[SelectionDAG] move splat util functions up from x86 lowering

This was supposed to be NFC, but the change in SDLoc
definitions causes instruction scheduling changes.

There's nothing x86-specific in this code, and it can
likely be used from DAGCombiner's simplifyVBinOp().

llvm-svn: 358930

9 files changed, 101 insertions, 93 deletions

diff --git a/llvm/include/llvm/CodeGen/SelectionDAG.h b/llvm/include/llvm/CodeGen/SelectionDAG.h
index aac3e0f..f272c59 100644
--- a/llvm/include/llvm/CodeGen/SelectionDAG.h
+++ b/llvm/include/llvm/CodeGen/SelectionDAG.h
@@ -1539,6 +1539,13 @@ public:
   /// Test whether \p V has a splatted value.
   bool isSplatValue(SDValue V, bool AllowUndefs = false);
 
+  /// If V is a splatted value, return the source vector and its splat index.
+  SDValue getSplatSourceVector(SDValue V, int &SplatIndex);
+
+  /// If V is a splat vector, return its scalar source operand by extracting
+  /// that element from the source vector.
+  SDValue getSplatValue(SDValue V);
+
   /// Match a binop + shuffle pyramid that represents a horizontal reduction
   /// over the elements of a vector starting from the EXTRACT_VECTOR_ELT node /p
   /// Extract. The reduction must use one of the opcodes listed in /p
diff --git a/llvm/include/llvm/CodeGen/SelectionDAGNodes.h b/llvm/include/llvm/CodeGen/SelectionDAGNodes.h
index d9f930d..67fc3e9 100644
--- a/llvm/include/llvm/CodeGen/SelectionDAGNodes.h
+++ b/llvm/include/llvm/CodeGen/SelectionDAGNodes.h
@@ -1641,6 +1641,10 @@ SDValue peekThroughBitcasts(SDValue V);
 /// If \p V is not a bitcasted one-use value, it is returned as-is.
 SDValue peekThroughOneUseBitcasts(SDValue V);
 
+/// Return the non-extracted vector source operand of \p V if it exists.
+/// If \p V is not an extracted subvector, it is returned as-is.
+SDValue peekThroughExtractSubvectors(SDValue V);
+
 /// Returns true if \p V is a bitwise not operation. Assumes that an all ones
 /// constant is canonicalized to be operand 1.
 bool isBitwiseNot(SDValue V);
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index b2de2a4..4b21b96 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -2300,6 +2300,52 @@ bool SelectionDAG::isSplatValue(SDValue V, bool AllowUndefs) {
          (AllowUndefs || !UndefElts);
 }
 
+SDValue SelectionDAG::getSplatSourceVector(SDValue V, int &SplatIdx) {
+  V = peekThroughExtractSubvectors(V);
+
+  EVT VT = V.getValueType();
+  unsigned Opcode = V.getOpcode();
+  switch (Opcode) {
+  default: {
+    APInt UndefElts;
+    APInt DemandedElts = APInt::getAllOnesValue(VT.getVectorNumElements());
+    if (isSplatValue(V, DemandedElts, UndefElts)) {
+      // Handle case where all demanded elements are UNDEF.
+      if (DemandedElts.isSubsetOf(UndefElts)) {
+        SplatIdx = 0;
+        return getUNDEF(VT);
+      }
+      SplatIdx = (UndefElts & DemandedElts).countTrailingOnes();
+      return V;
+    }
+    break;
+  }
+  case ISD::VECTOR_SHUFFLE: {
+    // Check if this is a shuffle node doing a splat.
+    // TODO - remove this and rely purely on SelectionDAG::isSplatValue,
+    // getTargetVShiftNode currently struggles without the splat source.
+    auto *SVN = cast<ShuffleVectorSDNode>(V);
+    if (!SVN->isSplat())
+      break;
+    int Idx = SVN->getSplatIndex();
+    int NumElts = V.getValueType().getVectorNumElements();
+    SplatIdx = Idx % NumElts;
+    return V.getOperand(Idx / NumElts);
+  }
+  }
+
+  return SDValue();
+}
+
+SDValue SelectionDAG::getSplatValue(SDValue V) {
+  int SplatIdx;
+  if (SDValue SrcVector = getSplatSourceVector(V, SplatIdx))
+    return getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(V),
+                   SrcVector.getValueType().getScalarType(), SrcVector,
+                   getIntPtrConstant(SplatIdx, SDLoc(V)));
+  return SDValue();
+}
+
 /// If a SHL/SRA/SRL node has a constant or splat constant shift amount that
 /// is less than the element bit-width of the shift node, return it.
 static const APInt *getValidShiftAmountConstant(SDValue V) {
@@ -8585,6 +8631,12 @@ SDValue llvm::peekThroughOneUseBitcasts(SDValue V) {
   return V;
 }
 
