From fea1ddbc355040a2dd4c4d8616cfd2d0de93bfd6 Mon Sep 17 00:00:00 2001
From: Sanjay Patel <spatel@rotateright.com>
Date: Fri, 8 Mar 2019 19:20:28 +0000
Subject: [PATCH] [x86] prevent infinite looping from inverse shuffle
 transforms

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355713 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp           |  7 ++++++-
 test/CodeGen/X86/vector-shuffle-128-unpck.ll | 12 ++++++++++++
 2 files changed, 18 insertions(+), 1 deletion(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 49d30c7cbae..b7446002922 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -9850,11 +9850,16 @@ static bool is128BitUnpackShuffleMask(ArrayRef<int> Mask) {
   MVT EltVT = MVT::getIntegerVT(128 / Mask.size());
   MVT VT = MVT::getVectorVT(EltVT, Mask.size());
 
+  // We can't assume a canonical shuffle mask, so try the commuted version too.
+  SmallVector<int, 4> CommutedMask(Mask.begin(), Mask.end());
+  ShuffleVectorSDNode::commuteMask(CommutedMask);
+
   // Match any of unary/binary or low/high.
   for (unsigned i = 0; i != 4; ++i) {
     SmallVector<int, 16> UnpackMask;
     createUnpackShuffleMask(VT, UnpackMask, (i >> 1) % 2, i % 2);
-    if (isTargetShuffleEquivalent(Mask, UnpackMask))
+    if (isTargetShuffleEquivalent(Mask, UnpackMask) ||
+        isTargetShuffleEquivalent(CommutedMask, UnpackMask))
       return true;
   }
   return false;
diff --git a/test/CodeGen/X86/vector-shuffle-128-unpck.ll b/test/CodeGen/X86/vector-shuffle-128-unpck.ll
index 47d9c41e019..58e4c10232f 100644
--- a/test/CodeGen/X86/vector-shuffle-128-unpck.ll
+++ b/test/CodeGen/X86/vector-shuffle-128-unpck.ll
@@ -221,3 +221,15 @@ define <16 x i8> @unpckl_unary_extracted_v32i8(<32 x i8> %x) {
   ret <16 x i8> %r
 }
 
+; This would infinite loop because we did not recognize the unpack shuffle mask in commuted form.
+
+define <8 x i32> @extract_unpckl_v8i32(<8 x i32> %a) {
+; ALL-LABEL: extract_unpckl_v8i32:
+; ALL:       # %bb.0:
+; ALL-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; ALL-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; ALL-NEXT:    retq
+  %shuffle = shufflevector <8 x i32> %a, <8 x i32> undef, <8 x i32> <i32 4, i32 undef, i32 5, i32 1, i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <8 x i32> %shuffle
+}
+
-- 
2.11.0