From: Chandler Carruth <chandlerc@gmail.com>
Date: Thu, 2 Oct 2014 07:56:47 +0000 (+0000)
Subject: [x86] Merge the third combining test into the generic one and add proper
X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=cd912001b4419a20bab21709c7aabc68599c5c46;p=oota-llvm.git

[x86] Merge the third combining test into the generic one and add proper
checks for all the ISA variants.

If the SSE2 checks here terrify you, good. This is (in large part) the
kind of amazingly bad code that is holding LLVM back when vectorizing on
older ISAs.

At the same time, these tests seem increasingly dubious to me. There are
a very large number of tests and it isn't clear that they are
systematically covering a specific set of functionality. Anyways,
I don't want to reduce testing during the transition, I just want to
consolidate it to where it is easier to manage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218860 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/test/CodeGen/X86/combine-vec-shuffle-3.ll b/test/CodeGen/X86/combine-vec-shuffle-3.ll
deleted file mode 100644
index bd2d34ca189..00000000000
--- a/test/CodeGen/X86/combine-vec-shuffle-3.ll
+++ /dev/null
@@ -1,380 +0,0 @@
-; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 | FileCheck %s
-
-define <4 x float> @test1(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test1
-; Mask: [0,1,2,3]
-; CHECK: movaps
-; CHECK: ret
-
-define <4 x float> @test2(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
-  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test2
-; Mask: [0,5,6,7]
-; CHECK: movss
-; CHECK: ret
-
-define <4 x float> @test3(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 1, i32 7>
-  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 2, i32 4, i32 1>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test3
-; Mask: [0,1,4,5]
-; CHECK: movlhps
-; CHECK: ret
-
-define <4 x float> @test4(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 2, i32 3, i32 5, i32 5>
-  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test4
-; Mask: [6,7,2,3]
-; CHECK: movhlps
-; CHECK-NEXT: ret
-
-define <4 x float> @test5(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test5
-; Mask: [4,1,6,7]
-; CHECK: blendps $13
-; CHECK: ret
-
-
-define <4 x i32> @test6(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test6
-; Mask: [4,5,6,7]
-; CHECK: movaps
-; CHECK: ret
-
-define <4 x i32> @test7(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test7
-; Mask: [0,5,6,7]
-; CHECK: movss
-; CHECK: ret
-
-define <4 x i32> @test8(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 1, i32 7>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 0, i32 2, i32 4, i32 1>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test8
-; Mask: [0,1,4,5]
-; CHECK: movlhps
-; CHECK: ret
-
-define <4 x i32> @test9(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 3, i32 5, i32 5>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test9
-; Mask: [6,7,2,3]
-; CHECK: movhlps
-; CHECK-NEXT: ret
-
-define <4 x i32> @test10(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test10
-; Mask: [4,1,6,7]
-; CHECK: blendps
-; CHECK: ret
-
-define <4 x float> @test11(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x float> %1, <4 x float> %a, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test11
-; Mask: [0,1,2,3]
-; CHECK-NOT: movaps
-; CHECK-NOT: blendps
-; CHECK: ret
-
-define <4 x float> @test12(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 6, i32 7>
-  %2 = shufflevector <4 x float> %1, <4 x float> %a, <4 x i32> <i32 4, i32 1, i32 2, i32 3>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test12
-; Mask: [0,5,6,7]
-; CHECK: movss
-; CHECK: ret
-
-define <4 x float> @test13(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
-  %2 = shufflevector <4 x float> %1, <4 x float> %a, <4 x i32> <i32 4, i32 5, i32 2, i32 3>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test13
-; Mask: [0,1,4,5]
-; CHECK: movlhps
-; CHECK: ret
-
-define <4 x float> @test14(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 6, i32 7, i32 5, i32 5>
-  %2 = shufflevector <4 x float> %1, <4 x float> %a, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test14
-; Mask: [6,7,2,3]
-; CHECK: movhlps
-; CHECK: ret
-
-define <4 x float> @test15(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 7>
-  %2 = shufflevector <4 x float> %1, <4 x float> %a, <4 x i32> <i32 0, i32 5, i32 2, i32 3>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test15
-; Mask: [4,1,6,7]
-; CHECK: blendps $13
-; CHECK: ret
-
-define <4 x i32> @test16(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %a, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test16
-; Mask: [0,1,2,3]
-; CHECK-NOT: movaps
-; CHECK-NOT: blendps
-; CHECK: ret
-
-define <4 x i32> @test17(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 6, i32 7>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %a, <4 x i32> <i32 4, i32 1, i32 2, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test17
-; Mask: [0,5,6,7]
-; CHECK: movss
-; CHECK: ret
-
-define <4 x i32> @test18(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %a, <4 x i32> <i32 4, i32 5, i32 2, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test18
-; Mask: [0,1,4,5]
-; CHECK: movlhps
-; CHECK: ret
-
-define <4 x i32> @test19(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 6, i32 7, i32 5, i32 5>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %a, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test19
-; Mask: [6,7,2,3]
-; CHECK: movhlps
-; CHECK: ret
-
-define <4 x i32> @test20(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 7>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %a, <4 x i32> <i32 0, i32 5, i32 2, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test20
-; Mask: [4,1,6,7]
-; CHECK: blendps $13
-; CHECK: ret
-
-; Check some negative cases.
