From 84b3f53bcb753d886663dfda1423325bf4056720 Mon Sep 17 00:00:00 2001
From: Chandler Carruth <chandlerc@gmail.com>
Date: Thu, 2 Oct 2014 07:42:58 +0000
Subject: [PATCH] [x86] Merge the second set of vector combining tests into a
 common test file.

Some of these really don't make sense to test -- we're testing for the
*lack* of combining two shuffles into one, presumably because the two
would generate better shuffles in the end. But if you look at the
generated code shown here, in many cases the generated code is, frankly,
terrible. Or we combine any two generated shuffles back into a single
instruction! I've left a FIXME to revisit these decisions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@218859 91177308-0d34-0410-b5e6-96231b3b80d8
---
 test/CodeGen/X86/combine-vec-shuffle-2.ll    | 317 -------------
 test/CodeGen/X86/vector-shuffle-combining.ll | 470 +++++++++++++++++++
 2 files changed, 470 insertions(+), 317 deletions(-)
 delete mode 100644 test/CodeGen/X86/combine-vec-shuffle-2.ll

diff --git a/test/CodeGen/X86/combine-vec-shuffle-2.ll b/test/CodeGen/X86/combine-vec-shuffle-2.ll
deleted file mode 100644
index 872384ca0a1..00000000000
--- a/test/CodeGen/X86/combine-vec-shuffle-2.ll
+++ /dev/null
@@ -1,317 +0,0 @@
-; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 | FileCheck %s
-
-; Check that DAGCombiner correctly folds the following pairs of shuffles
-; using the following rules:
-;  1. shuffle(shuffle(x, y), undef) -> x
-;  2. shuffle(shuffle(x, y), undef) -> y
-;  3. shuffle(shuffle(x, y), undef) -> shuffle(x, undef)
-;  4. shuffle(shuffle(x, y), undef) -> shuffle(undef, y)
-;
-; Rules 3. and 4. are used only if the resulting shuffle mask is legal.
-
-define <4 x i32> @test1(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 4, i32 3, i32 1>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 4, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test1
-; Mask: [3,0,0,1]
-; CHECK: pshufd $67
-; CHECK-NEXT: ret
-
-
-define <4 x i32> @test2(<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 3>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 4, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test2
-; Mask: [2,0,0,3]
-; CHECK: pshufd $-62
-; CHECK-NEXT: ret
-
-
-define <4 x i32> @test3(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 6, i32 2, i32 3>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 4, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test3
-; Mask: [2,0,0,3]
-; CHECK: pshufd $-62
-; CHECK-NEXT: ret
-
-
-define <4 x i32> @test4(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 4, i32 7, i32 1>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 4, i32 4, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test4
-; Mask: [0,0,0,1]
-; CHECK: pshufd $64
-; CHECK-NEXT: ret
-
-
-define <4 x i32> @test5(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 5, i32 5, i32 2, i32 3>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 4, i32 4, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test5
-; Mask: [1,1]
-; CHECK: movhlps
-; CHECK-NEXT: 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 0, i32 6, i32 2, i32 4>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 4, i32 0, i32 4>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test6
-; Mask: [2,0,0,0]
-; CHECK: pshufd $2
-; CHECK-NEXT: 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> undef, <4 x i32> <i32 0, i32 2, i32 0, i32 2>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test7
-; Mask: [0,2,0,2]
-; CHECK: pshufd $-120
-; CHECK-NEXT: 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 4, i32 1, i32 6, i32 3>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 1, i32 4, i32 3, i32 4>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test8
-; Mask: [1,0,3,0]
-; CHECK: pshufd $49
-; CHECK-NEXT: 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 1, i32 3, i32 2, i32 5>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 4, i32 2>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test9
-; Mask: [1,3,0,2]
-; CHECK: pshufd $-115
-; 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 1, i32 1, i32 5, i32 5>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 4, i32 1, i32 4>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test10
-; Mask: [1,0,1,0]
-; CHECK: pshufd $17
-; CHECK-NEXT: ret
-
-
-define <4 x i32> @test11(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 1, i32 2, i32 5, i32 4>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 4, i32 1, i32 0>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test11
-; Mask: [1,0,2,1]
-; CHECK: pshufd $97
-; CHECK-NEXT: ret
-
-
-define <4 x i32> @test12(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 0, i32 2, i32 4>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 1, i32 4, i32 0, i32 4>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test12
-; Mask: [0,0,0,0]
-; CHECK: pshufd $0
-; CHECK-NEXT: ret
-
-
-; The following pair of shuffles is folded into vector %A.
