[Mlir-commits] [mlir] [MLIR][AArch64] Lower `vector.contract` to SVE FEAT_BF16 operations (PR #147052)

[Andrzej Warzyński]

================
@@ -23,4 +23,14 @@ def ApplyArmSVELowerContractionPatternsOp
   let assemblyFormat = "attr-dict";
 }
 
+def ApplyArmSVELowerContractionToBFMMLAPatternsOp
+    : Op<Transform_Dialect, "apply_patterns.arm_sve.vector_contract_to_bfmmla",
+         [DeclareOpInterfaceMethods<PatternDescriptorOpInterface>]> {
+  let description = [{
+    Indicates that vector contraction-like operations should be lowered to
+    finer-grained vector primitives using the ArmSVE dialect.
----------------
banach-space wrote:

ATM this is just a copy of the description for `ApplyArmSVELowerContractionToI8MMPatternsOp`. Could you extend it so that the difference between the two is clear? Thanks!

https://github.com/llvm/llvm-project/pull/147052