[Mlir-commits] [mlir] [MLIR][Math] Add lowering for isnormal and isinf (PR #184370)

[llvmlistbot at llvm.org]

maxxlife wrote:

I tested a PR at commit 56e8d0f1a6e3. A direct scalar f32 test with:
`mlir-opt math-isinf.mlir --convert-math-to-llvm`
successfully lowered` math.isinf` to:
`"llvm.intr.is.fpclass"(%arg0)`
I also ran the exact input and full gpu-lower-to-xevm-pipeline reproduction command from #213583. With this PR, mlir-opt completed successfully with exit code 0 isntead of failing during kernel serialization. This confirmas that the lowering works on the current head of this PR. Since the PR is based on an older LLVM revision with a different XeVM pipeline layout, I have not yet verified the result after rebasing onto the latest `main` or together with #213874 


https://github.com/llvm/llvm-project/pull/184370