From: Chandler Carruth <chandlerc@gmail.com>
Date: Thu, 19 Feb 2015 15:06:13 +0000 (+0000)
Subject: [x86] Teach the unpack lowering how to lower with an initial unpack in
X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=commitdiff_plain;h=c57e90422fe64d89c527a8a7886e0f7d875fef3f

[x86] Teach the unpack lowering how to lower with an initial unpack in
addition to lowering to trees rooted in an unpack.

This saves shuffles and or registers in many various ways, lets us
handle another class of v4i32 shuffles pre SSE4.1 without domain
crosses, etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@229856 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 537773d1cfb..45128fe032a 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -8467,7 +8467,7 @@ static SDValue lowerVectorShuffleAsUnpack(MVT VT, SDLoc DL, SDValue V1,
     return M >= 0 && M % Size < Size / 2;
   });
   int NumHiInputs = std::count_if(
-      Mask.begin(), Mask.end(), [Size](int M) { return M % Size > Size / 2; });
+      Mask.begin(), Mask.end(), [Size](int M) { return M % Size >= Size / 2; });
 
   bool UnpackLo = NumLoInputs >= NumHiInputs;
 
@@ -8494,6 +8494,12 @@ static SDValue lowerVectorShuffleAsUnpack(MVT VT, SDLoc DL, SDValue V1,
           Mask[i] % Size;
     }
 
+    // If we will have to shuffle both inputs to use the unpack, check whether
+    // we can just unpack first and shuffle the result. If so, skip this unpack.
+    if ((NumLoInputs == 0 || NumHiInputs == 0) && !isNoopShuffleMask(V1Mask) &&
+        !isNoopShuffleMask(V2Mask))
+      return SDValue();
+
     // Shuffle the inputs into place.
     V1 = DAG.getVectorShuffle(VT, DL, V1, DAG.getUNDEF(VT), V1Mask);
     V2 = DAG.getVectorShuffle(VT, DL, V2, DAG.getUNDEF(VT), V2Mask);
@@ -8520,6 +8526,35 @@ static SDValue lowerVectorShuffleAsUnpack(MVT VT, SDLoc DL, SDValue V1,
       return Unpack;
   }
 
+  // If none of the unpack-rooted lowerings worked (or were profitable) try an
+  // initial unpack.
+  if (NumLoInputs == 0 || NumHiInputs == 0) {
+    assert((NumLoInputs > 0 || NumHiInputs > 0) &&
+           "We have to have *some* inputs!");
+    int HalfOffset = NumLoInputs == 0 ? Size / 2 : 0;
+
+    // FIXME: We could consider the total complexity of the permute of each
+    // possible unpacking. Or at the least we should consider how many
+    // half-crossings are created.
+    // FIXME: We could consider commuting the unpacks.
+
+    SmallVector<int, 32> PermMask;
+    PermMask.assign(Size, -1);
+    for (int i = 0; i < Size; ++i) {
+      if (Mask[i] < 0)
+        continue;
+
+      assert(Mask[i] % Size >= HalfOffset && "Found input from wrong half!");
+
+      PermMask[i] =
+          2 * ((Mask[i] % Size) - HalfOffset) + (Mask[i] < Size ? 0 : 1);
+    }
+    return DAG.getVectorShuffle(
+        VT, DL, DAG.getNode(NumLoInputs == 0 ? X86ISD::UNPCKH : X86ISD::UNPCKL,
+                            DL, VT, V1, V2),
+        DAG.getUNDEF(VT), PermMask);
+  }
+
   return SDValue();
 }
 
