[DAGCombiner][x86] scalarize binop followed by extractelement

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). The transform is limited with a TLI hook because there's an
existing transform in CodeGenPrepare that tries to do the opposite transform.

Differential Revision: https://reviews.llvm.org/D55722

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@350354 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/CodeGen/TargetLowering.h b/include/llvm/CodeGen/TargetLowering.h
index 9cd4af9b22e8ef8e304cc16457be5ee363ec235e..eeff202fe2725df6f47345fb73dcda322df86205 100644 (file)
--- a/include/llvm/CodeGen/TargetLowering.h
+++ b/include/llvm/CodeGen/TargetLowering.h
@@ -2407,6 +2407,12 @@ public:
     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 {

diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index faa00d5f8ead1091dea92b8cea02e6c7f591fdaf..3d5c30ef8929a10b27ee830d20199241b98c781f 100644 (file)
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -915,9 +915,11 @@ static bool isConstantOrConstantVector(SDValue N, bool NoOpaques = false) {
 
 // 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,
@@ -15580,6 +15582,40 @@ SDValue DAGCombiner::scalarizeExtractedVectorLoad(SDNode *EVE, EVT InVecVT,
   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);
@@ -15670,6 +15706,9 @@ SDValue DAGCombiner::visitEXTRACT_VECTOR_ELT(SDNode *N) {
     }
   }
 
+  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
@@ -17055,8 +17094,8 @@ static SDValue combineShuffleOfScalars(ShuffleVectorSDNode *SVN,
     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()))

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 7aab1cdc954ffe9480428f3a8e70cd81e22292a9..13fb83588598751a09685e1959ede04e40fb334b 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -4875,6 +4875,18 @@ bool X86TargetLowering::isExtractSubvectorCheap(EVT ResVT, EVT SrcVT,
   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();

diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index 6dd6e9acc9ef3229d47dcc1fe104ec14a0c57e31..89685596e096e4c0252b03ca9a4ccd2d75d1b9a3 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@@ -1039,6 +1039,11 @@ namespace llvm {
     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

diff --git a/test/CodeGen/X86/and-load-fold.ll b/test/CodeGen/X86/and-load-fold.ll
index 4c49df1c5bc15cbf6f08f7088951d0d13befccca..27520f88a3ff166d168c76d60e7941373868d8fa 100644 (file)
--- a/test/CodeGen/X86/and-load-fold.ll
+++ b/test/CodeGen/X86/and-load-fold.ll
@@ -8,9 +8,8 @@ define i8 @foo(<4 x i8>* %V) {
 ; 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>*

diff --git a/test/CodeGen/X86/extract-fp.ll b/test/CodeGen/X86/extract-fp.ll
index f72764c044c10b3ceb3cb4870290b536b69e71ea..06ba30b603e50b7679b085074c4e13f944d63573 100644 (file)
--- a/test/CodeGen/X86/extract-fp.ll
+++ b/test/CodeGen/X86/extract-fp.ll
@@ -4,8 +4,8 @@
 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
@@ -15,9 +15,9 @@ define float @ext_fadd_v4f32(<4 x float> %x) {
 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
@@ -28,19 +28,20 @@ define float @ext_fsub_v4f32(<4 x float> %x) {
 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
@@ -50,9 +51,9 @@ define float @ext_fdiv_v4f32(<4 x float> %x) {
 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

diff --git a/test/CodeGen/X86/horizontal-reduce-smax.ll b/test/CodeGen/X86/horizontal-reduce-smax.ll
index 98fbd3620face30bed88f6dcf5b1239f2e9f2fe8..55bac985d6874520a4bb64742551ce6fff48d5b3 100644 (file)
--- a/test/CodeGen/X86/horizontal-reduce-smax.ll
+++ b/test/CodeGen/X86/horizontal-reduce-smax.ll
@@ -211,21 +211,19 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ;
 ; 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
 ;
@@ -244,21 +242,19 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ;
 ; 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>
@@ -309,26 +305,24 @@ define i8 @test_reduce_v16i8(<16 x i8> %a0) {
 ;
 ; 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
 ;
@@ -366,26 +360,24 @@ define i8 @test_reduce_v16i8(<16 x i8> %a0) {
 ;
 ; 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>
@@ -736,11 +728,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -748,11 +739,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -761,11 +751,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -787,11 +776,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -799,11 +787,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -812,11 +799,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -825,11 +811,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -890,14 +875,13 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -905,13 +889,12 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -920,13 +903,12 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -971,14 +953,13 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -986,13 +967,12 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1001,13 +981,12 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1016,13 +995,12 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1513,11 +1491,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1528,11 +1505,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1542,11 +1518,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1572,11 +1547,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1587,11 +1561,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1601,11 +1574,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1616,11 +1588,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1696,14 +1667,13 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -1714,13 +1684,12 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1730,13 +1699,12 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1793,14 +1761,13 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -1811,13 +1778,12 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1827,13 +1793,12 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1844,13 +1809,12 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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

