[X86][SSE] Support v4i32 rotations (PR37426)

[RKSimon]

Summary

As suggested by Fabian on PR37426, we can use PMULUDQ to perform v4i32 vector rotations as the upper 32bits of the multiply will contain the 'wrapped' bits of the rotation.

v8i16/v16i8 rotations would be straightforward to add to lowerRotate in the future - ideally we'd mostly share code with the vector shifts lowering.

[RKSimon]

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

1599

It might be better to take the cost of 2 loads to remove the 2xPSHUFD and fold directly into the PMULUDQs? There is a small increase in codesize.

[xbolva00]

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

1599

+1 for PMULUDQ

[craig.topper]

lib/Target/X86/X86ISelLowering.cpp

23739

Combine these two ifs into one condition?

[xbolva00]

lib/Target/X86/X86ISelLowering.cpp

23739

Yes

[RKSimon]

Merged ifs()

[craig.topper]

LGTM

lib/Target/X86/X86ISelLowering.cpp

if (!Subtarget.useSoftFloat() && Subtarget.hasSSE2()) {
 
     // In the customized shift lowering, the legal v4i32/v2i64 cases
     // in AVX2 will be recognized.
     for (auto VT : { MVT::v16i8, MVT::v8i16, MVT::v4i32, MVT::v2i64 }) {
       setOperationAction(ISD::SRL,              VT, Custom);
       setOperationAction(ISD::SHL,              VT, Custom);
       setOperationAction(ISD::SRA,              VT, Custom);
     }
+
+    setOperationAction(ISD::ROTL,               MVT::v4i32, Custom);
   }
 
   if (!Subtarget.useSoftFloat() && Subtarget.hasSSSE3()) {
     setOperationAction(ISD::ABS,                MVT::v16i8, Legal);
     setOperationAction(ISD::ABS,                MVT::v8i16, Legal);
     setOperationAction(ISD::ABS,                MVT::v4i32, Legal);
     setOperationAction(ISD::BITREVERSE,         MVT::v16i8, Custom);
     setOperationAction(ISD::CTLZ,               MVT::v16i8, Custom);
if (!Subtarget.useSoftFloat() && Subtarget.hasAVX()) {
     // In the customized shift lowering, the legal v8i32/v4i64 cases
     // in AVX2 will be recognized.
     for (auto VT : { MVT::v32i8, MVT::v16i16, MVT::v8i32, MVT::v4i64 }) {
       setOperationAction(ISD::SRL, VT, Custom);
       setOperationAction(ISD::SHL, VT, Custom);
       setOperationAction(ISD::SRA, VT, Custom);
     }
 
+    setOperationAction(ISD::ROTL,              MVT::v8i32, Custom);
+
     setOperationAction(ISD::SELECT,            MVT::v4f64, Custom);
     setOperationAction(ISD::SELECT,            MVT::v4i64, Custom);
     setOperationAction(ISD::SELECT,            MVT::v8f32, Custom);
 
     for (auto VT : { MVT::v16i16, MVT::v8i32, MVT::v4i64 }) {
       setOperationAction(ISD::SIGN_EXTEND,     VT, Custom);
       setOperationAction(ISD::ZERO_EXTEND,     VT, Custom);
       setOperationAction(ISD::ANY_EXTEND,      VT, Custom);
if (VT == MVT::v2i64  && Amt.getOpcode() == ISD::BITCAST &&
     }
 
     if (SupportedVectorShiftWithBaseAmnt(VT, Subtarget, Op.getOpcode()))
       return DAG.getNode(X86OpcV, dl, VT, R, Op.getOperand(1));
   }
   return SDValue();
 }
 
+// Convert a shift/rotate left amount to a multiplication scale factor.
+static SDValue convertShiftLeftToScale(SDValue Amt, const SDLoc &dl,
+                                       const X86Subtarget &Subtarget,
+                                       SelectionDAG &DAG) {
+  MVT VT = Amt.getSimpleValueType();
+  bool ConstantAmt = ISD::isBuildVectorOfConstantSDNodes(Amt.getNode());
+
+  if (ConstantAmt && (VT == MVT::v8i16 || VT == MVT::v4i32 ||
+                      (Subtarget.hasInt256() && VT == MVT::v16i16))) {
+    SmallVector<SDValue, 8> Elts;
+    MVT SVT = VT.getVectorElementType();
+    unsigned SVTBits = SVT.getSizeInBits();
+    APInt One(SVTBits, 1);
+    unsigned NumElems = VT.getVectorNumElements();
+
+    for (unsigned i = 0; i != NumElems; ++i) {
+      SDValue Op = Amt->getOperand(i);
+      if (Op->isUndef()) {
+        Elts.push_back(Op);
+        continue;
+      }
+
+      ConstantSDNode *ND = cast<ConstantSDNode>(Op);
+      APInt C(SVTBits, ND->getAPIntValue().getZExtValue());
+      uint64_t ShAmt = C.getZExtValue();
+      if (ShAmt >= SVTBits) {
+        Elts.push_back(DAG.getUNDEF(SVT));
+        continue;
+      }
+      Elts.push_back(DAG.getConstant(One.shl(ShAmt), dl, SVT));
+    }
+    return DAG.getBuildVector(VT, dl, Elts);
+  }
+
+  if (VT == MVT::v4i32) {
+    Amt = DAG.getNode(ISD::SHL, dl, VT, Amt, DAG.getConstant(23, dl, VT));
+    Amt = DAG.getNode(ISD::ADD, dl, VT, Amt,
+                      DAG.getConstant(0x3f800000U, dl, VT));
+    Amt = DAG.getBitcast(MVT::v4f32, Amt);
+    return DAG.getNode(ISD::FP_TO_SINT, dl, VT, Amt);
+  }
+
+  return SDValue();
+}
+
 static SDValue LowerShift(SDValue Op, const X86Subtarget &Subtarget,
                           SelectionDAG &DAG) {
   MVT VT = Op.getSimpleValueType();
   SDLoc dl(Op);
   SDValue R = Op.getOperand(0);
   SDValue Amt = Op.getOperand(1);
   bool ConstantAmt = ISD::isBuildVectorOfConstantSDNodes(Amt.getNode());
 
if ((VT == MVT::v2i64 || (VT == MVT::v4i64 && Subtarget.hasInt256())) &&
     R = DAG.getNode(ISD::SRL, dl, VT, R, Amt);
     R = DAG.getNode(ISD::XOR, dl, VT, R, M);
     R = DAG.getNode(ISD::SUB, dl, VT, R, M);
     return R;
   }
 
