[Mlir-commits] [mlir] [MLIR][XeGPU] Allow some nd ops to have argument shapes mismatch for … (PR #120566)

[Jianhui Li]

Jianhui-Li wrote:

> > That seems like an unnecessary demand because after SIMT we don't really need sg_map. Also what if some user wants to directly generate XeGPU SIMT?
> 
> Seems to me that `sg_map` is essentially required which, IMO, is fine on its own. We just need to clearly document and/or enforce it somehow such that users can easily figure out what to do.

I am OK to keep `sg_map` as required as well. It simplifies the IR validation, and the user (as a compiler component) should know the SIMT distribution rule exactly therefore won't be extra burden to attach `sg_map` attribute for XeGPU Ops.  

https://github.com/llvm/llvm-project/pull/120566