diff --git a/test/CodeGen/X86/horizontal-reduce-smin.ll b/test/CodeGen/X86/horizontal-reduce-smin.ll
index 3ac86148e85228fd8a79d5e294b871b793e59c37..7036d9375912317774131dc614fcc0bcb6020026 100644 (file)
--- a/test/CodeGen/X86/horizontal-reduce-smin.ll
+++ b/test/CodeGen/X86/horizontal-reduce-smin.ll
@@ -213,21 +213,19 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ;
 ; 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
 ;
@@ -246,21 +244,19 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ;
 ; 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>
@@ -311,26 +307,24 @@ define i8 @test_reduce_v16i8(<16 x i8> %a0) {
 ;
 ; 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
 ;
@@ -368,26 +362,24 @@ define i8 @test_reduce_v16i8(<16 x i8> %a0) {
 ;
 ; 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>
@@ -740,11 +732,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -752,11 +743,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -765,11 +755,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -791,11 +780,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -803,11 +791,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -816,11 +803,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -829,11 +815,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -894,14 +879,13 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -909,13 +893,12 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -924,13 +907,12 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -975,14 +957,13 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -990,13 +971,12 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1005,13 +985,12 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1020,13 +999,12 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1517,11 +1495,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1532,11 +1509,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1546,11 +1522,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1576,11 +1551,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1591,11 +1565,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1605,11 +1578,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1620,11 +1592,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1700,14 +1671,13 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -1718,13 +1688,12 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1734,13 +1703,12 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1797,14 +1765,13 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -1815,13 +1782,12 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1831,13 +1797,12 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1848,13 +1813,12 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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

diff --git a/test/CodeGen/X86/horizontal-reduce-umax.ll b/test/CodeGen/X86/horizontal-reduce-umax.ll
index 88f6b01131a2c8e32f528099284af229c037487c..2bc72158ff388a9e4937da09bb05fbcee7a8a77a 100644 (file)
--- a/test/CodeGen/X86/horizontal-reduce-umax.ll
+++ b/test/CodeGen/X86/horizontal-reduce-umax.ll
@@ -240,18 +240,18 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -260,8 +260,8 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -282,18 +282,18 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -302,8 +302,8 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -312,8 +312,8 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -321,8 +321,8 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ; 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>
@@ -358,13 +358,13 @@ define i8 @test_reduce_v16i8(<16 x i8> %a0) {
 ; 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
 ;
@@ -372,11 +372,11 @@ define i8 @test_reduce_v16i8(<16 x i8> %a0) {
 ; 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
 ;
@@ -399,13 +399,13 @@ define i8 @test_reduce_v16i8(<16 x i8> %a0) {
 ; 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
 ;
@@ -413,11 +413,11 @@ define i8 @test_reduce_v16i8(<16 x i8> %a0) {
 ; 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
 ;
@@ -425,11 +425,11 @@ define i8 @test_reduce_v16i8(<16 x i8> %a0) {
 ; 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
 ;
@@ -439,8 +439,8 @@ define i8 @test_reduce_v16i8(<16 x i8> %a0) {
 ; 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>
@@ -845,8 +845,8 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -854,10 +854,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -868,8 +868,8 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -881,8 +881,8 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -908,8 +908,8 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -917,10 +917,10 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -931,8 +931,8 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -944,8 +944,8 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -956,8 +956,8 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
@@ -999,13 +999,13 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -1015,11 +1015,11 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1030,11 +1030,11 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1060,13 +1060,13 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -1076,11 +1076,11 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1091,11 +1091,11 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1108,8 +1108,8 @@ define i8 @test_reduce_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1688,8 +1688,8 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1699,10 +1699,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1716,8 +1716,8 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1730,8 +1730,8 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1761,8 +1761,8 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1772,10 +1772,10 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1789,8 +1789,8 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1803,8 +1803,8 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1817,8 +1817,8 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1867,13 +1867,13 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -1886,11 +1886,11 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1902,11 +1902,11 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1936,13 +1936,13 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -1955,11 +1955,11 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1971,11 +1971,11 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1990,8 +1990,8 @@ define i8 @test_reduce_v64i8(<64 x i8> %a0) {
 ; 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