diff --git a/test/CodeGen/X86/sse3.ll b/test/CodeGen/X86/sse3.ll
index a3e80223feb..4223b790272 100644
--- a/test/CodeGen/X86/sse3.ll
+++ b/test/CodeGen/X86/sse3.ll
@@ -204,14 +204,9 @@ entry:
 define <8 x i16> @t12(<8 x i16> %T0, <8 x i16> %T1) nounwind readnone {
 ; X64-LABEL: t12:
 ; X64:       ## BB#0: ## %entry
-; X64-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[0,1,1,3]
-; X64-NEXT:    movdqa {{.*#+}} xmm1 = [65535,65535,65535,65535,65535,0,65535,65535]
-; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
-; X64-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,5,6,7]
-; X64-NEXT:    pand %xmm1, %xmm0
-; X64-NEXT:    pandn %xmm2, %xmm1
-; X64-NEXT:    por %xmm0, %xmm1
-; X64-NEXT:    movdqa %xmm1, %xmm0
+; X64-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; X64-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; X64-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,7,6,7]
 ; X64-NEXT:    retq
 entry:
 	%tmp9 = shufflevector <8 x i16> %T0, <8 x i16> %T1, <8 x i32> < i32 0, i32 1, i32 undef, i32 undef, i32 3, i32 11, i32 undef , i32 undef >
@@ -222,13 +217,9 @@ entry:
 define <8 x i16> @t13(<8 x i16> %T0, <8 x i16> %T1) nounwind readnone {
 ; X64-LABEL: t13:
 ; X64:       ## BB#0: ## %entry
-; X64-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,1,1,3]
-; X64-NEXT:    movdqa {{.*#+}} xmm0 = [65535,65535,65535,65535,65535,0,65535,65535]
-; X64-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
-; X64-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,5,6,7]
-; X64-NEXT:    pand %xmm0, %xmm1
-; X64-NEXT:    pandn %xmm2, %xmm0
-; X64-NEXT:    por %xmm1, %xmm0
+; X64-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; X64-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[0,2,2,3,4,5,6,7]
+; X64-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,7,6,7]
 ; X64-NEXT:    retq
 entry:
 	%tmp9 = shufflevector <8 x i16> %T0, <8 x i16> %T1, <8 x i32> < i32 8, i32 9, i32 undef, i32 undef, i32 11, i32 3, i32 undef , i32 undef >
diff --git a/test/CodeGen/X86/vector-shuffle-128-v16.ll b/test/CodeGen/X86/vector-shuffle-128-v16.ll
index 55495a285c2..3cc52fb8c6f 100644
--- a/test/CodeGen/X86/vector-shuffle-128-v16.ll
+++ b/test/CodeGen/X86/vector-shuffle-128-v16.ll
@@ -339,26 +339,20 @@ define <16 x i8> @shuffle_v16i8_03_02_01_00_07_06_05_04_19_18_17_16_23_22_21_20(
 ;
 ; SSSE3-LABEL: shuffle_v16i8_03_02_01_00_07_06_05_04_19_18_17_16_23_22_21_20:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    movdqa {{.*#+}} xmm2 = <3,2,1,0,7,6,5,4,u,u,u,u,u,u,u,u>
-; SSSE3-NEXT:    pshufb %xmm2, %xmm1
-; SSSE3-NEXT:    pshufb %xmm2, %xmm0
-; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSSE3-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]
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[6,4,2,0,14,12,10,8,7,5,3,1,15,13,11,9]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v16i8_03_02_01_00_07_06_05_04_19_18_17_16_23_22_21_20:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    movdqa {{.*#+}} xmm2 = <3,2,1,0,7,6,5,4,u,u,u,u,u,u,u,u>
-; SSE41-NEXT:    pshufb %xmm2, %xmm1
-; SSE41-NEXT:    pshufb %xmm2, %xmm0
-; SSE41-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE41-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]
+; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[6,4,2,0,14,12,10,8,7,5,3,1,15,13,11,9]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: shuffle_v16i8_03_02_01_00_07_06_05_04_19_18_17_16_23_22_21_20:
 ; AVX:       # BB#0:
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = <3,2,1,0,7,6,5,4,u,u,u,u,u,u,u,u>
-; AVX-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[6,4,2,0,14,12,10,8,7,5,3,1,15,13,11,9]
 ; AVX-NEXT:    retq
   %shuffle = shufflevector <16 x i8> %a, <16 x i8> %b, <16 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4, i32 19, i32 18, i32 17, i32 16, i32 23, i32 22, i32 21, i32 20>
   ret <16 x i8> %shuffle
diff --git a/test/CodeGen/X86/vector-shuffle-128-v4.ll b/test/CodeGen/X86/vector-shuffle-128-v4.ll
index 290019e7014..9ccb80a346e 100644
--- a/test/CodeGen/X86/vector-shuffle-128-v4.ll
+++ b/test/CodeGen/X86/vector-shuffle-128-v4.ll
@@ -485,20 +485,20 @@ define <4 x i32> @shuffle_v4i32_0145(<4 x i32> %a, <4 x i32> %b) {
 define <4 x i32> @shuffle_v4i32_0451(<4 x i32> %a, <4 x i32> %b) {
 ; SSE2-LABEL: shuffle_v4i32_0451:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,1]
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,3,1]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,3,2]
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: shuffle_v4i32_0451:
 ; SSE3:       # BB#0:
-; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,1]
-; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,3,1]
+; SSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,3,2]
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v4i32_0451:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,1]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,3,1]
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,3,2]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v4i32_0451:
@@ -541,20 +541,20 @@ define <4 x i32> @shuffle_v4i32_4501(<4 x i32> %a, <4 x i32> %b) {
 define <4 x i32> @shuffle_v4i32_4015(<4 x i32> %a, <4 x i32> %b) {
 ; SSE2-LABEL: shuffle_v4i32_4015:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,1]
-; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0,1,3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,0,2,3]
 ; SSE2-NEXT:    retq
 ;
 ; SSE3-LABEL: shuffle_v4i32_4015:
 ; SSE3:       # BB#0:
-; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,1]
-; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0,1,3]
+; SSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,0,2,3]
