[Mlir-commits] [mlir] [MLIR][XeGPU] Extend op definitions to support 3D+: dpas, dpas_mx (PR #199809)

[Sang Ik Lee]

================
@@ -930,19 +942,21 @@ def XeGPU_DpasOp : XeGPU_Op<"dpas", [Pure, AllElementTypesMatch<["lhs", "rhs"]>,
     Arguments:
 
     - `lhs`: A vector value representing the left-hand-side matrix tile (A) participating in the
-      matrix multiply.
+      matrix multiply. Can be 1D, 2D, 3D, or 4D where leading dimensions are batch dimensions.
 
-    - `rhs`: A vector value representing the right-hand-side matrix tile (B).
+    - `rhs`: A vector value representing the right-hand-side matrix tile (B). Can be 1D, 2D, 3D, or 4D
+      where leading dimensions are batch dimensions, plus an optional trailing dimension for VNNI packing.
 
     - `acc`: [optional] A vector value representing the accumulator matrix tile (C). When present, the
       result is computed as `lhs * rhs + acc`; otherwise, the accumulator is implicitly assumed to be zero.
+      Must have the same batch dimensions as lhs and rhs.
----------------
silee2 wrote:

I guess I found the answer to my previous question. This statement means A, B and C need to have same batch dimension. Correct?

https://github.com/llvm/llvm-project/pull/199809