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

[Cullen Rhodes]

================
@@ -474,13 +474,13 @@ struct MoveTileSliceToVectorArmSMELowering
 ///        vector<[4]xf32>) -> ()
 ///
 /// Currently only supports FMOPA and BFMOPA (non-widening).
-struct VectorOuterProductToArmSMELowering
-    : public ConvertOpToLLVMPattern<vector::OuterProductOp> {
-  using ConvertOpToLLVMPattern<vector::OuterProductOp>::ConvertOpToLLVMPattern;
+struct OuterProductToArmSMELowering
----------------
c-rhodes wrote:

Just an observation that's unrelated to this change, but I think this should be `OuterProductConversion` or something to distinguish `vector op -> higher-level ArmSME op -> intrinsic ArmSME op`. Likewise for other intrinsic conversion patterns here, e.g. `LoadTileSliceToArmSMELowering` -> `LoadTileSliceConversion`. Don't expect you to do anything about it as part of this patch just thought I'd mention.

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