   // If possible, lower this packed shift into a vector multiply instead of
   // expanding it into a sequence of scalar shifts.
-  // Do this only if the vector shift count is a constant build_vector.
-  if (ConstantAmt && Op.getOpcode() == ISD::SHL &&
-      (VT == MVT::v8i16 || VT == MVT::v4i32 ||
-       (Subtarget.hasInt256() && VT == MVT::v16i16))) {
-    SmallVector<SDValue, 8> Elts;
-    MVT SVT = VT.getVectorElementType();
-    unsigned SVTBits = SVT.getSizeInBits();
-    APInt One(SVTBits, 1);
-    unsigned NumElems = VT.getVectorNumElements();
-
-    for (unsigned i=0; i !=NumElems; ++i) {
-      SDValue Op = Amt->getOperand(i);
-      if (Op->isUndef()) {
-        Elts.push_back(Op);
-        continue;
-      }
-
-      ConstantSDNode *ND = cast<ConstantSDNode>(Op);
-      APInt C(SVTBits, ND->getAPIntValue().getZExtValue());
-      uint64_t ShAmt = C.getZExtValue();
-      if (ShAmt >= SVTBits) {
-        Elts.push_back(DAG.getUNDEF(SVT));
-        continue;
-      }
-      Elts.push_back(DAG.getConstant(One.shl(ShAmt), dl, SVT));
-    }
-    SDValue BV = DAG.getBuildVector(VT, dl, Elts);
-    return DAG.getNode(ISD::MUL, dl, VT, R, BV);
-  }
-
-  // Lower SHL with variable shift amount.
-  if (VT == MVT::v4i32 && Op->getOpcode() == ISD::SHL) {
-    Op = DAG.getNode(ISD::SHL, dl, VT, Amt, DAG.getConstant(23, dl, VT));
-
-    Op = DAG.getNode(ISD::ADD, dl, VT, Op,
-                     DAG.getConstant(0x3f800000U, dl, VT));
-    Op = DAG.getBitcast(MVT::v4f32, Op);
-    Op = DAG.getNode(ISD::FP_TO_SINT, dl, VT, Op);
-    return DAG.getNode(ISD::MUL, dl, VT, Op, R);
-  }
+  if (Op.getOpcode() == ISD::SHL)
+    if (SDValue Scale = convertShiftLeftToScale(Amt, dl, Subtarget, DAG))
+      return DAG.getNode(ISD::MUL, dl, VT, R, Scale);
 
   // If possible, lower this shift as a sequence of two shifts by
   // constant plus a MOVSS/MOVSD/PBLEND instead of scalarizing it.
   // Example:
   //   (v4i32 (srl A, (build_vector < X, Y, Y, Y>)))
   //
   // Could be rewritten as:
   //   (v4i32 (MOVSS (srl A, <Y,Y,Y,Y>), (srl A, <X,X,X,X>)))
if (VT.is256BitVector())
     return Lower256IntArith(Op, DAG);
 
   return SDValue();
 }
 
 static SDValue LowerRotate(SDValue Op, const X86Subtarget &Subtarget,
                            SelectionDAG &DAG) {
   MVT VT = Op.getSimpleValueType();
+  assert(VT.isVector() && "Custom lowering only for vector rotates!");
+
   SDLoc DL(Op);
   SDValue R = Op.getOperand(0);
   SDValue Amt = Op.getOperand(1);
   unsigned Opcode = Op.getOpcode();
   unsigned EltSizeInBits = VT.getScalarSizeInBits();
 
   if (Subtarget.hasAVX512()) {
     // Attempt to rotate by immediate.
if (getTargetConstantBitsFromNode(Amt, EltSizeInBits, UndefElts, EltBits)) {
                            DAG.getConstant(RotateAmt, DL, MVT::i8));
       }
     }
 
     // Else, fall-back on VPROLV/VPRORV.
     return Op;
   }
 
-  assert(VT.isVector() && "Custom lowering only for vector rotates!");
-  assert(Subtarget.hasXOP() && "XOP support required for vector rotates!");
   assert((Opcode == ISD::ROTL) && "Only ROTL supported");
 
   // XOP has 128-bit vector variable + immediate rotates.
   // +ve/-ve Amt = rotate left/right - just need to handle ISD::ROTL.
-
+  if (Subtarget.hasXOP()) {
     // Split 256-bit integers.
     if (VT.is256BitVector())
       return Lower256IntArith(Op, DAG);
-
     assert(VT.is128BitVector() && "Only rotate 128-bit vectors!");
 
     // Attempt to rotate by immediate.
     if (auto *BVAmt = dyn_cast<BuildVectorSDNode>(Amt)) {
       if (auto *RotateConst = BVAmt->getConstantSplatNode()) {
         uint64_t RotateAmt = RotateConst->getAPIntValue().getZExtValue();
         assert(RotateAmt < EltSizeInBits && "Rotation out of range");
         return DAG.getNode(X86ISD::VROTLI, DL, VT, R,
                            DAG.getConstant(RotateAmt, DL, MVT::i8));
       }
     }
 
