[Mlir-commits] [mlir] [MLIR][ROCDL] Add Scale Convert Packed FP4 <-> F32/BF16/F16 Support for GFX950 (PR #140676)

[Krzysztof Drewniak]

krzysz00 wrote:

@CRobeck Such verification happens, if applicable, while rewriting higher-level operations to intrinsics. Also, since MLIR doesn't have a coherent way to check which architecture a piece of code is targetting, there's nowhere to put the verifier you're interested in

Also, competing PR at #140801 

https://github.com/llvm/llvm-project/pull/140676