From 89cea3c36b451d5a06d51b3101052f9faea91211 Mon Sep 17 00:00:00 2001
From: Andrea Di Biagio <Andrea_DiBiagio@sn.scee.net>
Date: Sat, 16 Aug 2014 00:29:44 +0000
Subject: [PATCH] [DAGCombiner] Improve the folding of target independet
 shuffles to Undef.

When combining a pair of shuffle nodes, check if the combined shuffle mask is
trivially Undef. In case, immediately fold that pair of shuffles to Undef.

The lack of checks for undef masks was the root-cause of a poor-codegen bug
in the dag combiner.

Example:
  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 4, i32 1, i32 1, i32 6>
  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 4, i32 1, i32 6>
  %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 1, i32 5, i32 3, i32 3>

Before this patch, on x86 (with -mcpu=corei7) we failed to fold the entire
sequence to Undef value and therefore we generated:
  shufps $-123, %xmm1, $xmm0
  pshufd $-46, %xmm0, %xmm0

With this patch, the entire shuffle sequence is folded to Undef and no
shuffles are generated in the output assembly.

Added new test cases to test 'combine-vec-shuffle-5.ll'.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@215797 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/CodeGen/SelectionDAG/DAGCombiner.cpp  |  16 ++
 test/CodeGen/X86/combine-vec-shuffle-5.ll | 191 ++++++++++++++++++++++
 2 files changed, 207 insertions(+)

diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index c1ff817fc35..e9a38b1b437 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -10787,6 +10787,14 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) {
         Idx = OtherSV->getMaskElt(Idx);
       Mask.push_back(Idx);
     }
+
+    // Check if all indices in Mask are Undef. In case, propagate Undef.
+    bool isUndefMask = true;
+    for (unsigned i = 0; i != NumElts && isUndefMask; ++i)
+      isUndefMask &= Mask[i] < 0;
+
+    if (isUndefMask)
+      return DAG.getUNDEF(VT);
     
     bool CommuteOperands = false;
     if (N0.getOperand(1).getOpcode() != ISD::UNDEF) {
@@ -10932,6 +10940,14 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) {
       Mask.push_back(Idx);
     }
 
+    // Check if all indices in Mask are Undef. In case, propagate Undef.
+    bool isUndefMask = true;
+    for (unsigned i = 0; i != NumElts && isUndefMask; ++i)
+      isUndefMask &= Mask[i] < 0;
+
+    if (isUndefMask)
+      return DAG.getUNDEF(VT);
+
     // Avoid introducing shuffles with illegal mask.
     if (TLI.isShuffleMaskLegal(Mask, VT)) {
       if (IsSV1Undef)
diff --git a/test/CodeGen/X86/combine-vec-shuffle-5.ll b/test/CodeGen/X86/combine-vec-shuffle-5.ll
index 16c45efe4be..82b020e8cb4 100644
--- a/test/CodeGen/X86/combine-vec-shuffle-5.ll
+++ b/test/CodeGen/X86/combine-vec-shuffle-5.ll
@@ -255,3 +255,194 @@ define <4 x i8> @test4c(<4 x i8>* %a, <4 x i8>* %b) {
 ; CHECK: blendps $13
 ; CHECK: ret
 
