[X86][SSE] Enable commutation for SSE immediate blend instructions

Patch to allow (v)blendps, (v)blendpd, (v)pblendw and vpblendd instructions to be commuted - swaps the src registers and inverts the blend mask.

This is primarily to improve memory folding (see new tests), but it also improves the quality of shuffles (see modified tests).

Differential Revision: http://reviews.llvm.org/D6015

llvm-svn: 221313

6 files changed, 170 insertions, 82 deletions

diff --git a/llvm/test/CodeGen/X86/combine-or.ll b/llvm/test/CodeGen/X86/combine-or.ll
index ba9d48d6047..9539eae9d6a 100644
--- a/llvm/test/CodeGen/X86/combine-or.ll
+++ b/llvm/test/CodeGen/X86/combine-or.ll
@@ -19,8 +19,7 @@ define <2 x i64> @test1(<2 x i64> %a, <2 x i64> %b) {
 define <4 x i32> @test2(<4 x i32> %a, <4 x i32> %b) {
 ; CHECK-LABEL: test2:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
-; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
 ; CHECK-NEXT:    retq
   %shuf1 = shufflevector <4 x i32> %a, <4 x i32> zeroinitializer, <4 x i32><i32 4, i32 4, i32 2, i32 3>
   %shuf2 = shufflevector <4 x i32> %b, <4 x i32> zeroinitializer, <4 x i32><i32 0, i32 1, i32 4, i32 4>
@@ -32,8 +31,7 @@ define <4 x i32> @test2(<4 x i32> %a, <4 x i32> %b) {
 define <2 x i64> @test3(<2 x i64> %a, <2 x i64> %b) {
 ; CHECK-LABEL: test3:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
-; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
 ; CHECK-NEXT:    retq
   %shuf1 = shufflevector <2 x i64> %a, <2 x i64> zeroinitializer, <2 x i32><i32 2, i32 1>
   %shuf2 = shufflevector <2 x i64> %b, <2 x i64> zeroinitializer, <2 x i32><i32 0, i32 2>
@@ -45,8 +43,7 @@ define <2 x i64> @test3(<2 x i64> %a, <2 x i64> %b) {
 define <4 x i32> @test4(<4 x i32> %a, <4 x i32> %b) {
 ; CHECK-LABEL: test4:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3,4,5,6,7]
-; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
 ; CHECK-NEXT:    retq
   %shuf1 = shufflevector <4 x i32> %a, <4 x i32> zeroinitializer, <4 x i32><i32 0, i32 4, i32 4, i32 4>
   %shuf2 = shufflevector <4 x i32> %b, <4 x i32> zeroinitializer, <4 x i32><i32 4, i32 1, i32 2, i32 3>
@@ -106,8 +103,7 @@ define <2 x i64> @test8(<2 x i64> %a, <2 x i64> %b) {
 define <4 x i32> @test9(<4 x i32> %a, <4 x i32> %b) {
 ; CHECK-LABEL: test9:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
-; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
 ; CHECK-NEXT:    retq
   %and1 = and <4 x i32> %a, <i32 0, i32 0, i32 -1, i32 -1>
   %and2 = and <4 x i32> %b, <i32 -1, i32 -1, i32 0, i32 0>
@@ -119,8 +115,7 @@ define <4 x i32> @test9(<4 x i32> %a, <4 x i32> %b) {
 define <2 x i64> @test10(<2 x i64> %a, <2 x i64> %b) {
 ; CHECK-LABEL: test10:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
-; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
 ; CHECK-NEXT:    retq
   %and1 = and <2 x i64> %a, <i64 0, i64 -1>
   %and2 = and <2 x i64> %b, <i64 -1, i64 0>
@@ -132,8 +127,7 @@ define <2 x i64> @test10(<2 x i64> %a, <2 x i64> %b) {
 define <4 x i32> @test11(<4 x i32> %a, <4 x i32> %b) {
 ; CHECK-LABEL: test11:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3,4,5,6,7]
-; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
 ; CHECK-NEXT:    retq
   %and1 = and <4 x i32> %a, <i32 -1, i32 0, i32 0, i32 0>
   %and2 = and <4 x i32> %b, <i32 0, i32 -1, i32 -1, i32 -1>
@@ -230,12 +224,10 @@ define <4 x i32> @test18(<4 x i32> %a, <4 x i32> %b) {
 ; CHECK-LABEL: test18:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    xorps %xmm2, %xmm2
-; CHECK-NEXT:    xorps %xmm3, %xmm3
