Fix PR14161

- Check index being extracted to be constant 0 before simplfiying.
  Otherwise, retain the original sequence.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@166504 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 42 insertions, 1 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index a3973ed963..f73b9d64b9 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -6630,9 +6630,12 @@ X86TargetLowering::lowerVectorIntExtend(SDValue Op, SelectionDAG &DAG) const {
         .getOperand(0).getValueType().getSizeInBits() == SignificantBits) {
     // (bitcast (sclr2vec (ext_vec_elt x))) -> (bitcast x)
     SDValue V = V1.getOperand(0).getOperand(0).getOperand(0);
+    ConstantSDNode *CIdx =
+      dyn_cast<ConstantSDNode>(V1.getOperand(0).getOperand(0).getOperand(1));
     // If it's foldable, i.e. normal load with single use, we will let code
     // selection to fold it. Otherwise, we will short the conversion sequence.
-    if (!ISD::isNormalLoad(V.getNode()) || !V.hasOneUse())
+    if (CIdx && CIdx->getZExtValue() == 0 &&
+        (!ISD::isNormalLoad(V.getNode()) || !V.hasOneUse()))
       V1 = DAG.getNode(ISD::BITCAST, DL, V1.getValueType(), V);
   }
 
diff --git a/test/CodeGen/X86/pr14161.ll b/test/CodeGen/X86/pr14161.ll
new file mode 100644
index 0000000000..ff4532eac3
--- /dev/null
+++ b/test/CodeGen/X86/pr14161.ll
@@ -0,0 +1,38 @@
+; RUN: llc < %s -mtriple=x86_64-linux-pc -mcpu=corei7 | FileCheck %s
+
+declare <4 x i32> @llvm.x86.sse41.pminud(<4 x i32>, <4 x i32>)
+
+define <2 x i16> @good(<4 x i32>*, <4 x i8>*) {
+entry:
+  %2 = load <4 x i32>* %0, align 16
+  %3 = call <4 x i32> @llvm.x86.sse41.pminud(<4 x i32> %2, <4 x i32> <i32 127, i32 127, i32 127, i32 127>)
+  %4 = extractelement <4 x i32> %3, i32 0
+  %5 = extractelement <4 x i32> %3, i32 1
+  %6 = extractelement <4 x i32> %3, i32 2
+  %7 = extractelement <4 x i32> %3, i32 3
+  %8 = bitcast i32 %4 to <2 x i16>
+  %9 = bitcast i32 %5 to <2 x i16>
+  ret <2 x i16> %8
+; CHECK: good
+; CHECK: pminud
+; CHECK-NEXT: pmovzxwq
+; CHECK: ret
+}
+
+define <2 x i16> @bad(<4 x i32>*, <4 x i8>*) {
+entry:
+  %2 = load <4 x i32>* %0, align 16
+  %3 = call <4 x i32> @llvm.x86.sse41.pminud(<4 x i32> %2, <4 x i32> <i32 127, i32 127, i32 127, i32 127>)
+  %4 = extractelement <4 x i32> %3, i32 0
+  %5 = extractelement <4 x i32> %3, i32 1
+  %6 = extractelement <4 x i32> %3, i32 2
+  %7 = extractelement <4 x i32> %3, i32 3
+  %8 = bitcast i32 %4 to <2 x i16>
+  %9 = bitcast i32 %5 to <2 x i16>
+  ret <2 x i16> %9
+; CHECK: bad
+; CHECK: pminud
+; CHECK: pextrd
+; CHECK: pmovzxwq
+; CHECK: ret
+}