[Mlir-commits] [mlir] [mlir][vector] Sink vector.extract/splat into load/store ops (PR #134389)

[Andrzej Warzyński]

================
@@ -469,8 +469,29 @@ def ApplySinkVectorPatternsOp : Op<Transform_Dialect,
     %0 = arith.addf %a, %b : vector<4x2xf32>
     %r = vector.transpose %0, [1, 0] : vector<2x4xf32>
     ```
-    At the moment, these patterns are limited to vector.broadcast and
-    vector.transpose.
+    At the moment, these patterns are limited to vector.broadcast,
+    vector.transpose and vector.extract.
+  }];
+
+  let assemblyFormat = "attr-dict";
+}
+
+def ApplySinkVectorMemPatternsOp : Op<Transform_Dialect,
+    "apply_patterns.vector.sink_mem_ops",
+    [DeclareOpInterfaceMethods<PatternDescriptorOpInterface>]> {
+  let description = [{
+    Patterns that remove redundant Vector Ops by merging them with load/store
----------------
banach-space wrote:

[nit] `StoreOpFromSplatOrBroadcast` will replace `vector.store` with either `vector.store` or `memref.store`. Perhaps update to:

> Patterns that replace redundant Vector Ops (followed by vector.load/vector.store) with either vector.load/vector.store or memref.load/memref.store. Currently limited to 1-element vectors.

https://github.com/llvm/llvm-project/pull/134389