 ; SSE3-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v4i32_4015:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,1]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0,1,3]
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,0,2,3]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v4i32_4015:
diff --git a/test/CodeGen/X86/vector-shuffle-128-v8.ll b/test/CodeGen/X86/vector-shuffle-128-v8.ll
index 4b85337bf41..84b47faafaa 100644
--- a/test/CodeGen/X86/vector-shuffle-128-v8.ll
+++ b/test/CodeGen/X86/vector-shuffle-128-v8.ll
@@ -951,16 +951,16 @@ define <8 x i16> @shuffle_v8i16_0c1d6879(<8 x i16> %a, <8 x i16> %b) {
 define <8 x i16> @shuffle_v8i16_109832ba(<8 x i16> %a, <8 x i16> %b) {
 ; SSE2-LABEL: shuffle_v8i16_109832ba:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[1,0,3,2,4,5,6,7]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,0,3,2,4,5,6,7]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[2,0,3,1,4,5,6,7]
+; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,4,7,5]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuffle_v8i16_109832ba:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[1,0,3,2,4,5,6,7]
-; SSSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,0,3,2,4,5,6,7]
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[2,0,3,1,4,5,6,7]
+; SSSE3-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,4,7,5]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuffle_v8i16_109832ba:
diff --git a/test/CodeGen/X86/vector-shuffle-256-v32.ll b/test/CodeGen/X86/vector-shuffle-256-v32.ll
index d0942bfdd90..39d85d803a9 100644
--- a/test/CodeGen/X86/vector-shuffle-256-v32.ll
+++ b/test/CodeGen/X86/vector-shuffle-256-v32.ll
@@ -314,9 +314,8 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_16_
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[0]
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],zero
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; 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:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -339,9 +338,8 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_17_00_
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,0],zero,xmm0[0]
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; 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:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,0,3,0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -362,9 +360,8 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_18_00_00_
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[2],zero,zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0],zero,xmm0[0,0]
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; 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:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,5,0,0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -385,9 +382,8 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_19_00_00_00_
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[3],zero,zero,zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0],zero,xmm0[0,0,0]
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; 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:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,7,0,0,0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -408,9 +404,8 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_20_00_00_00_00_
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[4],zero,zero,zero,zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0],zero,xmm0[0,0,0,0]
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; 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:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,9,0,0,0,0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -430,9 +425,8 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_21_00_00_00_00_00_
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[5],zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0],zero,xmm0[0,0,0,0,0]
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; 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:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,11,0,0,0,0,0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -452,9 +446,8 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_00_22_00_00_00_00_00_00_
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm2[6],zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0],zero,xmm0[0,0,0,0,0,0]
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; 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:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,13,0,0,0,0,0,0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -474,9 +467,8 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_23_00_00_00_00_00_00_00_
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,xmm2[7],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0],zero,xmm0[0,0,0,0,0,0,0]
-; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
+; 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:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,15,0,0,0,0,0,0,0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -960,11 +952,8 @@ define <32 x i8> @shuffle_v32i8_32_01_34_03_36_05_38_07_40_09_42_11_44_13_46_15_
 define <32 x i8> @shuffle_v32i8_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32(<32 x i8> %a, <32 x i8> %b) {
 ; AVX1-LABEL: shuffle_v32i8_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
 ; AVX1-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -983,17 +972,12 @@ define <32 x i8> @shuffle_v32i8_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_