-; CHECK-NEXT:    blendps {{.*#+}} xmm3 = xmm0[0],xmm3[1,2,3]
-; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,0,1,1]
-; CHECK-NEXT:    blendps {{.*#+}} xmm2 = xmm1[0],xmm2[1,2,3]
-; CHECK-NEXT:    orps %xmm0, %xmm2
-; CHECK-NEXT:    movaps %xmm2, %xmm0
+; CHECK-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3]
+; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,0,1,1]
+; CHECK-NEXT:    blendps {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3]
+; CHECK-NEXT:    por %xmm1, %xmm0
 ; CHECK-NEXT:    retq
   %shuf1 = shufflevector <4 x i32> %a, <4 x i32> zeroinitializer, <4 x i32><i32 4, i32 0, i32 4, i32 4>
   %shuf2 = shufflevector <4 x i32> %b, <4 x i32> zeroinitializer, <4 x i32><i32 0, i32 4, i32 4, i32 4>
@@ -295,8 +287,7 @@ define <2 x i64> @test21(<2 x i64> %a, <2 x i64> %b) {
 define <4 x i8> @test_crash(<4 x i8> %a, <4 x i8> %b) {
 ; CHECK-LABEL: test_crash:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
-; CHECK-NEXT:    movdqa %xmm1, %xmm0
+; CHECK-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
 ; CHECK-NEXT:    retq
   %shuf1 = shufflevector <4 x i8> %a, <4 x i8> zeroinitializer, <4 x i32><i32 4, i32 4, i32 2, i32 3>
   %shuf2 = shufflevector <4 x i8> %b, <4 x i8> zeroinitializer, <4 x i32><i32 0, i32 1, i32 4, i32 4>
diff --git a/llvm/test/CodeGen/X86/commute-blend-avx2.ll b/llvm/test/CodeGen/X86/commute-blend-avx2.ll
new file mode 100644
index 00000000000..d06c6dad8db
--- /dev/null
+++ b/llvm/test/CodeGen/X86/commute-blend-avx2.ll
@@ -0,0 +1,89 @@
+; RUN: llc -O3 -mtriple=x86_64-unknown -mcpu=core-avx2 -mattr=avx2 < %s | FileCheck %s
+
+define <8 x i16> @commute_fold_vpblendw_128(<8 x i16> %a, <8 x i16>* %b) #0 {
+  %1 = load <8 x i16>* %b
+  %2 = call <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16> %1, <8 x i16> %a, i8 17)
+  ret <8 x i16> %2
+
+  ;LABEL:      commute_fold_vpblendw_128
+  ;CHECK:      vpblendw {{.*#+}} xmm0 = xmm0[0],mem[1,2,3],xmm0[4],mem[5,6,7]
+  ;CHECK-NEXT: retq
+}
+declare <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16>, <8 x i16>, i8) nounwind readnone
+
+define <16 x i16> @commute_fold_vpblendw_256(<16 x i16> %a, <16 x i16>* %b) #0 {
+  %1 = load <16 x i16>* %b
+  %2 = call <16 x i16> @llvm.x86.avx2.pblendw(<16 x i16> %1, <16 x i16> %a, i8 17)
+  ret <16 x i16> %2
+
+  ;LABEL:      commute_fold_vpblendw_256
+  ;CHECK:      vpblendw {{.*#+}} ymm0 = ymm0[0],mem[1,2,3],ymm0[4],mem[5,6,7],ymm0[8],mem[9,10,11],ymm0[12],mem[13,14,15]
+  ;CHECK-NEXT: retq
+}
+declare <16 x i16> @llvm.x86.avx2.pblendw(<16 x i16>, <16 x i16>, i8) nounwind readnone
+
+define <4 x i32> @commute_fold_vpblendd_128(<4 x i32> %a, <4 x i32>* %b) #0 {
+  %1 = load <4 x i32>* %b
+  %2 = call <4 x i32> @llvm.x86.avx2.pblendd.128(<4 x i32> %1, <4 x i32> %a, i8 1)
+  ret <4 x i32> %2
+
+  ;LABEL:      commute_fold_vpblendd_128
+  ;CHECK:      vpblendd {{.*#+}} xmm0 = xmm0[0],mem[1,2,3]
+  ;CHECK-NEXT: retq
+}
+declare <4 x i32> @llvm.x86.avx2.pblendd.128(<4 x i32>, <4 x i32>, i8) nounwind readnone
+
+define <8 x i32> @commute_fold_vpblendd_256(<8 x i32> %a, <8 x i32>* %b) #0 {
+  %1 = load <8 x i32>* %b
+  %2 = call <8 x i32> @llvm.x86.avx2.pblendd.256(<8 x i32> %1, <8 x i32> %a, i8 129)
+  ret <8 x i32> %2
+
+  ;LABEL:      commute_fold_vpblendd_256
+  ;CHECK:      vpblendd {{.*#+}} ymm0 = ymm0[0],mem[1,2,3,4,5,6],ymm0[7]
+  ;CHECK-NEXT: retq
+}
+declare <8 x i32> @llvm.x86.avx2.pblendd.256(<8 x i32>, <8 x i32>, i8) nounwind readnone
+
+define <4 x float> @commute_fold_vblendps_128(<4 x float> %a, <4 x float>* %b) #0 {
