[X86][SSE] Lower constant vXi8 ISD::SRL/ISD::SRA using PMULLW

[RKSimon]

Summary

Extending the concept introduced in D49562, this patch lowers constant vXi8 ISD::SRL/ISD::SRA by zero/sign extending to vXi16 and using PMULLW and then truncating the high 8 bits of the result.

[craig.topper]

lib/Target/X86/X86ISelLowering.cpp

23701 ↗

(On Diff #160806)

hasBWI is unnecessary here right? BWI is only truly legal under useBWIRegs not hasBWI. But if you have a v64i8 type here, it must already be legal.

23715 ↗

(On Diff #160806)

Can we do this for v32i8 and useBWIRegs?

[RKSimon]

lib/Target/X86/X86ISelLowering.cpp

23701 ↗

(On Diff #160806)

Yup - I can simplify that cheers!

23715 ↗

(On Diff #160806)

Its caught in code above it (line 32684).

[RKSimon]

lib/Target/X86/X86ISelLowering.cpp

23715 ↗

(On Diff #160806)

Sorry that's line 23684

[RKSimon]

Remove unnecessary hasBWI()

[craig.topper]

LGTM

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

static SDValue LowerShift(SDValue Op, const X86Subtarget &Subtarget,
   }
 
   // If possible, lower this packed shift into a vector multiply instead of
   // expanding it into a sequence of scalar shifts.
   if (Opc == ISD::SHL)
     if (SDValue Scale = convertShiftLeftToScale(Amt, dl, Subtarget, DAG))
       return DAG.getNode(ISD::MUL, dl, VT, R, Scale);
 
-  // Constant ISD::SRL can be performed efficiently on vXi8/vXi16 vectors as we
+  // Constant ISD::SRL can be performed efficiently on vXi16 vectors as we
   // can replace with ISD::MULHU, creating scale factor from (NumEltBits - Amt).
   // TODO: Improve support for the shift by zero special case.
   if (Opc == ISD::SRL && ConstantAmt &&
       ((Subtarget.hasSSE41() && VT == MVT::v8i16) ||
        DAG.isKnownNeverZero(Amt)) &&
-      (VT == MVT::v16i8 || VT == MVT::v8i16 ||
-       ((VT == MVT::v32i8 || VT == MVT::v16i16) && Subtarget.hasInt256()))) {
+      (VT == MVT::v8i16 || (VT == MVT::v16i16 && Subtarget.hasInt256()))) {
     SDValue EltBits = DAG.getConstant(VT.getScalarSizeInBits(), dl, VT);
     SDValue RAmt = DAG.getNode(ISD::SUB, dl, VT, EltBits, Amt);
     if (SDValue Scale = convertShiftLeftToScale(RAmt, dl, Subtarget, DAG)) {
       SDValue Zero = DAG.getConstant(0, dl, VT);
       SDValue ZAmt = DAG.getSetCC(dl, VT, Amt, Zero, ISD::SETEQ);
       SDValue Res = DAG.getNode(ISD::MULHU, dl, VT, R, Scale);
       return DAG.getSelect(dl, VT, ZAmt, R, Res);
     }
if ((Subtarget.hasInt256() && VT == MVT::v8i16) ||
     MVT ExtVT = MVT::getVectorVT(EvtSVT, VT.getVectorNumElements());
     unsigned ExtOpc = Opc == ISD::SRA ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
     R = DAG.getNode(ExtOpc, dl, ExtVT, R);
     Amt = DAG.getNode(ISD::ZERO_EXTEND, dl, ExtVT, Amt);
     return DAG.getNode(ISD::TRUNCATE, dl, VT,
                        DAG.getNode(Opc, dl, ExtVT, R, Amt));
   }
 
+  // Constant ISD::SRA/SRL can be performed efficiently on vXi8 vectors as we
+  // extend to vXi16 to perform a MUL scale effectively as a MUL_LOHI.
+  if (ConstantAmt && (Opc == ISD::SRA || Opc == ISD::SRL) &&
+      (VT == MVT::v16i8 || VT == MVT::v64i8 ||
+       (VT == MVT::v32i8 && Subtarget.hasInt256())) &&
+      !Subtarget.hasXOP()) {
+    int NumElts = VT.getVectorNumElements();
+    SDValue Cst8 = DAG.getConstant(8, dl, MVT::i8);
+
+    // Extend constant shift amount to vXi16 (it doesn't matter if the type
+    // isn't legal).
+    MVT ExVT = MVT::getVectorVT(MVT::i16, NumElts);
+    Amt = DAG.getZExtOrTrunc(Amt, dl, ExVT);
+    Amt = DAG.getNode(ISD::SUB, dl, ExVT, DAG.getConstant(8, dl, ExVT), Amt);
+    Amt = DAG.getNode(ISD::SHL, dl, ExVT, DAG.getConstant(1, dl, ExVT), Amt);
+    assert(ISD::isBuildVectorOfConstantSDNodes(Amt.getNode()) &&
+           "Constant build vector expected");
+
+    if (VT == MVT::v16i8 && Subtarget.hasInt256()) {
+      R = Opc == ISD::SRA ? DAG.getSExtOrTrunc(R, dl, ExVT)
+                          : DAG.getZExtOrTrunc(R, dl, ExVT);
+      R = DAG.getNode(ISD::MUL, dl, ExVT, R, Amt);
+      R = DAG.getNode(X86ISD::VSRLI, dl, ExVT, R, Cst8);
+      return DAG.getZExtOrTrunc(R, dl, VT);
+    }
+
+    SmallVector<SDValue, 16> LoAmt, HiAmt;
+    for (int i = 0; i != NumElts; i += 16) {
+      for (int j = 0; j != 8; ++j) {
+        LoAmt.push_back(Amt.getOperand(i + j));
+        HiAmt.push_back(Amt.getOperand(i + j + 8));
+      }
+    }
+
+    MVT VT16 = MVT::getVectorVT(MVT::i16, NumElts / 2);
+    SDValue LoA = DAG.getBuildVector(VT16, dl, LoAmt);
+    SDValue HiA = DAG.getBuildVector(VT16, dl, HiAmt);
+
+    unsigned ShiftOp = Opc == ISD::SRA ? X86ISD::VSRAI : X86ISD::VSRLI;
+    SDValue LoR = DAG.getBitcast(VT16, getUnpackl(DAG, dl, VT, R, R));
+    SDValue HiR = DAG.getBitcast(VT16, getUnpackh(DAG, dl, VT, R, R));
+    LoR = DAG.getNode(ShiftOp, dl, VT16, LoR, Cst8);
+    HiR = DAG.getNode(ShiftOp, dl, VT16, HiR, Cst8);
+    LoR = DAG.getNode(ISD::MUL, dl, VT16, LoR, LoA);
+    HiR = DAG.getNode(ISD::MUL, dl, VT16, HiR, HiA);
+    LoR = DAG.getNode(X86ISD::VSRLI, dl, VT16, LoR, Cst8);
+    HiR = DAG.getNode(X86ISD::VSRLI, dl, VT16, HiR, Cst8);
+    return DAG.getNode(X86ISD::PACKUS, dl, VT, LoR, HiR);
+  }
+
   if (VT == MVT::v16i8 ||
       (VT == MVT::v32i8 && Subtarget.hasInt256() && !Subtarget.hasXOP()) ||
       (VT == MVT::v64i8 && Subtarget.hasBWI())) {
     MVT ExtVT = MVT::getVectorVT(MVT::i16, VT.getVectorNumElements() / 2);
     unsigned ShiftOpcode = Op->getOpcode();
 
     auto SignBitSelect = [&](MVT SelVT, SDValue Sel, SDValue V0, SDValue V1) {
       if (VT.is512BitVector()) {

