llvm-project/mlir/lib/Dialect/Affine/Transforms/LoopUnrollAndJam.cpp
River Riddle a70aa7bb0d [mlir:Transforms] Move out the remaining non-dialect independent transforms and utilities
This has been a major TODO for a very long time, and is necessary for establishing a proper
dialect-free dependency layering for the Transforms library. Code was moved to effectively
two main locations:

* Affine/
There was quite a bit of affine dialect related code in Transforms/ do to historical reasons
(of a time way into MLIR's past). The following headers were moved to:
Transforms/LoopFusionUtils.h -> Dialect/Affine/LoopFusionUtils.h
Transforms/LoopUtils.h -> Dialect/Affine/LoopUtils.h
Transforms/Utils.h -> Dialect/Affine/Utils.h

The following transforms were also moved:
AffineLoopFusion, AffinePipelineDataTransfer, LoopCoalescing

* SCF/
Only one SCF pass was in Transforms/ (likely accidentally placed here): ParallelLoopCollapsing
The SCF specific utilities in LoopUtils have been moved to SCF/Utils.h

* Misc:
mlir::moveLoopInvariantCode was also moved to LoopLikeInterface.h given
that it is a simple utility defined in terms of LoopLikeOpInterface.

Differential Revision: https://reviews.llvm.org/D117848
2022-01-24 19:25:53 -08:00

83 lines
2.9 KiB
C++

//===- LoopUnrollAndJam.cpp - Code to perform loop unroll and jam ---------===//
//
// 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 loop unroll and jam. Unroll and jam is a transformation
// that improves locality, in particular, register reuse, while also improving
// operation level parallelism. The example below shows what it does in nearly
// the general case. Loop unroll and jam currently works if the bounds of the
// loops inner to the loop being unroll-jammed do not depend on the latter.
//
// Before After unroll and jam of i by factor 2:
//
// for i, step = 2
// for i S1(i);
// S1; S2(i);
// S2; S1(i+1);
// for j S2(i+1);
// S3; for j
// S4; S3(i, j);
// S5; S4(i, j);
// S6; S3(i+1, j)
// S4(i+1, j)
// S5(i);
// S6(i);
// S5(i+1);
// S6(i+1);
//
// Note: 'if/else' blocks are not jammed. So, if there are loops inside if
// op's, bodies of those loops will not be jammed.
//===----------------------------------------------------------------------===//
#include "PassDetail.h"
#include "mlir/Dialect/Affine/Analysis/AffineAnalysis.h"
#include "mlir/Dialect/Affine/Analysis/LoopAnalysis.h"
#include "mlir/Dialect/Affine/IR/AffineOps.h"
#include "mlir/Dialect/Affine/LoopUtils.h"
#include "mlir/Dialect/Affine/Passes.h"
#include "mlir/IR/AffineExpr.h"
#include "mlir/IR/AffineMap.h"
#include "mlir/IR/BlockAndValueMapping.h"
#include "mlir/IR/Builders.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/Support/CommandLine.h"
using namespace mlir;
#define DEBUG_TYPE "affine-loop-unroll-jam"
namespace {
/// Loop unroll jam pass. Currently, this just unroll jams the first
/// outer loop in a Function.
struct LoopUnrollAndJam : public AffineLoopUnrollAndJamBase<LoopUnrollAndJam> {
explicit LoopUnrollAndJam(Optional<unsigned> unrollJamFactor = None) {
if (unrollJamFactor)
this->unrollJamFactor = *unrollJamFactor;
}
void runOnOperation() override;
};
} // namespace
std::unique_ptr<OperationPass<FuncOp>>
mlir::createLoopUnrollAndJamPass(int unrollJamFactor) {
return std::make_unique<LoopUnrollAndJam>(
unrollJamFactor == -1 ? None : Optional<unsigned>(unrollJamFactor));
}
void LoopUnrollAndJam::runOnOperation() {
if (getOperation().isExternal())
return;
// Currently, just the outermost loop from the first loop nest is
// unroll-and-jammed by this pass. However, runOnAffineForOp can be called on
// any for operation.
auto &entryBlock = getOperation().front();
if (auto forOp = dyn_cast<AffineForOp>(entryBlock.front()))
(void)loopUnrollJamByFactor(forOp, unrollJamFactor);
}