     // Use general rotate by variable (per-element).
     return Op;
   }
 
+  // Split 256-bit integers on pre-AVX2 targets.
+  if (!Subtarget.hasAVX2())
+    if (VT.is256BitVector())
+      return Lower256IntArith(Op, DAG);
+
+  assert((VT == MVT::v4i32 || (VT == MVT::v8i32 && Subtarget.hasAVX2())) &&
+         "Only v4i32/v8i32 vector rotates supported");
+
+  // Rotate by an uniform constant - expand back to shifts.
+  // TODO - legalizers should be able to handle this.
+  if (auto *BVAmt = dyn_cast<BuildVectorSDNode>(Amt)) {
+    if (auto *RotateConst = BVAmt->getConstantSplatNode()) {
+      uint64_t RotateAmt = RotateConst->getAPIntValue().getZExtValue();
+      assert(RotateAmt < EltSizeInBits && "Rotation out of range");
+      if (RotateAmt == 0)
+        return R;
+
+      SDValue SHL = getTargetVShiftByConstNode(X86ISD::VSHLI, DL, VT, R,
+                                               RotateAmt, DAG);
+      SDValue SRL = getTargetVShiftByConstNode(X86ISD::VSRLI, DL, VT, R,
+                                               EltSizeInBits - RotateAmt, DAG);
+      return DAG.getNode(ISD::OR, DL, VT, SHL, SRL);
+    }
+  }
+
+  // AVX2 - best to fallback to variable shifts.
+  // TODO - legalizers should be able to handle this.
+  if (Subtarget.hasAVX2()) {
+    SDValue AmtR = DAG.getConstant(EltSizeInBits, DL, VT);
+    AmtR = DAG.getNode(ISD::SUB, DL, VT, AmtR, Amt);
+    SDValue SHL = DAG.getNode(ISD::SHL, DL, VT, R, Amt);
+    SDValue SRL = DAG.getNode(ISD::SRL, DL, VT, R, AmtR);
+    return DAG.getNode(ISD::OR, DL, VT, SHL, SRL);
+  }
+
+  // As with shifts, convert the rotation amount to a multiplication factor,
+  // and make use of the PMULUDQ instruction to multiply 2 lanes of v4i32
+  // to v2i64 results at a time. The upper 32-bits contain the wrapped bits
+  // that can then be OR'd with the lower 32-bits.
+  Amt = convertShiftLeftToScale(Amt, DL, Subtarget, DAG);
+
+  static const int OddMask[] = {1, -1, 3, -1};
+  SDValue R13 = DAG.getVectorShuffle(VT, DL, R, R, OddMask);
+  SDValue Amt13 = DAG.getVectorShuffle(VT, DL, Amt, Amt, OddMask);
+
+  SDValue Res02 = DAG.getNode(X86ISD::PMULUDQ, DL, MVT::v2i64,
+                              DAG.getBitcast(MVT::v2i64, R),
+                              DAG.getBitcast(MVT::v2i64, Amt));
+  SDValue Res13 = DAG.getNode(X86ISD::PMULUDQ, DL, MVT::v2i64,
+                              DAG.getBitcast(MVT::v2i64, R13),
+                              DAG.getBitcast(MVT::v2i64, Amt13));
+  Res02 = DAG.getBitcast(VT, Res02);
+  Res13 = DAG.getBitcast(VT, Res13);
+
+  return DAG.getNode(ISD::OR, DL, VT,
+                     DAG.getVectorShuffle(VT, DL, Res02, Res13, {0, 4, 2, 6}),
+                     DAG.getVectorShuffle(VT, DL, Res02, Res13, {1, 5, 3, 7}));
+}
+
 static SDValue LowerXALUO(SDValue Op, SelectionDAG &DAG) {
   // Lower the "add/sub/mul with overflow" instruction into a regular ins plus
   // a "setcc" instruction that checks the overflow flag. The "brcond" lowering
   // looks for this combo and may remove the "setcc" instruction if the "setcc"
   // has only one use.
   SDNode *N = Op.getNode();
   SDValue LHS = N->getOperand(0);
   SDValue RHS = N->getOperand(1);

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

; X32-SSE-NEXT:    retl
   %lshr = lshr <2 x i64> %a, %b64
   %or = or <2 x i64> %shl, %lshr
   ret <2 x i64> %or
 }
 
 define <4 x i32> @var_rotate_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; SSE2-LABEL: var_rotate_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [32,32,32,32]
-; SSE2-NEXT:    psubd %xmm1, %xmm2
 ; SSE2-NEXT:    pslld $23, %xmm1
 ; SSE2-NEXT:    paddd {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
-; SSE2-NEXT:    pmuludq %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[1,1,3,3]
-; SSE2-NEXT:    pmuludq %xmm3, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    psrlq $32, %xmm3
-; SSE2-NEXT:    movdqa %xmm0, %xmm4
-; SSE2-NEXT:    psrld %xmm3, %xmm4
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm0, %xmm5
-; SSE2-NEXT:    psrld %xmm3, %xmm5
-; SSE2-NEXT:    movsd {{.*#+}} xmm5 = xmm4[0],xmm5[1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[1,3,2,3]
-; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    movdqa %xmm2, %xmm5
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1]
-; SSE2-NEXT:    movdqa %xmm0, %xmm6
-; SSE2-NEXT:    psrld %xmm5, %xmm6
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
-; SSE2-NEXT:    psrld %xmm2, %xmm0
-; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm6[0],xmm0[1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,3,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm2, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,2,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
-; SSE2-NEXT:    por %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_rotate_v4i32:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [32,32,32,32]
-; SSE41-NEXT:    psubd %xmm1, %xmm2
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
 ; SSE41-NEXT:    pslld $23, %xmm1
 ; SSE41-NEXT:    paddd {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    cvttps2dq %xmm1, %xmm1
-; SSE41-NEXT:    pmulld %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm2, %xmm3
-; SSE41-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    movdqa %xmm0, %xmm4
-; SSE41-NEXT:    psrld %xmm3, %xmm4
-; SSE41-NEXT:    movdqa %xmm2, %xmm3
-; SSE41-NEXT:    psrlq $32, %xmm3
-; SSE41-NEXT:    movdqa %xmm0, %xmm5
-; SSE41-NEXT:    psrld %xmm3, %xmm5
-; SSE41-NEXT:    pblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm4[4,5,6,7]
-; SSE41-NEXT:    pxor %xmm3, %xmm3
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    psrld %xmm2, %xmm3
-; SSE41-NEXT:    psrld %xmm4, %xmm0
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4,5,6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3],xmm0[4,5],xmm5[6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; SSE41-NEXT:    pmuludq %xmm2, %xmm3
+; SSE41-NEXT:    pmuludq %xmm1, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7]
 ; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: var_rotate_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm1, %xmm2, %xmm2
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
 ; AVX1-NEXT:    vpslld $23, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vcvttps2dq %xmm1, %xmm1
-; AVX1-NEXT:    vpmulld %xmm0, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm2[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrld %xmm3, %xmm0, %xmm3
-; AVX1-NEXT:    vpsrlq $32, %xmm2, %xmm4
-; AVX1-NEXT:    vpsrld %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm4 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
-; AVX1-NEXT:    vpsrld %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vpsrld %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm4[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7]
-; AVX1-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; AVX1-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,2,2]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7]
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_rotate_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [32,32,32,32]
-; AVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm2
-; AVX2-NEXT:    vpsllvd %xmm1, %xmm0, %xmm1
-; AVX2-NEXT:    vpsrlvd %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpsllvd %xmm1, %xmm0, %xmm2
+; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [32,32,32,32]
+; AVX2-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
+; AVX2-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_rotate_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vprolvd %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_rotate_v4i32:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vprolvd %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; XOP-LABEL: var_rotate_v4i32:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: var_rotate_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [32,32,32,32]
-; X32-SSE-NEXT:    psubd %xmm1, %xmm2
 ; X32-SSE-NEXT:    pslld $23, %xmm1
 ; X32-SSE-NEXT:    paddd {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    cvttps2dq %xmm1, %xmm1
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
-; X32-SSE-NEXT:    pmuludq %xmm0, %xmm1
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[1,1,3,3]
-; X32-SSE-NEXT:    pmuludq %xmm3, %xmm4
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[0,2,2,3]
-; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
-; X32-SSE-NEXT:    psrlq $32, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm4
-; X32-SSE-NEXT:    psrld %xmm3, %xmm4
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
-; X32-SSE-NEXT:    psrld %xmm3, %xmm5
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm5 = xmm4[0],xmm5[1]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[1,3,2,3]
-; X32-SSE-NEXT:    pxor %xmm4, %xmm4
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm5
-; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm6
-; X32-SSE-NEXT:    psrld %xmm5, %xmm6
-; X32-SSE-NEXT:    punpckhdq {{.*#+}} xmm2 = xmm2[2],xmm4[2],xmm2[3],xmm4[3]
-; X32-SSE-NEXT:    psrld %xmm2, %xmm0
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm6[0],xmm0[1]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm1, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,3,2,3]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm2, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,2,3]
+; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
-; X32-SSE-NEXT:    por %xmm1, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %b32 = sub <4 x i32> <i32 32, i32 32, i32 32, i32 32>, %b
   %shl = shl <4 x i32> %a, %b
   %lshr = lshr <4 x i32> %a, %b32
   %or = or <4 x i32> %shl, %lshr
   ret <4 x i32> %or
 }
 
