[llvm] r289341 - [X86][SSE] Ensure UNPCK inputs are a consistent value type in LowerHorizontalByteSum
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Sat Dec 10 13:16:46 PST 2016
Author: rksimon
Date: Sat Dec 10 15:16:45 2016
New Revision: 289341
URL: http://llvm.org/viewvc/llvm-project?rev=289341&view=rev
Log:
[X86][SSE] Ensure UNPCK inputs are a consistent value type in LowerHorizontalByteSum
Modified:
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=289341&r1=289340&r2=289341&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Sat Dec 10 15:16:45 2016
@@ -21917,8 +21917,9 @@ static SDValue LowerHorizontalByteSum(SD
// two v2i64 vectors which concatenated are the 4 population counts. We can
// then use PACKUSWB to shrink and concatenate them into a v4i32 again.
SDValue Zeros = getZeroVector(VT, Subtarget, DAG, DL);
- SDValue Low = DAG.getNode(X86ISD::UNPCKL, DL, VT, V, Zeros);
- SDValue High = DAG.getNode(X86ISD::UNPCKH, DL, VT, V, Zeros);
+ SDValue V32 = DAG.getBitcast(VT, V);
+ SDValue Low = DAG.getNode(X86ISD::UNPCKL, DL, VT, V32, Zeros);
+ SDValue High = DAG.getNode(X86ISD::UNPCKH, DL, VT, V32, Zeros);
// Do the horizontal sums into two v2i64s.
Zeros = getZeroVector(ByteVecVT, Subtarget, DAG, DL);
More information about the llvm-commits
mailing list