+
+; Verify that the dag combiner correctly folds the following shuffle pairs to Undef.
+
+define <4 x i32> @test1b(<4 x i32> %A) {
+  %1 = shufflevector <4 x i32> %A, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 5, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 4, i32 5, i32 2, i32 7>
+  ret <4 x i32> %2
+}
+; CHECK-LABEL: test1b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test2b(<4 x i32> %A) {
+  %1 = shufflevector <4 x i32> %A, <4 x i32> undef, <4 x i32> <i32 4, i32 5, i32 1, i32 6>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 1, i32 0, i32 6, i32 7>
+  ret <4 x i32> %2
+}
+; CHECK-LABEL: test2b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test3b(<4 x i32> %A, <4 x i32> %B) {
+  %1 = shufflevector <4 x i32> %A, <4 x i32> undef, <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 1, i32 3, i32 1, i32 3>
+  ret <4 x i32> %2
+}
+; CHECK-LABEL: test3b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test4b(<4 x i32> %A, <4 x i32> %B) {
+  %1 = shufflevector <4 x i32> %A, <4 x i32> undef, <4 x i32> <i32 4, i32 1, i32 1, i32 6>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %B, <4 x i32> <i32 0, i32 3, i32 3, i32 0>
+  ret <4 x i32> %2
+}
+; CHECK-LABEL: test4b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test5b(<4 x i32> %A) {
+  %1 = shufflevector <4 x i32> %A, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 5, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 4, i32 5, i32 2, i32 7>
+  %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  ret <4 x i32> %3
+}
+; CHECK-LABEL: test5b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test6b(<4 x i32> %A) {
+  %1 = shufflevector <4 x i32> %A, <4 x i32> undef, <4 x i32> <i32 4, i32 5, i32 1, i32 6>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 1, i32 0, i32 6, i32 7>
+  %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x i32> %3
+}
+; CHECK-LABEL: test6b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test7b(<4 x i32> %A, <4 x i32> %B) {
+  %1 = shufflevector <4 x i32> %A, <4 x i32> undef, <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 1, i32 3, i32 1, i32 3>
+  %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 5, i32 1, i32 6>
+  ret <4 x i32> %3
+}
+; CHECK-LABEL: test7b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test8b(<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 1, i32 6>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 4, i32 1, i32 6>
+  %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 1, i32 5, i32 3, i32 3>
+  ret <4 x i32> %3
+}
+; CHECK-LABEL: test8b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test9b(<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 undef, i32 7>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 3, i32 4, i32 2, i32 1>
+  %3 = shufflevector <4 x i32> %2, <4 x i32> %A, <4 x i32> <i32 2, i32 1, i32 1, i32 2>
+  ret <4 x i32> %3
+}
+; CHECK-LABEL: test9b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test10b(<4 x i32> %A, <4 x i32> %B) {
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 undef, i32 undef, i32 1, i32 6>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %A, <4 x i32> <i32 0, i32 6, i32 1, i32 0>
+  %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 0, i32 2>
+  ret <4 x i32> %3
+}
+; CHECK-LABEL: test10b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test11b(<4 x i32> %A, <4 x i32> %B) {
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 0, i32 undef, i32 1, i32 undef>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %B, <4 x i32> <i32 1, i32 3, i32 1, i32 3>
+  %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 0, i32 5, i32 1, i32 6>
+  ret <4 x i32> %3
+}
+; CHECK-LABEL: test11b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <4 x i32> @test12b(<4 x i32> %A, <4 x i32> %B) {
+  %1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 undef, i32 1, i32 1, i32 undef>
+  %2 = shufflevector <4 x i32> %1, <4 x i32> %B, <4 x i32> <i32 0, i32 3, i32 3, i32 0>
+  %3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 1, i32 5, i32 1, i32 4>
+  ret <4 x i32> %3
+}
+; CHECK-LABEL: test12b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <8 x i32> @test13b(<8 x i32> %A, <8 x i32> %B) {
+  %1 = shufflevector <8 x i32> %A, <8 x i32> %B, <8 x i32> <i32 0, i32 undef, i32 1, i32 undef, i32 0, i32 undef, i32 1, i32 undef>
+  %2 = shufflevector <8 x i32> %1, <8 x i32> %B, <8 x i32> <i32 1, i32 3, i32 1, i32 3, i32 1, i32 3, i32 1, i32 3>
+  %3 = shufflevector <8 x i32> %2, <8 x i32> undef, <8 x i32> <i32 0, i32 9, i32 1, i32 10, i32 0, i32 9, i32 1, i32 10>
+  ret <8 x i32> %3
+}
+; CHECK-LABEL: test13b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <8 x i32> @test14b(<8 x i32> %A, <8 x i32> %B) {
+  %1 = shufflevector <8 x i32> %A, <8 x i32> %B, <8 x i32> <i32 undef, i32 1, i32 1, i32 undef, i32 undef, i32 1, i32 1, i32 undef>
+  %2 = shufflevector <8 x i32> %1, <8 x i32> %B, <8 x i32> <i32 0, i32 3, i32 3, i32 0, i32 0, i32 3, i32 3, i32 0>
+  %3 = shufflevector <8 x i32> %2, <8 x i32> undef, <8 x i32> <i32 1, i32 9, i32 1, i32 8, i32 1, i32 9, i32 1, i32 8>
+  ret <8 x i32> %3
+}
+; CHECK-LABEL: test14b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <8 x i32> @test15b(<8 x i32> %A, <8 x i32> %B) {
+  %1 = shufflevector <8 x i32> %A, <8 x i32> %B, <8 x i32> <i32 0, i32 1, i32 undef, i32 11, i32 0, i32 1, i32 undef, i32 11>
+  %2 = shufflevector <8 x i32> %1, <8 x i32> undef, <8 x i32> <i32 8, i32 9, i32 2, i32 11, i32 8, i32 9, i32 2, i32 11>
+  %3 = shufflevector <8 x i32> %2, <8 x i32> %A, <8 x i32> <i32 2, i32 2, i32 undef, i32 2, i32 2, i32 2, i32 undef, i32 2>
+  ret <8 x i32> %3
+}
+; CHECK-LABEL: test15b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
+define <8 x i32> @test16b(<8 x i32> %A, <8 x i32> %B) {
+  %1 = shufflevector <8 x i32> %A, <8 x i32> %B, <8 x i32> <i32 undef, i32 undef, i32 1, i32 10, i32 undef, i32 undef, i32 1, i32 10>
+  %2 = shufflevector <8 x i32> %1, <8 x i32> %A, <8 x i32> <i32 0, i32 10, i32 2, i32 11, i32 0, i32 10, i32 2, i32 11>
+  %3 = shufflevector <8 x i32> %2, <8 x i32> undef, <8 x i32> <i32 4, i32 9, i32 undef, i32 0, i32 4, i32 9, i32 undef, i32 0>
+  ret <8 x i32> %3
+}
+; CHECK-LABEL: test16b
+; CHECK-NOT: blendps
+; CHECK-NOT: pshufd
+; CHECK-NOT: movhlps
+; CHECK: ret
+
-- 
2.34.1