X32-SSE-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
 ; X32-SSE-NEXT:    pslld %xmm4, %xmm5
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [32,32,32,32]
-; X32-SSE-NEXT:    psubd %xmm2, %xmm4
-; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm2
-; X32-SSE-NEXT:    movss {{.*#+}} xmm3 = xmm4[0],xmm3[1,2,3]
-; X32-SSE-NEXT:    psrld %xmm3, %xmm1
+; X32-SSE-NEXT:    movd %xmm2, %eax
+; X32-SSE-NEXT:    movl $32, %ecx
+; X32-SSE-NEXT:    subl %eax, %ecx
+; X32-SSE-NEXT:    movd %ecx, %xmm4
+; X32-SSE-NEXT:    psrld %xmm4, %xmm1
 ; X32-SSE-NEXT:    por %xmm5, %xmm1
+; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm2
 ; X32-SSE-NEXT:    pand %xmm2, %xmm0
 ; X32-SSE-NEXT:    pandn %xmm1, %xmm2
 ; X32-SSE-NEXT:    por %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %amt, <4 x i32> undef, <4 x i32> zeroinitializer
   %res = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %splat)
   ret <4 x i32> %res
 }

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

; X32-SSE-NEXT:    retl
   %splat = shufflevector <2 x i64> %amt, <2 x i64> undef, <2 x i32> zeroinitializer
   %res = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %x, <2 x i64> %x, <2 x i64> %splat)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    xorps %xmm2, %xmm2
-; SSE2-NEXT:    xorps %xmm3, %xmm3
-; SSE2-NEXT:    movss {{.*#+}} xmm3 = xmm1[0],xmm3[1,2,3]
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    pslld %xmm3, %xmm4
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [32,32,32,32]
-; SSE2-NEXT:    psubd %xmm1, %xmm3
-; SSE2-NEXT:    movss {{.*#+}} xmm2 = xmm3[0],xmm2[1,2,3]
-; SSE2-NEXT:    psrld %xmm2, %xmm0
-; SSE2-NEXT:    por %xmm4, %xmm0
+; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    andl $31, %eax
+; SSE2-NEXT:    movd %eax, %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    pslld %xmm1, %xmm2
+; SSE2-NEXT:    movl $32, %ecx
+; SSE2-NEXT:    subl %eax, %ecx
+; SSE2-NEXT:    movd %ecx, %xmm1
+; SSE2-NEXT:    psrld %xmm1, %xmm0
+; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v4i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3

 ; XOPAVX2-LABEL: splatvar_funnnel_v4i32:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
-; X32-SSE-NEXT:    xorps %xmm2, %xmm2
-; X32-SSE-NEXT:    xorps %xmm3, %xmm3
-; X32-SSE-NEXT:    movss {{.*#+}} xmm3 = xmm1[0],xmm3[1,2,3]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
-; X32-SSE-NEXT:    pslld %xmm3, %xmm4
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [32,32,32,32]
-; X32-SSE-NEXT:    psubd %xmm1, %xmm3
-; X32-SSE-NEXT:    movss {{.*#+}} xmm2 = xmm3[0],xmm2[1,2,3]
-; X32-SSE-NEXT:    psrld %xmm2, %xmm0
-; X32-SSE-NEXT:    por %xmm4, %xmm0
+; X32-SSE-NEXT:    movd %xmm1, %eax
+; X32-SSE-NEXT:    andl $31, %eax
+; X32-SSE-NEXT:    movd %eax, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    pslld %xmm1, %xmm2
+; X32-SSE-NEXT:    movl $32, %ecx
+; X32-SSE-NEXT:    subl %eax, %ecx
+; X32-SSE-NEXT:    movd %ecx, %xmm1
+; X32-SSE-NEXT:    psrld %xmm1, %xmm0
+; X32-SSE-NEXT:    por %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %amt, <4 x i32> undef, <4 x i32> zeroinitializer
   %res = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> %splat)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @splatvar_funnnel_v8i16(<8 x i16> %x, <8 x i16> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v8i16:

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

 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm2
 ; SSE2-NEXT:    pxor %xmm3, %xmm3
 ; SSE2-NEXT:    xorps %xmm4, %xmm4
 ; SSE2-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
 ; SSE2-NEXT:    movdqa %xmm1, %xmm5
 ; SSE2-NEXT:    psrld %xmm4, %xmm5
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32,32,32,32]
-; SSE2-NEXT:    psubd %xmm2, %xmm4
-; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
-; SSE2-NEXT:    movss {{.*#+}} xmm3 = xmm4[0],xmm3[1,2,3]
-; SSE2-NEXT:    pslld %xmm3, %xmm0
+; SSE2-NEXT:    movd %xmm2, %eax
+; SSE2-NEXT:    movl $32, %ecx
+; SSE2-NEXT:    subl %eax, %ecx
+; SSE2-NEXT:    movd %ecx, %xmm4
+; SSE2-NEXT:    pslld %xmm4, %xmm0
 ; SSE2-NEXT:    por %xmm5, %xmm0
+; SSE2-NEXT:    pcmpeqd %xmm3, %xmm2
 ; SSE2-NEXT:    pand %xmm2, %xmm1
 ; SSE2-NEXT:    pandn %xmm0, %xmm2
 ; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v4i32:
 ; SSE41:       # %bb.0:

