[Mlir-commits] [mlir] [mlir][memref] Rewrite scalar `memref.copy` through reinterpret_cast into load/store (PR #186118)

[Andrzej Warzyński]

================
@@ -0,0 +1,205 @@
+//===- BypassReinterpretCast.cpp - Expansion patterns for MemRef operations
+//----------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/Arith/IR/Arith.h"
+#include "mlir/Dialect/Arith/Transforms/Passes.h"
+#include "mlir/Dialect/Arith/Utils/Utils.h"
+#include "mlir/Dialect/MemRef/IR/MemRef.h"
+#include "mlir/Dialect/MemRef/Transforms/Transforms.h"
+#include "mlir/IR/TypeUtilities.h"
+#include "mlir/Transforms/DialectConversion.h"
+
+namespace mlir {
+namespace memref {
+#define GEN_PASS_DEF_BYPASSREINTERPRETCASTPASS
+#include "mlir/Dialect/MemRef/Transforms/Passes.h.inc"
+} // namespace memref
+} // namespace mlir
+
+using namespace mlir;
+
+namespace {
+
+// Checks if strided memref is an (expanded) scalar view
+// of an (expanded) single row/column memref, with the same rank.
+static bool isScalarSlice(memref::ReinterpretCastOp rc) {
+  auto underlyingType = dyn_cast<MemRefType>(rc.getSource().getType());
+  auto stridedType = dyn_cast<MemRefType>(rc.getType());
----------------
banach-space wrote:

IIUC, one of these is the input type and the other is the output type? Perhaps rename? 
* `underlyingType` -> `rcInputTy`, 
* `stridedType` -> `rcOutputTy`?

Or something similar.

`stridedType` is a bit misleading as it can be used to create "identity layout views" with default strides that don't require printing, e.g. `%1` below:
```mlir
%0 = memref.alloc() : memref<3xf32>
%1 = memref.reinterpret_cast %0 to offset: [0], sizes: [3, 1, 1], strides: [1, 1, 1] : memref<3xf32> to memref<3x1x1xf32>
```

https://github.com/llvm/llvm-project/pull/186118