[llvm-commits] [llvm] r137194 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/avx-vpermil.ll
Bruno Cardoso Lopes
bruno.cardoso at gmail.com
Tue Aug 9 18:54:17 PDT 2011
Author: bruno
Date: Tue Aug 9 20:54:17 2011
New Revision: 137194
URL: http://llvm.org/viewvc/llvm-project?rev=137194&view=rev
Log:
Fix a bug in vpermilps mask checking. Fix PR10560
Modified:
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
llvm/trunk/test/CodeGen/X86/avx-vpermil.ll
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=137194&r1=137193&r2=137194&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Tue Aug 9 20:54:17 2011
@@ -3509,9 +3509,12 @@
int LaneSize = NumElts/NumLanes;
for (int i = 0; i < LaneSize; ++i) {
int HighElt = i+LaneSize;
- if (Mask[i] < 0 && (isUndefOrInRange(Mask[HighElt], LaneSize, NumElts)))
- continue;
- if (Mask[HighElt] < 0 && (isUndefOrInRange(Mask[i], 0, LaneSize)))
+ bool HighValid = isUndefOrInRange(Mask[HighElt], LaneSize, NumElts);
+ bool LowValid = isUndefOrInRange(Mask[i], 0, LaneSize);
+
+ if (!HighValid || !LowValid)
+ return false;
+ if (Mask[i] < 0 || Mask[HighElt] < 0)
continue;
if (Mask[HighElt]-Mask[i] != LaneSize)
return false;
Modified: llvm/trunk/test/CodeGen/X86/avx-vpermil.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx-vpermil.ll?rev=137194&r1=137193&r2=137194&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/avx-vpermil.ll (original)
+++ llvm/trunk/test/CodeGen/X86/avx-vpermil.ll Tue Aug 9 20:54:17 2011
@@ -37,3 +37,9 @@
ret <8 x float> %shuffle
}
+; CHECK-NOT: vpermilps
+define <8 x float> @funcF(<8 x float> %a) nounwind uwtable readnone ssp {
+entry:
+ %shuffle = shufflevector <8 x float> %a, <8 x float> zeroinitializer, <8 x i32> <i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9>
+ ret <8 x float> %shuffle
+}
More information about the llvm-commits
mailing list