From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Tue, 18 Dec 2018 17:31:11 +0000 (+0000)
Subject: [X86][SSE] Don't use 'sign bit select' vXi8 ROTL lowering for constant rotation amounts
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=1b35b9e855149e041d9be4c74fcd248252956e72;p=llvm

[X86][SSE] Don't use 'sign bit select' vXi8 ROTL lowering for constant rotation amounts

Noticed by @spatel on D55747 - we get much better codegen if we use the regular shift expansion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@349510 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 3d8b1787bb0..cc044e1dbf6 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -24818,6 +24818,9 @@ static SDValue LowerRotate(SDValue Op, const X86Subtarget &Subtarget,
   // v16i8/v32i8: Split rotation into rot4/rot2/rot1 stages and select by
   // the amount bit.
   if (EltSizeInBits == 8 && !IsSplatAmt) {
+    if (ISD::isBuildVectorOfConstantSDNodes(Amt.getNode()))
+      return SDValue();
+
     // We don't need ModuloAmt here as we just peek at individual bits.
     MVT ExtVT = MVT::getVectorVT(MVT::i16, NumElts / 2);
 
diff --git a/test/CodeGen/X86/vector-rotate-128.ll b/test/CodeGen/X86/vector-rotate-128.ll
index 650fe3b616c..6258229392c 100644
--- a/test/CodeGen/X86/vector-rotate-128.ll
+++ b/test/CodeGen/X86/vector-rotate-128.ll
@@ -1487,148 +1487,104 @@ define <8 x i16> @constant_rotate_v8i16(<8 x i16> %a) nounwind {
 define <16 x i8> @constant_rotate_v16i8(<16 x i8> %a) nounwind {
 ; SSE2-LABEL: constant_rotate_v16i8:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [8192,24640,41088,57536,57600,41152,24704,8256]
-; SSE2-NEXT:    pxor %xmm0, %xmm0
-; SSE2-NEXT:    pxor %xmm3, %xmm3
-; SSE2-NEXT:    pcmpgtb %xmm2, %xmm3
-; SSE2-NEXT:    movdqa %xmm1, %xmm4
-; SSE2-NEXT:    psrlw $4, %xmm4
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE2-NEXT:    movdqa %xmm1, %xmm5
-; SSE2-NEXT:    psllw $4, %xmm5
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm5
-; SSE2-NEXT:    por %xmm4, %xmm5
-; SSE2-NEXT:    pand %xmm3, %xmm5
-; SSE2-NEXT:    pandn %xmm1, %xmm3
-; SSE2-NEXT:    por %xmm5, %xmm3
-; SSE2-NEXT:    movdqa %xmm3, %xmm1
-; SSE2-NEXT:    psrlw $6, %xmm1
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    movdqa %xmm3, %xmm4
-; SSE2-NEXT:    psllw $2, %xmm4
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE2-NEXT:    por %xmm1, %xmm4
-; SSE2-NEXT:    paddb %xmm2, %xmm2
 ; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    pcmpgtb %xmm2, %xmm1
-; SSE2-NEXT:    pand %xmm1, %xmm4
-; SSE2-NEXT:    pandn %xmm3, %xmm1
-; SSE2-NEXT:    por %xmm4, %xmm1
-; SSE2-NEXT:    movdqa %xmm1, %xmm3
-; SSE2-NEXT:    paddb %xmm1, %xmm3
-; SSE2-NEXT:    movdqa %xmm1, %xmm4
-; SSE2-NEXT:    psrlw $7, %xmm4
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm4
-; SSE2-NEXT:    por %xmm3, %xmm4
-; SSE2-NEXT:    paddb %xmm2, %xmm2
-; SSE2-NEXT:    pcmpgtb %xmm2, %xmm0
-; SSE2-NEXT:    pand %xmm0, %xmm4
-; SSE2-NEXT:    pandn %xmm1, %xmm0
-; SSE2-NEXT:    por %xmm4, %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:    movdqa %xmm0, %xmm3
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3],xmm3[4],xmm1[4],xmm3[5],xmm1[5],xmm3[6],xmm1[6],xmm3[7],xmm1[7]
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm3
+; SSE2-NEXT:    psrlw $8, %xmm3
+; SSE2-NEXT:    packuswb %xmm2, %xmm3
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; 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:    pmullw {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
+; SSE2-NEXT:    pand %xmm2, %xmm1
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pmullw {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pand %xmm2, %xmm0
+; SSE2-NEXT:    packuswb %xmm1, %xmm0
+; SSE2-NEXT:    por %xmm3, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: constant_rotate_v16i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    psrlw $4, %xmm0
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm0
-; SSE41-NEXT:    movdqa %xmm1, %xmm2
-; SSE41-NEXT:    psllw $4, %xmm2
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    por %xmm0, %xmm2
-; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = [8192,24640,41088,57536,57600,41152,24704,8256]
-; SSE41-NEXT:    pblendvb %xmm0, %xmm2, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm2
-; SSE41-NEXT:    psrlw $6, %xmm2