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

; AVX-NEXT:    retq
   ret <4 x i32> %1
 }
 
 define <16 x i8> @combine_vec_sdiv_by_pow2b_v16i8(<16 x i8> %x) {
 ; SSE-LABEL: combine_vec_sdiv_by_pow2b_v16i8:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa %xmm0, %xmm1
 ; SSE-NEXT:    pxor %xmm0, %xmm0
-; SSE-NEXT:    pcmpgtb %xmm1, %xmm0
-; SSE-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [1,4,2,16,8,32,64,2]
-; SSE-NEXT:    pmullw %xmm2, %xmm3
+; SSE-NEXT:    pxor %xmm3, %xmm3
+; SSE-NEXT:    pcmpgtb %xmm1, %xmm3
+; SSE-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm0[8],xmm3[9],xmm0[9],xmm3[10],xmm0[10],xmm3[11],xmm0[11],xmm3[12],xmm0[12],xmm3[13],xmm0[13],xmm3[14],xmm0[14],xmm3[15],xmm0[15]
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [1,4,2,16,8,32,64,2]
+; SSE-NEXT:    pmullw %xmm0, %xmm3
 ; SSE-NEXT:    psrlw $8, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; SSE-NEXT:    pmullw %xmm2, %xmm0
+; SSE-NEXT:    pmullw %xmm0, %xmm2
+; SSE-NEXT:    psrlw $8, %xmm2
+; SSE-NEXT:    packuswb %xmm3, %xmm2
+; SSE-NEXT:    paddb %xmm1, %xmm2
+; SSE-NEXT:    movdqa %xmm2, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
+; SSE-NEXT:    psraw $8, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [256,64,128,16,32,8,4,128]
+; SSE-NEXT:    pmullw %xmm3, %xmm0
 ; SSE-NEXT:    psrlw $8, %xmm0
-; SSE-NEXT:    packuswb %xmm0, %xmm3
-; SSE-NEXT:    paddb %xmm1, %xmm3
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm3[8],xmm2[9],xmm3[9],xmm2[10],xmm3[10],xmm2[11],xmm3[11],xmm2[12],xmm3[12],xmm2[13],xmm3[13],xmm2[14],xmm3[14],xmm2[15],xmm3[15]
-; SSE-NEXT:    movdqa %xmm2, %xmm4
-; SSE-NEXT:    psraw $4, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [16384,32800,41056,8384,16384,32800,41056,8384]
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm5[8],xmm0[9],xmm5[9],xmm0[10],xmm5[10],xmm0[11],xmm5[11],xmm0[12],xmm5[12],xmm0[13],xmm5[13],xmm0[14],xmm5[14],xmm0[15],xmm5[15]
-; SSE-NEXT:    pblendvb %xmm0, %xmm4, %xmm2
-; SSE-NEXT:    movdqa %xmm2, %xmm4
-; SSE-NEXT:    psraw $2, %xmm4
-; SSE-NEXT:    paddw %xmm0, %xmm0
-; SSE-NEXT:    pblendvb %xmm0, %xmm4, %xmm2
-; SSE-NEXT:    movdqa %xmm2, %xmm4
-; SSE-NEXT:    psraw $1, %xmm4
-; SSE-NEXT:    paddw %xmm0, %xmm0
-; SSE-NEXT:    pblendvb %xmm0, %xmm4, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    psraw $8, %xmm2
+; SSE-NEXT:    pmullw %xmm3, %xmm2
 ; SSE-NEXT:    psrlw $8, %xmm2
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    movdqa %xmm3, %xmm4
-; SSE-NEXT:    psraw $4, %xmm4
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3],xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
-; SSE-NEXT:    pblendvb %xmm0, %xmm4, %xmm3
-; SSE-NEXT:    movdqa %xmm3, %xmm4
-; SSE-NEXT:    psraw $2, %xmm4
-; SSE-NEXT:    paddw %xmm0, %xmm0
-; SSE-NEXT:    pblendvb %xmm0, %xmm4, %xmm3
-; SSE-NEXT:    movdqa %xmm3, %xmm4
-; SSE-NEXT:    psraw $1, %xmm4
-; SSE-NEXT:    paddw %xmm0, %xmm0
-; SSE-NEXT:    pblendvb %xmm0, %xmm4, %xmm3
-; SSE-NEXT:    psrlw $8, %xmm3
-; SSE-NEXT:    packuswb %xmm2, %xmm3
+; SSE-NEXT:    packuswb %xmm0, %xmm2
 ; SSE-NEXT:    movaps {{.*#+}} xmm0 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
-; SSE-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
+; SSE-NEXT:    pblendvb %xmm0, %xmm2, %xmm1
 ; SSE-NEXT:    movdqa %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_vec_sdiv_by_pow2b_v16i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm1
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm2
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,4,2,16,8,32,64,2]
-; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX1-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackuswb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpaddb %xmm1, %xmm0, %xmm1
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
-; AVX1-NEXT:    vpsraw $4, %xmm2, %xmm3
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [16384,32800,41056,8384,16384,32800,41056,8384]
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm0[8],xmm4[8],xmm0[9],xmm4[9],xmm0[10],xmm4[10],xmm0[11],xmm4[11],xmm0[12],xmm4[12],xmm0[13],xmm4[13],xmm0[14],xmm4[14],xmm0[15],xmm4[15]
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpsraw $2, %xmm2, %xmm3
-; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpsraw $1, %xmm2, %xmm3
-; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vpsraw $8, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [256,64,128,16,32,8,4,128]
+; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX1-NEXT:    vpsraw $4, %xmm1, %xmm3
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpsraw $2, %xmm1, %xmm3
-; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpsraw $1, %xmm1, %xmm3
-; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm4
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpsraw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpmullw %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
 ; AVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: combine_vec_sdiv_by_pow2b_v16i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
 ; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
 ; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; AVX2-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vpaddb %xmm1, %xmm0, %xmm1
-; AVX2-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
-; AVX2-NEXT:    vpsraw $4, %xmm2, %xmm3
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm4 = [16384,32800,41056,8384,16384,32800,41056,8384]
-; AVX2-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm0[8],xmm4[8],xmm0[9],xmm4[9],xmm0[10],xmm4[10],xmm0[11],xmm4[11],xmm0[12],xmm4[12],xmm0[13],xmm4[13],xmm0[14],xmm4[14],xmm0[15],xmm4[15]
-; AVX2-NEXT:    vpblendvb %xmm5, %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpsraw $2, %xmm2, %xmm3
-; AVX2-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpblendvb %xmm5, %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpsraw $1, %xmm2, %xmm3
-; AVX2-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpblendvb %xmm5, %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpsrlw $8, %xmm2, %xmm2
-; AVX2-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX2-NEXT:    vpsraw $4, %xmm1, %xmm3
-; AVX2-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
-; AVX2-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
-; AVX2-NEXT:    vpsraw $2, %xmm1, %xmm3
-; AVX2-NEXT:    vpaddw %xmm4, %xmm4, %xmm4
-; AVX2-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
-; AVX2-NEXT:    vpsraw $1, %xmm1, %xmm3
-; AVX2-NEXT:    vpaddw %xmm4, %xmm4, %xmm4
-; AVX2-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
-; AVX2-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX2-NEXT:    vpmovsxbw %xmm1, %ymm1
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; AVX2-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
 ; AVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: combine_vec_sdiv_by_pow2b_v16i8:
 ; AVX512F:       # %bb.0:
; XOP-NEXT:    retq
   ret <16 x i8> %div
 }
 