-define <4 x float> @test1b(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 0>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test1b
-; CHECK: shufps
-; CHECK: shufps
-; CHECK: ret
-
-define <4 x float> @test2b(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 0, i32 5>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test2b
-; CHECK: shufps
-; CHECK: pshufd
-; CHECK: ret
-
-define <4 x float> @test3b(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 0, i32 6, i32 3>
-  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 7, i32 2, i32 7>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test3b
-; CHECK: shufps
-; CHECK: shufps
-; CHECK: ret
-
-define <4 x float> @test4b(<4 x float> %a, <4 x float> %b) {
-  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 5, i32 5, i32 2, i32 7>
-  ret <4 x float> %2
-}
-; CHECK-LABEL: test4b
-; CHECK: shufps
-; CHECK: shufps
-; CHECK: ret
-
-
-; Verify that we correctly fold shuffles even when we use illegal vector types.
-define <4 x i8> @test1c(<4 x i8>* %a, <4 x i8>* %b) {
-  %A = load <4 x i8>* %a
-  %B = load <4 x i8>* %b
-  %1 = shufflevector <4 x i8> %A, <4 x i8> %B, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
-  %2 = shufflevector <4 x i8> %1, <4 x i8> %B, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
-  ret <4 x i8> %2
-}
-; CHECK-LABEL: test1c
-; Mask: [0,5,6,7]
-; CHECK: movss
-; CHECK-NEXT: ret
-
-define <4 x i8> @test2c(<4 x i8>* %a, <4 x i8>* %b) {
-  %A = load <4 x i8>* %a
-  %B = load <4 x i8>* %b
-  %1 = shufflevector <4 x i8> %A, <4 x i8> %B, <4 x i32> <i32 0, i32 5, i32 1, i32 5>
-  %2 = shufflevector <4 x i8> %1, <4 x i8> %B, <4 x i32> <i32 0, i32 2, i32 4, i32 1>
-  ret <4 x i8> %2
-}
-; CHECK-LABEL: test2c
-; Mask: [0,1,4,5]
-; CHECK: movlhps
-; CHECK-NEXT: ret
-
-define <4 x i8> @test3c(<4 x i8>* %a, <4 x i8>* %b) {
-  %A = load <4 x i8>* %a
-  %B = load <4 x i8>* %b
-  %1 = shufflevector <4 x i8> %A, <4 x i8> %B, <4 x i32> <i32 2, i32 3, i32 5, i32 5>
-  %2 = shufflevector <4 x i8> %1, <4 x i8> %B, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
-  ret <4 x i8> %2
-}
-; CHECK-LABEL: test3c
-; Mask: [6,7,2,3]
-; CHECK: movhlps
-; CHECK-NEXT: ret
-
-define <4 x i8> @test4c(<4 x i8>* %a, <4 x i8>* %b) {
-  %A = load <4 x i8>* %a
-  %B = load <4 x i8>* %b
-  %1 = shufflevector <4 x i8> %A, <4 x i8> %B, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x i8> %1, <4 x i8> %B, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
-  ret <4 x i8> %2
-}
-; CHECK-LABEL: test4c
-; Mask: [4,1,6,7]
-; CHECK: blendps $13
-; CHECK: ret
-
-; The following test cases are generated from this C++ code
-; 
-;__m128 blend_01(__m128 a, __m128 b)
-;{
-;  __m128 s = a;
-;  s = _mm_blend_ps( s, b, 1<<0 );
-;  s = _mm_blend_ps( s, b, 1<<1 );
-;  return s;
-;}
-;
-;__m128 blend_02(__m128 a, __m128 b)
-;{
-;  __m128 s = a;
-;  s = _mm_blend_ps( s, b, 1<<0 );
-;  s = _mm_blend_ps( s, b, 1<<2 );
-;  return s;
-;}
-;
-;__m128 blend_123(__m128 a, __m128 b)
-;{
-;  __m128 s = a;
-;  s = _mm_blend_ps( s, b, 1<<1 );
-;  s = _mm_blend_ps( s, b, 1<<2 );
-;  s = _mm_blend_ps( s, b, 1<<3 );
-;  return s;
-;}
-
-; Ideally, we should collapse the following shuffles into a single one.
