diff options
| author | Craig Topper <craig.topper@intel.com> | 2018-02-19 22:07:31 +0000 |
|---|---|---|
| committer | Craig Topper <craig.topper@intel.com> | 2018-02-19 22:07:31 +0000 |
| commit | b195ed8ce3ce2f31684b275640b9f7abaad8eeac (patch) | |
| tree | f8c4b087897cdb0cf9340b6866f10bf2d441bc2f /llvm/lib/Target | |
| parent | 1d14779aed200632328b38b04a89575b0a89c38b (diff) | |
| download | bcm5719-llvm-b195ed8ce3ce2f31684b275640b9f7abaad8eeac.tar.gz bcm5719-llvm-b195ed8ce3ce2f31684b275640b9f7abaad8eeac.zip | |
[X86] Use vpmovq2m/vpmovd2m for truncate to vXi1 when possible.
Previously we used vptestmd, but the scheduling data for SKX says vpmovq2m/vpmovd2m is lower latency. We already used vpmovb2m/vpmovw2m for byte/word truncates. So this is more consistent anyway.
llvm-svn: 325534
Diffstat (limited to 'llvm/lib/Target')
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 4 |
1 files changed, 4 insertions, 0 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index b8b3fbec732..c6916fdf1a1 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -16762,6 +16762,10 @@ static SDValue LowerTruncateVecI1(SDValue Op, SelectionDAG &DAG, In = DAG.getNode(ISD::SHL, DL, InVT, In, DAG.getConstant(ShiftInx, DL, InVT)); } + // If we have DQI, emit a pattern that will be iseled as vpmovq2m/vpmovd2m. + if (Subtarget.hasDQI()) + return DAG.getNode(X86ISD::CMPM, DL, VT, DAG.getConstant(0, DL, InVT), + In, DAG.getConstant(6, DL, MVT::i8)); return DAG.getNode(X86ISD::CMPM, DL, VT, In, getZeroVector(InVT, Subtarget, DAG, DL), DAG.getConstant(4, DL, MVT::i8)); |
