[Mlir-commits] [mlir] [mlir][ArmSME] Support lowering masked vector.outerproduct ops to SME (PR #69604)

[Andrzej Warzyński]

================
@@ -427,12 +427,88 @@ struct TransposeOpToArmSMELowering
   }
 };
 
+/// Conversion pattern for vector.outerproduct.
+///
+/// If the vector.outerproduct is masked (and the mask from a
+/// vector.create_mask), then the mask is decomposed into two 1-D masks for the
+/// operands.
----------------
banach-space wrote:

And what if it isn't masked?

https://github.com/llvm/llvm-project/pull/69604