[Mlir-commits] [mlir] [mlir][scf] Implement conversion from scf.forall to scf.parallel (PR #94109)

[Spenser Bauman]

================
@@ -0,0 +1,82 @@
+//===- ForallToParallel.cpp - scf.forall to scf.parallel loop conversion --===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// Transforms SCF.ForallOp's into SCF.ParallelOps's.
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Dialect/SCF/IR/SCF.h"
+#include "mlir/Dialect/SCF/Transforms/Passes.h"
+#include "mlir/Dialect/SCF/Transforms/Transforms.h"
+#include "mlir/IR/PatternMatch.h"
+
+namespace mlir {
+#define GEN_PASS_DEF_SCFFORALLTOPARALLELLOOP
+#include "mlir/Dialect/SCF/Transforms/Passes.h.inc"
+} // namespace mlir
+
+using namespace mlir;
+
+LogicalResult mlir::scf::forallToParallelLoop(RewriterBase &rewriter,
----------------
sabauma wrote:

I've added a test to lock down preserving the attribute. Formalizing the `mapping` attribute on the `scf.parallel` op has some fallout on the GPU dialect, so I opted not to do so. Some of the GPU transforms use the `mapping` attribute with values other than `DeviceMappingArrayAttr`, which is what `scf.forall` requires.

https://github.com/llvm/llvm-project/pull/94109