[Mlir-commits] [mlir] [mlir][vector] Add vector.to_elements unrolling (PR #157142)

[Erick Ochoa Lopez]

================
@@ -808,6 +808,28 @@ struct TestUnrollVectorFromElements
   }
 };
 
+struct TestUnrollVectorToElements
+    : public PassWrapper<TestUnrollVectorToElements,
+                         OperationPass<func::FuncOp>> {
+  MLIR_DEFINE_EXPLICIT_INTERNAL_INLINE_TYPE_ID(TestUnrollVectorToElements)
+
+  StringRef getArgument() const final {
+    return "test-unroll-vector-to-elements";
+  }
+  StringRef getDescription() const final {
+    return "Test unrolling patterns for to_elements ops";
+  }
+  void getDependentDialects(DialectRegistry &registry) const override {
+    registry.insert<func::FuncDialect, vector::VectorDialect>();
+  }
+
+  void runOnOperation() override {
+    RewritePatternSet patterns(&getContext());
+    populateVectorToElementsLoweringPatterns(patterns);
+    (void)applyPatternsGreedily(getOperation(), std::move(patterns));
+  }
+};
+
----------------
amd-eochoalo wrote:

The reason why this test pass was added is to keep the style consistent with https://github.com/llvm/llvm-project/pull/151175 . Sorry, I am not familiar with TD as an acronym. Do you mean the test dialect or the transform dialect? I am assuming you are asking about adding a transform dialect operation for this pass? I have no experience with them and how scaling is different between the transform dialect and these test passes here. Happy to add it if needed, but I prefer to maintain the consistency with the vector.from_elements unrolling PR.

https://github.com/llvm/llvm-project/pull/157142