[Mlir-commits] [mlir] [mlir][vector] Add a new TD op to wrap unit-dim collapsing patterns (PR #157507)

[Andrzej Warzyński]

================
@@ -1,4 +1,7 @@
 // RUN: mlir-opt %s -test-vector-transfer-collapse-inner-most-dims -split-input-file | FileCheck %s
+// RUN: mlir-opt -split-input-file \
----------------
banach-space wrote:

> Do we necessarily need the other -test-vector-transfer-collapse-inner-most-dims since we have the transform op now?

We don't - I can just remove it, yes. See this [commit](https://github.com/llvm/llvm-project/pull/157507/commits/01f3b48cc337513d24f0504b87fea20f4adf64f5)

> If not, then the function on the other file could also just live here in this file I guess.

Not quite :) Having a dedicated TD file means that we don't have to repeat the pattern for every example/"split" :)

Btw, using separate TD files is not a completely new pattern: https://github.com/llvm/llvm-project/blob/193df2a12bb28d1e04217c1cdb39048c5d19aed7/mlir/test/Dialect/Linalg/vectorization/extract-with-patterns.mlir#L1-L3

https://github.com/llvm/llvm-project/pull/157507