diff --git a/test/CodeGen/X86/horizontal-reduce-umin.ll b/test/CodeGen/X86/horizontal-reduce-umin.ll
index 482d08260374e4c94426b7a744961e6f32abb09f..e4b92239c23f982b8e600044b95080cd108635d7 100644 (file)
--- a/test/CodeGen/X86/horizontal-reduce-umin.ll
+++ b/test/CodeGen/X86/horizontal-reduce-umin.ll
@@ -242,8 +242,8 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -278,8 +278,8 @@ define i16 @test_reduce_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -785,8 +785,8 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -839,8 +839,8 @@ define i16 @test_reduce_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1592,8 +1592,8 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1656,8 +1656,8 @@ define i16 @test_reduce_v32i16(<32 x i16> %a0) {
 ; 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
 ;

diff --git a/test/CodeGen/X86/known-bits-vector.ll b/test/CodeGen/X86/known-bits-vector.ll
index cedc740e9eb8baa0173cf5ff7c6a95f2a0e86250..df31b223f09acbb76ddcf6d1d781b0435fe53fd7 100644 (file)
--- a/test/CodeGen/X86/known-bits-vector.ll
+++ b/test/CodeGen/X86/known-bits-vector.ll
@@ -5,18 +5,14 @@
 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
@@ -38,8 +34,8 @@ define float @knownbits_mask_extract_uitofp(<2 x i64> %a0) nounwind {
 ;
 ; 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>

diff --git a/test/CodeGen/X86/known-signbits-vector.ll b/test/CodeGen/X86/known-signbits-vector.ll
index 0e27c71362b8a76f5048a211b34d675ddb9ab445..a6d976ed7cd67efa878460c186d206789d04fa64 100644 (file)
--- a/test/CodeGen/X86/known-signbits-vector.ll
+++ b/test/CodeGen/X86/known-signbits-vector.ll
@@ -74,8 +74,8 @@ define float @signbits_ashr_extract_sitofp_0(<2 x i64> %a0) nounwind {
 ;
 ; 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>
@@ -101,12 +101,9 @@ define float @signbits_ashr_extract_sitofp_1(<2 x i64> %a0) nounwind {
 ;
 ; 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
@@ -132,13 +129,10 @@ define float @signbits_ashr_shl_extract_sitofp(<2 x i64> %a0) nounwind {
 ;
 ; 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>
@@ -168,10 +162,8 @@ define float @signbits_ashr_insert_ashr_extract_sitofp(i64 %a0, i64 %a1) nounwin
 ; 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

diff --git a/test/CodeGen/X86/pr30511.ll b/test/CodeGen/X86/pr30511.ll
index 69d0a94cc11caef8a35e2305e13f9ee0bb0d03bc..6764656e3337ceccb1d820683fee7cd9957de4f0 100644 (file)
--- a/test/CodeGen/X86/pr30511.ll
+++ b/test/CodeGen/X86/pr30511.ll
@@ -9,7 +9,7 @@ define i64 @PR30511(<2 x double> %a) {
 ; 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>

diff --git a/test/CodeGen/X86/setcc-combine.ll b/test/CodeGen/X86/setcc-combine.ll
index e8ebce9e56a489baebdf2203732bf3ac60fe0d62..9d2753cc29ae4bd65e293f229db859188cf2b014 100644 (file)
--- a/test/CodeGen/X86/setcc-combine.ll
+++ b/test/CodeGen/X86/setcc-combine.ll
@@ -6,18 +6,16 @@ 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>
@@ -65,18 +63,16 @@ 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>
@@ -124,18 +120,16 @@ 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>
@@ -170,18 +164,16 @@ 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>

