diff options
| author | Hyunsung Lee <ita9naiwa@gmail.com> | 2025-01-29 12:56:43 +0900 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-01-28 22:56:43 -0500 |
| commit | 3a3377579f137a0a6e14b60d891a9736707e7e8d (patch) | |
| tree | 3b567abbb6043eab18d4f334920c3ddd6d28b43c /llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp | |
| parent | 3bd3e06f3fe418e24af65457877f40cee0544f9d (diff) | |
| download | llvm-3a3377579f137a0a6e14b60d891a9736707e7e8d.zip llvm-3a3377579f137a0a6e14b60d891a9736707e7e8d.tar.gz llvm-3a3377579f137a0a6e14b60d891a9736707e7e8d.tar.bz2 | |
[mlir][math]Update `convertPowfOp` `ExpandPatterns.cpp` (#124402)
The current implementation of `convertPowfOp` requires a calculation of
`a * a` but, max\<fp16\> ~= 65,504, and if `a` is about 16, it will
overflow so get INF in fp8 or fp16 easily.
Remove support when `a < 0`. Overhead of handling negative value of `a`
is large and easy to overflow;
- related issue in iree:
https://github.com/iree-org/iree/issues/15936
Diffstat (limited to 'llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp')
0 files changed, 0 insertions, 0 deletions