[Mlir-commits] [mlir] [MLIR][ROCDL] Remove patterns for ops supported as intrinsics in the AMDGPU backend (PR #102971)

[Krzysztof Drewniak]

krzysz00 wrote:

@akuegel You'll want to run `--arith-emulate-unsupported-floats` to expand out the bf16 operations, I think.

Alternatively, I'll take a followup patch to lower BF16 `math.*` to the equivalent f32 calls (LLVM intrinsics or OCML ones as relevant) with `extf` and `truncf`

https://github.com/llvm/llvm-project/pull/102971