From 0b94b5f52b11092a69267159dfe0df3acdfcabd7 Mon Sep 17 00:00:00 2001
From: Nadav Rotem <nadav.rotem@intel.com>
Date: Tue, 17 Jan 2012 09:13:19 +0000
Subject: [PATCH] Fix 11769.

In CanXFormVExtractWithShuffleIntoLoad we assumed that EXTRACT_VECTOR_ELT can be later handled by the DAGCombiner.
However, in some cases on AVX, the EXTRACT_VECTOR_ELT is legalized to EXTRACT_SUBVECTOR + EXTRACT_VECTOR_ELT, which
currently is not handled by the DAGCombiner. In this patch I added a check that we only extract from the XMM part.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@148298 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp |  7 +++++++
 test/CodeGen/X86/avx-shuffle.ll    | 13 ++++++++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 043110867f0..8f69b0e314f 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -6241,6 +6241,13 @@ bool CanXFormVExtractWithShuffleIntoLoad(SDValue V, SelectionDAG &DAG,
   int Idx = (Elt > NumElems) ? -1 : SVOp->getMaskElt(Elt);
   V = (Idx < (int)NumElems) ? V.getOperand(0) : V.getOperand(1);
 
+  // If we are accessing the upper part of a YMM register
+  // then the EXTRACT_VECTOR_ELT is likely to be legalized to a sequence of
+  // EXTRACT_SUBVECTOR + EXTRACT_VECTOR_ELT, which are not detected at this point
+  // because the legalization of N did not happen yet.
+  if (Idx >= NumElems/2 && VT.getSizeInBits() == 256)
+    return false;
+
   // Skip one more bit_convert if necessary
   if (V.getOpcode() == ISD::BITCAST)
     V = V.getOperand(0);
diff --git a/test/CodeGen/X86/avx-shuffle.ll b/test/CodeGen/X86/avx-shuffle.ll
index a059a4d3575..232fc785120 100644
--- a/test/CodeGen/X86/avx-shuffle.ll
+++ b/test/CodeGen/X86/avx-shuffle.ll
@@ -60,7 +60,18 @@ entry:
 
 define <16 x i16> @test7(<4 x i16> %a) nounwind {
 ; CHECK: test7
-
   %b = shufflevector <4 x i16> %a, <4 x i16> undef, <16 x i32> <i32 1, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK: ret
   ret <16 x i16> %b
 }
+
+; CHECK: test8
+define void @test8() {
+entry:
+  %0 = load <16 x i64> addrspace(1)* null, align 128
+  %1 = shufflevector <16 x i64> <i64 undef, i64 undef, i64 0, i64 undef, i64 0, i64 0, i64 0, i64 0, i64 0, i64 0, i64 undef, i64 0, i64 undef, i64 undef, i64 undef, i64 undef>, <16 x i64> %0, <16 x i32> <i32 17, i32 18, i32 2, i32 undef, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 undef, i32 11, i32 undef, i32 undef, i32 undef, i32 26>
+  %2 = shufflevector <16 x i64> %1, <16 x i64> %0, <16 x i32> <i32 0, i32 1, i32 2, i32 30, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 undef, i32 11, i32 undef, i32 22, i32 20, i32 15>
+  store <16 x i64> %2, <16 x i64> addrspace(1)* undef, align 128
+; CHECK: ret
+  ret void
+}
-- 
2.34.1