 define <16 x i8> @non_splat_minus_one_divisor_1(<16 x i8> %A) {
 ; SSE-LABEL: non_splat_minus_one_divisor_1:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa %xmm0, %xmm1
 ; SSE-NEXT:    pxor %xmm2, %xmm2
-; SSE-NEXT:    pxor %xmm0, %xmm0
-; SSE-NEXT:    pcmpgtb %xmm1, %xmm0
-; SSE-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; SSE-NEXT:    pmullw {{.*}}(%rip), %xmm4
-; SSE-NEXT:    psrlw $8, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; SSE-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE-NEXT:    pxor %xmm3, %xmm3
+; SSE-NEXT:    pcmpgtb %xmm0, %xmm3
+; SSE-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
+; SSE-NEXT:    movdqa %xmm4, %xmm0
+; SSE-NEXT:    psllw $1, %xmm0
+; SSE-NEXT:    pblendw {{.*#+}} xmm0 = xmm4[0,1],xmm0[2],xmm4[3,4,5],xmm0[6],xmm4[7]
 ; SSE-NEXT:    psrlw $8, %xmm0
-; SSE-NEXT:    packuswb %xmm0, %xmm4
-; SSE-NEXT:    paddb %xmm1, %xmm4
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm4[8],xmm3[9],xmm4[9],xmm3[10],xmm4[10],xmm3[11],xmm4[11],xmm3[12],xmm4[12],xmm3[13],xmm4[13],xmm3[14],xmm4[14],xmm3[15],xmm4[15]
-; SSE-NEXT:    movdqa %xmm3, %xmm5
-; SSE-NEXT:    psraw $4, %xmm5
-; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [0,32,0,32,8192,8224,57376,57376]
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm6[8],xmm0[9],xmm6[9],xmm0[10],xmm6[10],xmm0[11],xmm6[11],xmm0[12],xmm6[12],xmm0[13],xmm6[13],xmm0[14],xmm6[14],xmm0[15],xmm6[15]
-; SSE-NEXT:    pblendvb %xmm0, %xmm5, %xmm3
-; SSE-NEXT:    movdqa %xmm3, %xmm5
-; SSE-NEXT:    psraw $2, %xmm5
-; SSE-NEXT:    paddw %xmm0, %xmm0
-; SSE-NEXT:    pblendvb %xmm0, %xmm5, %xmm3
-; SSE-NEXT:    movdqa %xmm3, %xmm5
-; SSE-NEXT:    psraw $1, %xmm5
-; SSE-NEXT:    paddw %xmm0, %xmm0
-; SSE-NEXT:    pblendvb %xmm0, %xmm5, %xmm3
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
+; SSE-NEXT:    pmullw {{.*}}(%rip), %xmm3
 ; SSE-NEXT:    psrlw $8, %xmm3
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    movdqa %xmm4, %xmm5
-; SSE-NEXT:    psraw $4, %xmm5
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm6[0],xmm0[1],xmm6[1],xmm0[2],xmm6[2],xmm0[3],xmm6[3],xmm0[4],xmm6[4],xmm0[5],xmm6[5],xmm0[6],xmm6[6],xmm0[7],xmm6[7]
-; SSE-NEXT:    pblendvb %xmm0, %xmm5, %xmm4
-; SSE-NEXT:    movdqa %xmm4, %xmm5
-; SSE-NEXT:    psraw $2, %xmm5
-; SSE-NEXT:    paddw %xmm0, %xmm0
-; SSE-NEXT:    pblendvb %xmm0, %xmm5, %xmm4
-; SSE-NEXT:    movdqa %xmm4, %xmm5
-; SSE-NEXT:    psraw $1, %xmm5
-; SSE-NEXT:    paddw %xmm0, %xmm0
-; SSE-NEXT:    pblendvb %xmm0, %xmm5, %xmm4
-; SSE-NEXT:    psrlw $8, %xmm4
-; SSE-NEXT:    packuswb %xmm3, %xmm4
+; SSE-NEXT:    packuswb %xmm3, %xmm0
+; SSE-NEXT:    paddb %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm3
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
+; SSE-NEXT:    psraw $8, %xmm3
+; SSE-NEXT:    movdqa %xmm3, %xmm4
+; SSE-NEXT:    psllw $7, %xmm4
+; SSE-NEXT:    psllw $8, %xmm3
+; SSE-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0,1],xmm4[2],xmm3[3,4,5],xmm4[6],xmm3[7]
+; SSE-NEXT:    psrlw $8, %xmm3
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE-NEXT:    psraw $8, %xmm0
+; SSE-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    packuswb %xmm0, %xmm3
 ; SSE-NEXT:    movaps {{.*#+}} xmm0 = [0,0,255,0,0,0,255,0,0,255,255,255,255,255,255,255]
-; SSE-NEXT:    pblendvb %xmm0, %xmm4, %xmm1
+; SSE-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
 ; SSE-NEXT:    psubb %xmm1, %xmm2
 ; SSE-NEXT:    movaps {{.*#+}} xmm0 = [255,255,0,255,255,255,0,255,255,0,0,0,0,255,0,255]
 ; SSE-NEXT:    pblendvb %xmm0, %xmm2, %xmm1
 ; SSE-NEXT:    movdqa %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: non_splat_minus_one_divisor_1:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm2
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm3 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $1, %xmm3, %xmm4
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1],xmm4[2],xmm3[3,4,5],xmm4[6],xmm3[7]
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
 ; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpaddb %xmm2, %xmm0, %xmm2
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
-; AVX1-NEXT:    vpsraw $4, %xmm3, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [0,32,0,32,8192,8224,57376,57376]
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm0[8],xmm5[8],xmm0[9],xmm5[9],xmm0[10],xmm5[10],xmm0[11],xmm5[11],xmm0[12],xmm5[12],xmm0[13],xmm5[13],xmm0[14],xmm5[14],xmm0[15],xmm5[15]
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpsraw $2, %xmm3, %xmm4
-; AVX1-NEXT:    vpaddw %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpsraw $1, %xmm3, %xmm4
-; AVX1-NEXT:    vpaddw %xmm6, %xmm6, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpsraw $8, %xmm3, %xmm3
+; AVX1-NEXT:    vpsllw $7, %xmm3, %xmm4
+; AVX1-NEXT:    vpsllw $8, %xmm3, %xmm3
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0,1],xmm4[2],xmm3[3,4,5],xmm4[6],xmm3[7]
 ; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; AVX1-NEXT:    vpsraw $4, %xmm2, %xmm4
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm5 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3],xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpsraw $2, %xmm2, %xmm4
-; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm2, %xmm2
-; AVX1-NEXT:    vpsraw $1, %xmm2, %xmm4
-; AVX1-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm5, %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vpsraw $8, %xmm2, %xmm2
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
-; AVX1-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpackuswb %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,0,255,0,0,0,255,0,0,255,255,255,255,255,255,255]
 ; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsubb %xmm0, %xmm1, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,255,0,255,255,255,0,255,255,0,0,0,0,255,0,255]
 ; AVX1-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: non_splat_minus_one_divisor_1:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm2
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero
 ; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
 ; AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX2-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpaddb %xmm2, %xmm0, %xmm2
-; AVX2-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
-; AVX2-NEXT:    vpsraw $4, %xmm3, %xmm4
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm5 = [0,32,0,32,8192,8224,57376,57376]
-; AVX2-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm0[8],xmm5[8],xmm0[9],xmm5[9],xmm0[10],xmm5[10],xmm0[11],xmm5[11],xmm0[12],xmm5[12],xmm0[13],xmm5[13],xmm0[14],xmm5[14],xmm0[15],xmm5[15]
-; AVX2-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
-; AVX2-NEXT:    vpsraw $2, %xmm3, %xmm4
-; AVX2-NEXT:    vpaddw %xmm6, %xmm6, %xmm6
-; AVX2-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
-; AVX2-NEXT:    vpsraw $1, %xmm3, %xmm4
-; AVX2-NEXT:    vpaddw %xmm6, %xmm6, %xmm6
-; AVX2-NEXT:    vpblendvb %xmm6, %xmm4, %xmm3, %xmm3
-; AVX2-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX2-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; AVX2-NEXT:    vpsraw $4, %xmm2, %xmm4
-; AVX2-NEXT:    vpunpcklbw {{.*#+}} xmm5 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3],xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
-; AVX2-NEXT:    vpblendvb %xmm5, %xmm4, %xmm2, %xmm2
-; AVX2-NEXT:    vpsraw $2, %xmm2, %xmm4
-; AVX2-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpblendvb %xmm5, %xmm4, %xmm2, %xmm2
-; AVX2-NEXT:    vpsraw $1, %xmm2, %xmm4
-; AVX2-NEXT:    vpaddw %xmm5, %xmm5, %xmm5
-; AVX2-NEXT:    vpblendvb %xmm5, %xmm4, %xmm2, %xmm2
-; AVX2-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; AVX2-NEXT:    vpmovsxbw %xmm2, %ymm2
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX2-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,0,255,0,0,0,255,0,0,255,255,255,255,255,255,255]
 ; AVX2-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
 ; AVX2-NEXT:    vpsubb %xmm0, %xmm1, %xmm1
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,255,0,255,255,255,0,255,255,0,0,0,0,255,0,255]
 ; AVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq

llvm/trunk/test/CodeGen/X86/vector-rotate-512.ll

 ; AVX512BW-NEXT:    vpsllw $2, %zmm2, %zmm3
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm2 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536]
-; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm3
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
-; AVX512BW-NEXT:    vpsrlw $2, %zmm0, %zmm3
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
-; AVX512BW-NEXT:    vpsrlw $1, %zmm0, %zmm3
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
+; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
+; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_rotate_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256]
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm2
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k1}
 ; AVX512VLBW-NEXT:    vpsllw $2, %zmm2, %zmm3
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm2 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536]
-; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512VLBW-NEXT:    vpsrlw $4, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
-; AVX512VLBW-NEXT:    vpsrlw $2, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
-; AVX512VLBW-NEXT:    vpsrlw $1, %zmm0, %zmm3
-; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
+; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %shl = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>
   %lshr = lshr <64 x i8> %a, <i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7>
   %or = or <64 x i8> %shl, %lshr
   ret <64 x i8> %or
 }
 

llvm/trunk/test/CodeGen/X86/vector-shift-ashr-128.ll

 ; X32-SSE-NEXT:    andnps %xmm2, %xmm1
 ; X32-SSE-NEXT:    orps %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
   ret <8 x i16> %shift
 }
 