diff --git a/test/CodeGen/X86/shrink_vmul-widen.ll b/test/CodeGen/X86/shrink_vmul-widen.ll
index 7a24743d5620b0602158d1f1c89d5a7f69f91ca5..d0fad23394bd48047c0e4d90065bcffeec560d62 100644 (file)
--- a/test/CodeGen/X86/shrink_vmul-widen.ll
+++ b/test/CodeGen/X86/shrink_vmul-widen.ll
@@ -2096,7 +2096,7 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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]
@@ -2110,10 +2110,10 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2122,7 +2122,7 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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]
@@ -2130,60 +2130,57 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
 ;
@@ -2252,14 +2249,13 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2326,12 +2322,11 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2345,8 +2340,8 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2355,10 +2350,10 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2367,7 +2362,7 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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]
@@ -2375,16 +2370,16 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2392,13 +2387,13 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2409,24 +2404,21 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2493,11 +2485,10 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2557,12 +2548,11 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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

diff --git a/test/CodeGen/X86/shrink_vmul.ll b/test/CodeGen/X86/shrink_vmul.ll
index b2fa8192abf46bbd8cabf91d7ea490386283ed21..8a8a396775ba1ac03b08fd45f48c76fff50f408d 100644 (file)
--- a/test/CodeGen/X86/shrink_vmul.ll
+++ b/test/CodeGen/X86/shrink_vmul.ll
@@ -2079,7 +2079,7 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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]
@@ -2093,10 +2093,10 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2105,7 +2105,7 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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]
@@ -2113,60 +2113,57 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
 ;
@@ -2235,14 +2232,13 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2309,12 +2305,11 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2328,8 +2323,8 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2338,10 +2333,10 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2350,7 +2345,7 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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]
@@ -2358,16 +2353,16 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2375,13 +2370,13 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2392,24 +2387,21 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2476,11 +2468,10 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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
@@ -2540,12 +2531,11 @@ define void @PR34947(<9 x i16>* %p0, <9 x i32>* %p1) nounwind {
 ; 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

diff --git a/test/CodeGen/X86/vector-fshl-128.ll b/test/CodeGen/X86/vector-fshl-128.ll
index eb3dadf0c27561859c17f907babf61a7aa6a34fe..fd93f447079216e0a887f0dc89a0c989be7a136a 100644 (file)
--- a/test/CodeGen/X86/vector-fshl-128.ll
+++ b/test/CodeGen/X86/vector-fshl-128.ll
@@ -1294,12 +1294,13 @@ define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %
 ; 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
@@ -1471,12 +1472,13 @@ define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %
 ; 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

diff --git a/test/CodeGen/X86/vector-fshl-rot-128.ll b/test/CodeGen/X86/vector-fshl-rot-128.ll
index c831f249c12255a1bcf5d713c3a99ec324754a55..2e12eeabffa0e16f18600317ba95bac03f56570a 100644 (file)
--- a/test/CodeGen/X86/vector-fshl-rot-128.ll
+++ b/test/CodeGen/X86/vector-fshl-rot-128.ll
@@ -783,18 +783,16 @@ define <2 x i64> @splatvar_funnnel_v2i64(<2 x i64> %x, <2 x i64> %amt) nounwind
 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:
@@ -882,18 +880,16 @@ define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %amt) nounwind
 ;
 ; 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)

diff --git a/test/CodeGen/X86/vector-fshr-128.ll b/test/CodeGen/X86/vector-fshr-128.ll
index 616a3780e346f031affdd11cf2430c18b2f1081d..f6b07318328ed57b06ec303d5d54697af2bc2e54 100644 (file)
--- a/test/CodeGen/X86/vector-fshr-128.ll
+++ b/test/CodeGen/X86/vector-fshr-128.ll
@@ -1309,12 +1309,13 @@ define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %
 ; 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
@@ -1485,12 +1486,13 @@ define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %y, <4 x i32> %
 ; 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