 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
 ; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm2
 ; X32-SSE-NEXT:    pxor %xmm3, %xmm3
 ; X32-SSE-NEXT:    xorps %xmm4, %xmm4
 ; X32-SSE-NEXT:    movss {{.*#+}} xmm4 = xmm2[0],xmm4[1,2,3]
 ; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
 ; X32-SSE-NEXT:    psrld %xmm4, %xmm5
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [32,32,32,32]
-; X32-SSE-NEXT:    psubd %xmm2, %xmm4
-; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm2
-; X32-SSE-NEXT:    movss {{.*#+}} xmm3 = xmm4[0],xmm3[1,2,3]
-; X32-SSE-NEXT:    pslld %xmm3, %xmm0
+; X32-SSE-NEXT:    movd %xmm2, %eax
+; X32-SSE-NEXT:    movl $32, %ecx
+; X32-SSE-NEXT:    subl %eax, %ecx
+; X32-SSE-NEXT:    movd %ecx, %xmm4
+; X32-SSE-NEXT:    pslld %xmm4, %xmm0
 ; X32-SSE-NEXT:    por %xmm5, %xmm0
+; X32-SSE-NEXT:    pcmpeqd %xmm3, %xmm2
 ; X32-SSE-NEXT:    pand %xmm2, %xmm1
 ; X32-SSE-NEXT:    pandn %xmm0, %xmm2
 ; X32-SSE-NEXT:    por %xmm1, %xmm2
 ; X32-SSE-NEXT:    movdqa %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %amt, <4 x i32> undef, <4 x i32> zeroinitializer
   %res = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %splat)
   ret <4 x i32> %res

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

; X32-SSE-NEXT:    retl
   %splat = shufflevector <2 x i64> %amt, <2 x i64> undef, <2 x i32> zeroinitializer
   %res = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %x, <2 x i64> %x, <2 x i64> %splat)
   ret <2 x i64> %res
 }
 
 define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE2-NEXT:    xorps %xmm2, %xmm2
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    psubd %xmm1, %xmm3
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm3
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    movss {{.*#+}} xmm1 = xmm3[0],xmm1[1,2,3]
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    pslld %xmm1, %xmm4
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [32,32,32,32]
-; SSE2-NEXT:    psubd %xmm3, %xmm1
-; SSE2-NEXT:    movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
-; SSE2-NEXT:    psrld %xmm2, %xmm0
-; SSE2-NEXT:    por %xmm4, %xmm0
+; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    negl %eax
+; SSE2-NEXT:    andl $31, %eax
+; SSE2-NEXT:    movd %eax, %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    pslld %xmm1, %xmm2
+; SSE2-NEXT:    movl $32, %ecx
+; SSE2-NEXT:    subl %eax, %ecx
+; SSE2-NEXT:    movd %ecx, %xmm1
+; SSE2-NEXT:    psrld %xmm1, %xmm0
+; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_funnnel_v4i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE41-NEXT:    pxor %xmm2, %xmm2
 ; SSE41-NEXT:    psubd %xmm1, %xmm2
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2

 ; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; XOPAVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
 ; XOPAVX2-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_funnnel_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; X32-SSE-NEXT:    xorps %xmm2, %xmm2
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    psubd %xmm1, %xmm3
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm3
-; X32-SSE-NEXT:    pxor %xmm1, %xmm1
-; X32-SSE-NEXT:    movss {{.*#+}} xmm1 = xmm3[0],xmm1[1,2,3]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
-; X32-SSE-NEXT:    pslld %xmm1, %xmm4
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [32,32,32,32]
-; X32-SSE-NEXT:    psubd %xmm3, %xmm1
-; X32-SSE-NEXT:    movss {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
-; X32-SSE-NEXT:    psrld %xmm2, %xmm0
-; X32-SSE-NEXT:    por %xmm4, %xmm0
+; X32-SSE-NEXT:    movd %xmm1, %eax
+; X32-SSE-NEXT:    negl %eax
+; X32-SSE-NEXT:    andl $31, %eax
+; X32-SSE-NEXT:    movd %eax, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    pslld %xmm1, %xmm2
+; X32-SSE-NEXT:    movl $32, %ecx
+; X32-SSE-NEXT:    subl %eax, %ecx
+; X32-SSE-NEXT:    movd %ecx, %xmm1
+; X32-SSE-NEXT:    psrld %xmm1, %xmm0
+; X32-SSE-NEXT:    por %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %amt, <4 x i32> undef, <4 x i32> zeroinitializer
   %res = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %x, <4 x i32> %x, <4 x i32> %splat)
   ret <4 x i32> %res
 }
 