+  %1 = load <4 x float>* %b
+  %2 = call <4 x float> @llvm.x86.sse41.blendps(<4 x float> %1, <4 x float> %a, i8 3)
+  ret <4 x float> %2
+
+  ;LABEL:      commute_fold_vblendps_128
+  ;CHECK:      vblendps {{.*#+}} xmm0 = xmm0[0,1],mem[2,3]
+  ;CHECK-NEXT: retq
+}
+declare <4 x float> @llvm.x86.sse41.blendps(<4 x float>, <4 x float>, i8) nounwind readnone
+
+define <8 x float> @commute_fold_vblendps_256(<8 x float> %a, <8 x float>* %b) #0 {
+  %1 = load <8 x float>* %b
+  %2 = call <8 x float> @llvm.x86.avx.blend.ps.256(<8 x float> %1, <8 x float> %a, i8 7)
+  ret <8 x float> %2
+
+  ;LABEL:      commute_fold_vblendps_256
+  ;CHECK:      vblendps {{.*#+}} ymm0 = ymm0[0,1,2],mem[3,4,5,6,7]
+  ;CHECK-NEXT: retq
+}
+declare <8 x float> @llvm.x86.avx.blend.ps.256(<8 x float>, <8 x float>, i8) nounwind readnone
+
+define <2 x double> @commute_fold_vblendpd_128(<2 x double> %a, <2 x double>* %b) #0 {
+  %1 = load <2 x double>* %b
+  %2 = call <2 x double> @llvm.x86.sse41.blendpd(<2 x double> %1, <2 x double> %a, i8 1)
+  ret <2 x double> %2
+
+  ;LABEL:      commute_fold_vblendpd_128
+  ;CHECK:      vblendpd {{.*#+}} xmm0 = xmm0[0],mem[1]
+  ;CHECK-NEXT: retq
+}
+declare <2 x double> @llvm.x86.sse41.blendpd(<2 x double>, <2 x double>, i8) nounwind readnone
+
+define <4 x double> @commute_fold_vblendpd_256(<4 x double> %a, <4 x double>* %b) #0 {
+  %1 = load <4 x double>* %b
+  %2 = call <4 x double> @llvm.x86.avx.blend.pd.256(<4 x double> %1, <4 x double> %a, i8 7)
+  ret <4 x double> %2
+
+  ;LABEL:      commute_fold_vblendpd_256
+  ;CHECK:      vblendpd {{.*#+}} ymm0 = ymm0[0,1,2],mem[3]
+  ;CHECK-NEXT: retq
+}
+declare <4 x double> @llvm.x86.avx.blend.pd.256(<4 x double>, <4 x double>, i8) nounwind readnone
diff --git a/llvm/test/CodeGen/X86/commute-blend-sse41.ll b/llvm/test/CodeGen/X86/commute-blend-sse41.ll
new file mode 100644
index 00000000000..59fef8c3a29
--- /dev/null
+++ b/llvm/test/CodeGen/X86/commute-blend-sse41.ll
@@ -0,0 +1,34 @@
+; RUN: llc -O3 -mtriple=x86_64-unknown -mcpu=corei7 < %s | FileCheck %s
+
+define <8 x i16> @commute_fold_pblendw(<8 x i16> %a, <8 x i16>* %b) #0 {
+  %1 = load <8 x i16>* %b
+  %2 = call <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16> %1, <8 x i16> %a, i8 17)
+  ret <8 x i16> %2
+
+  ;LABEL:      commute_fold_pblendw
+  ;CHECK:      pblendw {{.*#+}} xmm0 = xmm0[0],mem[1,2,3],xmm0[4],mem[5,6,7]
+  ;CHECK-NEXT: retq
+}
+declare <8 x i16> @llvm.x86.sse41.pblendw(<8 x i16>, <8 x i16>, i8) nounwind readnone
+
+define <4 x float> @commute_fold_blendps(<4 x float> %a, <4 x float>* %b) #0 {
+  %1 = load <4 x float>* %b
+  %2 = call <4 x float> @llvm.x86.sse41.blendps(<4 x float> %1, <4 x float> %a, i8 3)
+  ret <4 x float> %2
+
+  ;LABEL:      commute_fold_blendps
+  ;CHECK:      blendps {{.*#+}} xmm0 = xmm0[0,1],mem[2,3]
+  ;CHECK-NEXT: retq
+}
+declare <4 x float> @llvm.x86.sse41.blendps(<4 x float>, <4 x float>, i8) nounwind readnone
+
+define <2 x double> @commute_fold_blendpd(<2 x double> %a, <2 x double>* %b) #0 {
+  %1 = load <2 x double>* %b
+  %2 = call <2 x double> @llvm.x86.sse41.blendpd(<2 x double> %1, <2 x double> %a, i8 1)
+  ret <2 x double> %2
+
+  ;LABEL:      commute_fold_vblendpd
+  ;CHECK:      blendpd {{.*#+}} xmm0 = xmm0[0],mem[1]
+  ;CHECK-NEXT: retq
+}
+declare <2 x double> @llvm.x86.sse41.blendpd(<2 x double>, <2 x double>, i8) nounwind readnone
diff --git a/llvm/test/CodeGen/X86/vector-blend.ll b/llvm/test/CodeGen/X86/vector-blend.ll
index b66ed708b0d..7022b71336f 100644
--- a/llvm/test/CodeGen/X86/vector-blend.ll
