[Mlir-commits] [mlir] [mlir][vector] Disable CompressStoreOp/ExpandLoadOp for scalable vectors (PR #117538)

[Andrzej Warzyński]

================
@@ -24,6 +24,9 @@ include "mlir/IR/DialectBase.td"
 // Explicitly disallow 0-D vectors for now until we have good enough coverage.
 def IsVectorOfNonZeroRankTypePred : And<[CPred<"::llvm::isa<::mlir::VectorType>($_self)">,
                                          CPred<"::llvm::cast<::mlir::VectorType>($_self).getRank() > 0">]>;
+def IsFixedVectorOfNonZeroRankTypePred : And<[CPred<"::llvm::isa<::mlir::VectorType>($_self)">,
+                                              CPred<"::llvm::cast<::mlir::VectorType>($_self).getRank() > 0">,
+                                              CPred<"!::llvm::cast<VectorType>($_self).isScalable()">]>;
----------------
banach-space wrote:

I've implemented it and you've approved it:
* https://github.com/llvm/llvm-project/pull/87986

:) I've not landed it yet - we've been discussing how "scalability" is modelled and I wanted to avoid merging it pre-maturely. And the life happened 🤷🏻‍♂️ 

I will land it this week, unless there's some new comments.

https://github.com/llvm/llvm-project/pull/117538