summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Target
diff options
context:
space:
mode:
authorSimon Pilgrim <llvm-dev@redking.me.uk>2018-10-08 18:40:50 +0000
committerSimon Pilgrim <llvm-dev@redking.me.uk>2018-10-08 18:40:50 +0000
commit6fc8d055656bf6e278dc50973672594ddd9f5edf (patch)
treef5de5663c82564ea3f825a01d4727a5b166ed901 /llvm/lib/Target
parent869e5d1fca1caaddfc6c3f8fc7de447e874f49ee (diff)
downloadbcm5719-llvm-6fc8d055656bf6e278dc50973672594ddd9f5edf.tar.gz
bcm5719-llvm-6fc8d055656bf6e278dc50973672594ddd9f5edf.zip
[X86][AVX2] Enable ZERO_EXTEND_VECTOR_INREG lowering of 256-bit vectors
Some necessary yak shaving before lowering *_EXTEND_VECTOR_INREG 256-bit vectors on AVX1 targets as suggested by D52964. Differential Revision: https://reviews.llvm.org/D52970 llvm-svn: 343991
Diffstat (limited to 'llvm/lib/Target')
-rw-r--r--llvm/lib/Target/X86/X86ISelLowering.cpp12
1 files changed, 5 insertions, 7 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index ef8bf06c173..556dcd3f402 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1125,9 +1125,10 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
}
if (HasInt256) {
- setOperationAction(ISD::SIGN_EXTEND_VECTOR_INREG, MVT::v4i64, Custom);
- setOperationAction(ISD::SIGN_EXTEND_VECTOR_INREG, MVT::v8i32, Custom);
- setOperationAction(ISD::SIGN_EXTEND_VECTOR_INREG, MVT::v16i16, Custom);
+ for (auto VT : { MVT::v16i16, MVT::v8i32, MVT::v4i64 }) {
+ setOperationAction(ISD::SIGN_EXTEND_VECTOR_INREG, VT, Custom);
+ setOperationAction(ISD::ZERO_EXTEND_VECTOR_INREG, VT, Custom);
+ }
// The custom lowering for UINT_TO_FP for v8i32 becomes interesting
// when we have a 256bit-wide blend with immediate.
@@ -19727,10 +19728,7 @@ static SDValue LowerEXTEND_VECTOR_INREG(SDValue Op,
In = extractSubVector(In, 0, DAG, dl, std::max(InSize, 128));
}
- assert((Op.getOpcode() != ISD::ZERO_EXTEND_VECTOR_INREG ||
- InVT == MVT::v64i8) && "Zero extend only for v64i8 input!");
-
- // SSE41 targets can use the pmovsx* instructions directly for 128-bit results,
+ // SSE41 targets can use the pmov[sz]x* instructions directly for 128-bit results,
// so are legal and shouldn't occur here. AVX2/AVX512 pmovsx* instructions still
// need to be handled here for 256/512-bit results.
if (Subtarget.hasInt256()) {
OpenPOWER on IntegriCloud