; X32-SSE-NEXT:    retl
   %lshr = lshr <2 x i64> %a, %splat64
   %or = or <2 x i64> %shl, %lshr
   ret <2 x i64> %or
 }
 
 define <4 x i32> @splatvar_rotate_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; SSE2-LABEL: splatvar_rotate_v4i32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    xorps %xmm3, %xmm3
-; SSE2-NEXT:    movss {{.*#+}} xmm3 = xmm1[0],xmm3[1,2,3]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [32,32,32,32]
-; SSE2-NEXT:    psubd %xmm1, %xmm4
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pslld %xmm3, %xmm1
-; SSE2-NEXT:    movdqa %xmm4, %xmm3
-; SSE2-NEXT:    psrlq $32, %xmm3
-; SSE2-NEXT:    movdqa %xmm0, %xmm5
-; SSE2-NEXT:    psrld %xmm3, %xmm5
-; SSE2-NEXT:    movdqa %xmm4, %xmm3
-; SSE2-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE2-NEXT:    movdqa %xmm0, %xmm6
-; SSE2-NEXT:    psrld %xmm3, %xmm6
-; SSE2-NEXT:    movsd {{.*#+}} xmm6 = xmm5[0],xmm6[1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm6[1,3,2,3]
-; SSE2-NEXT:    movdqa %xmm4, %xmm5
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm2[0],xmm5[1],xmm2[1]
-; SSE2-NEXT:    movdqa %xmm0, %xmm6
-; SSE2-NEXT:    psrld %xmm5, %xmm6
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm2[2],xmm4[3],xmm2[3]
-; SSE2-NEXT:    psrld %xmm4, %xmm0
-; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm6[0],xmm0[1]
+; SSE2-NEXT:    pslld $23, %xmm1
+; SSE2-NEXT:    paddd {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,3,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm2, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,2,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
-; SSE2-NEXT:    por %xmm0, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: splatvar_rotate_v4i32:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [32,32,32,32]
-; SSE41-NEXT:    psubd %xmm1, %xmm3
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pslld %xmm2, %xmm1
-; SSE41-NEXT:    movdqa %xmm3, %xmm2
-; SSE41-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    movdqa %xmm0, %xmm4
-; SSE41-NEXT:    psrld %xmm2, %xmm4
-; SSE41-NEXT:    movdqa %xmm3, %xmm2
-; SSE41-NEXT:    psrlq $32, %xmm2
-; SSE41-NEXT:    movdqa %xmm0, %xmm5
-; SSE41-NEXT:    psrld %xmm2, %xmm5
-; SSE41-NEXT:    pblendw {{.*#+}} xmm5 = xmm5[0,1,2,3],xmm4[4,5,6,7]
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
-; SSE41-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrld %xmm3, %xmm2
-; SSE41-NEXT:    psrld %xmm4, %xmm0
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2,3],xmm0[4,5],xmm5[6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; SSE41-NEXT:    pslld $23, %xmm1
+; SSE41-NEXT:    paddd {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    cvttps2dq %xmm1, %xmm1
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; SSE41-NEXT:    pmuludq %xmm2, %xmm3
+; SSE41-NEXT:    pmuludq %xmm1, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7]
 ; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: splatvar_rotate_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
-; AVX1-NEXT:    vpslld %xmm2, %xmm0, %xmm2
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrld %xmm3, %xmm0, %xmm3
-; AVX1-NEXT:    vpsrlq $32, %xmm1, %xmm4
-; AVX1-NEXT:    vpsrld %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm4 = xmm1[2],xmm4[2],xmm1[3],xmm4[3]
-; AVX1-NEXT:    vpsrld %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
-; AVX1-NEXT:    vpsrld %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm4[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7]
-; AVX1-NEXT:    vpor %xmm0, %xmm2, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpslld $23, %xmm1, %xmm1
+; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vcvttps2dq %xmm1, %xmm1
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; AVX1-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,2,2]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7]
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_rotate_v4i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
+; AVX2-NEXT:    vpslld %xmm2, %xmm0, %xmm2
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [32,32,32,32]
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm3, %xmm1
-; AVX2-NEXT:    vpslld %xmm2, %xmm0, %xmm2
 ; AVX2-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpor %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_rotate_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vpbroadcastd %xmm1, %xmm1

