[Mlir-commits] [mlir] [mlir][tosa] Support variadic batch dimensions in matmul ops (PR #226179)

[Deeptanshu Sekhri]

================
@@ -454,16 +461,21 @@ def Tosa_MatMulTOp : Tosa_InferShapedTypeOp<"matmul_t", [NoMemoryEffect]> {
   let summary = "Matrix multiplication with transposed B operator.";
 
   let description = [{
-    Performs two dimensional matrix multiplications. `A` matrix is of shape
-    `N x H x C`. `B` matrix is of shape `D x W x C`. This is effectively a
-    matrix multiply of `A` by the transposed `B` matrix. If the batched
-    dimension of input `B` is of size 1, the `B` matrix is broadcast.
+    Performs batched matrix multiplication of `A` with shape `[..., H, C]` and
----------------
deeptanshusekhri wrote:

Same as above, updated to match spec.

https://github.com/llvm/llvm-project/pull/226179