-
-define <4 x float> @blend_01(<4 x float> %a, <4 x float> %b) {
-  %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 undef, i32 2, i32 3>
-  %shuffle6 = shufflevector <4 x float> %shuffle, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 3>
-  ret <4 x float> %shuffle6
-}
-; CHECK-LABEL: blend_01
-; CHECK: movsd
-; CHECK-NEXT: ret
-
-define <4 x float> @blend_02(<4 x float> %a, <4 x float> %b) {
-  %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 undef, i32 3>
-  %shuffle6 = shufflevector <4 x float> %shuffle, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
-  ret <4 x float> %shuffle6
-}
-; CHECK-LABEL: blend_02
-; CHECK: blendps $5
-; CHECK-NEXT: ret
-
-define <4 x float> @blend_123(<4 x float> %a, <4 x float> %b) {
-  %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 undef, i32 undef>
-  %shuffle6 = shufflevector <4 x float> %shuffle, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 6, i32 undef>
-  %shuffle12 = shufflevector <4 x float> %shuffle6, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
-  ret <4 x float> %shuffle12
-}
-; CHECK-LABEL: blend_123
-; CHECK: movss
-; CHECK: ret
-
-define <4 x i32> @test_movhl_1(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 7, i32 5, i32 3>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 6, i32 1, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test_movhl_1
-; CHECK: movhlps
-; CHECK-NEXT: ret
-
-define <4 x i32> @test_movhl_2(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 0, i32 3, i32 6>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 3, i32 7, i32 0, i32 2>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test_movhl_2
-; CHECK: movhlps
-; CHECK-NEXT: ret
-
-define <4 x i32> @test_movhl_3(<4 x i32> %a, <4 x i32> %b) {
-  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 7, i32 6, i32 3, i32 2>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 6, i32 0, i32 3, i32 2>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test_movhl_3
-; CHECK: movhlps
-; CHECK-NEXT: ret
-
diff --git a/test/CodeGen/X86/vector-shuffle-combining.ll b/test/CodeGen/X86/vector-shuffle-combining.ll
index d8e6cf2b8c1..dd0961769e7 100644
--- a/test/CodeGen/X86/vector-shuffle-combining.ll
+++ b/test/CodeGen/X86/vector-shuffle-combining.ll
@@ -1097,3 +1097,1004 @@ define <4 x i32> @combine_nested_undef_test28(<4 x i32> %A, <4 x i32> %B) {
   %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 3, i32 2>
   ret <4 x i32> %2
 }
+
+define <4 x float> @combine_test1(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_test1:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test1:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movaps %xmm1, %xmm0
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test1:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test1:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovaps %xmm1, %xmm0
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test2(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_test2:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test2:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test2:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    movss %xmm0, %xmm1
+; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test2:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovss %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test3(<4 x float> %a, <4 x float> %b) {
+; SSE-LABEL: combine_test3:
+; SSE:       # BB#0:
+; SSE-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test3:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 1, i32 7>
+  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 2, i32 4, i32 1>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test4(<4 x float> %a, <4 x float> %b) {
+; SSE-LABEL: combine_test4:
+; SSE:       # BB#0:
+; SSE-NEXT:    movhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test4:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 2, i32 3, i32 5, i32 5>
+  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test5(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_test5:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test5:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test5:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test5:
+; AVX:       # BB#0:
+; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
+  ret <4 x float> %2
+}
+
+define <4 x i32> @combine_test6(<4 x i32> %a, <4 x i32> %b) {
+; SSE2-LABEL: combine_test6:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test6:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movaps %xmm1, %xmm0
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test6:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test6:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovaps %xmm1, %xmm0
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_test7(<4 x i32> %a, <4 x i32> %b) {
+; SSE2-LABEL: combine_test7:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test7:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test7:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    movss %xmm0, %xmm1
+; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test7:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovss %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_test8(<4 x i32> %a, <4 x i32> %b) {
+; SSE-LABEL: combine_test8:
+; SSE:       # BB#0:
+; SSE-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test8:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 1, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 0, i32 2, i32 4, i32 1>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_test9(<4 x i32> %a, <4 x i32> %b) {
+; SSE-LABEL: combine_test9:
+; SSE:       # BB#0:
+; SSE-NEXT:    movhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test9:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 3, i32 5, i32 5>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_test10(<4 x i32> %a, <4 x i32> %b) {
+; SSE2-LABEL: combine_test10:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test10:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test10:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; SSE41-NEXT:    retq
+;
+; AVX1-LABEL: combine_test10:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: combine_test10:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; AVX2-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