 define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) nounwind {
-; SSE2-LABEL: constant_shift_v16i8:
-; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [8192,24640,41088,57536,49376,32928,16480,32]
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm3[8],xmm4[9],xmm3[9],xmm4[10],xmm3[10],xmm4[11],xmm3[11],xmm4[12],xmm3[12],xmm4[13],xmm3[13],xmm4[14],xmm3[14],xmm4[15],xmm3[15]
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    pxor %xmm5, %xmm5
-; SSE2-NEXT:    pcmpgtw %xmm4, %xmm5
-; SSE2-NEXT:    movdqa %xmm5, %xmm6
-; SSE2-NEXT:    pandn %xmm1, %xmm6
-; SSE2-NEXT:    psraw $4, %xmm1
-; SSE2-NEXT:    pand %xmm5, %xmm1
-; SSE2-NEXT:    por %xmm6, %xmm1
-; SSE2-NEXT:    paddw %xmm4, %xmm4
-; SSE2-NEXT:    pxor %xmm5, %xmm5
-; SSE2-NEXT:    pcmpgtw %xmm4, %xmm5
-; SSE2-NEXT:    movdqa %xmm5, %xmm6
-; SSE2-NEXT:    pandn %xmm1, %xmm6
-; SSE2-NEXT:    psraw $2, %xmm1
-; SSE2-NEXT:    pand %xmm5, %xmm1
-; SSE2-NEXT:    por %xmm6, %xmm1
-; SSE2-NEXT:    paddw %xmm4, %xmm4
-; SSE2-NEXT:    pxor %xmm5, %xmm5
-; SSE2-NEXT:    pcmpgtw %xmm4, %xmm5
-; SSE2-NEXT:    movdqa %xmm5, %xmm4
-; SSE2-NEXT:    pandn %xmm1, %xmm4
-; SSE2-NEXT:    psraw $1, %xmm1
-; SSE2-NEXT:    pand %xmm5, %xmm1
-; SSE2-NEXT:    por %xmm4, %xmm1
-; SSE2-NEXT:    psrlw $8, %xmm1
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    pcmpgtw %xmm3, %xmm4
-; SSE2-NEXT:    movdqa %xmm4, %xmm5
-; SSE2-NEXT:    pandn %xmm0, %xmm5
-; SSE2-NEXT:    psraw $4, %xmm0
-; SSE2-NEXT:    pand %xmm4, %xmm0
-; SSE2-NEXT:    por %xmm5, %xmm0
-; SSE2-NEXT:    paddw %xmm3, %xmm3
-; SSE2-NEXT:    pxor %xmm4, %xmm4
-; SSE2-NEXT:    pcmpgtw %xmm3, %xmm4
-; SSE2-NEXT:    movdqa %xmm4, %xmm5
-; SSE2-NEXT:    pandn %xmm0, %xmm5
-; SSE2-NEXT:    psraw $2, %xmm0
-; SSE2-NEXT:    pand %xmm4, %xmm0
-; SSE2-NEXT:    por %xmm5, %xmm0
-; SSE2-NEXT:    paddw %xmm3, %xmm3
-; SSE2-NEXT:    pcmpgtw %xmm3, %xmm2
-; SSE2-NEXT:    movdqa %xmm2, %xmm3
-; SSE2-NEXT:    pandn %xmm0, %xmm3
-; SSE2-NEXT:    psraw $1, %xmm0
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    por %xmm3, %xmm0
-; SSE2-NEXT:    psrlw $8, %xmm0
-; SSE2-NEXT:    packuswb %xmm1, %xmm0
-; SSE2-NEXT:    retq
+; SSE-LABEL: constant_shift_v16i8:
+; SSE:       # %bb.0:
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
+; SSE-NEXT:    psraw $8, %xmm1
+; SSE-NEXT:    pmullw {{.*}}(%rip), %xmm1
+; SSE-NEXT:    psrlw $8, %xmm1
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    psraw $8, %xmm0
+; SSE-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE-NEXT:    psrlw $8, %xmm0
+; SSE-NEXT:    packuswb %xmm1, %xmm0
+; SSE-NEXT:    retq
 ;
-; SSE41-LABEL: constant_shift_v16i8:
-; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [8192,24640,41088,57536,49376,32928,16480,32]
-; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm3[8],xmm0[9],xmm3[9],xmm0[10],xmm3[10],xmm0[11],xmm3[11],xmm0[12],xmm3[12],xmm0[13],xmm3[13],xmm0[14],xmm3[14],xmm0[15],xmm3[15]
-; SSE41-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
-; SSE41-NEXT:    movdqa %xmm2, %xmm4
-; SSE41-NEXT:    psraw $4, %xmm4
-; SSE41-NEXT:    pblendvb %xmm0, %xmm4, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm4
-; SSE41-NEXT:    psraw $2, %xmm4
-; SSE41-NEXT:    paddw %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm4, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm4
-; SSE41-NEXT:    psraw $1, %xmm4
-; SSE41-NEXT:    paddw %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm4, %xmm2
-; SSE41-NEXT:    psrlw $8, %xmm2
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    psraw $4, %xmm3
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    psraw $2, %xmm3
-; SSE41-NEXT:    paddw %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    psraw $1, %xmm3
-; SSE41-NEXT:    paddw %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
-; SSE41-NEXT:    psrlw $8, %xmm1
-; SSE41-NEXT:    packuswb %xmm2, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
-; SSE41-NEXT:    retq
+; AVX1-LABEL: constant_shift_v16i8:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vpsraw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpsraw $8, %xmm0, %xmm0
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
+; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    retq
 ;
-; AVX-LABEL: constant_shift_v16i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [8192,24640,41088,57536,49376,32928,16480,32]
-; AVX-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
-; AVX-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; AVX-NEXT:    vpsraw $4, %xmm3, %xmm4
-; AVX-NEXT:    vpblendvb %xmm2, %xmm4, %xmm3, %xmm3
-; AVX-NEXT:    vpsraw $2, %xmm3, %xmm4
-; AVX-NEXT:    vpaddw %xmm2, %xmm2, %xmm2
-; AVX-NEXT:    vpblendvb %xmm2, %xmm4, %xmm3, %xmm3
-; AVX-NEXT:    vpsraw $1, %xmm3, %xmm4
-; AVX-NEXT:    vpaddw %xmm2, %xmm2, %xmm2
-; AVX-NEXT:    vpblendvb %xmm2, %xmm4, %xmm3, %xmm2
-; AVX-NEXT:    vpsrlw $8, %xmm2, %xmm2
-; AVX-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX-NEXT:    vpsraw $4, %xmm0, %xmm3
-; AVX-NEXT:    vpblendvb %xmm1, %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vpsraw $2, %xmm0, %xmm3
-; AVX-NEXT:    vpaddw %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpblendvb %xmm1, %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vpsraw $1, %xmm0, %xmm3
-; AVX-NEXT:    vpaddw %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpblendvb %xmm1, %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $8, %xmm0, %xmm0
-; AVX-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    retq
+; AVX2-LABEL: constant_shift_v16i8:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpmovsxbw %xmm0, %ymm0
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: constant_shift_v16i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpshab {{.*}}(%rip), %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: constant_shift_v16i8:
 ; AVX512DQ:       # %bb.0:

 ; AVX512BWVL-NEXT:    vpmovsxbw %xmm0, %ymm0
 ; AVX512BWVL-NEXT:    vpsravw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_shift_v16i8:
 ; X32-SSE:       # %bb.0:
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm3 = [8192,24640,41088,57536,49376,32928,16480,32]
-; X32-SSE-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm3[8],xmm4[9],xmm3[9],xmm4[10],xmm3[10],xmm4[11],xmm3[11],xmm4[12],xmm3[12],xmm4[13],xmm3[13],xmm4[14],xmm3[14],xmm4[15],xmm3[15]
-; X32-SSE-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE-NEXT:    pxor %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpgtw %xmm4, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm5, %xmm6
-; X32-SSE-NEXT:    pandn %xmm1, %xmm6
-; X32-SSE-NEXT:    psraw $4, %xmm1
-; X32-SSE-NEXT:    pand %xmm5, %xmm1
-; X32-SSE-NEXT:    por %xmm6, %xmm1
-; X32-SSE-NEXT:    paddw %xmm4, %xmm4
-; X32-SSE-NEXT:    pxor %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpgtw %xmm4, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm5, %xmm6
-; X32-SSE-NEXT:    pandn %xmm1, %xmm6
-; X32-SSE-NEXT:    psraw $2, %xmm1
-; X32-SSE-NEXT:    pand %xmm5, %xmm1
-; X32-SSE-NEXT:    por %xmm6, %xmm1
-; X32-SSE-NEXT:    paddw %xmm4, %xmm4
-; X32-SSE-NEXT:    pxor %xmm5, %xmm5
-; X32-SSE-NEXT:    pcmpgtw %xmm4, %xmm5
-; X32-SSE-NEXT:    movdqa %xmm5, %xmm4
-; X32-SSE-NEXT:    pandn %xmm1, %xmm4
-; X32-SSE-NEXT:    psraw $1, %xmm1
-; X32-SSE-NEXT:    pand %xmm5, %xmm1
-; X32-SSE-NEXT:    por %xmm4, %xmm1
+; X32-SSE-NEXT:    psraw $8, %xmm1
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm1
 ; X32-SSE-NEXT:    psrlw $8, %xmm1
 ; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE-NEXT:    pxor %xmm4, %xmm4
-; X32-SSE-NEXT:    pcmpgtw %xmm3, %xmm4
-; X32-SSE-NEXT:    movdqa %xmm4, %xmm5
-; X32-SSE-NEXT:    pandn %xmm0, %xmm5
-; X32-SSE-NEXT:    psraw $4, %xmm0
-; X32-SSE-NEXT:    pand %xmm4, %xmm0
-; X32-SSE-NEXT:    por %xmm5, %xmm0
-; X32-SSE-NEXT:    paddw %xmm3, %xmm3
-; X32-SSE-NEXT:    pxor %xmm4, %xmm4
-; X32-SSE-NEXT:    pcmpgtw %xmm3, %xmm4
-; X32-SSE-NEXT:    movdqa %xmm4, %xmm5
-; X32-SSE-NEXT:    pandn %xmm0, %xmm5
-; X32-SSE-NEXT:    psraw $2, %xmm0
-; X32-SSE-NEXT:    pand %xmm4, %xmm0
-; X32-SSE-NEXT:    por %xmm5, %xmm0
-; X32-SSE-NEXT:    paddw %xmm3, %xmm3
-; X32-SSE-NEXT:    pcmpgtw %xmm3, %xmm2
-; X32-SSE-NEXT:    movdqa %xmm2, %xmm3
-; X32-SSE-NEXT:    pandn %xmm0, %xmm3
-; X32-SSE-NEXT:    psraw $1, %xmm0
-; X32-SSE-NEXT:    pand %xmm2, %xmm0
-; X32-SSE-NEXT:    por %xmm3, %xmm0
+; X32-SSE-NEXT:    psraw $8, %xmm0
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm0
 ; X32-SSE-NEXT:    psrlw $8, %xmm0
 ; X32-SSE-NEXT:    packuswb %xmm1, %xmm0
 ; X32-SSE-NEXT:    retl
   %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
   ret <16 x i8> %shift
 }
 