 ; XOPAVX2-LABEL: splatvar_rotate_v4i32:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vprotd %xmm1, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatvar_rotate_v4i32:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    xorps %xmm3, %xmm3
-; X32-SSE-NEXT:    movss {{.*#+}} xmm3 = xmm1[0],xmm3[1,2,3]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm4 = [32,32,32,32]
-; X32-SSE-NEXT:    psubd %xmm1, %xmm4
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    pslld %xmm3, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm4, %xmm3
-; X32-SSE-NEXT:    psrlq $32, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm5
-; X32-SSE-NEXT:    psrld %xmm3, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm4, %xmm3
-; X32-SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm6
-; X32-SSE-NEXT:    psrld %xmm3, %xmm6
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm6 = xmm5[0],xmm6[1]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm6[1,3,2,3]
-; X32-SSE-NEXT:    movdqa %xmm4, %xmm5
-; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm2[0],xmm5[1],xmm2[1]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm6
-; X32-SSE-NEXT:    psrld %xmm5, %xmm6
-; X32-SSE-NEXT:    punpckhdq {{.*#+}} xmm4 = xmm4[2],xmm2[2],xmm4[3],xmm2[3]
-; X32-SSE-NEXT:    psrld %xmm4, %xmm0
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm6[0],xmm0[1]
+; X32-SSE-NEXT:    pslld $23, %xmm1
+; X32-SSE-NEXT:    paddd {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    cvttps2dq %xmm1, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm1, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,3,2,3]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm2, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,2,3]
+; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
-; X32-SSE-NEXT:    por %xmm0, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %splat = shufflevector <4 x i32> %b, <4 x i32> undef, <4 x i32> zeroinitializer
   %splat32 = sub <4 x i32> <i32 32, i32 32, i32 32, i32 32>, %splat
   %shl = shl <4 x i32> %a, %splat
   %lshr = lshr <4 x i32> %a, %splat32
   %or = or <4 x i32> %shl, %lshr
   ret <4 x i32> %or
 }
; X32-SSE-NEXT:    retl
   %or = or <2 x i64> %shl, %lshr
   ret <2 x i64> %or
 }
 
 define <4 x i32> @constant_rotate_v4i32(<4 x i32> %a) nounwind {
 ; SSE2-LABEL: constant_rotate_v4i32:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [16,32,64,128]
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    pmuludq %xmm1, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,3,2,3]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; SSE2-NEXT:    pmuludq %xmm1, %xmm3
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    psrld $27, %xmm1
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    psrld $25, %xmm3
-; SSE2-NEXT:    movsd {{.*#+}} xmm3 = xmm1[0],xmm3[1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
-; SSE2-NEXT:    movdqa %xmm0, %xmm3
-; SSE2-NEXT:    psrld $28, %xmm3
-; SSE2-NEXT:    psrld $26, %xmm0
-; SSE2-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
+; SSE2-NEXT:    pmuludq %xmm2, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,2,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT:    por %xmm2, %xmm0
+; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: constant_rotate_v4i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [16,32,64,128]
-; SSE41-NEXT:    pmulld %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrld $25, %xmm2
-; SSE41-NEXT:    movdqa %xmm0, %xmm3
-; SSE41-NEXT:    psrld $27, %xmm3
-; SSE41-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrld $26, %xmm2
-; SSE41-NEXT:    psrld $28, %xmm0
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; SSE41-NEXT:    pmuludq %xmm2, %xmm3
+; SSE41-NEXT:    pmuludq %xmm1, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,0,2,2]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7]
 ; SSE41-NEXT:    por %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: constant_rotate_v4i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm0, %xmm1
-; AVX1-NEXT:    vpsrld $25, %xmm0, %xmm2
-; AVX1-NEXT:    vpsrld $27, %xmm0, %xmm3
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7]
-; AVX1-NEXT:    vpsrld $26, %xmm0, %xmm3
-; AVX1-NEXT:    vpsrld $28, %xmm0, %xmm0
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm3[4,5,6,7]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [16,32,64,128]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpmuludq %xmm2, %xmm3, %xmm2
+; AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,2,2]
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7]
-; AVX1-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_rotate_v4i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsllvd {{.*}}(%rip), %xmm0, %xmm1
-; AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm0
-; AVX2-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %xmm0, %xmm1
+; AVX2-NEXT:    vpsllvd {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_rotate_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,5,6,7]
 ; AVX512BW-NEXT:    vprolvd %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_rotate_v4i32:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vprolvd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; XOP-LABEL: constant_rotate_v4i32:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotd {{.*}}(%rip), %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_rotate_v4i32:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa {{.*#+}} xmm1 = [16,32,64,128]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
-; X32-SSE-NEXT:    pmuludq %xmm1, %xmm2
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm1, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,3,2,3]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
-; X32-SSE-NEXT:    pmuludq %xmm1, %xmm3
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[0,2,2,3]
-; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    psrld $27, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    psrld $25, %xmm3
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm3 = xmm1[0],xmm3[1]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,3,2,3]
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm3
-; X32-SSE-NEXT:    psrld $28, %xmm3
-; X32-SSE-NEXT:    psrld $26, %xmm0
-; X32-SSE-NEXT:    movsd {{.*#+}} xmm0 = xmm3[0],xmm0[1]
+; X32-SSE-NEXT:    pmuludq %xmm2, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,2,3]
+; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; X32-SSE-NEXT:    por %xmm2, %xmm0
+; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %shl = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
   %lshr = lshr <4 x i32> %a, <i32 28, i32 27, i32 26, i32 25>
   %or = or <4 x i32> %shl, %lshr
   ret <4 x i32> %or
 }
 
 define <8 x i16> @constant_rotate_v8i16(<8 x i16> %a) nounwind {
; X32-SSE-NEXT:    retl
   %or = or <2 x i64> %shl, %lshr
   ret <2 x i64> %or
 }
 
