diff options
| author | Simon Pilgrim <llvm-dev@redking.me.uk> | 2015-07-29 20:31:45 +0000 |
|---|---|---|
| committer | Simon Pilgrim <llvm-dev@redking.me.uk> | 2015-07-29 20:31:45 +0000 |
| commit | 86478c6909ebd71667413fb27b483d3bc0bff3dc (patch) | |
| tree | 610104db9faf7f3eaf6feb55aeefa88e36d42629 /llvm/lib/Target/X86/X86TargetTransformInfo.cpp | |
| parent | 46682630f41e0ed95b7f5ea73f2445357b16c67b (diff) | |
| download | bcm5719-llvm-86478c6909ebd71667413fb27b483d3bc0bff3dc.tar.gz bcm5719-llvm-86478c6909ebd71667413fb27b483d3bc0bff3dc.zip | |
[X86][SSE] Vectorize i64 ASHR operations
This patch vectorizes the v2i64/v4i64 ASHR shift operations - the last remaining integer vector shifts that are still being transferred to/from the scalar unit to be completed.
Differential Revision: http://reviews.llvm.org/D11439
llvm-svn: 243569
Diffstat (limited to 'llvm/lib/Target/X86/X86TargetTransformInfo.cpp')
| -rw-r--r-- | llvm/lib/Target/X86/X86TargetTransformInfo.cpp | 5 |
1 files changed, 3 insertions, 2 deletions
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp index a2d0884c4d4..04c16f8dfc7 100644 --- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp +++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp @@ -163,7 +163,8 @@ unsigned X86TTIImpl::getArithmeticInstrCost( { ISD::SRA, MVT::v32i8, 24 }, // vpblendvb sequence. { ISD::SRA, MVT::v16i16, 10 }, // extend/vpsravd/pack sequence. - { ISD::SRA, MVT::v4i64, 4*10 }, // Scalarized. + { ISD::SRA, MVT::v2i64, 4 }, // srl/xor/sub sequence. + { ISD::SRA, MVT::v4i64, 4 }, // srl/xor/sub sequence. // Vectorizing division is a bad idea. See the SSE2 table for more comments. { ISD::SDIV, MVT::v32i8, 32*20 }, @@ -270,7 +271,7 @@ unsigned X86TTIImpl::getArithmeticInstrCost( { ISD::SRA, MVT::v16i8, 54 }, // unpacked cmpgtb sequence. { ISD::SRA, MVT::v8i16, 32 }, // cmpgtb sequence. { ISD::SRA, MVT::v4i32, 16 }, // Shift each lane + blend. - { ISD::SRA, MVT::v2i64, 2*10 }, // Scalarized. + { ISD::SRA, MVT::v2i64, 12 }, // srl/xor/sub sequence. // It is not a good idea to vectorize division. We have to scalarize it and // in the process we will often end up having to spilling regular |