diff --git a/test/CodeGen/X86/vector-fshr-rot-128.ll b/test/CodeGen/X86/vector-fshr-rot-128.ll
index 2343f39668015b4dd7be0fefa1c227640f9665f1..fc915c2ea8f60f79e02c01ea837ec3c01c6b87a0 100644 (file)
--- a/test/CodeGen/X86/vector-fshr-rot-128.ll
+++ b/test/CodeGen/X86/vector-fshr-rot-128.ll
@@ -828,20 +828,17 @@ define <2 x i64> @splatvar_funnnel_v2i64(<2 x i64> %x, <2 x i64> %amt) nounwind
 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:
@@ -939,20 +936,17 @@ define <4 x i32> @splatvar_funnnel_v4i32(<4 x i32> %x, <4 x i32> %amt) nounwind
 ;
 ; 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)

diff --git a/test/CodeGen/X86/vector-gep.ll b/test/CodeGen/X86/vector-gep.ll
index 6b3e709e40ee89b9d004caa446701a27ac3847b4..8f62fe53825642e75bea361cb36cda4d4064cff1 100644 (file)
--- a/test/CodeGen/X86/vector-gep.ll
+++ b/test/CodeGen/X86/vector-gep.ll
@@ -20,9 +20,8 @@ define <4 x i32*> @AGEP0(i32* %ptr) nounwind {
 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

diff --git a/test/CodeGen/X86/vector-reduce-smax-widen.ll b/test/CodeGen/X86/vector-reduce-smax-widen.ll
index b31c2cf5dbb3ec0073ee0c8ce69f5851eae1a4d4..7be51758b1ca708153eac1a94c5eda85946148c7 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-smax-widen.ll
+++ b/test/CodeGen/X86/vector-reduce-smax-widen.ll
@@ -1120,31 +1120,28 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ;
 ; 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)
@@ -1169,11 +1166,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1181,11 +1177,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1194,11 +1189,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1207,11 +1201,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1241,11 +1234,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1256,11 +1248,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1270,11 +1261,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1285,11 +1275,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1327,11 +1316,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
 ;
@@ -1348,11 +1336,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1364,11 +1351,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1380,11 +1366,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1597,38 +1582,35 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ;
 ; 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)
@@ -1676,14 +1658,13 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -1691,13 +1672,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1706,13 +1686,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1721,13 +1700,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1788,14 +1766,13 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -1806,13 +1783,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1822,13 +1798,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1839,13 +1814,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1930,14 +1904,13 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
 ;
@@ -1954,13 +1927,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -1972,13 +1944,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -1990,13 +1961,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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

diff --git a/test/CodeGen/X86/vector-reduce-smax.ll b/test/CodeGen/X86/vector-reduce-smax.ll
index 01f453bd982d59413cb2dd904285e8be678a6e24..19ac789e269a66acb0663aa922c3cd1ec8bb55bd 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-smax.ll
+++ b/test/CodeGen/X86/vector-reduce-smax.ll
@@ -1340,31 +1340,28 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ;
 ; 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)
@@ -1389,11 +1386,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1401,11 +1397,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1414,11 +1409,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1427,11 +1421,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1461,11 +1454,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1476,11 +1468,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1490,11 +1481,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1505,11 +1495,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1547,11 +1536,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
 ;
@@ -1568,11 +1556,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1584,11 +1571,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1600,11 +1586,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1947,38 +1932,35 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ;
 ; 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)
@@ -2026,14 +2008,13 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -2041,13 +2022,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -2056,13 +2036,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -2071,13 +2050,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -2138,14 +2116,13 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -2156,13 +2133,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2172,13 +2148,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2189,13 +2164,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2280,14 +2254,13 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
 ;
@@ -2304,13 +2277,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -2322,13 +2294,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -2340,13 +2311,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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

diff --git a/test/CodeGen/X86/vector-reduce-smin-widen.ll b/test/CodeGen/X86/vector-reduce-smin-widen.ll
index 140df76e8ec2d9f2a37f7c54c228eb053f6fff75..dcc522ad2881ed35040825a50d8515578e691095 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-smin-widen.ll
+++ b/test/CodeGen/X86/vector-reduce-smin-widen.ll
@@ -1119,31 +1119,28 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ;
 ; 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)
@@ -1168,11 +1165,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1180,11 +1176,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1193,11 +1188,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1206,11 +1200,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1240,11 +1233,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1255,11 +1247,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1269,11 +1260,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1284,11 +1274,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1326,11 +1315,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
 ;
@@ -1347,11 +1335,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1363,11 +1350,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1379,11 +1365,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1596,38 +1581,35 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ;
 ; 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)
@@ -1675,14 +1657,13 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -1690,13 +1671,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1705,13 +1685,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1720,13 +1699,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1787,14 +1765,13 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -1805,13 +1782,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1821,13 +1797,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1838,13 +1813,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -1929,14 +1903,13 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
 ;
@@ -1953,13 +1926,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -1971,13 +1943,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -1989,13 +1960,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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

