[Mlir-commits] [mlir] [MLIR][XeGPU] Support vector.contract transpose_a/transpose_b via 'vector-to-gpu' patterns (PR #182885)

[Jianhui Li]

================
@@ -844,6 +845,11 @@ struct ContractionLowering : public OpRewritePattern<vector::ContractionOp> {
 struct ConvertVectorToXeGPUPass
     : public impl::ConvertVectorToXeGPUBase<ConvertVectorToXeGPUPass> {
   void runOnOperation() override {
+    RewritePatternSet prep(&getContext());
+    populatePrepareVectorToMMAPatterns(prep);
----------------
Jianhui-Li wrote:

I think you only need to add in the exisitng rewrite pass:
patterns.add<PrepareContractToGPUMMA>(
        patterns.getContext());


https://github.com/llvm/llvm-project/pull/182885