[Mlir-commits] [mlir] [mlir][xegpu] Remove chunk_size attribute; infer from types (PR #205122)

[Jianhui Li]

================
@@ -708,15 +708,12 @@ def XeGPU_LoadGatherOp : XeGPU_Op<"load", [MemoryEffects<[MemRead]>, AnchorLayou
 
   let description = [{ It (aka. load) load data per each lane. The output
     describes the data being loaded at the subgroup level, so its size is
-    consistent with the number of lanes in a subgroup. When the chunk size
-    is larger than 2, the output vector is a 2D vector, with dim-0 correspoding
-    to lanes, and dim-1 corresponding to the chunk size loaded by each lane.
+    consistent with the number of lanes in a subgroup. At lane level each lane
+    may load a 1D vector of contiguous elements; that per-lane width is implied
+    by the result type and does not need a separate attribute.
----------------
Jianhui-Li wrote:

Please add:
For each level, the mask, offsets, and result can be nD vectors, and their shape must be same. 

I don't think we need to describe this "each lane may load 1d vector" in the description.  The lane level could load nd vector with leading unit dims. 


https://github.com/llvm/llvm-project/pull/205122