 define <8 x i16> @splatvar_funnnel_v8i16(<8 x i16> %x, <8 x i16> %amt) nounwind {
 ; SSE2-LABEL: splatvar_funnnel_v8i16:

llvm/trunk/test/CodeGen/X86/vector-gep.ll

; CHECK-NEXT:    retl
   %A2 = getelementptr i32, <4 x i32*> %vecinit6.i, <4 x i32> <i32 1, i32 2, i32 3, i32 4>
   %A3 = getelementptr i32, <4 x i32*> %A2, <4 x i32> <i32 10, i32 14, i32 19, i32 233>
   ret <4 x i32*> %A3
 }
 
 define i32 @AGEP1(<4 x i32*> %param) nounwind {
 ; CHECK-LABEL: AGEP1:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpaddd {{\.LCPI.*}}, %xmm0, %xmm0
-; CHECK-NEXT:    vpextrd $3, %xmm0, %eax
-; CHECK-NEXT:    movl (%eax), %eax
+; CHECK-NEXT:    vextractps $3, %xmm0, %eax
+; CHECK-NEXT:    movl 16(%eax), %eax
 ; CHECK-NEXT:    retl
   %A2 = getelementptr i32, <4 x i32*> %param, <4 x i32> <i32 1, i32 2, i32 3, i32 4>
   %k = extractelement <4 x i32*> %A2, i32 3
   %v = load i32, i32* %k
   ret i32 %v
 }
 
 define i32 @AGEP2(<4 x i32*> %param, <4 x i32> %off) nounwind {

llvm/trunk/test/CodeGen/X86/vector-reduce-smax-widen.ll

 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i16:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smax.i16.v8i16(<8 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v16i16(<16 x i16> %a0) {
 ; SSE2-LABEL: test_v16i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smax.i16.v16i16(<16 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v32i16(<32 x i16> %a0) {

 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsw %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxsw %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpmaxsw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsw %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smax.i16.v32i16(<32 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v64i16(<64 x i16> %a0) {

 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsw %xmm7, %xmm3
 ; SSE41-NEXT:    pmaxsw %xmm5, %xmm1
 ; SSE41-NEXT:    pmaxsw %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxsw %xmm6, %xmm2
 ; SSE41-NEXT:    pmaxsw %xmm4, %xmm0
 ; SSE41-NEXT:    pmaxsw %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpmaxsw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpmaxsw %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpmaxsw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpmaxsw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxsw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxsw %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmaxsw %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmaxsw %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smax.i16.v64i16(<64 x i16> %a0)
   ret i16 %1
 }
 
 ;

 ; SSE2-NEXT:    pandn %xmm0, %xmm1
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    xorb $127, %al
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i8:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.i8.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
 ; SSE2-LABEL: test_v32i8:

 ; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movd %xmm2, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $127, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $127, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.i8.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {

 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsb %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpmaxsb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $127, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $127, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.i8.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {

 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsb %xmm7, %xmm3
 ; SSE41-NEXT:    pmaxsb %xmm5, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm6, %xmm2
 ; SSE41-NEXT:    pmaxsb %xmm4, %xmm0
 ; SSE41-NEXT:    pmaxsb %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpmaxsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpmaxsb %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpmaxsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpmaxsb %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxsb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsb %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $127, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxsb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmaxsb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $127, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmaxsb %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.i8.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.smax.i64.v2i64(<2 x i64>)

llvm/trunk/test/CodeGen/X86/vector-reduce-smax.ll

 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i16:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smax.i16.v8i16(<8 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v16i16(<16 x i16> %a0) {
 ; SSE2-LABEL: test_v16i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smax.i16.v16i16(<16 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v32i16(<32 x i16> %a0) {

 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsw %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxsw %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpmaxsw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsw %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smax.i16.v32i16(<32 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v64i16(<64 x i16> %a0) {