 define <4 x i32> @splatconstant_rotate_v4i32(<4 x i32> %a) nounwind {
 ; SSE-LABEL: splatconstant_rotate_v4i32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    pslld $4, %xmm1
-; SSE-NEXT:    psrld $28, %xmm0
+; SSE-NEXT:    psrld $28, %xmm1
+; SSE-NEXT:    pslld $4, %xmm0
 ; SSE-NEXT:    por %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: splatconstant_rotate_v4i32:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpslld $4, %xmm0, %xmm1
-; AVX-NEXT:    vpsrld $28, %xmm0, %xmm0
-; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpsrld $28, %xmm0, %xmm1
+; AVX-NEXT:    vpslld $4, %xmm0, %xmm0
+; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_rotate_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vprold $4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VL-LABEL: splatconstant_rotate_v4i32:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vprold $4, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; XOP-LABEL: splatconstant_rotate_v4i32:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotd $4, %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatconstant_rotate_v4i32:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    pslld $4, %xmm1
-; X32-SSE-NEXT:    psrld $28, %xmm0
+; X32-SSE-NEXT:    psrld $28, %xmm1
+; X32-SSE-NEXT:    pslld $4, %xmm0
 ; X32-SSE-NEXT:    por %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %shl = shl <4 x i32> %a, <i32 4, i32 4, i32 4, i32 4>
   %lshr = lshr <4 x i32> %a, <i32 28, i32 28, i32 28, i32 28>
   %or = or <4 x i32> %shl, %lshr
   ret <4 x i32> %or
 }
 
; X32-SSE-NEXT:    retl
   %or = or <2 x i64> %lmask, %rmask
   ret <2 x i64> %or
 }
 
 define <4 x i32> @splatconstant_rotate_mask_v4i32(<4 x i32> %a) nounwind {
 ; SSE-LABEL: splatconstant_rotate_mask_v4i32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    pslld $4, %xmm1
-; SSE-NEXT:    psrld $28, %xmm0
-; SSE-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    psrld $28, %xmm1
+; SSE-NEXT:    pslld $4, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    pand {{.*}}(%rip), %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: splatconstant_rotate_mask_v4i32:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpslld $4, %xmm0, %xmm1
-; AVX-NEXT:    vpsrld $28, %xmm0, %xmm0
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpsrld $28, %xmm0, %xmm1
+; AVX-NEXT:    vpslld $4, %xmm0, %xmm0
+; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_rotate_mask_v4i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512BW-NEXT:    vprold $4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512BW-NEXT:    vzeroupper

 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vprotd $4, %xmm0, %xmm0
 ; XOP-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; X32-SSE-LABEL: splatconstant_rotate_mask_v4i32:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    pslld $4, %xmm1
-; X32-SSE-NEXT:    psrld $28, %xmm0
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
-; X32-SSE-NEXT:    por %xmm0, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
+; X32-SSE-NEXT:    psrld $28, %xmm1
+; X32-SSE-NEXT:    pslld $4, %xmm0
+; X32-SSE-NEXT:    por %xmm1, %xmm0
+; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    retl
   %shl = shl <4 x i32> %a, <i32 4, i32 4, i32 4, i32 4>
   %lshr = lshr <4 x i32> %a, <i32 28, i32 28, i32 28, i32 28>
   %rmask = and <4 x i32> %lshr, <i32 127, i32 255, i32 511, i32 1023>
   %lmask = and <4 x i32> %shl, <i32 1023, i32 511, i32 255, i32 127>
   %or = or <4 x i32> %lmask, %rmask
   ret <4 x i32> %or
 }

test/CodeGen/X86/vector-rotate-256.ll

; XOPAVX2-NEXT:    retq
   %lshr = lshr <4 x i64> %a, %b64
   %or = or <4 x i64> %shl, %lshr
   ret <4 x i64> %or
 }
 
 define <8 x i32> @var_rotate_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ; AVX1-LABEL: var_rotate_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm1, %xmm3, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX1-NEXT:    vpsubd %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpslld $23, %xmm4, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [1065353216,1065353216,1065353216,1065353216]
-; AVX1-NEXT:    vpaddd %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vcvttps2dq %xmm4, %xmm4
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm6
-; AVX1-NEXT:    vpmulld %xmm6, %xmm4, %xmm4
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [1065353216,1065353216,1065353216,1065353216]
+; AVX1-NEXT:    vpaddd %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm2[1,1,3,3]
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm6 = xmm5[1,1,3,3]
+; AVX1-NEXT:    vpmuludq %xmm4, %xmm6, %xmm4
+; AVX1-NEXT:    vpmuludq %xmm2, %xmm5, %xmm2
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[1,1,3,3]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1],xmm4[2,3],xmm5[4,5],xmm4[6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[0,0,2,2]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm4[2,3],xmm2[4,5],xmm4[6,7]
+; AVX1-NEXT:    vpor %xmm5, %xmm2, %xmm2
 ; AVX1-NEXT:    vpslld $23, %xmm1, %xmm1
-; AVX1-NEXT:    vpaddd %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vpaddd %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vcvttps2dq %xmm1, %xmm1
-; AVX1-NEXT:    vpmulld %xmm0, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm4 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrld %xmm4, %xmm6, %xmm4
-; AVX1-NEXT:    vpsrlq $32, %xmm3, %xmm5
-; AVX1-NEXT:    vpsrld %xmm5, %xmm6, %xmm5
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm5[0,1,2,3],xmm4[4,5,6,7]
-; AVX1-NEXT:    vpxor %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm7 = xmm3[2],xmm5[2],xmm3[3],xmm5[3]
-; AVX1-NEXT:    vpsrld %xmm7, %xmm6, %xmm7
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero
-; AVX1-NEXT:    vpsrld %xmm3, %xmm6, %xmm3
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm7[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1],xmm4[2,3],xmm3[4,5],xmm4[6,7]
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm4 = xmm2[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrld %xmm4, %xmm0, %xmm4
-; AVX1-NEXT:    vpsrlq $32, %xmm2, %xmm6
-; AVX1-NEXT:    vpsrld %xmm6, %xmm0, %xmm6
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm4 = xmm6[0,1,2,3],xmm4[4,5,6,7]
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm5 = xmm2[2],xmm5[2],xmm2[3],xmm5[3]
-; AVX1-NEXT:    vpsrld %xmm5, %xmm0, %xmm5
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vpsrld %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm5[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm4[2,3],xmm0[4,5],xmm4[6,7]
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpmuludq %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,2,2]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7]
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_rotate_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [32,32,32,32,32,32,32,32]
-; AVX2-NEXT:    vpsubd %ymm1, %ymm2, %ymm2
-; AVX2-NEXT:    vpsllvd %ymm1, %ymm0, %ymm1
-; AVX2-NEXT:    vpsrlvd %ymm2, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpsllvd %ymm1, %ymm0, %ymm2
+; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [32,32,32,32,32,32,32,32]
+; AVX2-NEXT:    vpsubd %ymm1, %ymm3, %ymm1
+; AVX2-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: var_rotate_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vprolvd %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
; XOPAVX2-NEXT:    retq
   %lshr = lshr <4 x i64> %a, %splat64
   %or = or <4 x i64> %shl, %lshr
   ret <4 x i64> %or
 }
 