+SDValue llvm::peekThroughExtractSubvectors(SDValue V) {
+  while (V.getOpcode() == ISD::EXTRACT_SUBVECTOR)
+    V = V.getOperand(0);
+  return V;
+}
+
 bool llvm::isBitwiseNot(SDValue V) {
   if (V.getOpcode() != ISD::XOR)
     return false;
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 2534948..f13e114 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -5711,13 +5711,6 @@ static SDValue getShuffleVectorZeroOrUndef(SDValue V2, int Idx,
   return DAG.getVectorShuffle(VT, SDLoc(V2), V1, V2, MaskVec);
 }
 
-// Peek through EXTRACT_SUBVECTORs - typically used for AVX1 256-bit intops.
-static SDValue peekThroughEXTRACT_SUBVECTORs(SDValue V) {
-  while (V.getOpcode() == ISD::EXTRACT_SUBVECTOR)
-    V = V.getOperand(0);
-  return V;
-}
-
 static const Constant *getTargetConstantFromNode(SDValue Op) {
   Op = peekThroughBitcasts(Op);
 
@@ -24722,54 +24715,6 @@ static SDValue LowerScalarImmediateShift(SDValue Op, SelectionDAG &DAG,
   return SDValue();
 }
 
-// If V is a splat value, return the source vector and splat index;
-static SDValue IsSplatVector(SDValue V, int &SplatIdx, SelectionDAG &DAG) {
-  V = peekThroughEXTRACT_SUBVECTORs(V);
-
-  EVT VT = V.getValueType();
-  unsigned Opcode = V.getOpcode();
-  switch (Opcode) {
-  default: {
-    APInt UndefElts;
-    APInt DemandedElts = APInt::getAllOnesValue(VT.getVectorNumElements());
-    if (DAG.isSplatValue(V, DemandedElts, UndefElts)) {
-      // Handle case where all demanded elements are UNDEF.
-      if (DemandedElts.isSubsetOf(UndefElts)) {
-        SplatIdx = 0;
-        return DAG.getUNDEF(VT);
-      }
-      SplatIdx = (UndefElts & DemandedElts).countTrailingOnes();
-      return V;
-    }
-    break;
-  }
-  case ISD::VECTOR_SHUFFLE: {
-    // Check if this is a shuffle node doing a splat.
-    // TODO - remove this and rely purely on SelectionDAG::isSplatValue,
-    // getTargetVShiftNode currently struggles without the splat source.
-    auto *SVN = cast<ShuffleVectorSDNode>(V);
-    if (!SVN->isSplat())
-      break;
-    int Idx = SVN->getSplatIndex();
-    int NumElts = V.getValueType().getVectorNumElements();
-    SplatIdx = Idx % NumElts;
-    return V.getOperand(Idx / NumElts);
-  }
-  }
-
-  return SDValue();
-}
-
-static SDValue GetSplatValue(SDValue V, const SDLoc &dl,
-                             SelectionDAG &DAG) {
-  int SplatIdx;
-  if (SDValue SrcVector = IsSplatVector(V, SplatIdx, DAG))
-    return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl,
-                       SrcVector.getValueType().getScalarType(), SrcVector,
-                       DAG.getIntPtrConstant(SplatIdx, dl));
-  return SDValue();
-}
-
 static SDValue LowerScalarVariableShift(SDValue Op, SelectionDAG &DAG,
                                         const X86Subtarget &Subtarget) {
   MVT VT = Op.getSimpleValueType();
@@ -24780,7 +24725,7 @@ static SDValue LowerScalarVariableShift(SDValue Op, SelectionDAG &DAG,
   unsigned X86OpcI = getTargetVShiftUniformOpcode(Opcode, false);
   unsigned X86OpcV = getTargetVShiftUniformOpcode(Opcode, true);
 