 ;

llvm/trunk/test/CodeGen/X86/vector-shift-ashr-256.ll

 ; X32-AVX2-NEXT:    retl
   %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
   ret <16 x i16> %shift
 }
 
 define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) nounwind {
 ; AVX1-LABEL: constant_shift_v32i8:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [8192,24640,41088,57536,49376,32928,16480,32]
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm0[8],xmm3[8],xmm0[9],xmm3[9],xmm0[10],xmm3[10],xmm0[11],xmm3[11],xmm0[12],xmm3[12],xmm0[13],xmm3[13],xmm0[14],xmm3[14],xmm0[15],xmm3[15]
-; AVX1-NEXT:    vpsraw $4, %xmm4, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm2, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsraw $2, %xmm4, %xmm5
-; AVX1-NEXT:    vpaddw %xmm2, %xmm2, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsraw $1, %xmm4, %xmm5
-; AVX1-NEXT:    vpaddw %xmm6, %xmm6, %xmm9
-; AVX1-NEXT:    vpblendvb %xmm9, %xmm5, %xmm4, %xmm4
-; AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm8
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
-; AVX1-NEXT:    vpsraw $4, %xmm3, %xmm5
-; AVX1-NEXT:    vpblendvb %xmm1, %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpsraw $2, %xmm3, %xmm5
-; AVX1-NEXT:    vpaddw %xmm1, %xmm1, %xmm4
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpsraw $1, %xmm3, %xmm5
-; AVX1-NEXT:    vpaddw %xmm4, %xmm4, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm5, %xmm3, %xmm3
-; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; AVX1-NEXT:    vpackuswb %xmm8, %xmm3, %xmm8
-; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; AVX1-NEXT:    vpsraw $4, %xmm5, %xmm3
-; AVX1-NEXT:    vpblendvb %xmm2, %xmm3, %xmm5, %xmm2
-; AVX1-NEXT:    vpsraw $2, %xmm2, %xmm3
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vpsraw $1, %xmm2, %xmm3
-; AVX1-NEXT:    vpblendvb %xmm9, %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vpsraw $8, %xmm2, %xmm2
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [2,4,8,16,32,64,128,256]
+; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
+; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; AVX1-NEXT:    vpsraw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [256,128,64,32,16,8,4,2]
+; AVX1-NEXT:    vpmullw %xmm4, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vpsraw $8, %xmm2, %xmm2
+; AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
 ; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpsraw $4, %xmm0, %xmm3
-; AVX1-NEXT:    vpblendvb %xmm1, %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpsraw $2, %xmm0, %xmm1
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpsraw $1, %xmm0, %xmm1
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpsraw $8, %xmm0, %xmm0
+; AVX1-NEXT:    vpmullw %xmm4, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vinsertf128 $1, %xmm8, %ymm0, %ymm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_shift_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
-; AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX2-NEXT:    vpsraw $4, %ymm3, %ymm4
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; AVX2-NEXT:    vpsraw $2, %ymm3, %ymm4
-; AVX2-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; AVX2-NEXT:    vpsraw $1, %ymm3, %ymm4
-; AVX2-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm2
-; AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23]
+; AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX2-NEXT:    vpsraw $8, %ymm1, %ymm1
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
+; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX2-NEXT:    vpsraw $4, %ymm0, %ymm3
-; AVX2-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; AVX2-NEXT:    vpsraw $2, %ymm0, %ymm3
-; AVX2-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; AVX2-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; AVX2-NEXT:    vpsraw $1, %ymm0, %ymm3
-; AVX2-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; AVX2-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpsraw $8, %ymm0, %ymm0
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX2-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_shift_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,254,253,252,251,250,249,249,250,251,252,253,254,255,0]
 ; XOPAVX1-NEXT:    vpshab %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpshab %xmm2, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: constant_shift_v32i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,254,253,252,251,250,249,249,250,251,252,253,254,255,0]
 ; XOPAVX2-NEXT:    vpshab %xmm2, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpshab %xmm2, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: constant_shift_v32i8:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm1 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512DQ-NEXT:    vpsraw $4, %ymm3, %ymm4
-; AVX512DQ-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpsraw $2, %ymm3, %ymm4
-; AVX512DQ-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpsraw $1, %ymm3, %ymm4
-; AVX512DQ-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm2
-; AVX512DQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23]
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512DQ-NEXT:    vpsraw $8, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512DQ-NEXT:    vpsraw $4, %ymm0, %ymm3
-; AVX512DQ-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsraw $2, %ymm0, %ymm3
-; AVX512DQ-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsraw $1, %ymm0, %ymm3
-; AVX512DQ-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpsraw $8, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_shift_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovsxbw %ymm0, %zmm0
 ; AVX512BW-NEXT:    vpsravw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQVL-LABEL: constant_shift_v32i8:
 ; AVX512DQVL:       # %bb.0:
-; AVX512DQVL-NEXT:    vmovdqa {{.*#+}} ymm1 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; AVX512DQVL-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
-; AVX512DQVL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512DQVL-NEXT:    vpsraw $4, %ymm3, %ymm4
-; AVX512DQVL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; AVX512DQVL-NEXT:    vpsraw $2, %ymm3, %ymm4
-; AVX512DQVL-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; AVX512DQVL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; AVX512DQVL-NEXT:    vpsraw $1, %ymm3, %ymm4
-; AVX512DQVL-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; AVX512DQVL-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm2
-; AVX512DQVL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23]
+; AVX512DQVL-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512DQVL-NEXT:    vpsraw $8, %ymm1, %ymm1
+; AVX512DQVL-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
+; AVX512DQVL-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512DQVL-NEXT:    vpsraw $4, %ymm0, %ymm3
-; AVX512DQVL-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; AVX512DQVL-NEXT:    vpsraw $2, %ymm0, %ymm3
-; AVX512DQVL-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; AVX512DQVL-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; AVX512DQVL-NEXT:    vpsraw $1, %ymm0, %ymm3
-; AVX512DQVL-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; AVX512DQVL-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
+; AVX512DQVL-NEXT:    vpsraw $8, %ymm0, %ymm0
+; AVX512DQVL-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512DQVL-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512DQVL-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; AVX512DQVL-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
 ; AVX512DQVL-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: constant_shift_v32i8:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    vpmovsxbw %ymm0, %zmm0
 ; AVX512BWVL-NEXT:    vpsravw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BWVL-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BWVL-NEXT:    retq
 ;
 ; X32-AVX1-LABEL: constant_shift_v32i8:
 ; X32-AVX1:       # %bb.0:
-; X32-AVX1-NEXT:    vmovdqa {{.*#+}} xmm1 = [8192,24640,41088,57536,49376,32928,16480,32]
-; X32-AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
-; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; X32-AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm0[8],xmm3[8],xmm0[9],xmm3[9],xmm0[10],xmm3[10],xmm0[11],xmm3[11],xmm0[12],xmm3[12],xmm0[13],xmm3[13],xmm0[14],xmm3[14],xmm0[15],xmm3[15]
-; X32-AVX1-NEXT:    vpsraw $4, %xmm4, %xmm5
-; X32-AVX1-NEXT:    vpblendvb %xmm2, %xmm5, %xmm4, %xmm4
-; X32-AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; X32-AVX1-NEXT:    vpsraw $4, %xmm5, %xmm6
-; X32-AVX1-NEXT:    vpblendvb %xmm2, %xmm6, %xmm5, %xmm5
-; X32-AVX1-NEXT:    vpsraw $2, %xmm4, %xmm6
-; X32-AVX1-NEXT:    vpaddw %xmm2, %xmm2, %xmm2
-; X32-AVX1-NEXT:    vpblendvb %xmm2, %xmm6, %xmm4, %xmm4
-; X32-AVX1-NEXT:    vpsraw $2, %xmm5, %xmm6
-; X32-AVX1-NEXT:    vpblendvb %xmm2, %xmm6, %xmm5, %xmm5
-; X32-AVX1-NEXT:    vpsraw $1, %xmm4, %xmm6
-; X32-AVX1-NEXT:    vpaddw %xmm2, %xmm2, %xmm2
-; X32-AVX1-NEXT:    vpblendvb %xmm2, %xmm6, %xmm4, %xmm4
-; X32-AVX1-NEXT:    vpsraw $1, %xmm5, %xmm6
-; X32-AVX1-NEXT:    vpblendvb %xmm2, %xmm6, %xmm5, %xmm2
-; X32-AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; X32-AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
-; X32-AVX1-NEXT:    vpsraw $4, %xmm3, %xmm5
-; X32-AVX1-NEXT:    vpblendvb %xmm1, %xmm5, %xmm3, %xmm3
-; X32-AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-AVX1-NEXT:    vpsraw $4, %xmm0, %xmm5
-; X32-AVX1-NEXT:    vpblendvb %xmm1, %xmm5, %xmm0, %xmm0
-; X32-AVX1-NEXT:    vpsraw $2, %xmm3, %xmm5
-; X32-AVX1-NEXT:    vpaddw %xmm1, %xmm1, %xmm1
-; X32-AVX1-NEXT:    vpblendvb %xmm1, %xmm5, %xmm3, %xmm3
-; X32-AVX1-NEXT:    vpsraw $2, %xmm0, %xmm5
-; X32-AVX1-NEXT:    vpblendvb %xmm1, %xmm5, %xmm0, %xmm0
-; X32-AVX1-NEXT:    vpsraw $1, %xmm3, %xmm5
-; X32-AVX1-NEXT:    vpaddw %xmm1, %xmm1, %xmm1
-; X32-AVX1-NEXT:    vpblendvb %xmm1, %xmm5, %xmm3, %xmm3
-; X32-AVX1-NEXT:    vpsraw $1, %xmm0, %xmm5
-; X32-AVX1-NEXT:    vpblendvb %xmm1, %xmm5, %xmm0, %xmm0
-; X32-AVX1-NEXT:    vpsrlw $8, %xmm4, %xmm1
-; X32-AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
-; X32-AVX1-NEXT:    vpackuswb %xmm1, %xmm3, %xmm1
+; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; X32-AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X32-AVX1-NEXT:    vpsraw $8, %xmm2, %xmm2
+; X32-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [2,4,8,16,32,64,128,256]
+; X32-AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
 ; X32-AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; X32-AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-AVX1-NEXT:    vpsraw $8, %xmm1, %xmm1
+; X32-AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [256,128,64,32,16,8,4,2]
+; X32-AVX1-NEXT:    vpmullw %xmm4, %xmm1, %xmm1
+; X32-AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; X32-AVX1-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
+; X32-AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; X32-AVX1-NEXT:    vpsraw $8, %xmm2, %xmm2
+; X32-AVX1-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
+; X32-AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; X32-AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-AVX1-NEXT:    vpsraw $8, %xmm0, %xmm0
+; X32-AVX1-NEXT:    vpmullw %xmm4, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; X32-AVX1-NEXT:    retl
 ;
 ; X32-AVX2-LABEL: constant_shift_v32i8:
 ; X32-AVX2:       # %bb.0:
-; X32-AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; X32-AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
-; X32-AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; X32-AVX2-NEXT:    vpsraw $4, %ymm3, %ymm4
-; X32-AVX2-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; X32-AVX2-NEXT:    vpsraw $2, %ymm3, %ymm4
-; X32-AVX2-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; X32-AVX2-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; X32-AVX2-NEXT:    vpsraw $1, %ymm3, %ymm4
-; X32-AVX2-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; X32-AVX2-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm2
-; X32-AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; X32-AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23]
+; X32-AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; X32-AVX2-NEXT:    vpsraw $8, %ymm1, %ymm1
+; X32-AVX2-NEXT:    vpmullw {{\.LCPI.*}}, %ymm1, %ymm1
+; X32-AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; X32-AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; X32-AVX2-NEXT:    vpsraw $4, %ymm0, %ymm3
-; X32-AVX2-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vpsraw $2, %ymm0, %ymm3
-; X32-AVX2-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vpsraw $1, %ymm0, %ymm3
-; X32-AVX2-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vpsraw $8, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vpmullw {{\.LCPI.*}}, %ymm0, %ymm0
 ; X32-AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
 ; X32-AVX2-NEXT:    retl
   %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
   ret <32 x i8> %shift
 }
 
 ;
 ; Uniform Constant Shifts
 ;

llvm/trunk/test/CodeGen/X86/vector-shift-ashr-512.ll

 ; AVX512BW-NEXT:    retq
   %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
   ret <32 x i16> %shift
 }
 
 define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) nounwind {
 ; AVX512DQ-LABEL: constant_shift_v64i8:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8],ymm2[8],ymm0[9],ymm2[9],ymm0[10],ymm2[10],ymm0[11],ymm2[11],ymm0[12],ymm2[12],ymm0[13],ymm2[13],ymm0[14],ymm2[14],ymm0[15],ymm2[15],ymm0[24],ymm2[24],ymm0[25],ymm2[25],ymm0[26],ymm2[26],ymm0[27],ymm2[27],ymm0[28],ymm2[28],ymm0[29],ymm2[29],ymm0[30],ymm2[30],ymm0[31],ymm2[31]
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512DQ-NEXT:    vpsraw $4, %ymm4, %ymm5
-; AVX512DQ-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
-; AVX512DQ-NEXT:    vpsraw $2, %ymm4, %ymm5
-; AVX512DQ-NEXT:    vpaddw %ymm3, %ymm3, %ymm6
-; AVX512DQ-NEXT:    vpblendvb %ymm6, %ymm5, %ymm4, %ymm4
-; AVX512DQ-NEXT:    vpsraw $1, %ymm4, %ymm5
-; AVX512DQ-NEXT:    vpaddw %ymm6, %ymm6, %ymm7
-; AVX512DQ-NEXT:    vpblendvb %ymm7, %ymm5, %ymm4, %ymm4
-; AVX512DQ-NEXT:    vpsrlw $8, %ymm4, %ymm4
-; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm0[0],ymm2[0],ymm0[1],ymm2[1],ymm0[2],ymm2[2],ymm0[3],ymm2[3],ymm0[4],ymm2[4],ymm0[5],ymm2[5],ymm0[6],ymm2[6],ymm0[7],ymm2[7],ymm0[16],ymm2[16],ymm0[17],ymm2[17],ymm0[18],ymm2[18],ymm0[19],ymm2[19],ymm0[20],ymm2[20],ymm0[21],ymm2[21],ymm0[22],ymm2[22],ymm0[23],ymm2[23]
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512DQ-NEXT:    vpsraw $8, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [2,4,8,16,32,64,128,256,2,4,8,16,32,64,128,256]
+; AVX512DQ-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpmullw %ymm3, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
 ; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512DQ-NEXT:    vpsraw $4, %ymm0, %ymm5
-; AVX512DQ-NEXT:    vpblendvb %ymm2, %ymm5, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsraw $2, %ymm0, %ymm5
-; AVX512DQ-NEXT:    vpaddw %ymm2, %ymm2, %ymm8
-; AVX512DQ-NEXT:    vpblendvb %ymm8, %ymm5, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsraw $1, %ymm0, %ymm5
-; AVX512DQ-NEXT:    vpaddw %ymm8, %ymm8, %ymm9
-; AVX512DQ-NEXT:    vpblendvb %ymm9, %ymm5, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpsraw $8, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512DQ-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpmullw %ymm4, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpackuswb %ymm4, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
-; AVX512DQ-NEXT:    vpsraw $4, %ymm4, %ymm5
-; AVX512DQ-NEXT:    vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
-; AVX512DQ-NEXT:    vpsraw $2, %ymm3, %ymm4
-; AVX512DQ-NEXT:    vpblendvb %ymm6, %ymm4, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpsraw $1, %ymm3, %ymm4
-; AVX512DQ-NEXT:    vpblendvb %ymm7, %ymm4, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23]
-; AVX512DQ-NEXT:    vpsraw $4, %ymm1, %ymm4
-; AVX512DQ-NEXT:    vpblendvb %ymm2, %ymm4, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpsraw $2, %ymm1, %ymm2
-; AVX512DQ-NEXT:    vpblendvb %ymm8, %ymm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpsraw $1, %ymm1, %ymm2
-; AVX512DQ-NEXT:    vpblendvb %ymm9, %ymm2, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512DQ-NEXT:    vpsraw $8, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpmullw %ymm3, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
+; AVX512DQ-NEXT:    vpsraw $8, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpmullw %ymm4, %ymm1, %ymm1
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpackuswb %ymm2, %ymm1, %ymm1
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_shift_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsraw $4, %zmm1, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm0[8],zmm3[8],zmm0[9],zmm3[9],zmm0[10],zmm3[10],zmm0[11],zmm3[11],zmm0[12],zmm3[12],zmm0[13],zmm3[13],zmm0[14],zmm3[14],zmm0[15],zmm3[15],zmm0[24],zmm3[24],zmm0[25],zmm3[25],zmm0[26],zmm3[26],zmm0[27],zmm3[27],zmm0[28],zmm3[28],zmm0[29],zmm3[29],zmm0[30],zmm3[30],zmm0[31],zmm3[31],zmm0[40],zmm3[40],zmm0[41],zmm3[41],zmm0[42],zmm3[42],zmm0[43],zmm3[43],zmm0[44],zmm3[44],zmm0[45],zmm3[45],zmm0[46],zmm3[46],zmm0[47],zmm3[47],zmm0[56],zmm3[56],zmm0[57],zmm3[57],zmm0[58],zmm3[58],zmm0[59],zmm3[59],zmm0[60],zmm3[60],zmm0[61],zmm3[61],zmm0[62],zmm3[62],zmm0[63],zmm3[63]
-; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm1 {%k1}
-; AVX512BW-NEXT:    vpsraw $2, %zmm1, %zmm2
-; AVX512BW-NEXT:    vpaddw %zmm4, %zmm4, %zmm4
-; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm1 {%k1}
-; AVX512BW-NEXT:    vpsraw $1, %zmm1, %zmm2
-; AVX512BW-NEXT:    vpaddw %zmm4, %zmm4, %zmm4
-; AVX512BW-NEXT:    vpmovb2m %zmm4, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm1 {%k1}
+; AVX512BW-NEXT:    vpsraw $8, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsraw $4, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm0[0],zmm3[0],zmm0[1],zmm3[1],zmm0[2],zmm3[2],zmm0[3],zmm3[3],zmm0[4],zmm3[4],zmm0[5],zmm3[5],zmm0[6],zmm3[6],zmm0[7],zmm3[7],zmm0[16],zmm3[16],zmm0[17],zmm3[17],zmm0[18],zmm3[18],zmm0[19],zmm3[19],zmm0[20],zmm3[20],zmm0[21],zmm3[21],zmm0[22],zmm3[22],zmm0[23],zmm3[23],zmm0[32],zmm3[32],zmm0[33],zmm3[33],zmm0[34],zmm3[34],zmm0[35],zmm3[35],zmm0[36],zmm3[36],zmm0[37],zmm3[37],zmm0[38],zmm3[38],zmm0[39],zmm3[39],zmm0[48],zmm3[48],zmm0[49],zmm3[49],zmm0[50],zmm3[50],zmm0[51],zmm3[51],zmm0[52],zmm3[52],zmm0[53],zmm3[53],zmm0[54],zmm3[54],zmm0[55],zmm3[55]
-; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
-; AVX512BW-NEXT:    vpsraw $2, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpaddw %zmm3, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
-; AVX512BW-NEXT:    vpsraw $1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpaddw %zmm3, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpmovb2m %zmm3, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpsraw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
   %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
   ret <64 x i8> %shift
 }
 
 ;

llvm/trunk/test/CodeGen/X86/vector-shift-lshr-128.ll

 ; X32-SSE-NEXT:    retl
   %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
   ret <8 x i16> %shift
 }
 
 define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) nounwind {
 ; SSE2-LABEL: constant_shift_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [8192,24640,41088,57536,49376,32928,16480,32]
 ; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpgtb %xmm2, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm4
-; SSE2-NEXT:    pandn %xmm0, %xmm4
-; SSE2-NEXT:    psrlw $4, %xmm0
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    por %xmm4, %xmm0
-; SSE2-NEXT:    paddb %xmm2, %xmm2
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpgtb %xmm2, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm4
-; SSE2-NEXT:    pandn %xmm0, %xmm4
-; SSE2-NEXT:    psrlw $2, %xmm0
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT:    pand %xmm3, %xmm0
-; SSE2-NEXT:    por %xmm4, %xmm0
-; SSE2-NEXT:    paddb %xmm2, %xmm2
-; SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm2
-; SSE2-NEXT:    pandn %xmm0, %xmm2
-; SSE2-NEXT:    psrlw $1, %xmm0
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
-; SSE2-NEXT:    pand %xmm1, %xmm0
-; SSE2-NEXT:    por %xmm2, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm2
+; SSE2-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm2
+; SSE2-NEXT:    psrlw $8, %xmm2
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    psrlw $8, %xmm0
+; SSE2-NEXT:    packuswb %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: constant_shift_v16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    movdqa %xmm0, %xmm2
-; SSE41-NEXT:    psrlw $4, %xmm2
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [8192,24640,41088,57536,49376,32928,16480,32]
-; SSE41-NEXT:    pblendvb %xmm0, %xmm2, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm2
-; SSE41-NEXT:    psrlw $2, %xmm2
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    paddb %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm2, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm2
-; SSE41-NEXT:    psrlw $1, %xmm2
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    paddb %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm2, %xmm1
+; SSE41-NEXT:    pxor %xmm2, %xmm2
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE41-NEXT:    psrlw $8, %xmm0
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    packuswb %xmm0, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: constant_shift_v16i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm1
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [8192,24640,41088,57536,49376,32928,16480,32]
-; AVX-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $2, %xmm0, %xmm1
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
-; AVX-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $1, %xmm0, %xmm1
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
-; AVX-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    retq
+; AVX1-LABEL: constant_shift_v16i8:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
+; AVX1-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: constant_shift_v16i8:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
 ;
 ; XOP-LABEL: constant_shift_v16i8:
 ; XOP:       # %bb.0:
 ; XOP-NEXT:    vpshlb {{.*}}(%rip), %xmm0, %xmm0
 ; XOP-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: constant_shift_v16i8:
 ; AVX512DQ:       # %bb.0:

 ; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX512BWVL-NEXT:    vpsrlvw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
 ;
 ; X32-SSE-LABEL: constant_shift_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [8192,24640,41088,57536,49376,32928,16480,32]
 ; X32-SSE-NEXT:    pxor %xmm1, %xmm1
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    pandn %xmm0, %xmm4
-; X32-SSE-NEXT:    psrlw $4, %xmm0
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
-; X32-SSE-NEXT:    pand %xmm3, %xmm0
-; X32-SSE-NEXT:    por %xmm4, %xmm0
-; X32-SSE-NEXT:    paddb %xmm2, %xmm2
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    pandn %xmm0, %xmm4
-; X32-SSE-NEXT:    psrlw $2, %xmm0
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
-; X32-SSE-NEXT:    pand %xmm3, %xmm0
-; X32-SSE-NEXT:    por %xmm4, %xmm0
-; X32-SSE-NEXT:    paddb %xmm2, %xmm2
-; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm2
-; X32-SSE-NEXT:    pandn %xmm0, %xmm2
-; X32-SSE-NEXT:    psrlw $1, %xmm0
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm0
-; X32-SSE-NEXT:    pand %xmm1, %xmm0
-; X32-SSE-NEXT:    por %xmm2, %xmm0
+; X32-SSE-NEXT:    movdqa %xmm0, %xmm2
+; X32-SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm2
+; X32-SSE-NEXT:    psrlw $8, %xmm2
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    psrlw $8, %xmm0
+; X32-SSE-NEXT:    packuswb %xmm2, %xmm0
 ; X32-SSE-NEXT:    retl
   %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
   ret <16 x i8> %shift
 }
 
 ;
 ; Uniform Constant Shifts
 ;

llvm/trunk/test/CodeGen/X86/vector-shift-lshr-256.ll

; X32-AVX2-NEXT:    retl
   %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
   ret <16 x i16> %shift
 }
 
 define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) nounwind {
 ; AVX1-LABEL: constant_shift_v32i8:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX1-NEXT:    vpand %xmm8, %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [8192,24640,41088,57536,49376,32928,16480,32]
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX1-NEXT:    vpand %xmm7, %xmm2, %xmm2
-; AVX1-NEXT:    vpaddb %xmm6, %xmm6, %xmm3
-; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm2
-; AVX1-NEXT:    vpand %xmm8, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $2, %xmm0, %xmm2
-; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $1, %xmm0, %xmm2
-; AVX1-NEXT:    vpand %xmm7, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm3, %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm2[8],xmm1[9],xmm2[9],xmm1[10],xmm2[10],xmm1[11],xmm2[11],xmm1[12],xmm2[12],xmm1[13],xmm2[13],xmm1[14],xmm2[14],xmm1[15],xmm2[15]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [2,4,8,16,32,64,128,256]
+; AVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [256,128,64,32,16,8,4,2]
+; AVX1-NEXT:    vpmullw %xmm5, %xmm1, %xmm1
+; AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
+; AVX1-NEXT:    vpmullw %xmm4, %xmm2, %xmm2
+; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX1-NEXT:    vpmullw %xmm5, %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
+; AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_shift_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm1
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpsrlw $2, %ymm0, %ymm1
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX2-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpsrlw $1, %ymm0, %ymm1
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX2-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
+; AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23]
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX2-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: constant_shift_v32i8:
 ; XOPAVX1:       # %bb.0:
 ; XOPAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; XOPAVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,254,253,252,251,250,249,249,250,251,252,253,254,255,0]
 ; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
 ; XOPAVX1-NEXT:    vpshlb %xmm2, %xmm0, %xmm0
 ; XOPAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; XOPAVX1-NEXT:    retq
 ;
 ; XOPAVX2-LABEL: constant_shift_v32i8:
 ; XOPAVX2:       # %bb.0:
 ; XOPAVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; XOPAVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,255,254,253,252,251,250,249,249,250,251,252,253,254,255,0]
 ; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm1, %xmm1
 ; XOPAVX2-NEXT:    vpshlb %xmm2, %xmm0, %xmm0
 ; XOPAVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; XOPAVX2-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: constant_shift_v32i8:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm1
-; AVX512DQ-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; AVX512DQ-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsrlw $2, %ymm0, %ymm1
-; AVX512DQ-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsrlw $1, %ymm0, %ymm1
-; AVX512DQ-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
+; AVX512DQ-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23]
+; AVX512DQ-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_shift_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQVL-LABEL: constant_shift_v32i8:
 ; AVX512DQVL:       # %bb.0:
-; AVX512DQVL-NEXT:    vpsrlw $4, %ymm0, %ymm1
-; AVX512DQVL-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512DQVL-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; AVX512DQVL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX512DQVL-NEXT:    vpsrlw $2, %ymm0, %ymm1
-; AVX512DQVL-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512DQVL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX512DQVL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX512DQVL-NEXT:    vpsrlw $1, %ymm0, %ymm1
-; AVX512DQVL-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512DQVL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX512DQVL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512DQVL-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512DQVL-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512DQVL-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
+; AVX512DQVL-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
+; AVX512DQVL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512DQVL-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512DQVL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512DQVL-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
 ; AVX512DQVL-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: constant_shift_v32i8:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BWVL-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BWVL-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BWVL-NEXT:    retq
 ;
 ; X32-AVX1-LABEL: constant_shift_v32i8:
 ; X32-AVX1:       # %bb.0:
 ; X32-AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; X32-AVX1-NEXT:    vpsrlw $4, %xmm1, %xmm2
-; X32-AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; X32-AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [8192,24640,41088,57536,49376,32928,16480,32]
-; X32-AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm1, %xmm1
-; X32-AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm2
-; X32-AVX1-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; X32-AVX1-NEXT:    vpsrlw $2, %xmm1, %xmm3
-; X32-AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm0, %xmm0
-; X32-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; X32-AVX1-NEXT:    vpand %xmm2, %xmm3, %xmm3
-; X32-AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
-; X32-AVX1-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
-; X32-AVX1-NEXT:    vpsrlw $2, %xmm0, %xmm3
-; X32-AVX1-NEXT:    vpand %xmm2, %xmm3, %xmm2
-; X32-AVX1-NEXT:    vpsrlw $1, %xmm1, %xmm3
-; X32-AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm0, %xmm0
-; X32-AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; X32-AVX1-NEXT:    vpand %xmm2, %xmm3, %xmm3
-; X32-AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm4
-; X32-AVX1-NEXT:    vpblendvb %xmm4, %xmm3, %xmm1, %xmm1
-; X32-AVX1-NEXT:    vpsrlw $1, %xmm0, %xmm3
-; X32-AVX1-NEXT:    vpand %xmm2, %xmm3, %xmm2
-; X32-AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm0, %xmm0
+; X32-AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; X32-AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm2[8],xmm1[9],xmm2[9],xmm1[10],xmm2[10],xmm1[11],xmm2[11],xmm1[12],xmm2[12],xmm1[13],xmm2[13],xmm1[14],xmm2[14],xmm1[15],xmm2[15]
+; X32-AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [2,4,8,16,32,64,128,256]
+; X32-AVX1-NEXT:    vpmullw %xmm4, %xmm3, %xmm3
+; X32-AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
+; X32-AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; X32-AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [256,128,64,32,16,8,4,2]
+; X32-AVX1-NEXT:    vpmullw %xmm5, %xmm1, %xmm1
+; X32-AVX1-NEXT:    vpsrlw $8, %xmm1, %xmm1
+; X32-AVX1-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
+; X32-AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
+; X32-AVX1-NEXT:    vpmullw %xmm4, %xmm2, %xmm2
+; X32-AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; X32-AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; X32-AVX1-NEXT:    vpmullw %xmm5, %xmm0, %xmm0
+; X32-AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
+; X32-AVX1-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; X32-AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; X32-AVX1-NEXT:    retl
 ;
 ; X32-AVX2-LABEL: constant_shift_v32i8:
 ; X32-AVX2:       # %bb.0:
-; X32-AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm1
-; X32-AVX2-NEXT:    vpand {{\.LCPI.*}}, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; X32-AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vpsrlw $2, %ymm0, %ymm1
-; X32-AVX2-NEXT:    vpand {{\.LCPI.*}}, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; X32-AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; X32-AVX2-NEXT:    vpsrlw $1, %ymm0, %ymm1
-; X32-AVX2-NEXT:    vpand {{\.LCPI.*}}, %ymm1, %ymm1
-; X32-AVX2-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; X32-AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X32-AVX2-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
+; X32-AVX2-NEXT:    vpmullw {{\.LCPI.*}}, %ymm2, %ymm2
+; X32-AVX2-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; X32-AVX2-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23]
+; X32-AVX2-NEXT:    vpmullw {{\.LCPI.*}}, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; X32-AVX2-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; X32-AVX2-NEXT:    retl
   %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
   ret <32 x i8> %shift
 }
 
 ;
 ; Uniform Constant Shifts
 ;

llvm/trunk/test/CodeGen/X86/vector-shift-lshr-512.ll

 ; AVX512BW-NEXT:    retq
   %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
   ret <32 x i16> %shift
 }
 
 define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) nounwind {
 ; AVX512DQ-LABEL: constant_shift_v64i8:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpsrlw $4, %ymm0, %ymm2
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512DQ-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm4 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; AVX512DQ-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsrlw $2, %ymm0, %ymm2
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm5 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
-; AVX512DQ-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpaddb %ymm4, %ymm4, %ymm6
-; AVX512DQ-NEXT:    vpblendvb %ymm6, %ymm2, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsrlw $1, %ymm0, %ymm2
-; AVX512DQ-NEXT:    vmovdqa {{.*#+}} ymm7 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; AVX512DQ-NEXT:    vpand %ymm7, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpaddb %ymm6, %ymm6, %ymm8
-; AVX512DQ-NEXT:    vpblendvb %ymm8, %ymm2, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpsrlw $4, %ymm1, %ymm2
-; AVX512DQ-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpblendvb %ymm4, %ymm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpsrlw $2, %ymm1, %ymm2
-; AVX512DQ-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpblendvb %ymm6, %ymm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpsrlw $1, %ymm1, %ymm2
-; AVX512DQ-NEXT:    vpand %ymm7, %ymm2, %ymm2
-; AVX512DQ-NEXT:    vpblendvb %ymm8, %ymm2, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8],ymm2[8],ymm0[9],ymm2[9],ymm0[10],ymm2[10],ymm0[11],ymm2[11],ymm0[12],ymm2[12],ymm0[13],ymm2[13],ymm0[14],ymm2[14],ymm0[15],ymm2[15],ymm0[24],ymm2[24],ymm0[25],ymm2[25],ymm0[26],ymm2[26],ymm0[27],ymm2[27],ymm0[28],ymm2[28],ymm0[29],ymm2[29],ymm0[30],ymm2[30],ymm0[31],ymm2[31]
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [2,4,8,16,32,64,128,256,2,4,8,16,32,64,128,256]
+; AVX512DQ-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpmullw %ymm4, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm2[0],ymm0[1],ymm2[1],ymm0[2],ymm2[2],ymm0[3],ymm2[3],ymm0[4],ymm2[4],ymm0[5],ymm2[5],ymm0[6],ymm2[6],ymm0[7],ymm2[7],ymm0[16],ymm2[16],ymm0[17],ymm2[17],ymm0[18],ymm2[18],ymm0[19],ymm2[19],ymm0[20],ymm2[20],ymm0[21],ymm2[21],ymm0[22],ymm2[22],ymm0[23],ymm2[23]
+; AVX512DQ-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512DQ-NEXT:    # ymm5 = mem[0,1,0,1]
+; AVX512DQ-NEXT:    vpmullw %ymm5, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
+; AVX512DQ-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15],ymm1[24],ymm2[24],ymm1[25],ymm2[25],ymm1[26],ymm2[26],ymm1[27],ymm2[27],ymm1[28],ymm2[28],ymm1[29],ymm2[29],ymm1[30],ymm2[30],ymm1[31],ymm2[31]
+; AVX512DQ-NEXT:    vpmullw %ymm4, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
+; AVX512DQ-NEXT:    vpmullw %ymm5, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512DQ-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_shift_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32]
-; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512BW-NEXT:    vpsrlw $4, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
-; AVX512BW-NEXT:    vpsrlw $2, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
-; AVX512BW-NEXT:    vpsrlw $1, %zmm0, %zmm2
-; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
-; AVX512BW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k1}
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
+; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
+; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
+; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
   %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
   ret <64 x i8> %shift
 }
 
 ;
 ; Uniform Constant Shifts
 ;