 define <8 x i32> @splatvar_rotate_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ; AVX1-LABEL: splatvar_rotate_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[0,0,0,0]
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [32,32,32,32]
-; AVX1-NEXT:    vpsubd %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
-; AVX1-NEXT:    vpslld %xmm1, %xmm3, %xmm4
-; AVX1-NEXT:    vpslld %xmm1, %xmm0, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm8
-; AVX1-NEXT:    vpsrldq {{.*#+}} xmm4 = xmm2[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpsrld %xmm4, %xmm3, %xmm5
-; AVX1-NEXT:    vpsrlq $32, %xmm2, %xmm6
-; AVX1-NEXT:    vpsrld %xmm6, %xmm3, %xmm7
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm7[0,1,2,3],xmm5[4,5,6,7]
-; AVX1-NEXT:    vpxor %xmm7, %xmm7, %xmm7
-; AVX1-NEXT:    vpunpckhdq {{.*#+}} xmm7 = xmm2[2],xmm7[2],xmm2[3],xmm7[3]
-; AVX1-NEXT:    vpsrld %xmm7, %xmm3, %xmm1
-; AVX1-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero
-; AVX1-NEXT:    vpsrld %xmm2, %xmm3, %xmm3
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm3[0,1,2,3],xmm1[4,5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm5[2,3],xmm1[4,5],xmm5[6,7]
-; AVX1-NEXT:    vpsrld %xmm4, %xmm0, %xmm3
-; AVX1-NEXT:    vpsrld %xmm6, %xmm0, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpsrld %xmm7, %xmm0, %xmm4
-; AVX1-NEXT:    vpsrld %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm4[4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; AVX1-NEXT:    vpslld $23, %xmm1, %xmm1
+; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vcvttps2dq %xmm1, %xmm1
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm1[1,1,3,3]
+; AVX1-NEXT:    vpmuludq %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpmuludq %xmm1, %xmm2, %xmm2
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[1,1,3,3]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1],xmm3[2,3],xmm5[4,5],xmm3[6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,2,2]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2,3],xmm2[4,5],xmm3[6,7]
+; AVX1-NEXT:    vpor %xmm5, %xmm2, %xmm2
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpmuludq %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3],xmm1[4,5],xmm3[6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,2,2]
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7]
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm0, %ymm8, %ymm0
+; AVX1-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatvar_rotate_v8i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpbroadcastd %xmm1, %ymm2
-; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [32,32,32,32,32,32,32,32]
-; AVX2-NEXT:    vpsubd %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero
 ; AVX2-NEXT:    vpslld %xmm1, %ymm0, %ymm1
+; AVX2-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [32,32,32,32,32,32,32,32]
+; AVX2-NEXT:    vpsubd %ymm2, %ymm3, %ymm2
 ; AVX2-NEXT:    vpsrlvd %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatvar_rotate_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vpbroadcastd %xmm1, %ymm1
; XOPAVX2-NEXT:    retq
   %lshr = lshr <4 x i64> %a, <i64 60, i64 50, i64 14, i64 4>
   %or = or <4 x i64> %shl, %lshr
   ret <4 x i64> %or
 }
 
 define <8 x i32> @constant_rotate_v8i32(<8 x i32> %a) nounwind {
 ; AVX1-LABEL: constant_rotate_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm0, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpmulld {{.*}}(%rip), %xmm2, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; AVX1-NEXT:    vpsrld $21, %xmm2, %xmm3
-; AVX1-NEXT:    vpsrld $23, %xmm2, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpsrld $22, %xmm2, %xmm4
-; AVX1-NEXT:    vpsrld $24, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm4[4,5,6,7]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [256,512,1024,2048]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[1,1,3,3]
+; AVX1-NEXT:    vpmuludq %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpmuludq %xmm1, %xmm3, %xmm1
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[0,0,2,2]
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
+; AVX1-NEXT:    vpor %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [16,32,64,128]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm0[1,1,3,3]
+; AVX1-NEXT:    vpmuludq %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpmuludq %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm3[2,3],xmm2[4,5],xmm3[6,7]
-; AVX1-NEXT:    vpsrld $25, %xmm0, %xmm3
-; AVX1-NEXT:    vpsrld $27, %xmm0, %xmm4
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7]
-; AVX1-NEXT:    vpsrld $26, %xmm0, %xmm4
-; AVX1-NEXT:    vpsrld $28, %xmm0, %xmm0
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm4[4,5,6,7]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,2,2]
 ; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7]
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_rotate_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsllvd {{.*}}(%rip), %ymm0, %ymm1
-; AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpsrlvd {{.*}}(%rip), %ymm0, %ymm1
+; AVX2-NEXT:    vpsllvd {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_rotate_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,5,6,7,8,9,10,11]
 ; AVX512BW-NEXT:    vprolvd %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
; XOPAVX2-NEXT:    retq
   %lshr = lshr <4 x i64> %a, <i64 50, i64 50, i64 50, i64 50>
   %or = or <4 x i64> %shl, %lshr
   ret <4 x i64> %or
 }
 