-  if (SDValue BaseShAmt = GetSplatValue(Amt, dl, DAG)) {
+  if (SDValue BaseShAmt = DAG.getSplatValue(Amt)) {
     if (SupportedVectorShiftWithBaseAmnt(VT, Subtarget, Opcode)) {
       MVT EltVT = VT.getVectorElementType();
       assert(EltVT.bitsLE(MVT::i64) && "Unexpected element type!");
diff --git a/llvm/test/CodeGen/X86/vector-rotate-256.ll b/llvm/test/CodeGen/X86/vector-rotate-256.ll
index 3746c8e..ac50c36 100644
--- a/llvm/test/CodeGen/X86/vector-rotate-256.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-256.ll
@@ -527,11 +527,11 @@ define <4 x i64> @splatvar_rotate_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
 ;
 ; AVX2-LABEL: splatvar_rotate_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsllq %xmm1, %ymm0, %ymm2
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [64,64]
-; AVX2-NEXT:    vpsubq %xmm1, %xmm3, %xmm1
-; AVX2-NEXT:    vpsrlq %xmm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [64,64]
+; AVX2-NEXT:    vpsubq %xmm1, %xmm2, %xmm2
+; AVX2-NEXT:    vpsllq %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vpsrlq %xmm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: splatvar_rotate_v4i64:
diff --git a/llvm/test/CodeGen/X86/vector-rotate-512.ll b/llvm/test/CodeGen/X86/vector-rotate-512.ll
index f1118b1..01b4676 100644
--- a/llvm/test/CodeGen/X86/vector-rotate-512.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-512.ll
@@ -344,10 +344,10 @@ define <32 x i16> @splatvar_rotate_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind
 ; AVX512BW-LABEL: splatvar_rotate_v32i16:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
-; AVX512BW-NEXT:    vpsllw %xmm2, %zmm0, %zmm2
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX512BW-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX512BW-NEXT:    vpsllw %xmm2, %zmm0, %zmm2
 ; AVX512BW-NEXT:    vpsrlw %xmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512BW-NEXT:    retq
@@ -355,10 +355,10 @@ define <32 x i16> @splatvar_rotate_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind
 ; AVX512VLBW-LABEL: splatvar_rotate_v32i16:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
-; AVX512VLBW-NEXT:    vpsllw %xmm2, %zmm0, %zmm2
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16]
 ; AVX512VLBW-NEXT:    vpsubw %xmm1, %xmm3, %xmm1
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX512VLBW-NEXT:    vpsllw %xmm2, %zmm0, %zmm2
 ; AVX512VLBW-NEXT:    vpsrlw %xmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512VLBW-NEXT:    retq
@@ -426,14 +426,14 @@ define <64 x i8> @splatvar_rotate_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
 ; AVX512BW-LABEL: splatvar_rotate_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512BW-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
+; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512BW-NEXT:    vpsllw %xmm2, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4
 ; AVX512BW-NEXT:    vpsllw %xmm2, %zmm4, %zmm2
 ; AVX512BW-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512BW-NEXT:    vpandq %zmm2, %zmm3, %zmm2
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512BW-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
-; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512BW-NEXT:    vpsrlw %xmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw %xmm1, %zmm4, %zmm1
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
@@ -445,14 +445,14 @@ define <64 x i8> @splatvar_rotate_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
 ; AVX512VLBW-LABEL: splatvar_rotate_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
+; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
+; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsllw %xmm2, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpternlogd $255, %zmm4, %zmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpsllw %xmm2, %zmm4, %zmm2
 ; AVX512VLBW-NEXT:    vpbroadcastb %xmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpandq %zmm2, %zmm3, %zmm2
-; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8]
-; AVX512VLBW-NEXT:    vpsubb %xmm1, %xmm3, %xmm1
-; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VLBW-NEXT:    vpsrlw %xmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpsrlw %xmm1, %zmm4, %zmm1
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
diff --git a/llvm/test/CodeGen/X86/vector-shift-ashr-256.ll b/llvm/test/CodeGen/X86/vector-shift-ashr-256.ll
index e895ac9..fdf7f4a 100644
--- a/llvm/test/CodeGen/X86/vector-shift-ashr-256.ll
+++ b/llvm/test/CodeGen/X86/vector-shift-ashr-256.ll
@@ -737,8 +737,8 @@ define <4 x i64> @splatvar_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
 define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ; AVX1-LABEL: splatvar_shift_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsrad %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrad %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -752,8 +752,8 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpsrad %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsrad %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -779,8 +779,8 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ;
 ; X32-AVX1-LABEL: splatvar_shift_v8i32:
 ; X32-AVX1:       # %bb.0:
-; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpsrad %xmm1, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vpsrad %xmm1, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -799,8 +799,8 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind {
 ; AVX1-LABEL: splatvar_shift_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsraw %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsraw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -814,8 +814,8 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpsraw %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsraw %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -841,8 +841,8 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 ;
 ; X32-AVX1-LABEL: splatvar_shift_v16i16:
 ; X32-AVX1:       # %bb.0:
-; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpsraw %xmm1, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vpsraw %xmm1, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -861,8 +861,8 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ; AVX1-LABEL: splatvar_shift_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm3, %xmm3
@@ -968,8 +968,8 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ;
 ; X32-AVX1-LABEL: splatvar_shift_v32i8:
 ; X32-AVX1:       # %bb.0:
-; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpsrlw %xmm1, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
 ; X32-AVX1-NEXT:    vpsrlw %xmm1, %xmm3, %xmm3
diff --git a/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll b/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll
index 3212c78..ae087ba 100644
--- a/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll
+++ b/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll
@@ -580,8 +580,8 @@ define <4 x i64> @splatvar_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
 define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ; AVX1-LABEL: splatvar_shift_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -595,8 +595,8 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpsrld %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -622,8 +622,8 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ;
 ; X32-AVX1-LABEL: splatvar_shift_v8i32:
 ; X32-AVX1:       # %bb.0:
-; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpsrld %xmm1, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -642,8 +642,8 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind {
 ; AVX1-LABEL: splatvar_shift_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -657,8 +657,8 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpsrlw %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -684,8 +684,8 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 ;
 ; X32-AVX1-LABEL: splatvar_shift_v16i16:
 ; X32-AVX1:       # %bb.0:
-; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpsrlw %xmm1, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vpsrlw %xmm1, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -704,8 +704,8 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ; AVX1-LABEL: splatvar_shift_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsrlw %xmm1, %xmm3, %xmm3
@@ -793,8 +793,8 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ;
 ; X32-AVX1-LABEL: splatvar_shift_v32i8:
 ; X32-AVX1:       # %bb.0:
-; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpsrlw %xmm1, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
 ; X32-AVX1-NEXT:    vpsrlw %xmm1, %xmm3, %xmm3
diff --git a/llvm/test/CodeGen/X86/vector-shift-shl-256.ll b/llvm/test/CodeGen/X86/vector-shift-shl-256.ll
index dcf6630..bcd24b1 100644
--- a/llvm/test/CodeGen/X86/vector-shift-shl-256.ll
+++ b/llvm/test/CodeGen/X86/vector-shift-shl-256.ll
@@ -512,8 +512,8 @@ define <4 x i64> @splatvar_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind {
 define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ; AVX1-LABEL: splatvar_shift_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpslld %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpslld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -527,8 +527,8 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v8i32:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpslld %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpslld %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -554,8 +554,8 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ;
 ; X32-AVX1-LABEL: splatvar_shift_v8i32:
 ; X32-AVX1:       # %bb.0:
-; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpslld %xmm1, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vpslld %xmm1, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -574,8 +574,8 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind {
 ; AVX1-LABEL: splatvar_shift_v16i16:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsllw %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsllw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -589,8 +589,8 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 ;
 ; XOPAVX1-LABEL: splatvar_shift_v16i16:
 ; XOPAVX1:       # %bb.0:
-; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; XOPAVX1-NEXT:    vpsllw %xmm1, %xmm2, %xmm2
 ; XOPAVX1-NEXT:    vpsllw %xmm1, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -616,8 +616,8 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 ;
 ; X32-AVX1-LABEL: splatvar_shift_v16i16:
 ; X32-AVX1:       # %bb.0:
-; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpsllw %xmm1, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vpsllw %xmm1, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
@@ -636,8 +636,8 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ; AVX1-LABEL: splatvar_shift_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpsllw %xmm1, %xmm2, %xmm2
 ; AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsllw %xmm1, %xmm3, %xmm3
@@ -719,8 +719,8 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ;
 ; X32-AVX1-LABEL: splatvar_shift_v32i8:
 ; X32-AVX1:       # %bb.0:
-; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; X32-AVX1-NEXT:    vpsllw %xmm1, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
 ; X32-AVX1-NEXT:    vpsllw %xmm1, %xmm3, %xmm3