[Mlir-commits] [mlir] [mlir][amdgpu] Promote gpu.shuffle to amdgpu.dpp (PR #155158)

[Fabian Mora]

fabianmcg wrote:

I haven't checked the patch in full, but, can you add a `GPUAMDGPUTransformOps.td` and a new C++ library and move the transform there? I'm not blocking as the layering issue is already there, but a cleanup would be nice, otherwise I'll do it later.

The issue is that having these transforms in `GPUTranformOps` forces the loading of the AMDGPU dialect at Transform Dialect Extension registration time, making the whole library undesirable to use with other targets, or in other circumstances.

https://github.com/llvm/llvm-project/pull/155158