+  ret <4 x i32> %2
+}
+
+define <4 x float> @combine_test11(<4 x float> %a, <4 x float> %b) {
+; ALL-LABEL: combine_test11:
+; ALL:       # BB#0:
+; ALL-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  %2 = shufflevector <4 x float> %1, <4 x float> %a, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test12(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_test12:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movss %xmm0, %xmm1
+; SSE2-NEXT:    movss %xmm0, %xmm1
+; SSE2-NEXT:    movaps %xmm1, %xmm0
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test12:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movss %xmm0, %xmm1
+; SSSE3-NEXT:    movss %xmm0, %xmm1
+; SSSE3-NEXT:    movaps %xmm1, %xmm0
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test12:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    movss %xmm0, %xmm1
+; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test12:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovss %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 6, i32 7>
+  %2 = shufflevector <4 x float> %1, <4 x float> %a, <4 x i32> <i32 4, i32 1, i32 2, i32 3>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test13(<4 x float> %a, <4 x float> %b) {
+; SSE-LABEL: combine_test13:
+; SSE:       # BB#0:
+; SSE-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test13:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+  %2 = shufflevector <4 x float> %1, <4 x float> %a, <4 x i32> <i32 4, i32 5, i32 2, i32 3>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test14(<4 x float> %a, <4 x float> %b) {
+; SSE-LABEL: combine_test14:
+; SSE:       # BB#0:
+; SSE-NEXT:    movhlps {{.*#+}} xmm1 = xmm0[1],xmm1[1]
+; SSE-NEXT:    movaps %xmm1, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test14:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovhlps {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 6, i32 7, i32 5, i32 5>
+  %2 = shufflevector <4 x float> %1, <4 x float> %a, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test15(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_test15:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movaps %xmm0, %xmm2
+; SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm1[0,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[2,3]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm2[0,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm2[2,3]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test15:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movaps %xmm0, %xmm2
+; SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm1[0,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[2,3]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm2[0,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm2[2,3]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test15:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test15:
+; AVX:       # BB#0:
+; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 7>
+  %2 = shufflevector <4 x float> %1, <4 x float> %a, <4 x i32> <i32 0, i32 5, i32 2, i32 3>
+  ret <4 x float> %2
+}
+
+define <4 x i32> @combine_test16(<4 x i32> %a, <4 x i32> %b) {
+; ALL-LABEL: combine_test16:
+; ALL:       # BB#0:
+; ALL-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %a, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_test17(<4 x i32> %a, <4 x i32> %b) {
+; SSE2-LABEL: combine_test17:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movss %xmm0, %xmm1
+; SSE2-NEXT:    movss %xmm0, %xmm1
+; SSE2-NEXT:    movaps %xmm1, %xmm0
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test17:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movss %xmm0, %xmm1
+; SSSE3-NEXT:    movss %xmm0, %xmm1
+; SSSE3-NEXT:    movaps %xmm1, %xmm0
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test17:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    movss %xmm0, %xmm1
+; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test17:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovss %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 6, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %a, <4 x i32> <i32 4, i32 1, i32 2, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_test18(<4 x i32> %a, <4 x i32> %b) {
+; SSE-LABEL: combine_test18:
+; SSE:       # BB#0:
+; SSE-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test18:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %a, <4 x i32> <i32 4, i32 5, i32 2, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_test19(<4 x i32> %a, <4 x i32> %b) {
+; SSE-LABEL: combine_test19:
+; SSE:       # BB#0:
+; SSE-NEXT:    movhlps {{.*#+}} xmm1 = xmm0[1],xmm1[1]
+; SSE-NEXT:    movaps %xmm1, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test19:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovhlps {{.*#+}} xmm0 = xmm0[1],xmm1[1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 6, i32 7, i32 5, i32 5>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %a, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_test20(<4 x i32> %a, <4 x i32> %b) {
+; SSE2-LABEL: combine_test20:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movaps %xmm0, %xmm2
+; SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm1[0,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[2,3]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm2[0,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm2[2,3]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test20:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movaps %xmm0, %xmm2
+; SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,0],xmm1[0,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm1[2,3]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,0],xmm2[0,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm2[2,3]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test20:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; SSE41-NEXT:    retq
+;
+; AVX1-LABEL: combine_test20:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: combine_test20:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; AVX2-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %a, <4 x i32> <i32 0, i32 5, i32 2, i32 3>
+  ret <4 x i32> %2
+}
+
+
+; Check some negative cases.
