If 2.5 ulp is acceptable, denormals are not required, and
isn't a reciprocal which will already be handled, replace
with a faster fdiv.
Simplify the lowering tests by using per function
subtarget features.
Paths
| Differential D21765
AMDGPU: Change fdiv lowering based on !fpmath metadata ClosedPublic Authored by arsenm on Jun 27 2016, 11:54 AM.
Details
Summary If 2.5 ulp is acceptable, denormals are not required, and Simplify the lowering tests by using per function
Diff Detail Event Timelinearsenm updated this object. This revision is now accepted and ready to land.Jul 18 2016, 6:57 AM
Revision Contents
Diff 61997 lib/Target/AMDGPU/AMDGPU.h
lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
lib/Target/AMDGPU/AMDGPUIntrinsicInfo.h
lib/Target/AMDGPU/AMDGPUIntrinsicInfo.cpp
lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
lib/Target/AMDGPU/SIISelLowering.h
lib/Target/AMDGPU/SIISelLowering.cpp
lib/Target/AMDGPU/SIIntrinsics.td
test/CodeGen/AMDGPU/amdgpu-codegenprepare.ll
test/CodeGen/AMDGPU/fdiv.ll
test/CodeGen/AMDGPU/llvm.amdgcn.fdiv.fast.ll
|