[Mlir-commits] [mlir] [mlir][linalg] fix specialization of transposed matmul variants (PR #181387)

[Adam Siemieniuk]

================
@@ -314,6 +314,73 @@ func.func @negative_op_multi_reduction(%A: tensor<10x20x30xf32>,
 
 // -----
 
+// Both A and B transposed: no named op variant exists for this combination.
+#mapABt0 = affine_map<(d0, d1, d2) -> (d2, d0)>
+#mapABt1 = affine_map<(d0, d1, d2) -> (d1, d2)>
+#mapABt2 = affine_map<(d0, d1, d2) -> (d0, d1)>
+func.func @negative_matmul_transpose_a_and_b(%A: tensor<?x?xf32>, %B: tensor<?x?xf32>,
----------------
adam-smnk wrote:

Yes, that is supported. It'd be great to relax this check.
I think `linalg.(batch_)matmul` also supports output transpose.

https://github.com/llvm/llvm-project/pull/181387