 define <8 x i32> @splatconstant_rotate_v8i32(<8 x i32> %a) nounwind {
 ; AVX1-LABEL: splatconstant_rotate_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpslld $4, %xmm0, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpslld $4, %xmm2, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; AVX1-NEXT:    vpsrld $28, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrld $28, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpsrld $28, %xmm1, %xmm2
+; AVX1-NEXT:    vpslld $4, %xmm1, %xmm1
+; AVX1-NEXT:    vpor %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrld $28, %xmm0, %xmm2
+; AVX1-NEXT:    vpslld $4, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatconstant_rotate_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpslld $4, %ymm0, %ymm1
-; AVX2-NEXT:    vpsrld $28, %ymm0, %ymm0
-; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpsrld $28, %ymm0, %ymm1
+; AVX2-NEXT:    vpslld $4, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_rotate_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vprold $4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512BW-NEXT:    retq
; XOPAVX2-NEXT:    retq
   %lmask = and <4 x i64> %shl, <i64 33, i64 65, i64 129, i64 257>
   %or = or <4 x i64> %lmask, %rmask
   ret <4 x i64> %or
 }
 
 define <8 x i32> @splatconstant_rotate_mask_v8i32(<8 x i32> %a) nounwind {
 ; AVX1-LABEL: splatconstant_rotate_mask_v8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpslld $4, %xmm0, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpslld $4, %xmm2, %xmm3
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
-; AVX1-NEXT:    vpsrld $28, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrld $28, %xmm2, %xmm2
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpsrld $28, %xmm1, %xmm2
+; AVX1-NEXT:    vpslld $4, %xmm1, %xmm1
+; AVX1-NEXT:    vpor %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrld $28, %xmm0, %xmm2
+; AVX1-NEXT:    vpslld $4, %xmm0, %xmm0
+; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm0, %ymm0
-; AVX1-NEXT:    vandps {{.*}}(%rip), %ymm1, %ymm1
-; AVX1-NEXT:    vorps %ymm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: splatconstant_rotate_mask_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpslld $4, %ymm0, %ymm1
-; AVX2-NEXT:    vpsrld $28, %ymm0, %ymm0
+; AVX2-NEXT:    vpsrld $28, %ymm0, %ymm1
+; AVX2-NEXT:    vpslld $4, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512BW-LABEL: splatconstant_rotate_mask_v8i32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
 ; AVX512BW-NEXT:    vprold $4, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512BW-NEXT:    retq

test/CodeGen/X86/vector-shift-shl-128.ll

 }
 
 define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; SSE2-LABEL: var_shift_v4i32:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pslld $23, %xmm1
 ; SSE2-NEXT:    paddd {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; SSE2-NEXT:    pmuludq %xmm0, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; SSE2-NEXT:    pmuludq %xmm2, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm1, %xmm0
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; SSE2-NEXT:    pmuludq %xmm2, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: var_shift_v4i32:
 ; SSE41:       # %bb.0:
 ; SSE41-NEXT:    pslld $23, %xmm1
 ; SSE41-NEXT:    paddd {{.*}}(%rip), %xmm1
 ; SSE41-NEXT:    cvttps2dq %xmm1, %xmm1
 ; SSE41-NEXT:    pmulld %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: var_shift_v4i32:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpslld $23, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddd {{.*}}(%rip), %xmm1, %xmm1
 ; AVX1-NEXT:    vcvttps2dq %xmm1, %xmm1
-; AVX1-NEXT:    vpmulld %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_shift_v4i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpsllvd %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: var_shift_v4i32:

 ; AVX512VL-NEXT:    vpsllvd %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: var_shift_v4i32:
 ; X32-SSE:       # %bb.0:
 ; X32-SSE-NEXT:    pslld $23, %xmm1
 ; X32-SSE-NEXT:    paddd {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    cvttps2dq %xmm1, %xmm1
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,3,3]
-; X32-SSE-NEXT:    pmuludq %xmm0, %xmm1
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; X32-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; X32-SSE-NEXT:    pmuludq %xmm2, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm1, %xmm0
 ; X32-SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm0
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
+; X32-SSE-NEXT:    pmuludq %xmm2, %xmm1
+; X32-SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; X32-SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; X32-SSE-NEXT:    retl
   %shift = shl <4 x i32> %a, %b
   ret <4 x i32> %shift
 }
 
 define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
 ; SSE2-LABEL: var_shift_v8i16:
 ; SSE2:       # %bb.0:

test/CodeGen/X86/vector-shift-shl-256.ll

 ; AVX1-LABEL: var_shift_v8i32:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [1065353216,1065353216,1065353216,1065353216]
 ; AVX1-NEXT:    vpaddd %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; AVX1-NEXT:    vpmulld %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpmulld %xmm2, %xmm4, %xmm2
 ; AVX1-NEXT:    vpslld $23, %xmm1, %xmm1
 ; AVX1-NEXT:    vpaddd %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vcvttps2dq %xmm1, %xmm1
-; AVX1-NEXT:    vpmulld %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: var_shift_v8i32:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpsllvd %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;

 ; X32-AVX1-LABEL: var_shift_v8i32:
 ; X32-AVX1:       # %bb.0:
 ; X32-AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; X32-AVX1-NEXT:    vpslld $23, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [1065353216,1065353216,1065353216,1065353216]
 ; X32-AVX1-NEXT:    vpaddd %xmm3, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vcvttps2dq %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
-; X32-AVX1-NEXT:    vpmulld %xmm4, %xmm2, %xmm2
+; X32-AVX1-NEXT:    vpmulld %xmm2, %xmm4, %xmm2
 ; X32-AVX1-NEXT:    vpslld $23, %xmm1, %xmm1
 ; X32-AVX1-NEXT:    vpaddd %xmm3, %xmm1, %xmm1
 ; X32-AVX1-NEXT:    vcvttps2dq %xmm1, %xmm1
-; X32-AVX1-NEXT:    vpmulld %xmm0, %xmm1, %xmm0
+; X32-AVX1-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; X32-AVX1-NEXT:    retl
 ;
 ; X32-AVX2-LABEL: var_shift_v8i32:
 ; X32-AVX2:       # %bb.0:
 ; X32-AVX2-NEXT:    vpsllvd %ymm1, %ymm0, %ymm0
 ; X32-AVX2-NEXT:    retl
   %shift = shl <8 x i32> %a, %b