| 1 | //===- LoopRotation.cpp - Loop Rotation Pass ------------------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // This file implements Loop Rotation Pass. |
| 10 | // |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | |
| 13 | #include "llvm/Transforms/Scalar/LoopRotation.h" |
| 14 | #include "llvm/Analysis/AssumptionCache.h" |
| 15 | #include "llvm/Analysis/InstructionSimplify.h" |
| 16 | #include "llvm/Analysis/LazyBlockFrequencyInfo.h" |
| 17 | #include "llvm/Analysis/LoopInfo.h" |
| 18 | #include "llvm/Analysis/LoopPass.h" |
| 19 | #include "llvm/Analysis/MemorySSA.h" |
| 20 | #include "llvm/Analysis/MemorySSAUpdater.h" |
| 21 | #include "llvm/Analysis/ScalarEvolution.h" |
| 22 | #include "llvm/Analysis/TargetTransformInfo.h" |
| 23 | #include "llvm/Support/CommandLine.h" |
| 24 | #include "llvm/Transforms/Scalar.h" |
| 25 | #include "llvm/Transforms/Utils/LoopRotationUtils.h" |
| 26 | #include "llvm/Transforms/Utils/LoopUtils.h" |
| 27 | #include <optional> |
| 28 | using namespace llvm; |
| 29 | |
| 30 | #define DEBUG_TYPE "loop-rotate" |
| 31 | |
| 32 | static cl::opt<unsigned> DefaultRotationThreshold( |
| 33 | "rotation-max-header-size" , cl::init(Val: 16), cl::Hidden, |
| 34 | cl::desc("The default maximum header size for automatic loop rotation" )); |
| 35 | |
| 36 | static cl::opt<bool> PrepareForLTOOption( |
| 37 | "rotation-prepare-for-lto" , cl::init(Val: false), cl::Hidden, |
| 38 | cl::desc("Run loop-rotation in the prepare-for-lto stage. This option " |
| 39 | "should be used for testing only." )); |
| 40 | |
| 41 | // Experimentally allow loop header duplication. This should allow for better |
| 42 | // optimization at Oz, since loop-idiom recognition can then recognize things |
| 43 | // like memcpy. If this ends up being useful for many targets, we should drop |
| 44 | // this flag and make a code generation option that can be controlled |
| 45 | // independent of the opt level and exposed through the frontend. |
| 46 | static cl::opt<bool> ( |
| 47 | "enable-loop-header-duplication-at-minsize" , cl::init(Val: false), cl::Hidden, |
| 48 | cl::desc("Enable loop header duplication even for minsize" )); |
| 49 | |
| 50 | LoopRotatePass::LoopRotatePass(bool , bool PrepareForLTO, |
| 51 | bool CheckExitCount) |
| 52 | : EnableHeaderDuplication(EnableHeaderDuplication), |
| 53 | PrepareForLTO(PrepareForLTO), CheckExitCount(CheckExitCount) {} |
| 54 | |
| 55 | void LoopRotatePass::printPipeline( |
| 56 | raw_ostream &OS, function_ref<StringRef(StringRef)> MapClassName2PassName) { |
| 57 | static_cast<PassInfoMixin<LoopRotatePass> *>(this)->printPipeline( |
| 58 | OS, MapClassName2PassName); |
| 59 | OS << "<" ; |
| 60 | if (!EnableHeaderDuplication) |
| 61 | OS << "no-" ; |
| 62 | OS << "header-duplication;" ; |
| 63 | |
| 64 | if (!PrepareForLTO) |
| 65 | OS << "no-" ; |
| 66 | OS << "prepare-for-lto;" ; |
| 67 | |
| 68 | if (!CheckExitCount) |
| 69 | OS << "no-" ; |
| 70 | OS << "check-exit-count" ; |
| 71 | OS << ">" ; |
| 72 | } |
| 73 | |
| 74 | PreservedAnalyses LoopRotatePass::run(Loop &L, LoopAnalysisManager &AM, |
| 75 | LoopStandardAnalysisResults &AR, |
| 76 | LPMUpdater &) { |
| 77 | // Vectorization requires loop-rotation. Use default threshold for loops the |
| 78 | // user explicitly marked for vectorization, even when header duplication is |
| 79 | // disabled. |
| 80 | int Threshold = EnableHeaderDuplication && |
| 81 | (!L.getHeader()->getParent()->hasMinSize() || |
| 82 | EnableLoopHeaderDuplicationAtMinSize || |
| 83 | hasVectorizeTransformation(L: &L) == TM_ForcedByUser) |
| 84 | ? DefaultRotationThreshold |
| 85 | : 0; |
| 86 | const DataLayout &DL = L.getHeader()->getDataLayout(); |
| 87 | const SimplifyQuery SQ = getBestSimplifyQuery(AR, DL); |
| 88 | |
| 89 | std::optional<MemorySSAUpdater> MSSAU; |
| 90 | if (AR.MSSA) |
| 91 | MSSAU = MemorySSAUpdater(AR.MSSA); |
| 92 | bool Changed = |
| 93 | LoopRotation(L: &L, LI: &AR.LI, TTI: &AR.TTI, AC: &AR.AC, DT: &AR.DT, SE: &AR.SE, |
| 94 | MSSAU: MSSAU ? &*MSSAU : nullptr, SQ, RotationOnly: false, Threshold, IsUtilMode: false, |
| 95 | PrepareForLTO: PrepareForLTO || PrepareForLTOOption, CheckExitCount); |
| 96 | |
| 97 | if (!Changed) |
| 98 | return PreservedAnalyses::all(); |
| 99 | |
| 100 | if (AR.MSSA && VerifyMemorySSA) |
| 101 | AR.MSSA->verifyMemorySSA(); |
| 102 | |
| 103 | auto PA = getLoopPassPreservedAnalyses(); |
| 104 | if (AR.MSSA) |
| 105 | PA.preserve<MemorySSAAnalysis>(); |
| 106 | return PA; |
| 107 | } |
| 108 | |