[Mlir-commits] [mlir] [MLIR][XeGPU] Update XeGPU create_tdesc, update_offset, load, store and prefetch. (PR #154653)

[Adam Siemieniuk]

================
@@ -510,6 +511,8 @@ def XeGPU_CreateDescOp: XeGPU_Op<"create_tdesc", [Pure, ViewLikeOpInterface]> {
     match the dimension of offsets. It may also has a second dimension corresponding to
     the chunk_size if the chunk size is larger than 1.
 
+    This op is not available in SIMT mode.
----------------
adam-smnk wrote:

Why not?

It seems a bit odd that `nd_desc` is fine across all abstractions but not this one. And distribution logic now has to suddenly fold ops in this case.
Such cleanup would be better left for a folder at SIMT level or XeGPU lowering.

https://github.com/llvm/llvm-project/pull/154653