+; FIXME: Do any of these really make sense? Are they redundant with the above tests?
+
+define <4 x float> @combine_test1b(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_test1b:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,0]
+; SSE2-NEXT:    movaps %xmm1, %xmm0
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test1b:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,0]
+; SSSE3-NEXT:    movaps %xmm1, %xmm0
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test1b:
+; SSE41:       # BB#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
+;
+; AVX-LABEL: combine_test1b:
+; AVX:       # BB#0:
+; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
+; AVX-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[2,0],xmm0[2,0]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 0>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test2b(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_test2b:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[1,1]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test2b:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[1,1]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test2b:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
+; SSE41-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[1,1]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test2b:
+; AVX:       # BB#0:
+; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,0],xmm1[1,1]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 0, i32 5>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test3b(<4 x float> %a, <4 x float> %b) {
+; SSE-LABEL: combine_test3b:
+; SSE:       # BB#0:
+; SSE-NEXT:    movaps %xmm1, %xmm2
+; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm0[3,0]
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,0],xmm2[0,2]
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[3,3]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test3b:
+; AVX:       # BB#0:
+; AVX-NEXT:    vshufps {{.*#+}} xmm2 = xmm1[2,0],xmm0[3,0]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,0],xmm2[0,2]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[3,3]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 0, i32 6, i32 3>
+  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 0, i32 7, i32 2, i32 7>
+  ret <4 x float> %2
+}
+
+define <4 x float> @combine_test4b(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_test4b:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[3,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[0,2]
+; SSE2-NEXT:    movaps %xmm1, %xmm0
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test4b:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[3,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,1],xmm0[0,2]
+; SSSE3-NEXT:    movaps %xmm1, %xmm0
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test4b:
+; SSE41:       # BB#0:
+; 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
+;
+; AVX-LABEL: combine_test4b:
+; AVX:       # BB#0:
+; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,0],xmm1[3,0]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[1,1],xmm0[0,2]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  %2 = shufflevector <4 x float> %1, <4 x float> %b, <4 x i32> <i32 5, i32 5, i32 2, i32 7>
+  ret <4 x float> %2
+}
+
+
+; Verify that we correctly fold shuffles even when we use illegal vector types.
+
+define <4 x i8> @combine_test1c(<4 x i8>* %a, <4 x i8>* %b) {
+; SSE2-LABEL: combine_test1c:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movl (%rdi), %eax
+; SSE2-NEXT:    movd %eax, %xmm0
+; SSE2-NEXT:    pextrw $1, %xmm0, %ecx
+; SSE2-NEXT:    pinsrw $0, %eax, %xmm0
+; SSE2-NEXT:    movzbl %ah, %eax
+; SSE2-NEXT:    pinsrw $2, %eax, %xmm0
+; SSE2-NEXT:    pinsrw $4, %ecx, %xmm0
+; SSE2-NEXT:    shrl $8, %ecx
+; SSE2-NEXT:    pinsrw $6, %ecx, %xmm0
+; SSE2-NEXT:    movl (%rsi), %eax
+; SSE2-NEXT:    movd %eax, %xmm1
+; SSE2-NEXT:    pextrw $1, %xmm1, %ecx
+; SSE2-NEXT:    pinsrw $0, %eax, %xmm1
+; SSE2-NEXT:    movzbl %ah, %eax
+; SSE2-NEXT:    pinsrw $2, %eax, %xmm1
+; SSE2-NEXT:    pinsrw $4, %ecx, %xmm1
+; SSE2-NEXT:    shrl $8, %ecx
