[Mlir-commits] [mlir] [MLIR] Folding unpack and pack sequence in data layout propagation from padded domain (PR #138332)

Han-Chung Wang llvmlistbot at llvm.org
Tue May 6 14:23:03 PDT 2025


================
@@ -447,8 +469,10 @@ bubbleUpPackOpThroughGenericOp(RewriterBase &rewriter, linalg::PackOp packOp,
                             .getDefiningOp<tensor::EmptyOp>()) {
     dest = packOpDest;
   }
+  // Here pack(unpack) isn't naively foldable because the unpack op can be from
+  // an arbitrary domain so we need to keep both.
----------------
hanhanW wrote:

picky nit: remove `Here`. I have context, so I can see why you have "here" in this comment. (IIUC, you want to show the difference between two `packGenericOp` calls) However, it looks a little weird to me, if we read the code from the start of this function. How about removing `Here` from the comment?

```suggestion
  // pack(unpack) isn't naively foldable because the unpack op can be from
  // an arbitrary domain so we need to keep both.
```

https://github.com/llvm/llvm-project/pull/138332


More information about the Mlir-commits mailing list