diff --git a/test/CodeGen/X86/vector-reduce-smin.ll b/test/CodeGen/X86/vector-reduce-smin.ll
index e8b6a242b8bf83387d3043ad21b273e2063e808b..0b09c944347b1aab4324639ab9d2ed8995e82b38 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-smin.ll
+++ b/test/CodeGen/X86/vector-reduce-smin.ll
@@ -1339,31 +1339,28 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ;
 ; 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)
@@ -1388,11 +1385,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1400,11 +1396,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1413,11 +1408,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1426,11 +1420,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1460,11 +1453,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1475,11 +1467,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1489,11 +1480,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1504,11 +1494,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1546,11 +1535,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
 ;
@@ -1567,11 +1555,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1583,11 +1570,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1599,11 +1585,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1946,38 +1931,35 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ;
 ; 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)
@@ -2025,14 +2007,13 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -2040,13 +2021,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -2055,13 +2035,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -2070,13 +2049,12 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -2137,14 +2115,13 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -2155,13 +2132,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2171,13 +2147,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2188,13 +2163,12 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2279,14 +2253,13 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
 ;
@@ -2303,13 +2276,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -2321,13 +2293,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -2339,13 +2310,12 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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

diff --git a/test/CodeGen/X86/vector-reduce-umax-widen.ll b/test/CodeGen/X86/vector-reduce-umax-widen.ll
index 3b6cdad368277c17d9bfb1d73dece02273942e4d..46c95994b5199679f3cbf8010ee42896ea7578f3 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-umax-widen.ll
+++ b/test/CodeGen/X86/vector-reduce-umax-widen.ll
@@ -1166,8 +1166,8 @@ define i16 @test_v2i16(<2 x i16> %a0) {
 ; 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
 ;
@@ -1212,8 +1212,8 @@ define i16 @test_v4i16(<4 x i16> %a0) {
 ; 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
 ;
@@ -1269,18 +1269,18 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -1289,8 +1289,8 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -1299,8 +1299,8 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ; 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
@@ -1309,8 +1309,8 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ; 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)
@@ -1339,8 +1339,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1348,10 +1348,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1362,8 +1362,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1375,8 +1375,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1387,8 +1387,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1399,8 +1399,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1434,8 +1434,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1445,10 +1445,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1462,8 +1462,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1476,8 +1476,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1490,8 +1490,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1504,8 +1504,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1547,8 +1547,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
 ;
@@ -1562,10 +1562,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
 ;
@@ -1585,8 +1585,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1601,8 +1601,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1616,8 +1616,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1631,8 +1631,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1806,13 +1806,13 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; 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
 ;
@@ -1820,11 +1820,11 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; 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
 ;
@@ -1835,8 +1835,8 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; 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
@@ -1847,8 +1847,8 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; 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)
@@ -1877,13 +1877,13 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -1893,11 +1893,11 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1908,11 +1908,11 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1925,8 +1925,8 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1939,8 +1939,8 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -1974,13 +1974,13 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -1993,11 +1993,11 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2009,11 +2009,11 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2028,8 +2028,8 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2044,8 +2044,8 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2087,13 +2087,13 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
 ;
@@ -2112,11 +2112,11 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -2130,11 +2130,11 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -2150,8 +2150,8 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -2167,8 +2167,8 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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

