[Mlir-commits] [mlir] [mlir][vector] Add vector.to_elements unrolling (PR #157142)

[Erick Ochoa Lopez]

================
@@ -0,0 +1,54 @@
+//===- LowerVectorToElements.cpp - Lower 'vector.to_elements' op ----------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements target-independent rewrites and utilities to lower the
+// 'vector.to_elements' operation.
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/Vector/IR/VectorOps.h"
+#include "mlir/Dialect/Vector/Transforms/LoweringPatterns.h"
+
+#define DEBUG_TYPE "lower-vector-to-elements"
+
+using namespace mlir;
+
+namespace {
+
+struct UnrollToElements final : OpRewritePattern<vector::ToElementsOp> {
+  using OpRewritePattern::OpRewritePattern;
+
+  LogicalResult matchAndRewrite(vector::ToElementsOp op,
+                                PatternRewriter &rewriter) const override {
+
+    TypedValue<VectorType> source = op.getSource();
+    FailureOr<SmallVector<Value>> result =
+        vector::unrollVectorValue(source, rewriter);
+    if (failed(result)) {
+      return failure();
+    }
+    SmallVector<Value> vectors = *result;
+
+    // May be a large vector.
+    SmallVector<Value, 0> results;
----------------
amd-eochoalo wrote:

The comment above and my belief that this vector may be large, made me originally consider not using `SmallVector<Value>` which as the name implies is optimized for small vectors. I originally had `std::vector<Value>` due to its possibly large size. Jakub suggested this approach motivated by the following line 

> SmallVector has grown a few other minor advantages over std::vector, causing SmallVector<Type, 0> to be preferred over std::vector<Type>.



https://github.com/llvm/llvm-project/pull/157142