[Mlir-commits] [mlir] [mlir][linalg] Add a test to demonstrate peeling + vectorisation (PR #77590)

[Benjamin Maxwell]

================
@@ -0,0 +1,86 @@
+// RUN: mlir-opt %s --transform-interpreter --split-input-file -canonicalize | FileCheck %s
+
+// Demonstrates what happens when peeling the middle loop (2nd parallel
+// dimension) followed by vectorization in the presence of _scalable_ vectors
+// (these are introduced through scalable tiling). The main goal is to verify
+// that canonicalizations fold away the masks in the main loop.
+
+func.func @matmul(%A: tensor<1024x512xf32>,
+                  %B: tensor<512x2000xf32>,
+                  %C:tensor<1024x2000xf32>) -> tensor<1024x2000xf32> {
----------------
MacDue wrote:

nit: alignment
```suggestion
                  %C: tensor<1024x2000xf32>) -> tensor<1024x2000xf32> {
```

https://github.com/llvm/llvm-project/pull/77590