+++ b/llvm/test/CodeGen/X86/vector-blend.ll
@@ -202,8 +202,7 @@ define <8 x i16> @vsel_8xi16(<8 x i16> %v1, <8 x i16> %v2) {
 ;
 ; SSE41-LABEL: vsel_8xi16:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: vsel_8xi16:
@@ -518,10 +517,8 @@ define <8 x float> @constant_blendvps_avx(<8 x float> %xyzw, <8 x float> %abcd)
 ;
 ; SSE41-LABEL: constant_blendvps_avx:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    blendps {{.*#+}} xmm2 = xmm2[0,1,2],xmm0[3]
-; SSE41-NEXT:    blendps {{.*#+}} xmm3 = xmm3[0,1,2],xmm1[3]
-; SSE41-NEXT:    movaps %xmm2, %xmm0
-; SSE41-NEXT:    movaps %xmm3, %xmm1
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm2[0,1,2],xmm0[3]
+; SSE41-NEXT:    blendps {{.*#+}} xmm1 = xmm3[0,1,2],xmm1[3]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: constant_blendvps_avx:
@@ -637,10 +634,8 @@ define <8 x float> @blend_shufflevector_8xfloat(<8 x float> %a, <8 x float> %b)
 ;
 ; SSE41-LABEL: blend_shufflevector_8xfloat:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    blendps {{.*#+}} xmm2 = xmm0[0],xmm2[1,2,3]
-; SSE41-NEXT:    blendps {{.*#+}} xmm3 = xmm3[0,1],xmm1[2],xmm3[3]
-; SSE41-NEXT:    movaps %xmm2, %xmm0
-; SSE41-NEXT:    movaps %xmm3, %xmm1
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3]
+; SSE41-NEXT:    blendps {{.*#+}} xmm1 = xmm3[0,1],xmm1[2],xmm3[3]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: blend_shufflevector_8xfloat:
@@ -694,8 +689,7 @@ define <4 x i64> @blend_shufflevector_4xi64(<4 x i64> %a, <4 x i64> %b) {
 ;
 ; SSE41-LABEL: blend_shufflevector_4xi64:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm0[4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
 ; SSE41-NEXT:    movaps %xmm3, %xmm1
 ; SSE41-NEXT:    retq
 ;
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-128-v2.ll b/llvm/test/CodeGen/X86/vector-shuffle-128-v2.ll
index e743194ef70..6e11c0e1079 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-128-v2.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-128-v2.ll
@@ -260,8 +260,7 @@ define <2 x double> @shuffle_v2f64_21(<2 x double> %a, <2 x double> %b) {
 ;
 ; SSE41-LABEL: shuffle_v2f64_21:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    blendpd {{.*#+}} xmm1 = xmm1[0],xmm0[1]
-; SSE41-NEXT:    movapd %xmm1, %xmm0
+; SSE41-NEXT:    blendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: shuffle_v2f64_21:
@@ -508,8 +507,7 @@ define <2 x i64> @shuffle_v2i64_21(<2 x i64> %a, <2 x i64> %b) {
 ;
 ; SSE41-LABEL: shuffle_v2i64_21:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm0[4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: shuffle_v2i64_21:
@@ -545,8 +543,8 @@ define <2 x i64> @shuffle_v2i64_21_copy(<2 x i64> %nonce, <2 x i64> %a, <2 x i64
 ;
 ; SSE41-LABEL: shuffle_v2i64_21_copy:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm1[4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7]
+; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: shuffle_v2i64_21_copy:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining.ll
index 8ad9e5a11d9..446812df50e 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining.ll
@@ -369,7 +369,8 @@ define <4 x i32> @combine_bitwise_ops_test3b(<4 x i32> %a, <4 x i32> %b, <4 x i3
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    pxor %xmm1, %xmm0
 ; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
+; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
+; SSE41-NEXT:    movdqa	%xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_bitwise_ops_test3b:
@@ -411,8 +412,7 @@ define <4 x i32> @combine_bitwise_ops_test4b(<4 x i32> %a, <4 x i32> %b, <4 x i3
 ; SSE41-LABEL: combine_bitwise_ops_test4b:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    pand %xmm1, %xmm0
-; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm0[2,3],xmm2[4,5],xmm0[6,7]
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3],xmm2[4,5],xmm0[6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_bitwise_ops_test4b:
@@ -452,8 +452,7 @@ define <4 x i32> @combine_bitwise_ops_test5b(<4 x i32> %a, <4 x i32> %b, <4 x i3
 ; SSE41-LABEL: combine_bitwise_ops_test5b:
 ; SSE41:       # BB#0:
 ; SSE41-NEXT:    por %xmm1, %xmm0
-; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm2[0,1],xmm0[2,3],xmm2[4,5],xmm0[6,7]
-; SSE41-NEXT:    movdqa %xmm2, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3],xmm2[4,5],xmm0[6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_bitwise_ops_test5b:
@@ -1170,8 +1169,7 @@ define <4 x float> @combine_test2(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_test2:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    blendps {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
-; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: combine_test2:
@@ -1237,8 +1235,7 @@ define <4 x float> @combine_test5(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_test5:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    blendps {{.*#+}} xmm1 = xmm1[0],xmm0[1],xmm1[2,3]
-; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: combine_test5:
@@ -1299,8 +1296,7 @@ define <4 x i32> @combine_test7(<4 x i32> %a, <4 x i32> %b) {
 ;
 ; SSE41-LABEL: combine_test7:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_test7:
@@ -1371,8 +1367,7 @@ define <4 x i32> @combine_test10(<4 x i32> %a, <4 x i32> %b) {
 ;
 ; SSE41-LABEL: combine_test10:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3],xmm1[4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3],xmm1[4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_test10:
@@ -1415,8 +1410,7 @@ define <4 x float> @combine_test12(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_test12:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    blendps {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
-; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: combine_test12:
@@ -1479,8 +1473,7 @@ define <4 x float> @combine_test15(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_test15:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    blendps {{.*#+}} xmm1 = xmm1[0],xmm0[1],xmm1[2,3]
-; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: combine_test15:
@@ -1518,8 +1511,7 @@ define <4 x i32> @combine_test17(<4 x i32> %a, <4 x i32> %b) {
 ;
 ; SSE41-LABEL: combine_test17:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0,1],xmm1[2,3,4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_test17:
@@ -1587,8 +1579,7 @@ define <4 x i32> @combine_test20(<4 x i32> %a, <4 x i32> %b) {
 ;
 ; SSE41-LABEL: combine_test20:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0,1],xmm0[2,3],xmm1[4,5,6,7]
-; SSE41-NEXT:    movdqa %xmm1, %xmm0
+; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3],xmm1[4,5,6,7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: combine_test20:
@@ -1632,10 +1623,9 @@ define <4 x float> @combine_test1b(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_test1b:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    movaps %xmm1, %xmm2
-; SSE41-NEXT:    blendps {{.*#+}} xmm2 = xmm2[0],xmm0[1],xmm2[2],xmm0[3]
-; SSE41-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm2[0,0]
-; SSE41-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm2[2,0]
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
+; SSE41-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
+; SSE41-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,0]
 ; SSE41-NEXT:    movaps %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
@@ -1673,11 +1663,9 @@ define <4 x float> @combine_test2b(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_test2b:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    movaps %xmm1, %xmm2
-; SSE41-NEXT:    blendps {{.*#+}} xmm2 = xmm2[0],xmm0[1],xmm2[2],xmm0[3]
-; SSE41-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,0],xmm1[1,1]
-; SSE41-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2,1,3]
-; SSE41-NEXT:    movaps %xmm2, %xmm0
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
+; SSE41-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[1,1]
+; SSE41-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2,1,3]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: combine_test2b:
@@ -1736,10 +1724,9 @@ define <4 x float> @combine_test4b(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_test4b:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    movaps %xmm1, %xmm2
-; SSE41-NEXT:    blendps {{.*#+}} xmm2 = xmm2[0],xmm0[1],xmm2[2],xmm0[3]
-; SSE41-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[3,0]
-; SSE41-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1],xmm2[0,2]
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
+; SSE41-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[3,0]
+; SSE41-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[0,2]
 ; SSE41-NEXT:    movaps %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
@@ -2006,8 +1993,7 @@ define <4 x float> @combine_blend_01(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_blend_01:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    blendpd {{.*#+}} xmm1 = xmm1[0],xmm0[1]
-; SSE41-NEXT:    movapd %xmm1, %xmm0
+; SSE41-NEXT:    blendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: combine_blend_01:
@@ -2036,8 +2022,7 @@ define <4 x float> @combine_blend_02(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_blend_02:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    blendps {{.*#+}} xmm1 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
-; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: combine_blend_02:
@@ -2070,8 +2055,7 @@ define <4 x float> @combine_blend_123(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_blend_123:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    blendps {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3]
-; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: combine_blend_123:
@@ -2153,8 +2137,7 @@ define <4 x float> @combine_undef_input_test1(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_undef_input_test1:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    blendpd {{.*#+}} xmm1 = xmm1[0],xmm0[1]
-; SSE41-NEXT:    movapd %xmm1, %xmm0
+; SSE41-NEXT:    blendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: combine_undef_input_test1:
@@ -2343,8 +2326,7 @@ define <4 x float> @combine_undef_input_test11(<4 x float> %a, <4 x float> %b) {
 ;
 ; SSE41-LABEL: combine_undef_input_test11:
 ; SSE41:       # BB#0:
-; SSE41-NEXT:    blendpd {{.*#+}} xmm1 = xmm1[0],xmm0[1]
-; SSE41-NEXT:    movapd %xmm1, %xmm0
+; SSE41-NEXT:    blendpd {{.*#+}} xmm0 = xmm1[0],xmm0[1]
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: combine_undef_input_test11: