From 49cf7619254ff6529c14ece0a6fa3a02cb3fbfcf Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sun, 25 Oct 2015 17:40:54 +0000
Subject: [PATCH] [X86][SSE4A] Fix for EXTRQI shuffle lowering.

Incorrect range test - found during fuzz testing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251245 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp       |  4 ++--
 test/CodeGen/X86/vector-shuffle-sse4a.ll | 23 +++++++++++++++++++++++
 2 files changed, 25 insertions(+), 2 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index d1879570a03..c978fe08476 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -7334,7 +7334,7 @@ static SDValue lowerVectorShuffleWithSSE4A(SDLoc DL, MVT VT, SDValue V1,
     // Determine the extraction length from the part of the
     // lower half that isn't zeroable.
     int Len = HalfSize;
-    for (; Len >= 0; --Len)
+    for (; Len > 0; --Len)
       if (!Zeroable[Len - 1])
         break;
     assert(Len > 0 && "Zeroable shuffle mask");
@@ -7350,7 +7350,7 @@ static SDValue lowerVectorShuffleWithSSE4A(SDLoc DL, MVT VT, SDValue V1,
       M = M % Size;
 
       // All mask elements must be in the lower half.
-      if (M > HalfSize)
+      if (M >= HalfSize)
         return SDValue();
 
       if (Idx < 0 || (Src == V && Idx == (M - i))) {
diff --git a/test/CodeGen/X86/vector-shuffle-sse4a.ll b/test/CodeGen/X86/vector-shuffle-sse4a.ll
index 58807b9c0fd..2dd43e2852a 100644
--- a/test/CodeGen/X86/vector-shuffle-sse4a.ll
+++ b/test/CodeGen/X86/vector-shuffle-sse4a.ll
@@ -296,5 +296,28 @@ define <8 x i16> @shuf_089uuuuu(<8 x i16> %a0, <8 x i16> %a1) {
   ret <8 x i16> %s
 }
 
+;
+; Special Cases
+;
+
+; Out of range.
+define <16 x i8> @shuffle_8_18_uuuuuuuuuuuuuu(<16 x i8> %a, <16 x i8> %b) {
+; BTVER1-LABEL: shuffle_8_18_uuuuuuuuuuuuuu:
+; BTVER1:       # BB#0:
+; BTVER1-NEXT:    psrld $16, %xmm1
+; BTVER1-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; BTVER1-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; BTVER1-NEXT:    retq
+;
+; BTVER2-LABEL: shuffle_8_18_uuuuuuuuuuuuuu:
+; BTVER2:       # BB#0:
+; BTVER2-NEXT:    vpsrld $16, %xmm1, %xmm1
+; BTVER2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; BTVER2-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; BTVER2-NEXT:    retq
+  %1 = shufflevector <16 x i8> %a, <16 x i8> %b, <16 x i32> <i32 8, i32 18, 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>
+  ret <16 x i8> %1
+}
+
 declare <2 x i64> @llvm.x86.sse4a.extrqi(<2 x i64>, i8, i8) nounwind
 declare <2 x i64> @llvm.x86.sse4a.insertqi(<2 x i64>, <2 x i64>, i8, i8) nounwind
-- 
2.34.1