 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsw %xmm7, %xmm3
 ; SSE41-NEXT:    pmaxsw %xmm5, %xmm1
 ; SSE41-NEXT:    pmaxsw %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxsw %xmm6, %xmm2
 ; SSE41-NEXT:    pmaxsw %xmm4, %xmm0
 ; SSE41-NEXT:    pmaxsw %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpmaxsw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpmaxsw %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpmaxsw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpmaxsw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxsw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxsw %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmaxsw %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmaxsw %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpmaxsw %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32767,32767,32767,32767,32767,32767,32767,32767]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32767, %eax # imm = 0x7FFF
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smax.i16.v64i16(<64 x i16> %a0)
   ret i16 %1
 }
 
 ;

 ; SSE2-NEXT:    pandn %xmm0, %xmm1
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    xorb $127, %al
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i8:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.i8.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
 ; SSE2-LABEL: test_v32i8:

 ; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movd %xmm2, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $127, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $127, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.i8.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {

 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsb %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpmaxsb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $127, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $127, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.i8.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {

 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxsb %xmm7, %xmm3
 ; SSE41-NEXT:    pmaxsb %xmm5, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxsb %xmm6, %xmm2
 ; SSE41-NEXT:    pmaxsb %xmm4, %xmm0
 ; SSE41-NEXT:    pmaxsb %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $127, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpmaxsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpmaxsb %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpmaxsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpmaxsb %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxsb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxsb %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $127, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxsb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmaxsb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $127, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpmaxsb %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpmaxsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpmaxsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $127, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smax.i8.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.smax.i64.v2i64(<2 x i64>)

llvm/trunk/test/CodeGen/X86/vector-reduce-smin-widen.ll

 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i16:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smin.i16.v8i16(<8 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v16i16(<16 x i16> %a0) {
 ; SSE2-LABEL: test_v16i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smin.i16.v16i16(<16 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v32i16(<32 x i16> %a0) {

 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsw %xmm3, %xmm1
 ; SSE41-NEXT:    pminsw %xmm2, %xmm0
 ; SSE41-NEXT:    pminsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpminsw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsw %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smin.i16.v32i16(<32 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v64i16(<64 x i16> %a0) {

 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsw %xmm7, %xmm3
 ; SSE41-NEXT:    pminsw %xmm5, %xmm1
 ; SSE41-NEXT:    pminsw %xmm3, %xmm1
 ; SSE41-NEXT:    pminsw %xmm6, %xmm2
 ; SSE41-NEXT:    pminsw %xmm4, %xmm0
 ; SSE41-NEXT:    pminsw %xmm2, %xmm0
 ; SSE41-NEXT:    pminsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpminsw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpminsw %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpminsw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpminsw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpminsw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminsw %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpminsw %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpminsw %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smin.i16.v64i16(<64 x i16> %a0)
   ret i16 %1
 }
 
 ;

 ; SSE2-NEXT:    pandn %xmm0, %xmm1
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    xorb $-128, %al
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i8:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.i8.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
 ; SSE2-LABEL: test_v32i8:

 ; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movd %xmm2, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $-128, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $-128, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.i8.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {

 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsb %xmm3, %xmm1
 ; SSE41-NEXT:    pminsb %xmm2, %xmm0
 ; SSE41-NEXT:    pminsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpminsb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $-128, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $-128, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.i8.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {

 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsb %xmm7, %xmm3
 ; SSE41-NEXT:    pminsb %xmm5, %xmm1
 ; SSE41-NEXT:    pminsb %xmm3, %xmm1
 ; SSE41-NEXT:    pminsb %xmm6, %xmm2
 ; SSE41-NEXT:    pminsb %xmm4, %xmm0
 ; SSE41-NEXT:    pminsb %xmm2, %xmm0
 ; SSE41-NEXT:    pminsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpminsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpminsb %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpminsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpminsb %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpminsb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsb %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $-128, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminsb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpminsb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $-128, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpminsb %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.i8.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.smin.i64.v2i64(<2 x i64>)

llvm/trunk/test/CodeGen/X86/vector-reduce-smin.ll

 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i16:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v8i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smin.i16.v8i16(<8 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v16i16(<16 x i16> %a0) {
 ; SSE2-LABEL: test_v16i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smin.i16.v16i16(<16 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v32i16(<32 x i16> %a0) {

 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsw %xmm3, %xmm1
 ; SSE41-NEXT:    pminsw %xmm2, %xmm0
 ; SSE41-NEXT:    pminsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpminsw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsw %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smin.i16.v32i16(<32 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v64i16(<64 x i16> %a0) {

 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsw %xmm7, %xmm3
 ; SSE41-NEXT:    pminsw %xmm5, %xmm1
 ; SSE41-NEXT:    pminsw %xmm3, %xmm1
 ; SSE41-NEXT:    pminsw %xmm6, %xmm2
 ; SSE41-NEXT:    pminsw %xmm4, %xmm0
 ; SSE41-NEXT:    pminsw %xmm2, %xmm0
 ; SSE41-NEXT:    pminsw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
 ; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpminsw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpminsw %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpminsw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpminsw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpminsw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsw %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminsw %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpminsw %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpminsw %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminsw %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [32768,32768,32768,32768,32768,32768,32768,32768]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; AVX512-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.smin.i16.v64i16(<64 x i16> %a0)
   ret i16 %1
 }
 
 ;

 ; SSE2-NEXT:    pandn %xmm0, %xmm1
 ; SSE2-NEXT:    por %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    xorb $-128, %al
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v16i8:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.i8.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
 ; SSE2-LABEL: test_v32i8:

 ; SSE2-NEXT:    por %xmm1, %xmm2
 ; SSE2-NEXT:    movd %xmm2, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $-128, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $-128, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v32i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.i8.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {

 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsb %xmm3, %xmm1
 ; SSE41-NEXT:    pminsb %xmm2, %xmm0
 ; SSE41-NEXT:    pminsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpminsb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $-128, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $-128, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v64i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.i8.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {

 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminsb %xmm7, %xmm3
 ; SSE41-NEXT:    pminsb %xmm5, %xmm1
 ; SSE41-NEXT:    pminsb %xmm3, %xmm1
 ; SSE41-NEXT:    pminsb %xmm6, %xmm2
 ; SSE41-NEXT:    pminsb %xmm4, %xmm0
 ; SSE41-NEXT:    pminsb %xmm2, %xmm0
 ; SSE41-NEXT:    pminsb %xmm1, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pminub %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    xorb $-128, %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpminsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpminsb %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpminsb %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpminsb %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpminsb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpminsb %xmm4, %xmm0, %xmm0
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    xorb $-128, %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpminsb %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpminsb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    xorb $-128, %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v128i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpminsb %zmm1, %zmm0, %zmm0
 ; AVX512-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512-NEXT:    vpminsb %ymm1, %ymm0, %ymm0
 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512-NEXT:    vpminsb %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa {{.*#+}} xmm1 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX512-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX512-NEXT:    vpxor {{.*}}(%rip), %xmm0, %xmm0
+; AVX512-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX512-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512-NEXT:    xorb $-128, %al
 ; AVX512-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.smin.i8.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.smin.i64.v2i64(<2 x i64>)

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

 ; SSE2-LABEL: test_v2i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrld $16, %xmm1
 ; SSE41-NEXT:    pmaxuw %xmm0, %xmm1

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

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i16:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    notl %eax
 ; AVX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v8i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
+; AVX512BW-NEXT:    notl %eax
 ; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v8i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
+; AVX512VL-NEXT:    notl %eax
 ; AVX512VL-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v8i16(<8 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v16i16(<16 x i16> %a0) {
 ; SSE2-LABEL: test_v16i16:

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxuw %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    notl %eax
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    notl %eax
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v16i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
+; AVX512BW-NEXT:    notl %eax
 ; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v16i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
+; AVX512VL-NEXT:    notl %eax
 ; AVX512VL-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v16i16(<16 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v32i16(<32 x i16> %a0) {

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxuw %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxuw %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxuw %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpmaxuw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxuw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    notl %eax
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    notl %eax
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v32i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
+; AVX512BW-NEXT:    notl %eax
 ; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v32i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
+; AVX512VL-NEXT:    notl %eax
 ; AVX512VL-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v32i16(<32 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v64i16(<64 x i16> %a0) {

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    psrld $16, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxuw %xmm7, %xmm3
 ; SSE41-NEXT:    pmaxuw %xmm5, %xmm1
 ; SSE41-NEXT:    pmaxuw %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxuw %xmm6, %xmm2
 ; SSE41-NEXT:    pmaxuw %xmm4, %xmm0
 ; SSE41-NEXT:    pmaxuw %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxuw %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpmaxuw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpmaxuw %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpmaxuw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpmaxuw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxuw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxuw %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    notl %eax
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxuw %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmaxuw %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    notl %eax
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v64i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmaxuw %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
+; AVX512BW-NEXT:    notl %eax
 ; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v64i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpmaxuw %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
+; AVX512VL-NEXT:    notl %eax
 ; AVX512VL-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v64i16(<64 x i16> %a0)
   ret i16 %1
 }
 
 ;

 ; SSE2-NEXT:    pmaxub %xmm1, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    pminub %xmm1, %xmm0
+; SSE41-NEXT:    phminposuw %xmm0, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    notb %al
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v16i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
 ; SSE2-LABEL: test_v32i8:

 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxub %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    pminub %xmm1, %xmm0
+; SSE41-NEXT:    phminposuw %xmm0, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    notb %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    notb %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v32i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {

 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxub %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxub %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxub %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    pminub %xmm1, %xmm0
+; SSE41-NEXT:    phminposuw %xmm0, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpmaxub %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxub %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    notb %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    notb %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {

 ; SSE41-NEXT:    pmaxub %xmm7, %xmm3
 ; SSE41-NEXT:    pmaxub %xmm5, %xmm1
 ; SSE41-NEXT:    pmaxub %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxub %xmm6, %xmm2
 ; SSE41-NEXT:    pmaxub %xmm4, %xmm0
 ; SSE41-NEXT:    pmaxub %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxub %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    pminub %xmm1, %xmm0
+; SSE41-NEXT:    phminposuw %xmm0, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpmaxub %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpmaxub %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpmaxub %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpmaxub %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxub %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxub %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    notb %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxub %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmaxub %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    notb %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v128i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmaxub %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v128i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpmaxub %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.umax.i64.v2i64(<2 x i64>)

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

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v8i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i16:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    notl %eax
 ; AVX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v8i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
+; AVX512BW-NEXT:    notl %eax
 ; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v8i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
+; AVX512VL-NEXT:    notl %eax
 ; AVX512VL-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v8i16(<8 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v16i16(<16 x i16> %a0) {
 ; SSE2-LABEL: test_v16i16:

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxuw %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v16i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    notl %eax
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v16i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    notl %eax
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v16i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
+; AVX512BW-NEXT:    notl %eax
 ; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v16i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
+; AVX512VL-NEXT:    notl %eax
 ; AVX512VL-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v16i16(<16 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v32i16(<32 x i16> %a0) {

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxuw %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxuw %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxuw %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpmaxuw %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxuw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    notl %eax
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    notl %eax
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v32i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
+; AVX512BW-NEXT:    notl %eax
 ; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v32i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
+; AVX512VL-NEXT:    notl %eax
 ; AVX512VL-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v32i16(<32 x i16> %a0)
   ret i16 %1
 }
 
 define i16 @test_v64i16(<64 x i16> %a0) {

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pmaxsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    psrld $16, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pmaxsw %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxuw %xmm7, %xmm3
 ; SSE41-NEXT:    pmaxuw %xmm5, %xmm1
 ; SSE41-NEXT:    pmaxuw %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxuw %xmm6, %xmm2
 ; SSE41-NEXT:    pmaxuw %xmm4, %xmm0
 ; SSE41-NEXT:    pmaxuw %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxuw %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    phminposuw %xmm0, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    phminposuw %xmm1, %xmm0
 ; SSE41-NEXT:    movd %xmm0, %eax
+; SSE41-NEXT:    notl %eax
 ; SSE41-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpmaxuw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpmaxuw %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpmaxuw %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpmaxuw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxuw %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxuw %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovd %xmm0, %eax
+; AVX1-NEXT:    notl %eax
 ; AVX1-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i16:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxuw %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmaxuw %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vmovd %xmm0, %eax
+; AVX2-NEXT:    notl %eax
 ; AVX2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v64i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmaxuw %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovd %xmm0, %eax
+; AVX512BW-NEXT:    notl %eax
 ; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v64i16:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpmaxuw %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpmaxuw %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxuw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovd %xmm0, %eax
+; AVX512VL-NEXT:    notl %eax
 ; AVX512VL-NEXT:    # kill: def $ax killed $ax killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i16 @llvm.experimental.vector.reduce.umax.i16.v64i16(<64 x i16> %a0)
   ret i16 %1
 }
 
 ;

 ; SSE2-NEXT:    pmaxub %xmm1, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v16i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    pminub %xmm1, %xmm0
+; SSE41-NEXT:    phminposuw %xmm0, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v16i8:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX-NEXT:    notb %al
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v16i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v16i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v16i8(<16 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v32i8(<32 x i8> %a0) {
 ; SSE2-LABEL: test_v32i8:

 ; SSE2-NEXT:    movd %xmm0, %eax
 ; SSE2-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxub %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    pminub %xmm1, %xmm0
+; SSE41-NEXT:    phminposuw %xmm0, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    notb %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v32i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    notb %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v32i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v32i8(<32 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v64i8(<64 x i8> %a0) {

 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i8:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pmaxub %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxub %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxub %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    pminub %xmm1, %xmm0
+; SSE41-NEXT:    phminposuw %xmm0, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v64i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
 ; AVX1-NEXT:    vpmaxub %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxub %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    notb %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v64i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    notb %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v64i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v64i8(<64 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v128i8(<128 x i8> %a0) {

 ; SSE41-NEXT:    pmaxub %xmm7, %xmm3
 ; SSE41-NEXT:    pmaxub %xmm5, %xmm1
 ; SSE41-NEXT:    pmaxub %xmm3, %xmm1
 ; SSE41-NEXT:    pmaxub %xmm6, %xmm2
 ; SSE41-NEXT:    pmaxub %xmm4, %xmm0
 ; SSE41-NEXT:    pmaxub %xmm2, %xmm0
 ; SSE41-NEXT:    pmaxub %xmm1, %xmm0
 ; SSE41-NEXT:    pcmpeqd %xmm1, %xmm1
-; SSE41-NEXT:    pxor %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    pminub %xmm0, %xmm2
-; SSE41-NEXT:    phminposuw %xmm2, %xmm0
-; SSE41-NEXT:    pxor %xmm1, %xmm0
+; SSE41-NEXT:    pxor %xmm0, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    pminub %xmm1, %xmm0
+; SSE41-NEXT:    phminposuw %xmm0, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; SSE41-NEXT:    notb %al
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v128i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm5
 ; AVX1-NEXT:    vpmaxub %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm5
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
 ; AVX1-NEXT:    vpmaxub %xmm5, %xmm6, %xmm5
 ; AVX1-NEXT:    vpmaxub %xmm4, %xmm5, %xmm4
 ; AVX1-NEXT:    vpmaxub %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmaxub %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmaxub %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX1-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX1-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX1-NEXT:    notb %al
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: test_v128i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmaxub %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vpmaxub %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm2
-; AVX2-NEXT:    vpminub %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpsrlw $8, %xmm0, %xmm1
+; AVX2-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX2-NEXT:    vpxor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX2-NEXT:    notb %al
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_v128i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmaxub %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512BW-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BW-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512BW-NEXT:    vpternlogq $15, %zmm0, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    notb %al
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: test_v128i8:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpmaxub %zmm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
 ; AVX512VL-NEXT:    vpmaxub %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512VL-NEXT:    vpmaxub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512VL-NEXT:    vpminub %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vphminposuw %xmm0, %xmm0
-; AVX512VL-NEXT:    vpternlogq $15, %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    notb %al
 ; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.umax.i8.v128i8(<128 x i8> %a0)
   ret i8 %1
 }
 
 declare i64 @llvm.experimental.vector.reduce.umax.i64.v2i64(<2 x i64>)

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

 ; SSE2-LABEL: test_v2i16:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32768,32768,32768,32768,32768,32768,32768,32768]
 ; SSE2-NEXT:    pxor %xmm2, %xmm0
 ; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v2i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrld $16, %xmm1
 ; SSE41-NEXT:    pminuw %xmm0, %xmm1

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

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

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

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminuw %xmm3, %xmm1
 ; SSE41-NEXT:    pminuw %xmm2, %xmm0
 ; SSE41-NEXT:    pminuw %xmm1, %xmm0

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    psrld $16, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminuw %xmm7, %xmm3
 ; SSE41-NEXT:    pminuw %xmm5, %xmm1
 ; SSE41-NEXT:    pminuw %xmm3, %xmm1

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

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

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

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm4, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    psrld $16, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm4, %xmm1
 ; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v32i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminuw %xmm3, %xmm1
 ; SSE41-NEXT:    pminuw %xmm2, %xmm0
 ; SSE41-NEXT:    pminuw %xmm1, %xmm0

 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
 ; SSE2-NEXT:    pxor %xmm8, %xmm1
 ; SSE2-NEXT:    pminsw %xmm0, %xmm1
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    psrld $16, %xmm0
 ; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    pminsw %xmm1, %xmm0
-; SSE2-NEXT:    pxor %xmm8, %xmm0
 ; SSE2-NEXT:    movd %xmm0, %eax
+; SSE2-NEXT:    xorl $32768, %eax # imm = 0x8000
 ; SSE2-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: test_v64i16:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pminuw %xmm7, %xmm3
 ; SSE41-NEXT:    pminuw %xmm5, %xmm1
 ; SSE41-NEXT:    pminuw %xmm3, %xmm1

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

; X32-SSE-NEXT:    retl
   %lshr = lshr <2 x i64> %a, %splat64
   %or = or <2 x i64> %shl, %lshr
   ret <2 x i64> %or
 }
 
 define <4 x i32> @splatvar_rotate_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; SSE2-LABEL: splatvar_rotate_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    xorps %xmm2, %xmm2
-; SSE2-NEXT:    xorps %xmm3, %xmm3
-; SSE2-NEXT:    movss {{.*#+}} xmm3 = xmm1[0],xmm3[1,2,3]
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    pslld %xmm3, %xmm4
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [32,32,32,32]
-; SSE2-NEXT:    psubd %xmm1, %xmm3
-; SSE2-NEXT:    movss {{.*#+}} xmm2 = xmm3[0],xmm2[1,2,3]
-; SSE2-NEXT:    psrld %xmm2, %xmm0
-; SSE2-NEXT:    por %xmm4, %xmm0
+; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    andl $31, %eax
+; SSE2-NEXT:    movd %eax, %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    pslld %xmm1, %xmm2
+; SSE2-NEXT:    movl $32, %ecx
+; SSE2-NEXT:    subl %eax, %ecx
+; SSE2-NEXT:    movd %ecx, %xmm1
+; SSE2-NEXT:    psrld %xmm1, %xmm0
+; SSE2-NEXT:    por %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_rotate_v4i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE41-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; SSE41-NEXT:    movdqa %xmm0, %xmm3

 ; XOPAVX2-LABEL: splatvar_rotate_v4i32:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_rotate_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
-; X32-SSE-NEXT:    xorps %xmm2, %xmm2
-; X32-SSE-NEXT:    xorps %xmm3, %xmm3
-; X32-SSE-NEXT:    movss {{.*#+}} xmm3 = xmm1[0],xmm3[1,2,3]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
-; X32-SSE-NEXT:    pslld %xmm3, %xmm4
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [32,32,32,32]
-; X32-SSE-NEXT:    psubd %xmm1, %xmm3
-; X32-SSE-NEXT:    movss {{.*#+}} xmm2 = xmm3[0],xmm2[1,2,3]
-; X32-SSE-NEXT:    psrld %xmm2, %xmm0
-; X32-SSE-NEXT:    por %xmm4, %xmm0
+; X32-SSE-NEXT:    movd %xmm1, %eax
+; X32-SSE-NEXT:    andl $31, %eax
+; X32-SSE-NEXT:    movd %eax, %xmm1
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    pslld %xmm1, %xmm2
+; X32-SSE-NEXT:    movl $32, %ecx
+; X32-SSE-NEXT:    subl %eax, %ecx
+; X32-SSE-NEXT:    movd %ecx, %xmm1
+; X32-SSE-NEXT:    psrld %xmm1, %xmm0
+; X32-SSE-NEXT:    por %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %b, <4 x i32> undef, <4 x i32> zeroinitializer
   %splat32 = sub <4 x i32> <i32 32, i32 32, i32 32, i32 32>, %splat
   %shl = shl <4 x i32> %a, %splat
   %lshr = lshr <4 x i32> %a, %splat32
   %or = or <4 x i32> %shl, %lshr
   ret <4 x i32> %or
 }

llvm/trunk/test/CodeGen/X86/xor.ll

 ; X32-NEXT:    xorl %eax, %eax
 ; X32-NEXT:    orl %ecx, %edx
 ; X32-NEXT:    setne %al
 ; X32-NEXT:    retl
 ;
 ; X64-LIN-LABEL: PR17487:
 ; X64-LIN:       # %bb.0:
 ; X64-LIN-NEXT:    movd %edi, %xmm0
-; X64-LIN-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
-; X64-LIN-NEXT:    pandn {{.*}}(%rip), %xmm0
-; X64-LIN-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-LIN-NEXT:    movq %xmm0, %rcx
-; X64-LIN-NEXT:    xorl %eax, %eax
-; X64-LIN-NEXT:    cmpq $1, %rcx
-; X64-LIN-NEXT:    setne %al
+; X64-LIN-NEXT:    pextrw $0, %xmm0, %eax
+; X64-LIN-NEXT:    andl $1, %eax
 ; X64-LIN-NEXT:    retq
 ;
 ; X64-WIN-LABEL: PR17487:
 ; X64-WIN:       # %bb.0:
+; X64-WIN-NEXT:    andb $1, %cl
 ; X64-WIN-NEXT:    movzbl %cl, %eax
-; X64-WIN-NEXT:    movd %eax, %xmm0
-; X64-WIN-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
-; X64-WIN-NEXT:    pandn __xmm@{{.*}}(%rip), %xmm0
-; X64-WIN-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; X64-WIN-NEXT:    movq %xmm0, %rcx
-; X64-WIN-NEXT:    xorl %eax, %eax
-; X64-WIN-NEXT:    cmpq $1, %rcx
-; X64-WIN-NEXT:    setne %al
 ; X64-WIN-NEXT:    retq
   %tmp = insertelement <2 x i1> undef, i1 %tobool, i32 1
   %tmp1 = zext <2 x i1> %tmp to <2 x i64>
   %tmp2 = xor <2 x i64> %tmp1, <i64 1, i64 1>
   %tmp3 = extractelement <2 x i64> %tmp2, i32 1
   %add = add nsw i64 0, %tmp3
   %cmp6 = icmp ne i64 %add, 1
   %conv7 = zext i1 %cmp6 to i32