diff options
| author | Matt Arsenault <Matthew.Arsenault@amd.com> | 2023-07-16 08:32:08 -0400 |
|---|---|---|
| committer | Matt Arsenault <Matthew.Arsenault@amd.com> | 2023-07-21 18:55:42 -0400 |
| commit | 8406c3568aa5cd8256a2b359174eae82c747b162 (patch) | |
| tree | ca9b605228e354ba5831843f6f1adb064ddc61b6 /clang/lib/CodeGen/CodeGenModule.cpp | |
| parent | 6699c37028148c722de2a401c13eef2e92833a03 (diff) | |
| download | llvm-8406c3568aa5cd8256a2b359174eae82c747b162.zip llvm-8406c3568aa5cd8256a2b359174eae82c747b162.tar.gz llvm-8406c3568aa5cd8256a2b359174eae82c747b162.tar.bz2 | |
AMDGPU: Implement new 2ulp fdiv lowering
Extends the new frexp scaled reciprocal to the general case. The
reciprocal case is just the same thing when frexp of 1 is constant
folded. Could probably clean up the code to rely on that constant
folding.
Improves results for the IEEE path for the default OpenCL division. We
used to only emit the fdiv.fast intrinsic with a 2.5 ulp accuracy
threshold with DAZ, which uses explicit range checks. This gives us a
better fast option with the default IEEE behavior.
Diffstat (limited to 'clang/lib/CodeGen/CodeGenModule.cpp')
0 files changed, 0 insertions, 0 deletions