[Mlir-commits] [mlir] [MLIR][XeGPU] Add unroll patterns for XeGPU (1/N) (PR #137010)

[Chao Chen]

================
@@ -12,11 +12,18 @@
 namespace mlir {
 class RewritePatternSet;
 
+namespace vector {
+struct UnrollVectorOptions;
+} // namespace vector
+
 namespace xegpu {
 
 /// Appends patterns for folding aliasing ops into XeGPU ops into `patterns`.
 void populateXeGPUFoldAliasOpsPatterns(RewritePatternSet &patterns);
 
+void populateXeGPUUnrollPatterns(RewritePatternSet &patterns,
+                                 const vector::UnrollVectorOptions &options);
----------------
chencha3 wrote:

> It's better to decouple from the generic vector infrastructure and create XeGPU specific options (even if they end up being the same).

Even these patterns are to be used together with vector unroll patterns in the same pass?

https://github.com/llvm/llvm-project/pull/137010