 ; AVX1-LABEL: shuffle_v32i8_00_32_00_32_00_32_00_32_00_32_00_32_00_32_00_32_16_48_16_48_16_48_16_48_16_48_16_48_16_48_16_48:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,0,0,0,4,5,6,7]
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm3[0,0,0,0,4,5,6,7]
 ; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
-; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1]
+; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
 ; AVX1-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX1-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -1064,14 +1048,12 @@ define <32 x i8> @shuffle_v32i8_39_38_37_36_35_34_33_32_07_06_05_04_03_02_01_00_
 ; AVX1-LABEL: shuffle_v32i8_39_38_37_36_35_34_33_32_07_06_05_04_03_02_01_00_55_54_53_52_51_50_49_48_23_22_21_20_19_18_17_16:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = <7,6,5,4,3,2,1,0,u,u,u,u,u,u,u,u>
+; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
+; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [14,12,10,8,6,4,2,0,15,13,11,9,7,5,3,1]
 ; AVX1-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
-; AVX1-NEXT:    vpshufb %xmm3, %xmm4, %xmm4
-; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm4[0],xmm2[0]
+; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm0 = 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]
 ; AVX1-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
@@ -1485,16 +1467,15 @@ define <32 x i8> @shuffle_v32i8_42_45_12_13_35_35_60_40_17_22_29_44_33_12_48_51_
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm3 = xmm2[u,u,4,u,1,6],zero,zero,xmm2[0],zero,xmm2[11,u],zero,zero,zero,zero
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm4 = xmm1[u,u],zero,xmm1[u],zero,zero,xmm1[5,0],zero,xmm1[10],zero,xmm1[u,4,2,4,7]
 ; AVX1-NEXT:    vpor %xmm3, %xmm4, %xmm3
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm4 = xmm0[1,1,2,3]
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm6 = xmm5[4,3,u,3,u,u,u,u,u,u,u,u,u,u,u,u]
-; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm4 = xmm6[0],xmm4[0]
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm4
+; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm5 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3],xmm4[4],xmm0[4],xmm4[5],xmm0[5],xmm4[6],xmm0[6],xmm4[7],xmm0[7]
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[8,6,u,6,u,u,u,u,u,u,u,15,u,u,u,u]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm6 = [0,0,255,0,255,255,255,255,255,255,255,0,255,255,255,255]
-; AVX1-NEXT:    vpblendvb %xmm6, %xmm3, %xmm4, %xmm3
+; AVX1-NEXT:    vpblendvb %xmm6, %xmm3, %xmm5, %xmm3
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = zero,zero,xmm2[u,u],zero,zero,xmm2[12],zero,xmm2[u,u,u],zero,zero,xmm2[u,0,3]
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[10,13,u,u,3,3],zero,xmm1[8,u,u,u,12,1,u],zero,zero
 ; AVX1-NEXT:    vpor %xmm2, %xmm1, %xmm1
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = xmm5[u,u],zero,zero,xmm5[u,u,u,u,1,6,13,u,u],zero,xmm5[u,u]
+; AVX1-NEXT:    vpshufb {{.*#+}} xmm2 = xmm4[u,u],zero,zero,xmm4[u,u,u,u,1,6,13,u,u],zero,xmm4[u,u]
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[u,u,12,13,u,u,u,u],zero,zero,zero,xmm0[u,u,12,u,u]
 ; AVX1-NEXT:    vpor %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,255,0,0,255,255,255,255,0,0,0,255,255,0,255,255]
diff --git a/test/CodeGen/X86/vector-shuffle-combining.ll b/test/CodeGen/X86/vector-shuffle-combining.ll
index a1051c6610d..2fe78297f99 100644
--- a/test/CodeGen/X86/vector-shuffle-combining.ll
+++ b/test/CodeGen/X86/vector-shuffle-combining.ll
@@ -600,17 +600,17 @@ define <4 x i32> @combine_bitwise_ops_test3c(<4 x i32> %a, <4 x i32> %b, <4 x i3
 ; SSE2-LABEL: combine_bitwise_ops_test3c:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    pxor %xmm1, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,2]
 ; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: combine_bitwise_ops_test3c:
 ; SSSE3:       # BB#0:
 ; SSSE3-NEXT:    pxor %xmm1, %xmm0
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,2]
 ; SSSE3-NEXT:    pxor %xmm1, %xmm1
-; SSSE3-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSSE3-NEXT:    punpckhqdq {{.*#+}} xmm0 = xmm0[1],xmm1[1]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: combine_bitwise_ops_test3c:
@@ -1117,16 +1117,14 @@ define <4 x i32> @combine_nested_undef_test18(<4 x i32> %A, <4 x i32> %B) {
 define <4 x i32> @combine_nested_undef_test19(<4 x i32> %A, <4 x i32> %B) {
 ; SSE2-LABEL: combine_nested_undef_test19:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
-; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[0,0]
-; SSE2-NEXT:    movaps %xmm1, %xmm0
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,0,0,0]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: combine_nested_undef_test19:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
-; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[0,0]
-; SSSE3-NEXT:    movaps %xmm1, %xmm0
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,0,0,0]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: combine_nested_undef_test19:
@@ -1191,16 +1189,14 @@ define <4 x i32> @combine_nested_undef_test20(<4 x i32> %A, <4 x i32> %B) {
 define <4 x i32> @combine_nested_undef_test21(<4 x i32> %A, <4 x i32> %B) {
 ; SSE2-LABEL: combine_nested_undef_test21:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,0,1,1]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,3,0,3]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: combine_nested_undef_test21:
 ; SSSE3:       # BB#0:
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,0,1,1]
-; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,3,0,3]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: combine_nested_undef_test21: