From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sat, 19 Oct 2019 13:18:02 +0000 (+0000)
Subject: [X86][SSE] lowerV16I8Shuffle - tryToWidenViaDuplication - undef unpack args
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=200d59f918b7df9672f4872a3e4be2eaf3433b65;p=llvm

[X86][SSE] lowerV16I8Shuffle - tryToWidenViaDuplication - undef unpack args

tryToWidenViaDuplication lowers using the shuffle_v8i16(unpack_v16i8(shuffle_v8i16(x),shuffle_v8i16(x))) pattern, but the unpack only needs the even/odd 16i8 args if the original v16i8 shuffle mask references the even/odd elements - which isn't true for many extension style shuffles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@375342 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 49608d7c28a..1c9ee5d3ce6 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -14269,8 +14269,16 @@ static SDValue lowerV16I8Shuffle(const SDLoc &DL, ArrayRef<int> Mask,
                                DAG.getUNDEF(MVT::v8i16), PreDupI16Shuffle));
 
       // Unpack the bytes to form the i16s that will be shuffled into place.
+      bool EvenInUse = false, OddInUse = false;
+      for (int i = 0; i < 16; i += 2) {
+        EvenInUse |= (Mask[i + 0] >= 0);
+        OddInUse |= (Mask[i + 1] >= 0);
+        if (EvenInUse && OddInUse)
+          break;
+      }
       V1 = DAG.getNode(TargetLo ? X86ISD::UNPCKL : X86ISD::UNPCKH, DL,
-                       MVT::v16i8, V1, V1);
+                       MVT::v16i8, EvenInUse ? V1 : DAG.getUNDEF(MVT::v16i8),
+                       OddInUse ? V1 : DAG.getUNDEF(MVT::v16i8));
 
       int PostDupI16Shuffle[8] = {-1, -1, -1, -1, -1, -1, -1, -1};
       for (int i = 0; i < 16; ++i)
diff --git a/test/CodeGen/X86/bitcast-and-setcc-128.ll b/test/CodeGen/X86/bitcast-and-setcc-128.ll
index 8565d6ead1f..7d5886d7e59 100644
--- a/test/CodeGen/X86/bitcast-and-setcc-128.ll
+++ b/test/CodeGen/X86/bitcast-and-setcc-128.ll
@@ -185,8 +185,8 @@ define i2 @v2i8(<2 x i8> %a, <2 x i8> %b, <2 x i8> %c, <2 x i8> %d) {
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]
 ; SSE2-NEXT:    pcmpgtb %xmm3, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
 ; SSE2-NEXT:    pand %xmm0, %xmm1
 ; SSE2-NEXT:    movmskpd %xmm1, %eax
@@ -196,7 +196,7 @@ define i2 @v2i8(<2 x i8> %a, <2 x i8> %b, <2 x i8> %c, <2 x i8> %d) {
 ; SSSE3-LABEL: v2i8:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    pcmpgtb %xmm1, %xmm0
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = <u,u,0,0,u,u,0,0,u,u,1,1,u,u,1,1>
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = <u,u,u,0,u,u,u,0,u,u,u,1,u,u,u,1>
 ; SSSE3-NEXT:    pshufb %xmm1, %xmm0
 ; SSSE3-NEXT:    pcmpgtb %xmm3, %xmm2
 ; SSSE3-NEXT:    pshufb %xmm1, %xmm2
@@ -450,8 +450,8 @@ define i4 @v4i8(<4 x i8> %a, <4 x i8> %b, <4 x i8> %c, <4 x i8> %d) {
 ; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
 ; SSE2-SSSE3-NEXT:    pcmpgtb %xmm3, %xmm2
-; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; SSE2-SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; SSE2-SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3]
 ; SSE2-SSSE3-NEXT:    pand %xmm0, %xmm1
 ; SSE2-SSSE3-NEXT:    movmskps %xmm1, %eax
 ; SSE2-SSSE3-NEXT:    # kill: def $al killed $al killed $eax
diff --git a/test/CodeGen/X86/bitcast-setcc-128.ll b/test/CodeGen/X86/bitcast-setcc-128.ll
index 0dcdb3ec9ee..d7871ff21e8 100644
--- a/test/CodeGen/X86/bitcast-setcc-128.ll
+++ b/test/CodeGen/X86/bitcast-setcc-128.ll
@@ -157,7 +157,7 @@ define i2 @v2i8(<2 x i8> %a, <2 x i8> %b) {
 ; SSSE3-LABEL: v2i8:
 ; SSSE3:       # %bb.0:
 ; SSSE3-NEXT:    pcmpgtb %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[u,u,0,0,u,u,0,0,u,u,1,1,u,u,1,1]
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[u,u,u,0,u,u,u,0,u,u,u,1,u,u,u,1]
 ; SSSE3-NEXT:    movmskpd %xmm0, %eax
 ; SSSE3-NEXT:    # kill: def $al killed $al killed $eax
 ; SSSE3-NEXT:    retq
diff --git a/test/CodeGen/X86/vec_int_to_fp.ll b/test/CodeGen/X86/vec_int_to_fp.ll
index 6b863456dfa..269879e7f1a 100644
--- a/test/CodeGen/X86/vec_int_to_fp.ll
+++ b/test/CodeGen/X86/vec_int_to_fp.ll
@@ -1708,14 +1708,13 @@ define <8 x float> @sitofp_8i16_to_8f32(<8 x i16> %a) {
 define <8 x float> @sitofp_8i8_to_8f32(<16 x i8> %a) {
 ; SSE2-LABEL: sitofp_8i8_to_8f32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE2-NEXT:    psrad $24, %xmm1
-; SSE2-NEXT:    cvtdq2ps %xmm1, %xmm2
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; SSE2-NEXT:    psrad $24, %xmm0
-; SSE2-NEXT:    cvtdq2ps %xmm0, %xmm1
-; SSE2-NEXT:    movaps %xmm2, %xmm0
+; SSE2-NEXT:    cvtdq2ps %xmm0, %xmm0
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    psrad $24, %xmm1
+; SSE2-NEXT:    cvtdq2ps %xmm1, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: sitofp_8i8_to_8f32:
@@ -1756,14 +1755,13 @@ define <8 x float> @sitofp_8i8_to_8f32(<16 x i8> %a) {
 define <8 x float> @sitofp_16i8_to_8f32(<16 x i8> %a) {
 ; SSE2-LABEL: sitofp_16i8_to_8f32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE2-NEXT:    psrad $24, %xmm1
-; SSE2-NEXT:    cvtdq2ps %xmm1, %xmm2
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; SSE2-NEXT:    psrad $24, %xmm0
-; SSE2-NEXT:    cvtdq2ps %xmm0, %xmm1
-; SSE2-NEXT:    movaps %xmm2, %xmm0
+; SSE2-NEXT:    cvtdq2ps %xmm0, %xmm0
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    psrad $24, %xmm1
+; SSE2-NEXT:    cvtdq2ps %xmm1, %xmm1
 ; SSE2-NEXT:    retq
 ;
 ; SSE41-LABEL: sitofp_16i8_to_8f32:
@@ -4458,8 +4456,8 @@ define <8 x float> @sitofp_load_8i16_to_8f32(<8 x i16> *%a) {
 define <8 x float> @sitofp_load_8i8_to_8f32(<8 x i8> *%a) {
 ; SSE2-LABEL: sitofp_load_8i8_to_8f32:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; SSE2-NEXT:    psrad $24, %xmm0
 ; SSE2-NEXT:    cvtdq2ps %xmm0, %xmm0
diff --git a/test/CodeGen/X86/vector-reduce-mul.ll b/test/CodeGen/X86/vector-reduce-mul.ll
index 88b0f0b4679..927f51e9b1b 100644
--- a/test/CodeGen/X86/vector-reduce-mul.ll
+++ b/test/CodeGen/X86/vector-reduce-mul.ll
@@ -1587,40 +1587,60 @@ define i8 @test_v4i8(<4 x i8> %a0) {
 ;
 ; SSE41-LABEL: test_v4i8:
 ; SSE41:       # %bb.0:
-; 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:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; SSE41-NEXT:    pmullw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[2],zero,xmm1[6],zero,xmm1[u],zero,xmm1[u],zero,xmm1[u],zero,xmm1[u],zero,xmm1[u],zero,xmm1[u],zero
+; SSE41-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
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; SSE41-NEXT:    pmullw %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[2],zero,xmm0[6],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
+; SSE41-NEXT:    pmullw %xmm1, %xmm0
+; SSE41-NEXT:    pextrb $0, %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpmovzxbw {{.*#+}} 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
-; AVX-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX-NEXT:    vpmullw %xmm0, %xmm1, %xmm0
+; AVX-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
+; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[6],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
-; AVX512-LABEL: test_v4i8:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpmovzxbw {{.*#+}} 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
-; AVX512-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512-NEXT:    vpmullw %xmm0, %xmm1, %xmm0
-; AVX512-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[6],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
-; AVX512-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512-NEXT:    vpextrb $0, %xmm0, %eax
-; AVX512-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512-NEXT:    retq
+; AVX512BW-LABEL: test_v4i8:
+; AVX512BW:       # %bb.0:
+; AVX512BW-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
+; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[6],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
+; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    retq
+;
+; AVX512VL-LABEL: test_v4i8:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
+; AVX512VL-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
+; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[6],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
+; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512VL-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    retq
+;
+; AVX512DQ-LABEL: test_v4i8:
+; AVX512DQ:       # %bb.0:
+; AVX512DQ-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
+; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[6],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
+; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512DQ-NEXT:    vpextrb $0, %xmm0, %eax
+; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512DQ-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.mul.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
@@ -1645,28 +1665,25 @@ define i8 @test_v8i8(<8 x i8> %a0) {
 ;
 ; SSE41-LABEL: test_v8i8:
 ; SSE41:       # %bb.0:
-; 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:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmullw %xmm1, %xmm0
-; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
-; 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:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; SSE41-NEXT:    pmullw %xmm1, %xmm0
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[2],zero,xmm1[6],zero,xmm1[10],zero,xmm1[14],zero,xmm1[u],zero,xmm1[u],zero,xmm1[u],zero,xmm1[u],zero
+; SSE41-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
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; SSE41-NEXT:    pmullw %xmm0, %xmm1
-; SSE41-NEXT:    pextrb $0, %xmm1, %eax
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE41-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
+; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
+; SSE41-NEXT:    pmullw %xmm0, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[2],zero,xmm0[6],zero,xmm0[10],zero,xmm0[14],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
+; SSE41-NEXT:    pmullw %xmm1, %xmm0
+; SSE41-NEXT:    pextrb $0, %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: test_v8i8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpmovzxbw {{.*#+}} 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
-; AVX-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpmullw %xmm0, %xmm1, %xmm0
+; AVX-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
+; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[6],zero,xmm0[10],zero,xmm0[14],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
@@ -1677,10 +1694,9 @@ define i8 @test_v8i8(<8 x i8> %a0) {
 ;
 ; AVX512BW-LABEL: test_v8i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} 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
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpmullw %xmm0, %xmm1, %xmm0
+; AVX512BW-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
+; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[6],zero,xmm0[10],zero,xmm0[14],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
@@ -1691,10 +1707,10 @@ define i8 @test_v8i8(<8 x i8> %a0) {
 ;
 ; AVX512VL-LABEL: test_v8i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxbw {{.*#+}} 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
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512VL-NEXT:    vpmullw %xmm0, %xmm1, %xmm0
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; AVX512VL-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
+; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpalignr {{.*#+}} xmm1 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15,0,1,2,3]
 ; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[6],zero,xmm0[10],zero,xmm0[14],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
@@ -1705,10 +1721,9 @@ define i8 @test_v8i8(<8 x i8> %a0) {
 ;
 ; AVX512DQ-LABEL: test_v8i8:
 ; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} 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
-; AVX512DQ-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX512DQ-NEXT:    vpmullw %xmm0, %xmm1, %xmm0
+; AVX512DQ-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
+; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[6],zero,xmm0[10],zero,xmm0[14],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero,xmm0[u],zero
@@ -1755,21 +1770,19 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE41-NEXT:    pmullw %xmm1, %xmm0
 ; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
 ; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    packuswb %xmm2, %xmm0
-; SSE41-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
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; SSE41-NEXT:    pmullw %xmm3, %xmm0
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    packuswb %xmm3, %xmm0
+; SSE41-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
+; SSE41-NEXT:    pmullw %xmm2, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
 ; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    packuswb %xmm2, %xmm0
-; SSE41-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
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; SSE41-NEXT:    pmullw %xmm3, %xmm0
+; SSE41-NEXT:    packuswb %xmm3, %xmm0
+; SSE41-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
+; SSE41-NEXT:    pmullw %xmm2, %xmm0
 ; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    packuswb %xmm2, %xmm0
+; SSE41-NEXT:    packuswb %xmm3, %xmm0
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
 ; SSE41-NEXT:    psrlw $8, %xmm1
 ; SSE41-NEXT:    pmullw %xmm0, %xmm1
@@ -1782,9 +1795,9 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; AVX1-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; 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 %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8,8,10,10,12,12,14,14,0,0,2,2,4,4,6,6]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4,4,6,6,4,4,6,6,8,8,10,10,12,12,14,14]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
@@ -1948,30 +1961,31 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; SSE41-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]
 ; SSE41-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
 ; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmullw %xmm1, %xmm0
+; SSE41-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
+; SSE41-NEXT:    movdqa %xmm0, %xmm4
+; SSE41-NEXT:    pand %xmm1, %xmm4
 ; SSE41-NEXT:    pmullw %xmm2, %xmm3
-; SSE41-NEXT:    pshufb {{.*#+}} xmm3 = xmm3[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
-; SSE41-NEXT:    pmullw %xmm1, %xmm3
-; SSE41-NEXT:    pmullw %xmm0, %xmm3
-; SSE41-NEXT:    pand %xmm2, %xmm3
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    packuswb %xmm0, %xmm3
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE41-NEXT:    pmullw %xmm1, %xmm3
-; SSE41-NEXT:    pand %xmm2, %xmm3
-; SSE41-NEXT:    packuswb %xmm0, %xmm3
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,2,3]
-; SSE41-NEXT:    pmullw %xmm1, %xmm3
-; SSE41-NEXT:    pand %xmm2, %xmm3
-; SSE41-NEXT:    packuswb %xmm0, %xmm3
-; SSE41-NEXT:    movdqa %xmm3, %xmm0
+; SSE41-NEXT:    pand %xmm1, %xmm3
+; SSE41-NEXT:    packuswb %xmm4, %xmm3
+; SSE41-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
+; SSE41-NEXT:    pmullw %xmm0, %xmm2
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
+; SSE41-NEXT:    pand %xmm1, %xmm2
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    packuswb %xmm3, %xmm2
+; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
+; SSE41-NEXT:    pmullw %xmm0, %xmm2
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,3]
+; SSE41-NEXT:    pand %xmm1, %xmm2
+; SSE41-NEXT:    packuswb %xmm3, %xmm2
+; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
+; SSE41-NEXT:    pmullw %xmm0, %xmm2
+; SSE41-NEXT:    pand %xmm1, %xmm2
+; SSE41-NEXT:    packuswb %xmm3, %xmm2
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pmullw %xmm3, %xmm0
+; SSE41-NEXT:    pmullw %xmm2, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
@@ -1986,9 +2000,9 @@ define i8 @test_v32i8(<32 x i8> %a0) {
 ; AVX1-NEXT:    vpmullw %xmm2, %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 %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8,8,10,10,12,12,14,14,0,0,2,2,4,4,6,6]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4,4,6,6,4,4,6,6,8,8,10,10,12,12,14,14]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -2204,52 +2218,48 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ;
 ; SSE41-LABEL: test_v64i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; 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:    pmovzxbw {{.*#+}} 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
-; SSE41-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE41-NEXT:    pmullw %xmm2, %xmm0
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    movdqa %xmm0, %xmm6
-; SSE41-NEXT:    pand %xmm2, %xmm6
-; SSE41-NEXT:    pmullw %xmm4, %xmm5
-; SSE41-NEXT:    pand %xmm2, %xmm5
-; SSE41-NEXT:    packuswb %xmm6, %xmm5
 ; SSE41-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
 ; SSE41-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]
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm6 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
 ; SSE41-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]
 ; SSE41-NEXT:    pmullw %xmm3, %xmm1
-; SSE41-NEXT:    movdqa %xmm1, %xmm3
+; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
+; 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 %xmm1, %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[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE41-NEXT:    pmullw %xmm2, %xmm0
+; SSE41-NEXT:    pmullw %xmm3, %xmm1
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
+; SSE41-NEXT:    pshufb %xmm2, %xmm1
+; SSE41-NEXT:    pmullw %xmm4, %xmm5
+; SSE41-NEXT:    pshufb %xmm2, %xmm5
+; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
+; SSE41-NEXT:    movdqa %xmm0, %xmm3
 ; SSE41-NEXT:    pand %xmm2, %xmm3
-; SSE41-NEXT:    pmullw %xmm4, %xmm6
-; SSE41-NEXT:    pand %xmm2, %xmm6
-; SSE41-NEXT:    packuswb %xmm3, %xmm6
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm6[0],zero,xmm6[1],zero,xmm6[2],zero,xmm6[3],zero,xmm6[4],zero,xmm6[5],zero,xmm6[6],zero,xmm6[7],zero
 ; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm5[0],zero,xmm5[1],zero,xmm5[2],zero,xmm5[3],zero,xmm5[4],zero,xmm5[5],zero,xmm5[6],zero,xmm5[7],zero
-; SSE41-NEXT:    pmullw %xmm3, %xmm4
-; SSE41-NEXT:    pshufb {{.*#+}} xmm4 = xmm4[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
-; SSE41-NEXT:    pmullw %xmm1, %xmm3
-; SSE41-NEXT:    pmullw %xmm0, %xmm3
-; SSE41-NEXT:    pand %xmm2, %xmm3
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    packuswb %xmm0, %xmm3
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE41-NEXT:    pmullw %xmm1, %xmm3
-; SSE41-NEXT:    pand %xmm2, %xmm3
-; SSE41-NEXT:    packuswb %xmm0, %xmm3
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,2,3]
-; SSE41-NEXT:    pmullw %xmm1, %xmm3
-; SSE41-NEXT:    pand %xmm2, %xmm3
-; SSE41-NEXT:    packuswb %xmm0, %xmm3
-; SSE41-NEXT:    movdqa %xmm3, %xmm0
+; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
+; SSE41-NEXT:    pmullw %xmm4, %xmm1
+; SSE41-NEXT:    pand %xmm2, %xmm1
+; SSE41-NEXT:    packuswb %xmm3, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
+; SSE41-NEXT:    pmullw %xmm0, %xmm1
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE41-NEXT:    pand %xmm2, %xmm1
+; SSE41-NEXT:    pxor %xmm3, %xmm3
+; SSE41-NEXT:    packuswb %xmm3, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
+; SSE41-NEXT:    pmullw %xmm0, %xmm1
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE41-NEXT:    pand %xmm2, %xmm1
+; SSE41-NEXT:    packuswb %xmm3, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
+; SSE41-NEXT:    pmullw %xmm0, %xmm1
+; SSE41-NEXT:    pand %xmm2, %xmm1
+; SSE41-NEXT:    packuswb %xmm3, %xmm1
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pmullw %xmm3, %xmm0
+; SSE41-NEXT:    pmullw %xmm1, %xmm0
 ; SSE41-NEXT:    pextrb $0, %xmm0, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
@@ -2273,9 +2283,9 @@ define i8 @test_v64i8(<64 x i8> %a0) {
 ; AVX1-NEXT:    vpmullw %xmm3, %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 %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8,8,10,10,12,12,14,14,0,0,2,2,4,4,6,6]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4,4,6,6,4,4,6,6,8,8,10,10,12,12,14,14]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
@@ -2555,78 +2565,82 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ;
 ; SSE41-LABEL: test_v128i8:
 ; SSE41:       # %bb.0:
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm8 = xmm7[0],zero,xmm7[1],zero,xmm7[2],zero,xmm7[3],zero,xmm7[4],zero,xmm7[5],zero,xmm7[6],zero,xmm7[7],zero
-; SSE41-NEXT:    punpckhbw {{.*#+}} xmm7 = xmm7[8],xmm0[8],xmm7[9],xmm0[9],xmm7[10],xmm0[10],xmm7[11],xmm0[11],xmm7[12],xmm0[12],xmm7[13],xmm0[13],xmm7[14],xmm0[14],xmm7[15],xmm0[15]
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm9 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
-; SSE41-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]
-; SSE41-NEXT:    pmullw %xmm7, %xmm3
 ; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm10 = xmm5[0],zero,xmm5[1],zero,xmm5[2],zero,xmm5[3],zero,xmm5[4],zero,xmm5[5],zero,xmm5[6],zero,xmm5[7],zero
 ; SSE41-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm0[8],xmm5[9],xmm0[9],xmm5[10],xmm0[10],xmm5[11],xmm0[11],xmm5[12],xmm0[12],xmm5[13],xmm0[13],xmm5[14],xmm0[14],xmm5[15],xmm0[15]
-; SSE41-NEXT:    pmullw %xmm3, %xmm5
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm8 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; SSE41-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]
 ; SSE41-NEXT:    pmullw %xmm5, %xmm1
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm5 = xmm6[0],zero,xmm6[1],zero,xmm6[2],zero,xmm6[3],zero,xmm6[4],zero,xmm6[5],zero,xmm6[6],zero,xmm6[7],zero
-; SSE41-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm0[8],xmm6[9],xmm0[9],xmm6[10],xmm0[10],xmm6[11],xmm0[11],xmm6[12],xmm0[12],xmm6[13],xmm0[13],xmm6[14],xmm0[14],xmm6[15],xmm0[15]
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm7 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; 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 %xmm6, %xmm2
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm6 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
+; SSE41-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,255,255,255,255,255]
+; SSE41-NEXT:    pmullw %xmm10, %xmm8
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm10 = xmm7[0],zero,xmm7[1],zero,xmm7[2],zero,xmm7[3],zero,xmm7[4],zero,xmm7[5],zero,xmm7[6],zero,xmm7[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm7 = xmm7[8],xmm0[8],xmm7[9],xmm0[9],xmm7[10],xmm0[10],xmm7[11],xmm0[11],xmm7[12],xmm0[12],xmm7[13],xmm0[13],xmm7[14],xmm0[14],xmm7[15],xmm0[15]
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm5 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
+; SSE41-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]
+; SSE41-NEXT:    pmullw %xmm7, %xmm3
+; SSE41-NEXT:    movdqa %xmm3, %xmm7
+; SSE41-NEXT:    pmullw %xmm1, %xmm3
+; SSE41-NEXT:    pand %xmm9, %xmm1
+; SSE41-NEXT:    pand %xmm9, %xmm8
+; SSE41-NEXT:    packuswb %xmm1, %xmm8
+; SSE41-NEXT:    pand %xmm9, %xmm7
+; SSE41-NEXT:    pmullw %xmm10, %xmm5
+; SSE41-NEXT:    pand %xmm9, %xmm5
+; SSE41-NEXT:    packuswb %xmm7, %xmm5
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm10 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
 ; SSE41-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm0[8],xmm4[9],xmm0[9],xmm4[10],xmm0[10],xmm4[11],xmm0[11],xmm4[12],xmm0[12],xmm4[13],xmm0[13],xmm4[14],xmm0[14],xmm4[15],xmm0[15]
-; SSE41-NEXT:    pmullw %xmm2, %xmm4
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = 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:    pmovzxbw {{.*#+}} xmm7 = 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[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE41-NEXT:    pmullw %xmm4, %xmm0
-; SSE41-NEXT:    pmullw %xmm10, %xmm3
-; SSE41-NEXT:    movdqa {{.*#+}} xmm4 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
-; SSE41-NEXT:    pshufb %xmm4, %xmm3
-; SSE41-NEXT:    pmullw %xmm8, %xmm9
-; SSE41-NEXT:    pshufb %xmm4, %xmm9
-; SSE41-NEXT:    pmullw %xmm6, %xmm2
-; SSE41-NEXT:    pshufb %xmm4, %xmm2
-; SSE41-NEXT:    pmullw %xmm5, %xmm7
-; SSE41-NEXT:    pshufb %xmm4, %xmm7
-; SSE41-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,255,255]
-; SSE41-NEXT:    movdqa %xmm0, %xmm6
-; SSE41-NEXT:    pand %xmm5, %xmm6
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm7 = xmm7[0],zero,xmm7[1],zero,xmm7[2],zero,xmm7[3],zero,xmm7[4],zero,xmm7[5],zero,xmm7[6],zero,xmm7[7],zero
-; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
-; SSE41-NEXT:    pmullw %xmm7, %xmm2
-; SSE41-NEXT:    pand %xmm5, %xmm2
-; SSE41-NEXT:    packuswb %xmm6, %xmm2
-; SSE41-NEXT:    movdqa %xmm1, %xmm6
-; SSE41-NEXT:    pand %xmm5, %xmm6
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm7 = xmm9[0],zero,xmm9[1],zero,xmm9[2],zero,xmm9[3],zero,xmm9[4],zero,xmm9[5],zero,xmm9[6],zero,xmm9[7],zero
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
-; SSE41-NEXT:    pmullw %xmm7, %xmm3
-; SSE41-NEXT:    pand %xmm5, %xmm3
-; SSE41-NEXT:    packuswb %xmm6, %xmm3
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero,xmm3[2],zero,xmm3[3],zero,xmm3[4],zero,xmm3[5],zero,xmm3[6],zero,xmm3[7],zero
-; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
-; SSE41-NEXT:    pmullw %xmm3, %xmm2
-; SSE41-NEXT:    pshufb %xmm4, %xmm2
-; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
-; SSE41-NEXT:    pmullw %xmm1, %xmm2
-; SSE41-NEXT:    pmullw %xmm0, %xmm2
-; SSE41-NEXT:    pand %xmm5, %xmm2
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    packuswb %xmm0, %xmm2
-; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; SSE41-NEXT:    pmullw %xmm1, %xmm2
-; SSE41-NEXT:    pand %xmm5, %xmm2
-; SSE41-NEXT:    packuswb %xmm0, %xmm2
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm6[0],zero,xmm6[1],zero,xmm6[2],zero,xmm6[3],zero,xmm6[4],zero,xmm6[5],zero,xmm6[6],zero,xmm6[7],zero
+; SSE41-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm0[8],xmm6[9],xmm0[9],xmm6[10],xmm0[10],xmm6[11],xmm0[11],xmm6[12],xmm0[12],xmm6[13],xmm0[13],xmm6[14],xmm0[14],xmm6[15],xmm0[15]
 ; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,2,3]
-; SSE41-NEXT:    pmullw %xmm1, %xmm2
-; SSE41-NEXT:    pand %xmm5, %xmm2
-; SSE41-NEXT:    packuswb %xmm0, %xmm2
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
-; SSE41-NEXT:    psrlw $8, %xmm0
-; SSE41-NEXT:    pmullw %xmm2, %xmm0
-; SSE41-NEXT:    pextrb $0, %xmm0, %eax
+; 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 %xmm6, %xmm2
+; SSE41-NEXT:    pmullw %xmm2, %xmm3
+; SSE41-NEXT:    pmullw %xmm0, %xmm3
+; SSE41-NEXT:    pand %xmm9, %xmm0
+; SSE41-NEXT:    pmullw %xmm10, %xmm7
+; SSE41-NEXT:    pand %xmm9, %xmm7
+; SSE41-NEXT:    packuswb %xmm0, %xmm7
+; SSE41-NEXT:    pand %xmm9, %xmm2
+; SSE41-NEXT:    pmullw %xmm4, %xmm1
+; SSE41-NEXT:    pand %xmm9, %xmm1
+; SSE41-NEXT:    packuswb %xmm2, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm7[0],zero,xmm7[1],zero,xmm7[2],zero,xmm7[3],zero,xmm7[4],zero,xmm7[5],zero,xmm7[6],zero,xmm7[7],zero
+; SSE41-NEXT:    pmullw %xmm0, %xmm1
+; SSE41-NEXT:    movdqa {{.*#+}} xmm0 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
+; SSE41-NEXT:    pshufb %xmm0, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm5[0],zero,xmm5[1],zero,xmm5[2],zero,xmm5[3],zero,xmm5[4],zero,xmm5[5],zero,xmm5[6],zero,xmm5[7],zero
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm4 = xmm8[0],zero,xmm8[1],zero,xmm8[2],zero,xmm8[3],zero,xmm8[4],zero,xmm8[5],zero,xmm8[6],zero,xmm8[7],zero
+; SSE41-NEXT:    pmullw %xmm2, %xmm4
+; SSE41-NEXT:    pshufb %xmm0, %xmm4
+; SSE41-NEXT:    movdqa %xmm3, %xmm0
+; SSE41-NEXT:    pand %xmm9, %xmm0
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm2 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero
+; SSE41-NEXT:    pmovzxbw {{.*#+}} 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
+; SSE41-NEXT:    pmullw %xmm2, %xmm1
+; SSE41-NEXT:    pand %xmm9, %xmm1
+; SSE41-NEXT:    packuswb %xmm0, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    pmullw %xmm3, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE41-NEXT:    pand %xmm9, %xmm0
+; SSE41-NEXT:    pxor %xmm2, %xmm2
+; SSE41-NEXT:    packuswb %xmm2, %xmm0
+; SSE41-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
+; SSE41-NEXT:    pmullw %xmm1, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE41-NEXT:    pand %xmm9, %xmm0
+; SSE41-NEXT:    packuswb %xmm2, %xmm0
+; SSE41-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
+; SSE41-NEXT:    pmullw %xmm1, %xmm0
+; SSE41-NEXT:    pand %xmm9, %xmm0
+; SSE41-NEXT:    packuswb %xmm2, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrlw $8, %xmm1
+; SSE41-NEXT:    pmullw %xmm0, %xmm1
+; SSE41-NEXT:    pextrb $0, %xmm1, %eax
 ; SSE41-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE41-NEXT:    retq
 ;
@@ -2667,9 +2681,9 @@ define i8 @test_v128i8(<128 x i8> %a0) {
 ; AVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmullw %xmm4, %xmm1, %xmm1
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8,8,10,10,12,12,14,14,0,0,2,2,4,4,6,6]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4,4,6,6,4,4,6,6,8,8,10,10,12,12,14,14]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
diff --git a/test/CodeGen/X86/vector-sext.ll b/test/CodeGen/X86/vector-sext.ll
index 81071907584..48ff42319a4 100644
--- a/test/CodeGen/X86/vector-sext.ll
+++ b/test/CodeGen/X86/vector-sext.ll
@@ -267,22 +267,20 @@ entry:
 define <8 x i32> @sext_16i8_to_8i32(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: sext_16i8_to_8i32:
 ; SSE2:       # %bb.0: # %entry
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; SSE2-NEXT:    psrad $24, %xmm2
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT:    psrad $24, %xmm0
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
 ; SSE2-NEXT:    psrad $24, %xmm1
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: sext_16i8_to_8i32:
 ; SSSE3:       # %bb.0: # %entry
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; SSSE3-NEXT:    psrad $24, %xmm2
-; SSSE3-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSSE3-NEXT:    psrad $24, %xmm0
+; SSSE3-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
 ; SSSE3-NEXT:    psrad $24, %xmm1
-; SSSE3-NEXT:    movdqa %xmm2, %xmm0
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: sext_16i8_to_8i32:
@@ -313,12 +311,11 @@ define <8 x i32> @sext_16i8_to_8i32(<16 x i8> %A) nounwind uwtable readnone ssp
 ;
 ; X32-SSE2-LABEL: sext_16i8_to_8i32:
 ; X32-SSE2:       # %bb.0: # %entry
-; X32-SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
-; X32-SSE2-NEXT:    psrad $24, %xmm2
-; X32-SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; X32-SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; X32-SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; X32-SSE2-NEXT:    psrad $24, %xmm0
+; X32-SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
 ; X32-SSE2-NEXT:    psrad $24, %xmm1
-; X32-SSE2-NEXT:    movdqa %xmm2, %xmm0
 ; X32-SSE2-NEXT:    retl
 ;
 ; X32-SSE41-LABEL: sext_16i8_to_8i32:
@@ -337,7 +334,6 @@ entry:
 define <16 x i32> @sext_16i8_to_16i32(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: sext_16i8_to_16i32:
 ; SSE2:       # %bb.0: # %entry
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1],xmm4[2],xmm1[2],xmm4[3],xmm1[3]
 ; SSE2-NEXT:    psrad $24, %xmm4
@@ -353,7 +349,6 @@ define <16 x i32> @sext_16i8_to_16i32(<16 x i8> %A) nounwind uwtable readnone ss
 ;
 ; SSSE3-LABEL: sext_16i8_to_16i32:
 ; SSSE3:       # %bb.0: # %entry
-; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1],xmm4[2],xmm1[2],xmm4[3],xmm1[3]
 ; SSSE3-NEXT:    psrad $24, %xmm4
@@ -408,7 +403,6 @@ define <16 x i32> @sext_16i8_to_16i32(<16 x i8> %A) nounwind uwtable readnone ss
 ;
 ; X32-SSE2-LABEL: sext_16i8_to_16i32:
 ; X32-SSE2:       # %bb.0: # %entry
-; X32-SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; X32-SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; X32-SSE2-NEXT:    punpcklwd {{.*#+}} xmm4 = xmm4[0],xmm1[0],xmm4[1],xmm1[1],xmm4[2],xmm1[2],xmm4[3],xmm1[3]
 ; X32-SSE2-NEXT:    psrad $24, %xmm4
@@ -568,42 +562,40 @@ entry:
 define <8 x i64> @sext_16i8_to_8i64(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: sext_16i8_to_8i64:
 ; SSE2:       # %bb.0: # %entry
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; SSE2-NEXT:    psrad $24, %xmm1
-; SSE2-NEXT:    pxor %xmm5, %xmm5
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    pcmpgtd %xmm1, %xmm2
-; SSE2-NEXT:    movdqa %xmm1, %xmm4
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
+; SSE2-NEXT:    pxor %xmm4, %xmm4
+; SSE2-NEXT:    pxor %xmm3, %xmm3
+; SSE2-NEXT:    pcmpgtd %xmm1, %xmm3
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm3[2],xmm1[3],xmm3[3]
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
 ; SSE2-NEXT:    psrad $24, %xmm3
-; SSE2-NEXT:    pcmpgtd %xmm3, %xmm5
+; SSE2-NEXT:    pcmpgtd %xmm3, %xmm4
 ; SSE2-NEXT:    movdqa %xmm3, %xmm2
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1]
-; SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm5[2],xmm3[3],xmm5[3]
-; SSE2-NEXT:    movdqa %xmm4, %xmm0
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1]
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: sext_16i8_to_8i64:
 ; SSSE3:       # %bb.0: # %entry
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; SSSE3-NEXT:    psrad $24, %xmm1
-; SSSE3-NEXT:    pxor %xmm5, %xmm5
-; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    pcmpgtd %xmm1, %xmm2
-; SSSE3-NEXT:    movdqa %xmm1, %xmm4
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; SSSE3-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; SSSE3-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
+; SSSE3-NEXT:    pxor %xmm4, %xmm4
+; SSSE3-NEXT:    pxor %xmm3, %xmm3
+; SSSE3-NEXT:    pcmpgtd %xmm1, %xmm3
+; SSSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
+; SSSE3-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm3[2],xmm1[3],xmm3[3]
+; SSSE3-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
 ; SSSE3-NEXT:    psrad $24, %xmm3
-; SSSE3-NEXT:    pcmpgtd %xmm3, %xmm5
+; SSSE3-NEXT:    pcmpgtd %xmm3, %xmm4
 ; SSSE3-NEXT:    movdqa %xmm3, %xmm2
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1]
-; SSSE3-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm5[2],xmm3[3],xmm5[3]
-; SSSE3-NEXT:    movdqa %xmm4, %xmm0
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1]
+; SSSE3-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: sext_16i8_to_8i64:
@@ -648,22 +640,21 @@ define <8 x i64> @sext_16i8_to_8i64(<16 x i8> %A) nounwind uwtable readnone ssp
 ;
 ; X32-SSE2-LABEL: sext_16i8_to_8i64:
 ; X32-SSE2:       # %bb.0: # %entry
-; X32-SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X32-SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; X32-SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
+; X32-SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; X32-SSE2-NEXT:    psrad $24, %xmm1
-; X32-SSE2-NEXT:    pxor %xmm5, %xmm5
-; X32-SSE2-NEXT:    pxor %xmm2, %xmm2
-; X32-SSE2-NEXT:    pcmpgtd %xmm1, %xmm2
-; X32-SSE2-NEXT:    movdqa %xmm1, %xmm4
-; X32-SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; X32-SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
-; X32-SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
+; X32-SSE2-NEXT:    pxor %xmm4, %xmm4
+; X32-SSE2-NEXT:    pxor %xmm3, %xmm3
+; X32-SSE2-NEXT:    pcmpgtd %xmm1, %xmm3
+; X32-SSE2-NEXT:    movdqa %xmm1, %xmm0
+; X32-SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
+; X32-SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm3[2],xmm1[3],xmm3[3]
+; X32-SSE2-NEXT:    punpckhwd {{.*#+}} xmm3 = xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
 ; X32-SSE2-NEXT:    psrad $24, %xmm3
-; X32-SSE2-NEXT:    pcmpgtd %xmm3, %xmm5
+; X32-SSE2-NEXT:    pcmpgtd %xmm3, %xmm4
 ; X32-SSE2-NEXT:    movdqa %xmm3, %xmm2
-; X32-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1]
-; X32-SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm5[2],xmm3[3],xmm5[3]
-; X32-SSE2-NEXT:    movdqa %xmm4, %xmm0
+; X32-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1]
+; X32-SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
 ; X32-SSE2-NEXT:    retl
 ;
 ; X32-SSE41-LABEL: sext_16i8_to_8i64:
@@ -2134,8 +2125,8 @@ entry:
 define <8 x i64> @load_sext_8i8_to_8i64(<8 x i8> *%ptr) {
 ; SSE2-LABEL: load_sext_8i8_to_8i64:
 ; SSE2:       # %bb.0: # %entry
-; SSE2-NEXT:    movq {{.*#+}} xmm2 = mem[0],zero
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; SSE2-NEXT:    psrad $24, %xmm1
 ; SSE2-NEXT:    pxor %xmm4, %xmm4
@@ -2154,8 +2145,8 @@ define <8 x i64> @load_sext_8i8_to_8i64(<8 x i8> *%ptr) {
 ;
 ; SSSE3-LABEL: load_sext_8i8_to_8i64:
 ; SSSE3:       # %bb.0: # %entry
-; SSSE3-NEXT:    movq {{.*#+}} xmm2 = mem[0],zero
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSSE3-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; SSSE3-NEXT:    psrad $24, %xmm1
 ; SSSE3-NEXT:    pxor %xmm4, %xmm4
@@ -2204,8 +2195,8 @@ define <8 x i64> @load_sext_8i8_to_8i64(<8 x i8> *%ptr) {
 ; X32-SSE2-LABEL: load_sext_8i8_to_8i64:
 ; X32-SSE2:       # %bb.0: # %entry
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    movq {{.*#+}} xmm2 = mem[0],zero
-; X32-SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X32-SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
 ; X32-SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; X32-SSE2-NEXT:    psrad $24, %xmm1
 ; X32-SSE2-NEXT:    pxor %xmm4, %xmm4
@@ -2303,8 +2294,8 @@ entry:
 define <8 x i32> @load_sext_8i8_to_8i32(<8 x i8> *%ptr) {
 ; SSE2-LABEL: load_sext_8i8_to_8i32:
 ; SSE2:       # %bb.0: # %entry
-; SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; SSE2-NEXT:    psrad $24, %xmm0
 ; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
@@ -2313,8 +2304,8 @@ define <8 x i32> @load_sext_8i8_to_8i32(<8 x i8> *%ptr) {
 ;
 ; SSSE3-LABEL: load_sext_8i8_to_8i32:
 ; SSSE3:       # %bb.0: # %entry
-; SSSE3-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSSE3-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; SSSE3-NEXT:    psrad $24, %xmm0
 ; SSSE3-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
@@ -2347,8 +2338,8 @@ define <8 x i32> @load_sext_8i8_to_8i32(<8 x i8> *%ptr) {
 ; X32-SSE2-LABEL: load_sext_8i8_to_8i32:
 ; X32-SSE2:       # %bb.0: # %entry
 ; X32-SSE2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
-; X32-SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; X32-SSE2-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; X32-SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
 ; X32-SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; X32-SSE2-NEXT:    psrad $24, %xmm0
 ; X32-SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
diff --git a/test/CodeGen/X86/vector-shuffle-256-v32.ll b/test/CodeGen/X86/vector-shuffle-256-v32.ll
index 8dd211b5566..31c71354ee4 100644
--- a/test/CodeGen/X86/vector-shuffle-256-v32.ll
+++ b/test/CodeGen/X86/vector-shuffle-256-v32.ll
@@ -3106,16 +3106,14 @@ define <4 x i64> @PR28136(<32 x i8> %a0, <32 x i8> %a1) {
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm0 = 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:    vmovdqa {{.*#+}} xmm2 = [8,8,10,10,12,12,14,14,9,9,11,11,13,13,15,15]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm3
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3],xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
-; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm2
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = xmm0[8,u,10,u,12,u,14,u,9,u,11,u,13,u,15,u]
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm3 = xmm1[u,8,u,10,u,12,u,14,u,9,u,11,u,13,u,15]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
-; AVX1-NEXT:    vpblendvb %xmm4, %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,0,2,2,4,4,6,6,1,1,3,3,5,5,7,7]
-; AVX1-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX1-NEXT:    vpblendvb %xmm4, %xmm2, %xmm3, %xmm2
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,u,2,u,4,u,6,u,1,u,3,u,5,u,7,u]
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[u,0,u,2,u,4,u,6,u,1,u,3,u,5,u,7]
 ; AVX1-NEXT:    vpblendvb %xmm4, %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
diff --git a/test/CodeGen/X86/widen_conv-3.ll b/test/CodeGen/X86/widen_conv-3.ll
index 85f7c37b996..21e80c99dcf 100644
--- a/test/CodeGen/X86/widen_conv-3.ll
+++ b/test/CodeGen/X86/widen_conv-3.ll
@@ -60,8 +60,8 @@ define void @convert_v3i8_to_v3f32(<3 x float>* %dst.addr, <3 x i8>* %src.addr)
 ; X86-SSE2-NEXT:    pslld $16, %xmm2
 ; X86-SSE2-NEXT:    pandn %xmm2, %xmm1
 ; X86-SSE2-NEXT:    por %xmm0, %xmm1
-; X86-SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X86-SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; X86-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]
+; X86-SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
 ; X86-SSE2-NEXT:    psrad $24, %xmm0
 ; X86-SSE2-NEXT:    cvtdq2ps %xmm0, %xmm0
 ; X86-SSE2-NEXT:    movss %xmm0, (%eax)
@@ -97,8 +97,8 @@ define void @convert_v3i8_to_v3f32(<3 x float>* %dst.addr, <3 x i8>* %src.addr)
 ; X64-SSE2-NEXT:    pslld $16, %xmm2
 ; X64-SSE2-NEXT:    pandn %xmm2, %xmm1
 ; X64-SSE2-NEXT:    por %xmm0, %xmm1
-; X64-SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; X64-SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; X64-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]
+; X64-SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
 ; X64-SSE2-NEXT:    psrad $24, %xmm0
 ; X64-SSE2-NEXT:    cvtdq2ps %xmm0, %xmm0
 ; X64-SSE2-NEXT:    movlps %xmm0, (%rdi)