diff --git a/test/CodeGen/X86/vector-reduce-umax.ll b/test/CodeGen/X86/vector-reduce-umax.ll
index ffc98bb289d762543ff792ef1a6a020620cd2f8e..43cf9a9a662af157f25a5f83d31cec8f3bcdf271 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-umax.ll
+++ b/test/CodeGen/X86/vector-reduce-umax.ll
@@ -1389,18 +1389,18 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -1409,8 +1409,8 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -1419,8 +1419,8 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ; 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
@@ -1429,8 +1429,8 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ; 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)
@@ -1459,8 +1459,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1468,10 +1468,10 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1482,8 +1482,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1495,8 +1495,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1507,8 +1507,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1519,8 +1519,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
@@ -1554,8 +1554,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1565,10 +1565,10 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1582,8 +1582,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1596,8 +1596,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1610,8 +1610,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1624,8 +1624,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
@@ -1667,8 +1667,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
 ;
@@ -1682,10 +1682,10 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
 ;
@@ -1705,8 +1705,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1721,8 +1721,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1736,8 +1736,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -1751,8 +1751,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
@@ -2052,13 +2052,13 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; 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
 ;
@@ -2066,11 +2066,11 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; 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
 ;
@@ -2081,8 +2081,8 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; 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
@@ -2093,8 +2093,8 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; 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)
@@ -2123,13 +2123,13 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
 ;
@@ -2139,11 +2139,11 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -2154,11 +2154,11 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -2171,8 +2171,8 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -2185,8 +2185,8 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; 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
@@ -2220,13 +2220,13 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
 ;
@@ -2239,11 +2239,11 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2255,11 +2255,11 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2274,8 +2274,8 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2290,8 +2290,8 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; 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
@@ -2333,13 +2333,13 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
 ;
@@ -2358,11 +2358,11 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -2376,11 +2376,11 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -2396,8 +2396,8 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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
@@ -2413,8 +2413,8 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; 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

diff --git a/test/CodeGen/X86/vector-reduce-umin-widen.ll b/test/CodeGen/X86/vector-reduce-umin-widen.ll
index 51d2a25130d0a39cf0f69b797800d55317baef0c..9fac3bb10240dada3350cd3eb604b35010901b16 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-umin-widen.ll
+++ b/test/CodeGen/X86/vector-reduce-umin-widen.ll
@@ -1165,8 +1165,8 @@ define i16 @test_v2i16(<2 x i16> %a0) {
 ; 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
 ;
@@ -1211,8 +1211,8 @@ define i16 @test_v4i16(<4 x i16> %a0) {
 ; 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
 ;
@@ -1268,8 +1268,8 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -1319,8 +1319,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1391,8 +1391,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1479,8 +1479,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
 ;

diff --git a/test/CodeGen/X86/vector-reduce-umin.ll b/test/CodeGen/X86/vector-reduce-umin.ll
index a2116377563fb9a5f4007ae834e495f93c56c09f..ece96841a32134927b3fdca2087520c498a5c92d 100644 (file)
--- a/test/CodeGen/X86/vector-reduce-umin.ll
+++ b/test/CodeGen/X86/vector-reduce-umin.ll
@@ -1388,8 +1388,8 @@ define i16 @test_v8i16(<8 x i16> %a0) {
 ; 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
 ;
@@ -1439,8 +1439,8 @@ define i16 @test_v16i16(<16 x i16> %a0) {
 ; 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
 ;
@@ -1511,8 +1511,8 @@ define i16 @test_v32i16(<32 x i16> %a0) {
 ; 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
 ;
@@ -1599,8 +1599,8 @@ define i16 @test_v64i16(<64 x i16> %a0) {
 ; 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
 ;

diff --git a/test/CodeGen/X86/vector-rotate-128.ll b/test/CodeGen/X86/vector-rotate-128.ll
index 31d44126370ebd38220e1188fd15f20202c14099..4c528a55209832652e9d424fcd2335b3d2b6f1a8 100644 (file)
--- a/test/CodeGen/X86/vector-rotate-128.ll
+++ b/test/CodeGen/X86/vector-rotate-128.ll
@@ -738,18 +738,16 @@ define <2 x i64> @splatvar_rotate_v2i64(<2 x i64> %a, <2 x i64> %b) nounwind {
 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:
@@ -837,18 +835,16 @@ define <4 x i32> @splatvar_rotate_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
 ;
 ; 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

diff --git a/test/CodeGen/X86/xor.ll b/test/CodeGen/X86/xor.ll
index 65e5e83d7275dcf395d9ad2adb3f231c4b400972..f86b5812ace88917ec63292f16f8bc1b9ee94902 100644 (file)
--- a/test/CodeGen/X86/xor.ll
+++ b/test/CodeGen/X86/xor.ll
@@ -420,26 +420,14 @@ define i32 @PR17487(i1 %tobool) {
 ; 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>