+; SSE2-NEXT:    pinsrw $6, %ecx, %xmm1
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test1c:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movd (%rdi), %xmm0
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [0,128,128,128,1,128,128,128,2,128,128,128,3,128,128,128]
+; SSSE3-NEXT:    pshufb %xmm1, %xmm0
+; SSSE3-NEXT:    movd (%rsi), %xmm2
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[1,3]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,0],xmm0[3,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm2[0,2]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test1c:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    pmovzxbd (%rdi), %xmm1
+; SSE41-NEXT:    pmovzxbd (%rsi), %xmm0
+; SSE41-NEXT:    movss %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test1c:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpmovzxbd (%rdi), %xmm0
+; AVX-NEXT:    vpmovzxbd (%rsi), %xmm1
+; AVX-NEXT:    vmovss %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    retq
+  %A = load <4 x i8>* %a
+  %B = load <4 x i8>* %b
+  %1 = shufflevector <4 x i8> %A, <4 x i8> %B, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %2 = shufflevector <4 x i8> %1, <4 x i8> %B, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
+  ret <4 x i8> %2
+}
+
+define <4 x i8> @combine_test2c(<4 x i8>* %a, <4 x i8>* %b) {
+; SSE2-LABEL: combine_test2c:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movl (%rdi), %eax
+; SSE2-NEXT:    movd %eax, %xmm0
+; SSE2-NEXT:    pextrw $1, %xmm0, %ecx
+; SSE2-NEXT:    pinsrw $0, %eax, %xmm0
+; SSE2-NEXT:    movzbl %ah, %eax
+; SSE2-NEXT:    pinsrw $2, %eax, %xmm0
+; SSE2-NEXT:    pinsrw $4, %ecx, %xmm0
+; SSE2-NEXT:    shrl $8, %ecx
+; SSE2-NEXT:    pinsrw $6, %ecx, %xmm0
+; SSE2-NEXT:    movl (%rsi), %eax
+; SSE2-NEXT:    movd %eax, %xmm1
+; SSE2-NEXT:    pextrw $1, %xmm1, %ecx
+; SSE2-NEXT:    pinsrw $0, %eax, %xmm1
+; SSE2-NEXT:    movzbl %ah, %eax
+; SSE2-NEXT:    pinsrw $2, %eax, %xmm1
+; SSE2-NEXT:    pinsrw $4, %ecx, %xmm1
+; SSE2-NEXT:    shrl $8, %ecx
+; SSE2-NEXT:    pinsrw $6, %ecx, %xmm1
+; SSE2-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test2c:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movd (%rdi), %xmm0
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [0,128,128,128,1,128,128,128,2,128,128,128,3,128,128,128]
+; SSSE3-NEXT:    pshufb %xmm1, %xmm0
+; SSSE3-NEXT:    movd (%rsi), %xmm2
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm2[0]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test2c:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    pmovzxbd (%rdi), %xmm0
+; SSE41-NEXT:    pmovzxbd (%rsi), %xmm1
+; SSE41-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test2c:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpmovzxbd (%rdi), %xmm0
+; AVX-NEXT:    vpmovzxbd (%rsi), %xmm1
+; AVX-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX-NEXT:    retq
+  %A = load <4 x i8>* %a
+  %B = load <4 x i8>* %b
+  %1 = shufflevector <4 x i8> %A, <4 x i8> %B, <4 x i32> <i32 0, i32 5, i32 1, i32 5>
+  %2 = shufflevector <4 x i8> %1, <4 x i8> %B, <4 x i32> <i32 0, i32 2, i32 4, i32 1>
+  ret <4 x i8> %2
+}
+
+define <4 x i8> @combine_test3c(<4 x i8>* %a, <4 x i8>* %b) {
+; SSE2-LABEL: combine_test3c:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movl (%rdi), %eax
+; SSE2-NEXT:    movd %eax, %xmm0
+; SSE2-NEXT:    pextrw $1, %xmm0, %ecx
+; SSE2-NEXT:    pinsrw $0, %eax, %xmm0
+; SSE2-NEXT:    movzbl %ah, %eax
+; SSE2-NEXT:    pinsrw $2, %eax, %xmm0
+; SSE2-NEXT:    pinsrw $4, %ecx, %xmm0
+; SSE2-NEXT:    shrl $8, %ecx
+; SSE2-NEXT:    pinsrw $6, %ecx, %xmm0
+; SSE2-NEXT:    movl (%rsi), %eax
+; SSE2-NEXT:    movd %eax, %xmm1
+; SSE2-NEXT:    pextrw $1, %xmm1, %ecx
+; SSE2-NEXT:    pinsrw $0, %eax, %xmm1
+; SSE2-NEXT:    movzbl %ah, %eax
+; SSE2-NEXT:    pinsrw $2, %eax, %xmm1
+; SSE2-NEXT:    pinsrw $4, %ecx, %xmm1
+; SSE2-NEXT:    shrl $8, %ecx
+; SSE2-NEXT:    pinsrw $6, %ecx, %xmm1
+; SSE2-NEXT:    movhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test3c:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movd (%rdi), %xmm0
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [0,128,128,128,1,128,128,128,2,128,128,128,3,128,128,128]
+; SSSE3-NEXT:    pshufb %xmm1, %xmm0
+; SSSE3-NEXT:    movd (%rsi), %xmm2
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    movhlps {{.*#+}} xmm0 = xmm2[1],xmm0[1]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test3c:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    pmovzxbd (%rdi), %xmm0
+; SSE41-NEXT:    pmovzxbd (%rsi), %xmm1
+; SSE41-NEXT:    movhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_test3c:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpmovzxbd (%rdi), %xmm0
+; AVX-NEXT:    vpmovzxbd (%rsi), %xmm1
+; AVX-NEXT:    vmovhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; AVX-NEXT:    retq
+  %A = load <4 x i8>* %a
+  %B = load <4 x i8>* %b
+  %1 = shufflevector <4 x i8> %A, <4 x i8> %B, <4 x i32> <i32 2, i32 3, i32 5, i32 5>
+  %2 = shufflevector <4 x i8> %1, <4 x i8> %B, <4 x i32> <i32 6, i32 7, i32 0, i32 1>
+  ret <4 x i8> %2
+}
+
+define <4 x i8> @combine_test4c(<4 x i8>* %a, <4 x i8>* %b) {
+; SSE2-LABEL: combine_test4c:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movl (%rdi), %eax
+; SSE2-NEXT:    movd %eax, %xmm0
+; SSE2-NEXT:    pextrw $1, %xmm0, %ecx
+; SSE2-NEXT:    pinsrw $0, %eax, %xmm0
+; SSE2-NEXT:    movzbl %ah, %eax
+; SSE2-NEXT:    pinsrw $2, %eax, %xmm0
+; SSE2-NEXT:    pinsrw $4, %ecx, %xmm0
+; SSE2-NEXT:    shrl $8, %ecx
+; SSE2-NEXT:    pinsrw $6, %ecx, %xmm0
+; SSE2-NEXT:    movl (%rsi), %eax
+; SSE2-NEXT:    movd %eax, %xmm1
+; SSE2-NEXT:    pextrw $1, %xmm1, %ecx
+; SSE2-NEXT:    pinsrw $0, %eax, %xmm1
+; SSE2-NEXT:    movzbl %ah, %eax
+; SSE2-NEXT:    pinsrw $2, %eax, %xmm1
+; SSE2-NEXT:    pinsrw $4, %ecx, %xmm1
+; SSE2-NEXT:    shrl $8, %ecx
+; SSE2-NEXT:    pinsrw $6, %ecx, %xmm1
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[0,2]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[3,0],xmm0[2,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,0]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_test4c:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movd (%rdi), %xmm0
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm1 = [0,128,128,128,1,128,128,128,2,128,128,128,3,128,128,128]
+; SSSE3-NEXT:    pshufb %xmm1, %xmm0
+; SSSE3-NEXT:    movd (%rsi), %xmm2
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm2[0,2]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,3,1]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[3,0],xmm0[2,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,0]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_test4c:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    pmovzxbd (%rdi), %xmm0
+; SSE41-NEXT:    pmovzxbd (%rsi), %xmm1
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; SSE41-NEXT:    retq
+;
+; AVX1-LABEL: combine_test4c:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vpmovzxbd (%rdi), %xmm0
+; AVX1-NEXT:    vpmovzxbd (%rsi), %xmm1
+; AVX1-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: combine_test4c:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpmovzxbd (%rdi), %xmm0
+; AVX2-NEXT:    vpmovzxbd (%rsi), %xmm1
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3]
+; AVX2-NEXT:    retq
+  %A = load <4 x i8>* %a
+  %B = load <4 x i8>* %b
+  %1 = shufflevector <4 x i8> %A, <4 x i8> %B, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  %2 = shufflevector <4 x i8> %1, <4 x i8> %B, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
+  ret <4 x i8> %2
+}
+
+
+; The following test cases are generated from this C++ code
+;
+;__m128 blend_01(__m128 a, __m128 b)
+;{
+;  __m128 s = a;
+;  s = _mm_blend_ps( s, b, 1<<0 );
+;  s = _mm_blend_ps( s, b, 1<<1 );
+;  return s;
+;}
+;
+;__m128 blend_02(__m128 a, __m128 b)
+;{
+;  __m128 s = a;
+;  s = _mm_blend_ps( s, b, 1<<0 );
+;  s = _mm_blend_ps( s, b, 1<<2 );
+;  return s;
+;}
+;
+;__m128 blend_123(__m128 a, __m128 b)
+;{
+;  __m128 s = a;
+;  s = _mm_blend_ps( s, b, 1<<1 );
+;  s = _mm_blend_ps( s, b, 1<<2 );
+;  s = _mm_blend_ps( s, b, 1<<3 );
+;  return s;
+;}
+
+; Ideally, we should collapse the following shuffles into a single one.
+
+define <4 x float> @combine_blend_01(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_blend_01:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movss %xmm1, %xmm0
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
+; SSE2-NEXT:    movaps %xmm1, %xmm0
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_blend_01:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movss %xmm1, %xmm0
+; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[1,0],xmm0[0,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[2,3]
+; SSSE3-NEXT:    movaps %xmm1, %xmm0
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_blend_01:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    movsd %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_blend_01:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovsd %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    retq
+  %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 undef, i32 2, i32 3>
+  %shuffle6 = shufflevector <4 x float> %shuffle, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 3>
+  ret <4 x float> %shuffle6
+}
+
+define <4 x float> @combine_blend_02(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_blend_02:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movss %xmm1, %xmm0
+; SSE2-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_blend_02:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movss %xmm1, %xmm0
+; SSSE3-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,0]
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0,2]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_blend_02:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_blend_02:
+; AVX:       # BB#0:
+; AVX-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2],xmm0[3]
+; AVX-NEXT:    retq
+  %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 undef, i32 3>
+  %shuffle6 = shufflevector <4 x float> %shuffle, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
+  ret <4 x float> %shuffle6
+}
+
+define <4 x float> @combine_blend_123(<4 x float> %a, <4 x float> %b) {
+; SSE2-LABEL: combine_blend_123:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    movaps %xmm1, %xmm2
+; SSE2-NEXT:    movss %xmm0, %xmm2
+; SSE2-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3]
+; SSE2-NEXT:    movaps %xmm2, %xmm0
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_blend_123:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    movaps %xmm1, %xmm2
+; SSSE3-NEXT:    movss %xmm0, %xmm2
+; SSSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,1],xmm1[2,3]
+; SSSE3-NEXT:    movaps %xmm2, %xmm0
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_blend_123:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    movss %xmm0, %xmm1
+; SSE41-NEXT:    movaps %xmm1, %xmm0
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: combine_blend_123:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovss %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    retq
+  %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 undef, i32 undef>
+  %shuffle6 = shufflevector <4 x float> %shuffle, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 6, i32 undef>
+  %shuffle12 = shufflevector <4 x float> %shuffle6, <4 x float> %b, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
+  ret <4 x float> %shuffle12
+}
+
+define <4 x i32> @combine_test_movhl_1(<4 x i32> %a, <4 x i32> %b) {
+; SSE-LABEL: combine_test_movhl_1:
+; SSE:       # BB#0:
+; SSE-NEXT:    movhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test_movhl_1:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 7, i32 5, i32 3>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 6, i32 1, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_test_movhl_2(<4 x i32> %a, <4 x i32> %b) {
+; SSE-LABEL: combine_test_movhl_2:
+; SSE:       # BB#0:
+; SSE-NEXT:    movhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test_movhl_2:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 0, i32 3, i32 6>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 3, i32 7, i32 0, i32 2>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_test_movhl_3(<4 x i32> %a, <4 x i32> %b) {
+; SSE-LABEL: combine_test_movhl_3:
+; SSE:       # BB#0:
+; SSE-NEXT:    movhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_test_movhl_3:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovhlps {{.*#+}} xmm0 = xmm1[1],xmm0[1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 7, i32 6, i32 3, i32 2>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %b, <4 x i32> <i32 6, i32 0, i32 3, i32 2>
+  ret <4 x i32> %2
+}