-define <4 x i32> @test13(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 1, i32 4, i32 2, i32 6>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 4, i32 0, i32 2, i32 4>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test13
-; CHECK-NOT: pshufd
-; CHECK: ret
-
-
-; The following pair of shuffles is folded into vector %B.
-define <4 x i32> @test14(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 6, i32 2, i32 4>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 3, i32 4, i32 1, i32 4>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test14
-; CHECK-NOT: pshufd
-; CHECK: ret
-
-
-; Verify that we don't optimize the following cases. We expect more than one shuffle.
-
-define <4 x i32> @test15(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 4, i32 3, i32 1>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test15
-; CHECK: shufps $114
-; CHECK-NEXT: pshufd $-58
-; CHECK-NEXT: 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 0, i32 5, i32 2, i32 7>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test16
-; CHECK: blendps $10
-; CHECK-NEXT: pshufd $-58
-; CHECK-NEXT: 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 4, i32 1, i32 3, i32 1>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test17
-; CHECK: shufps $120
-; CHECK-NEXT: pshufd $-58
-; CHECK-NEXT: 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 4, i32 5, i32 2, i32 7>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test18
-; CHECK-NOT: blendps
-; CHECK: pshufd {{.*}} # xmm0 = xmm1[1,1,0,3]
-; CHECK-NEXT: 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 0, i32 4, i32 5, i32 6>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 0, i32 0, i32 0>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test19
-; CHECK: shufps $-104
-; CHECK-NEXT: pshufd $2
-; CHECK-NEXT: 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 3, i32 2, i32 4, i32 4>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test20
-; CHECK: shufps $11
-; CHECK-NEXT: pshufd $-58
-; CHECK-NEXT: ret
-
-
-define <4 x i32> @test21(<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 3, i32 1>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test21
-; CHECK: shufps $120
-; CHECK-NEXT: pshufd $-60
-; CHECK-NEXT: ret
-
-; Test that we correctly combine shuffles according to rule
-;  shuffle(shuffle(x, y), undef) -> shuffle(y, undef)
-
-define <4 x i32> @test22(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 4, i32 5, i32 2, i32 7>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test22
-; CHECK-NOT: blendps
-; CHECK: pshufd {{.*}} # xmm0 = xmm1[1,1,1,3]
-; CHECK-NEXT: ret
-
-define <4 x i32> @test23(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 4, i32 5, i32 2, i32 7>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test23
-; CHECK-NOT: blendps
-; CHECK: pshufd {{.*}} # xmm0 = xmm1[0,1,0,3]
-; CHECK-NEXT: ret
-
-define <4 x i32> @test24(<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> undef, <4 x i32> <i32 0, i32 3, i32 2, i32 4>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test24
-; CHECK-NOT: blendps
-; CHECK: pshufd {{.*}} # xmm0 = xmm1[0,3,2,0]
-; CHECK-NEXT: ret
-
-define <4 x i32> @test25(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %B, <4 x i32> %A, <4 x i32> <i32 1, i32 5, i32 2, i32 4>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 3, i32 1, i32 3, i32 1>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test25
-; CHECK-NOT:  shufps
-; CHECK: pshufd {{.*}} # xmm0 = xmm0[0,1,0,1]
-; CHECK-NEXT: ret
-
-define <4 x i32> @test26(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %B, <4 x i32> %A, <4 x i32> <i32 1, i32 2, i32 6, i32 7>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test26
-; CHECK-NOT: shufps
-; CHECK: movhlps {{.*}} # xmm0 = xmm0[1,1]
-; CHECK-NEXT: ret
-
-define <4 x i32> @test27(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %B, <4 x i32> %A, <4 x i32> <i32 2, i32 1, i32 5, i32 4>
-  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 3, i32 2, i32 3, i32 2>
-  ret <4 x i32> %2
-}
-; CHECK-LABEL: test27
-; CHECK-NOT: shufps
-; CHECK-NOT: movhlps
-; CHECK: pshufd {{.*}} # xmm0 = xmm0[0,1,0,1]
-; CHECK-NEXT: ret
-
-define <4 x i32> @test28(<4 x i32> %A, <4 x i32> %B) {
-  %1 = shufflevector <4 x i32> %B, <4 x i32> %A, <4 x i32> <i32 1, i32 2, i32 4, i32 5>
-  %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
-}
-; CHECK-LABEL: test28
-; CHECK-NOT: shufps
-; CHECK-NOT: movhlps
-; CHECK: pshufd {{.*}} # xmm0 = xmm0[0,1,1,0]
-; CHECK-NEXT: ret
-
diff --git a/test/CodeGen/X86/vector-shuffle-combining.ll b/test/CodeGen/X86/vector-shuffle-combining.ll
index 7255d8f7bc7..d8e6cf2b8c1 100644
--- a/test/CodeGen/X86/vector-shuffle-combining.ll
+++ b/test/CodeGen/X86/vector-shuffle-combining.ll
@@ -627,3 +627,473 @@ define <4 x i32> @combine_bitwise_ops_test6c(<4 x i32> %a, <4 x i32> %b, <4 x i3
   %xor = xor <4 x i32> %shuf1, %shuf2
   ret <4 x i32> %xor
 }
+
+define <4 x i32> @combine_nested_undef_test1(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test1:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,0,0,1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test1:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,0,0,1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 4, i32 3, i32 1>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 4, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test2(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test2:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,0,3]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test2:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,0,0,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 5, i32 2, i32 3>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 4, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test3(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test3:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,0,3]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test3:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,0,0,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 6, i32 2, i32 3>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 4, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test4(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test4:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test4:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 4, i32 7, i32 1>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 4, i32 4, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test5(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test5:
+; SSE:       # BB#0:
+; SSE-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test5:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovhlps {{.*#+}} xmm0 = xmm0[1,1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 5, i32 5, i32 2, i32 3>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 4, i32 4, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test6(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test6:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,0,0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test6:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,0,0,0]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 6, i32 2, i32 4>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 4, i32 0, i32 4>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test7(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test7:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,0,2]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test7:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,0,2]
+; 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> undef, <4 x i32> <i32 0, i32 2, i32 0, i32 2>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test8(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test8:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,0,3,0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test8:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,0,3,0]
+; 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> undef, <4 x i32> <i32 1, i32 4, i32 3, i32 4>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test9(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test9:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,3,0,2]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test9:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,3,0,2]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 1, i32 3, i32 2, i32 5>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 4, i32 2>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test10(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test10:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,0,1,0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test10:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,0,1,0]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 1, i32 1, i32 5, i32 5>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 4, i32 1, i32 4>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test11(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test11:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,0,2,1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test11:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,0,2,1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 1, i32 2, i32 5, i32 4>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 4, i32 1, i32 0>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test12(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test12:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; SSE-NEXT:    retq
+;
+; AVX1-LABEL: combine_nested_undef_test12:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: combine_nested_undef_test12:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vbroadcastss %xmm0, %xmm0
+; AVX2-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 0, i32 2, i32 4>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 1, i32 4, i32 0, i32 4>
+  ret <4 x i32> %2
+}
+
+; The following pair of shuffles is folded into vector %A.
+define <4 x i32> @combine_nested_undef_test13(<4 x i32> %A, <4 x i32> %B) {
+; ALL-LABEL: combine_nested_undef_test13:
+; ALL:       # BB#0:
+; ALL-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 1, i32 4, i32 2, i32 6>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 4, i32 0, i32 2, i32 4>
+  ret <4 x i32> %2
+}
+
+; The following pair of shuffles is folded into vector %B.
+define <4 x i32> @combine_nested_undef_test14(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test14:
+; SSE:       # BB#0:
+; SSE-NEXT:    movaps %xmm1, %xmm0
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test14:
+; 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 0, i32 6, i32 2, i32 4>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 3, i32 4, i32 1, i32 4>
+  ret <4 x i32> %2
+}
+
+
+; Verify that we don't optimize the following cases. We expect more than one shuffle.
+;
+; FIXME: Many of these already don't make sense, and the rest should stop
+; making sense with th enew vector shuffle lowering. Revisit at least testing for
+; it.
+
+define <4 x i32> @combine_nested_undef_test15(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test15:
+; SSE:       # BB#0:
+; SSE-NEXT:    movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0]
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,0],xmm0[3,1]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,1,0,3]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test15:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[2,0],xmm0[3,1]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,1,0,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 4, i32 3, i32 1>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test16(<4 x i32> %A, <4 x i32> %B) {
+; SSE2-LABEL: combine_nested_undef_test16:
+; SSE2:       # BB#0:
+; SSE2-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,2,0,3]
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: combine_nested_undef_test16:
+; SSSE3:       # BB#0:
+; SSSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,3]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,2,0,3]
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: combine_nested_undef_test16:
+; SSE41:       # BB#0:
+; SSE41-NEXT:    blendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,0,3]
+; SSE41-NEXT:    retq
+;
+; AVX1-LABEL: combine_nested_undef_test16:
+; AVX1:       # BB#0:
+; AVX1-NEXT:    vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,1,0,3]
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: combine_nested_undef_test16:
+; AVX2:       # BB#0:
+; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,1,0,3]
+; AVX2-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> undef, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test17(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test17:
+; SSE:       # BB#0:
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[1,0]
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[3,1]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,1,0,3]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test17:
+; AVX:       # BB#0:
+; AVX-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[1,0]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[0,2],xmm0[3,1]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,1,0,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 4, i32 1, i32 3, i32 1>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test18(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test18:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,0,3]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test18:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[1,1,0,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 4, i32 5, i32 2, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test19(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test19:
+; SSE:       # BB#0:
+; SSE-NEXT:    movlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,2]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,0,0,0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test19:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[1,2]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,0,0,0]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 4, i32 5, i32 6>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 0, i32 0, i32 0>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test20(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test20:
+; SSE:       # BB#0:
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,2],xmm1[0,0]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,0,3]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test20:
+; AVX:       # BB#0:
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,2],xmm1[0,0]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,1,0,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 3, i32 2, i32 4, i32 4>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 1, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test21(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test21:
+; SSE:       # BB#0:
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[1,0]
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,2],xmm0[3,1]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,3]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test21:
+; AVX:       # BB#0:
+; AVX-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,0],xmm0[1,0]
+; AVX-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[0,2],xmm0[3,1]
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 4, i32 1, i32 3, i32 1>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+
+; Test that we correctly combine shuffles according to rule
+;  shuffle(shuffle(x, y), undef) -> shuffle(y, undef)
+
+define <4 x i32> @combine_nested_undef_test22(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test22:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,1,3]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test22:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[1,1,1,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 4, i32 5, i32 2, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test23(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test23:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,0,3]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test23:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[0,1,0,3]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 4, i32 5, i32 2, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test24(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test24:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,3,2,0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test24:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm1[0,3,2,0]
+; AVX-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> undef, <4 x i32> <i32 0, i32 3, i32 2, i32 4>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test25(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test25:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test25:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %B, <4 x i32> %A, <4 x i32> <i32 1, i32 5, i32 2, i32 4>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 3, i32 1, i32 3, i32 1>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test26(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test26:
+; SSE:       # BB#0:
+; SSE-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test26:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovhlps {{.*#+}} xmm0 = xmm0[1,1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %B, <4 x i32> %A, <4 x i32> <i32 1, i32 2, i32 6, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test27(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test27:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test27:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %B, <4 x i32> %A, <4 x i32> <i32 2, i32 1, i32 5, i32 4>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 3, i32 2, i32 3, i32 2>
+  ret <4 x i32> %2
+}
+
+define <4 x i32> @combine_nested_undef_test28(<4 x i32> %A, <4 x i32> %B) {
+; SSE-LABEL: combine_nested_undef_test28:
+; SSE:       # BB#0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,0]
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: combine_nested_undef_test28:
+; AVX:       # BB#0:
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,1,0]
+; AVX-NEXT:    retq
+  %1 = shufflevector <4 x i32> %B, <4 x i32> %A, <4 x i32> <i32 1, i32 2, i32 4, i32 5>
+  %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
+}
-- 
2.34.1