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm2
-; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    psllw $2, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm3
-; SSE41-NEXT:    por %xmm2, %xmm3
-; SSE41-NEXT:    paddb %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm2
-; SSE41-NEXT:    paddb %xmm1, %xmm2
-; SSE41-NEXT:    movdqa %xmm1, %xmm3
-; SSE41-NEXT:    psrlw $7, %xmm3
-; SSE41-NEXT:    pand {{.*}}(%rip), %xmm3
-; SSE41-NEXT:    por %xmm2, %xmm3
-; SSE41-NEXT:    paddb %xmm0, %xmm0
-; SSE41-NEXT:    pblendvb %xmm0, %xmm3, %xmm1
+; SSE41-NEXT:    movdqa %xmm0, %xmm2
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm0[8],xmm2[9],xmm0[9],xmm2[10],xmm0[10],xmm2[11],xmm0[11],xmm2[12],xmm0[12],xmm2[13],xmm0[13],xmm2[14],xmm0[14],xmm2[15],xmm0[15]
+; SSE41-NEXT:    pmullw {{.*}}(%rip), %xmm2
+; SSE41-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
+; SSE41-NEXT:    pand %xmm3, %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:    movdqa {{.*#+}} xmm4 = [1,2,4,8,16,32,64,128]
+; SSE41-NEXT:    pmullw %xmm1, %xmm4
+; SSE41-NEXT:    pand %xmm3, %xmm4
+; SSE41-NEXT:    packuswb %xmm2, %xmm4
+; SSE41-NEXT:    pxor %xmm2, %xmm2
+; 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:    por %xmm4, %xmm1
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: constant_rotate_v16i8:
-; AVX:       # %bb.0:
-; AVX-NEXT:    vpsrlw $4, %xmm0, %xmm1
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX-NEXT:    vpsllw $4, %xmm0, %xmm2
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = [8192,24640,41088,57536,57600,41152,24704,8256]
-; AVX-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpsrlw $6, %xmm0, %xmm1
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX-NEXT:    vpsllw $2, %xmm0, %xmm3
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
-; AVX-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpaddb %xmm0, %xmm0, %xmm1
-; AVX-NEXT:    vpsrlw $7, %xmm0, %xmm3
-; AVX-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX-NEXT:    vpor %xmm3, %xmm1, %xmm1
-; AVX-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
-; AVX-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    retq
+; AVX1-LABEL: constant_rotate_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:    vpmullw {{.*}}(%rip), %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
+; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} 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
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm3, %xmm4
+; AVX1-NEXT:    vpand %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpackuswb %xmm1, %xmm2, %xmm1
+; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} 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]
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm0, %xmm0
+; AVX1-NEXT:    vpsrlw $8, %xmm0, %xmm0
+; AVX1-NEXT:    vpmullw {{.*}}(%rip), %xmm3, %xmm2
+; AVX1-NEXT:    vpsrlw $8, %xmm2, %xmm2
+; AVX1-NEXT:    vpackuswb %xmm0, %xmm2, %xmm0
+; AVX1-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: constant_rotate_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, %ymm1
+; AVX2-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
+; AVX2-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX2-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
+; AVX2-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_rotate_v16i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpsrlw $4, %xmm0, %xmm1
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX512F-NEXT:    vpsllw $4, %xmm0, %xmm2
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512F-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [8192,24640,41088,57536,57600,41152,24704,8256]
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX512F-NEXT:    vpsrlw $6, %xmm0, %xmm1
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX512F-NEXT:    vpsllw $2, %xmm0, %xmm3
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX512F-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512F-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX512F-NEXT:    vpaddb %xmm0, %xmm0, %xmm1
-; AVX512F-NEXT:    vpsrlw $7, %xmm0, %xmm3
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX512F-NEXT:    vpor %xmm3, %xmm1, %xmm1
-; AVX512F-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
-; AVX512F-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512F-NEXT:    vpsrlvd {{.*}}(%rip), %zmm0, %zmm1
+; AVX512F-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm0
+; AVX512F-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_rotate_v16i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsrlw $4, %xmm0, %xmm1
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX512VL-NEXT:    vpsllw $4, %xmm0, %xmm2
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
-; AVX512VL-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [8192,24640,41088,57536,57600,41152,24704,8256]
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpsrlw $6, %xmm0, %xmm1
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
-; AVX512VL-NEXT:    vpsllw $2, %xmm0, %xmm3
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX512VL-NEXT:    vpor %xmm1, %xmm3, %xmm1
-; AVX512VL-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpaddb %xmm0, %xmm0, %xmm1
-; AVX512VL-NEXT:    vpsrlw $7, %xmm0, %xmm3
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm3, %xmm3
-; AVX512VL-NEXT:    vpor %xmm3, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpaddb %xmm2, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; AVX512VL-NEXT:    vpsrlvd {{.*}}(%rip), %zmm0, %zmm1
+; AVX512VL-NEXT:    vpsllvd {{.*}}(%rip), %zmm0, %zmm0
+; AVX512VL-NEXT:    vpord %zmm1, %zmm0, %zmm0
+; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_rotate_v16i8:
@@ -1661,45 +1617,26 @@ define <16 x i8> @constant_rotate_v16i8(<16 x i8> %a) nounwind {
 ;
 ; X32-SSE-LABEL: constant_rotate_v16i8:
 ; X32-SSE:       # %bb.0:
-; X32-SSE-NEXT:    movdqa %xmm0, %xmm1
-; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [8192,24640,41088,57536,57600,41152,24704,8256]
-; X32-SSE-NEXT:    pxor %xmm0, %xmm0
-; X32-SSE-NEXT:    pxor %xmm3, %xmm3
-; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE-NEXT:    psrlw $4, %xmm4
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm4
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm5
-; X32-SSE-NEXT:    psllw $4, %xmm5
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm5
-; X32-SSE-NEXT:    por %xmm4, %xmm5
-; X32-SSE-NEXT:    pand %xmm3, %xmm5
-; X32-SSE-NEXT:    pandn %xmm1, %xmm3
-; X32-SSE-NEXT:    por %xmm5, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm1
-; X32-SSE-NEXT:    psrlw $6, %xmm1
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm3, %xmm4
-; X32-SSE-NEXT:    psllw $2, %xmm4
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm4
-; X32-SSE-NEXT:    por %xmm1, %xmm4
-; X32-SSE-NEXT:    paddb %xmm2, %xmm2
 ; X32-SSE-NEXT:    pxor %xmm1, %xmm1
-; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm1
-; X32-SSE-NEXT:    pand %xmm1, %xmm4
-; X32-SSE-NEXT:    pandn %xmm3, %xmm1
-; X32-SSE-NEXT:    por %xmm4, %xmm1
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm3
-; X32-SSE-NEXT:    paddb %xmm1, %xmm3
-; X32-SSE-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE-NEXT:    psrlw $7, %xmm4
-; X32-SSE-NEXT:    pand {{\.LCPI.*}}, %xmm4
-; X32-SSE-NEXT:    por %xmm3, %xmm4
-; X32-SSE-NEXT:    paddb %xmm2, %xmm2
-; X32-SSE-NEXT:    pcmpgtb %xmm2, %xmm0
-; X32-SSE-NEXT:    pand %xmm0, %xmm4
-; X32-SSE-NEXT:    pandn %xmm1, %xmm0
-; X32-SSE-NEXT:    por %xmm4, %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:    movdqa %xmm0, %xmm3
+; X32-SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3],xmm3[4],xmm1[4],xmm3[5],xmm1[5],xmm3[6],xmm1[6],xmm3[7],xmm1[7]
+; X32-SSE-NEXT:    pmullw {{\.LCPI.*}}, %xmm3
+; X32-SSE-NEXT:    psrlw $8, %xmm3
+; X32-SSE-NEXT:    packuswb %xmm2, %xmm3
+; 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:    pmullw {{\.LCPI.*}}, %xmm1
+; X32-SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
+; X32-SSE-NEXT:    pand %xmm2, %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:    pmullw {{\.LCPI.*}}, %xmm0
+; X32-SSE-NEXT:    pand %xmm2, %xmm0
+; X32-SSE-NEXT:    packuswb %xmm1, %xmm0
+; X32-SSE-NEXT:    por %xmm3, %xmm0
 ; X32-SSE-NEXT:    retl
   %shl = shl <16 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>
   %lshr = lshr <16 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>
diff --git a/test/CodeGen/X86/vector-rotate-256.ll b/test/CodeGen/X86/vector-rotate-256.ll
index 49dd2d2661d..b0d9cd9a3cb 100644
--- a/test/CodeGen/X86/vector-rotate-256.ll
+++ b/test/CodeGen/X86/vector-rotate-256.ll
@@ -1138,121 +1138,114 @@ define <32 x i8> @constant_rotate_v32i8(<32 x i8> %a) nounwind {
 ; AVX1-LABEL: constant_rotate_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:    vpsllw $4, %xmm1, %xmm4
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm9 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX1-NEXT:    vpand %xmm9, %xmm4, %xmm4
-; AVX1-NEXT:    vpor %xmm2, %xmm4, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [8192,24640,41088,57536,57600,41152,24704,8256]
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $6, %xmm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm10 = [3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3]
-; AVX1-NEXT:    vpand %xmm10, %xmm2, %xmm2
-; AVX1-NEXT:    vpsllw $2, %xmm1, %xmm7
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm11 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX1-NEXT:    vpand %xmm11, %xmm7, %xmm7
-; AVX1-NEXT:    vpor %xmm2, %xmm7, %xmm2
-; AVX1-NEXT:    vpaddb %xmm4, %xmm4, %xmm7
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $7, %xmm1, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm5 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX1-NEXT:    vpaddb %xmm1, %xmm1, %xmm6
-; AVX1-NEXT:    vpor %xmm2, %xmm6, %xmm2
-; AVX1-NEXT:    vpaddb %xmm7, %xmm7, %xmm6
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpsrlw $4, %xmm0, %xmm2
-; AVX1-NEXT:    vpand %xmm8, %xmm2, %xmm2
-; AVX1-NEXT:    vpsllw $4, %xmm0, %xmm3
-; AVX1-NEXT:    vpand %xmm9, %xmm3, %xmm3
-; AVX1-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $6, %xmm0, %xmm2
-; AVX1-NEXT:    vpand %xmm10, %xmm2, %xmm2
-; AVX1-NEXT:    vpsllw $2, %xmm0, %xmm3
-; AVX1-NEXT:    vpand %xmm11, %xmm3, %xmm3
-; AVX1-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm7, %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpsrlw $7, %xmm0, %xmm2
-; AVX1-NEXT:    vpand %xmm5, %xmm2, %xmm2
-; AVX1-NEXT:    vpaddb %xmm0, %xmm0, %xmm3
-; AVX1-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpxor %xmm8, %xmm8, %xmm8
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm8[8],xmm1[9],xmm8[9],xmm1[10],xmm8[10],xmm1[11],xmm8[11],xmm1[12],xmm8[12],xmm1[13],xmm8[13],xmm1[14],xmm8[14],xmm1[15],xmm8[15]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm9 = [256,128,64,32,16,8,4,2]
+; AVX1-NEXT:    vpmullw %xmm9, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm5 = 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 {{.*#+}} xmm6 = [256,2,4,8,16,32,64,128]
+; AVX1-NEXT:    vpmullw %xmm6, %xmm5, %xmm7
+; AVX1-NEXT:    vpsrlw $8, %xmm7, %xmm7
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm7, %xmm3
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} 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]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm7 = [1,128,64,32,16,8,4,2]
+; AVX1-NEXT:    vpmullw %xmm7, %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
+; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [1,2,4,8,16,32,64,128]
+; AVX1-NEXT:    vpmullw %xmm4, %xmm5, %xmm5
+; AVX1-NEXT:    vpand %xmm2, %xmm5, %xmm5
+; AVX1-NEXT:    vpackuswb %xmm1, %xmm5, %xmm1
+; AVX1-NEXT:    vpor %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8],xmm8[8],xmm0[9],xmm8[9],xmm0[10],xmm8[10],xmm0[11],xmm8[11],xmm0[12],xmm8[12],xmm0[13],xmm8[13],xmm0[14],xmm8[14],xmm0[15],xmm8[15]
+; AVX1-NEXT:    vpmullw %xmm9, %xmm3, %xmm3
+; AVX1-NEXT:    vpsrlw $8, %xmm3, %xmm3
+; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm5 = 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 %xmm6, %xmm5, %xmm6
+; AVX1-NEXT:    vpsrlw $8, %xmm6, %xmm6
+; AVX1-NEXT:    vpackuswb %xmm3, %xmm6, %xmm3
+; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; AVX1-NEXT:    vpmullw %xmm7, %xmm0, %xmm0
+; AVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
+; AVX1-NEXT:    vpmullw %xmm4, %xmm5, %xmm4
+; AVX1-NEXT:    vpand %xmm2, %xmm4, %xmm2
+; AVX1-NEXT:    vpackuswb %xmm0, %xmm2, %xmm0
+; AVX1-NEXT:    vpor %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: constant_rotate_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpsrlw $4, %ymm0, %ymm1
+; AVX2-NEXT:    vpsllw $4, %ymm0, %ymm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX2-NEXT:    vpsllw $4, %ymm0, %ymm2
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX2-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256]
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpsrlw $6, %ymm0, %ymm1
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX2-NEXT:    vpsllw $2, %ymm0, %ymm3
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
+; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm1
+; AVX2-NEXT:    vpsllw $2, %ymm1, %ymm3
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
-; AVX2-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX2-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX2-NEXT:    vpaddb %ymm0, %ymm0, %ymm1
-; AVX2-NEXT:    vpsrlw $7, %ymm0, %ymm3
-; AVX2-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
-; AVX2-NEXT:    vpor %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpaddb %ymm1, %ymm1, %ymm3
 ; AVX2-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX2-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
+; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX2-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]
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
+; AVX2-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX2-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]
+; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX2-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
+; AVX2-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: constant_rotate_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm1
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm1
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm2
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512F-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256]
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsrlw $6, %ymm0, %ymm1
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm3
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm1
+; AVX512F-NEXT:    vpsllw $2, %ymm1, %ymm3
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
-; AVX512F-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm1
-; AVX512F-NEXT:    vpsrlw $7, %ymm0, %ymm3
-; AVX512F-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
-; AVX512F-NEXT:    vpor %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm3
 ; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512F-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512F-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]
+; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512F-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]
+; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512F-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_rotate_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm1
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm2
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256]
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpsrlw $6, %ymm0, %ymm1
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm1
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm3
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm1
+; AVX512VL-NEXT:    vpsllw $2, %ymm1, %ymm3
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
-; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm1
-; AVX512VL-NEXT:    vpsrlw $7, %ymm0, %ymm3
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
-; AVX512VL-NEXT:    vpor %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
+; AVX512VL-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]
+; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512VL-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]
+; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_rotate_v32i8:
diff --git a/test/CodeGen/X86/vector-rotate-512.ll b/test/CodeGen/X86/vector-rotate-512.ll
index 0ff6fc1bed1..896d6c07387 100644
--- a/test/CodeGen/X86/vector-rotate-512.ll
+++ b/test/CodeGen/X86/vector-rotate-512.ll
@@ -545,94 +545,97 @@ define <32 x i16> @constant_rotate_v32i16(<32 x i16> %a) nounwind {
 define <64 x i8> @constant_rotate_v64i8(<64 x i8> %a) nounwind {
 ; AVX512F-LABEL: constant_rotate_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpsrlw $4, %ymm0, %ymm2
-; AVX512F-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]
+; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllw $4, %ymm0, %ymm4
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX512F-NEXT:    vpand %ymm5, %ymm4, %ymm4
-; AVX512F-NEXT:    vpor %ymm2, %ymm4, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256]
-; AVX512F-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsrlw $6, %ymm0, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm6 = [3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3]
-; AVX512F-NEXT:    vpand %ymm6, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllw $2, %ymm0, %ymm7
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm8 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX512F-NEXT:    vpand %ymm8, %ymm7, %ymm7
-; AVX512F-NEXT:    vpor %ymm2, %ymm7, %ymm2
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm4 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
+; AVX512F-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm2
+; AVX512F-NEXT:    vpsllw $2, %ymm2, %ymm5
+; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX512F-NEXT:    vpand %ymm6, %ymm5, %ymm5
 ; AVX512F-NEXT:    vpaddb %ymm4, %ymm4, %ymm7
-; AVX512F-NEXT:    vpblendvb %ymm7, %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsrlw $7, %ymm0, %ymm2
-; AVX512F-NEXT:    vmovdqa {{.*#+}} ymm9 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX512F-NEXT:    vpand %ymm9, %ymm2, %ymm2
-; AVX512F-NEXT:    vpaddb %ymm0, %ymm0, %ymm10
-; AVX512F-NEXT:    vpor %ymm2, %ymm10, %ymm2
-; AVX512F-NEXT:    vpaddb %ymm7, %ymm7, %ymm10
-; AVX512F-NEXT:    vpblendvb %ymm10, %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpsrlw $4, %ymm1, %ymm2
+; AVX512F-NEXT:    vpblendvb %ymm7, %ymm5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm5
+; AVX512F-NEXT:    vpaddb %ymm7, %ymm7, %ymm8
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm5, %ymm2, %ymm2
+; AVX512F-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm9 = ymm0[8],ymm5[8],ymm0[9],ymm5[9],ymm0[10],ymm5[10],ymm0[11],ymm5[11],ymm0[12],ymm5[12],ymm0[13],ymm5[13],ymm0[14],ymm5[14],ymm0[15],ymm5[15],ymm0[24],ymm5[24],ymm0[25],ymm5[25],ymm0[26],ymm5[26],ymm0[27],ymm5[27],ymm0[28],ymm5[28],ymm0[29],ymm5[29],ymm0[30],ymm5[30],ymm0[31],ymm5[31]
+; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512F-NEXT:    # ymm10 = mem[0,1,0,1]
+; AVX512F-NEXT:    vpmullw %ymm10, %ymm9, %ymm9
+; AVX512F-NEXT:    vpsrlw $8, %ymm9, %ymm9
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm5[0],ymm0[1],ymm5[1],ymm0[2],ymm5[2],ymm0[3],ymm5[3],ymm0[4],ymm5[4],ymm0[5],ymm5[5],ymm0[6],ymm5[6],ymm0[7],ymm5[7],ymm0[16],ymm5[16],ymm0[17],ymm5[17],ymm0[18],ymm5[18],ymm0[19],ymm5[19],ymm0[20],ymm5[20],ymm0[21],ymm5[21],ymm0[22],ymm5[22],ymm0[23],ymm5[23]
+; AVX512F-NEXT:    vbroadcasti128 {{.*#+}} ymm11 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512F-NEXT:    # ymm11 = mem[0,1,0,1]
+; AVX512F-NEXT:    vpmullw %ymm11, %ymm0, %ymm0
+; AVX512F-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512F-NEXT:    vpackuswb %ymm9, %ymm0, %ymm0
+; AVX512F-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512F-NEXT:    vpsllw $4, %ymm1, %ymm2
 ; AVX512F-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllw $4, %ymm1, %ymm3
-; AVX512F-NEXT:    vpand %ymm5, %ymm3, %ymm3
-; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm4, %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsrlw $6, %ymm1, %ymm2
-; AVX512F-NEXT:    vpand %ymm6, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllw $2, %ymm1, %ymm3
-; AVX512F-NEXT:    vpand %ymm8, %ymm3, %ymm3
-; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm7, %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpsrlw $7, %ymm1, %ymm2
-; AVX512F-NEXT:    vpand %ymm9, %ymm2, %ymm2
-; AVX512F-NEXT:    vpaddb %ymm1, %ymm1, %ymm3
-; AVX512F-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512F-NEXT:    vpblendvb %ymm10, %ymm2, %ymm1, %ymm1
+; AVX512F-NEXT:    vpblendvb %ymm4, %ymm2, %ymm1, %ymm2
+; AVX512F-NEXT:    vpsllw $2, %ymm2, %ymm3
+; AVX512F-NEXT:    vpand %ymm6, %ymm3, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm7, %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512F-NEXT:    vpblendvb %ymm8, %ymm3, %ymm2, %ymm2
+; AVX512F-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm5[8],ymm1[9],ymm5[9],ymm1[10],ymm5[10],ymm1[11],ymm5[11],ymm1[12],ymm5[12],ymm1[13],ymm5[13],ymm1[14],ymm5[14],ymm1[15],ymm5[15],ymm1[24],ymm5[24],ymm1[25],ymm5[25],ymm1[26],ymm5[26],ymm1[27],ymm5[27],ymm1[28],ymm5[28],ymm1[29],ymm5[29],ymm1[30],ymm5[30],ymm1[31],ymm5[31]
+; AVX512F-NEXT:    vpmullw %ymm10, %ymm3, %ymm3
+; AVX512F-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512F-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm5[0],ymm1[1],ymm5[1],ymm1[2],ymm5[2],ymm1[3],ymm5[3],ymm1[4],ymm5[4],ymm1[5],ymm5[5],ymm1[6],ymm5[6],ymm1[7],ymm5[7],ymm1[16],ymm5[16],ymm1[17],ymm5[17],ymm1[18],ymm5[18],ymm1[19],ymm5[19],ymm1[20],ymm5[20],ymm1[21],ymm5[21],ymm1[22],ymm5[22],ymm1[23],ymm5[23]
+; AVX512F-NEXT:    vpmullw %ymm11, %ymm1, %ymm1
+; AVX512F-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512F-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512F-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: constant_rotate_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpsrlw $4, %ymm0, %ymm2
-; AVX512VL-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]
+; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm4
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm5 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
-; AVX512VL-NEXT:    vpand %ymm5, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpor %ymm2, %ymm4, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256]
-; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpsrlw $6, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3]
-; AVX512VL-NEXT:    vpand %ymm6, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm7
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm8 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
-; AVX512VL-NEXT:    vpand %ymm8, %ymm7, %ymm7
-; AVX512VL-NEXT:    vpor %ymm2, %ymm7, %ymm2
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm4 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
+; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm2
+; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm5
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
+; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
 ; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm7
-; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpsrlw $7, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm9 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; AVX512VL-NEXT:    vpand %ymm9, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm10
-; AVX512VL-NEXT:    vpor %ymm2, %ymm10, %ymm2
-; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm10
-; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpsrlw $4, %ymm1, %ymm2
+; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm5
+; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm8
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm5, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm5 = 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]
+; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512VL-NEXT:    # ymm9 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm9, %ymm5, %ymm5
+; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
+; AVX512VL-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]
+; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    # ymm10 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm10, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpackuswb %ymm5, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512VL-NEXT:    vpsllw $4, %ymm1, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsllw $4, %ymm1, %ymm3
-; AVX512VL-NEXT:    vpand %ymm5, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm2, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsrlw $6, %ymm1, %ymm2
-; AVX512VL-NEXT:    vpand %ymm6, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsllw $2, %ymm1, %ymm3
-; AVX512VL-NEXT:    vpand %ymm8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm2, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpsrlw $7, %ymm1, %ymm2
-; AVX512VL-NEXT:    vpand %ymm9, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm3
-; AVX512VL-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm2, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm2, %ymm1, %ymm2
+; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm3
+; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
+; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = 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]
+; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpmullw %ymm9, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512VL-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]
+; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpmullw %ymm10, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_rotate_v64i8: