1 //===-- MachineFunctionSplitter.cpp - Split machine functions //-----------===// 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 // \file 10 // Uses profile information to split out cold blocks. 11 // 12 // This pass splits out cold machine basic blocks from the parent function. This 13 // implementation leverages the basic block section framework. Blocks marked 14 // cold by this pass are grouped together in a separate section prefixed with 15 // ".text.unlikely.*". The linker can then group these together as a cold 16 // section. The split part of the function is a contiguous region identified by 17 // the symbol "foo.cold". Grouping all cold blocks across functions together 18 // decreases fragmentation and improves icache and itlb utilization. Note that 19 // the overall changes to the binary size are negligible; only a small number of 20 // additional jump instructions may be introduced. 21 // 22 // For the original RFC of this pass please see 23 // https://groups.google.com/d/msg/llvm-dev/RUegaMg-iqc/wFAVxa6fCgAJ 24 //===----------------------------------------------------------------------===// 25 26 #include "llvm/ADT/SmallVector.h" 27 #include "llvm/Analysis/ProfileSummaryInfo.h" 28 #include "llvm/CodeGen/BasicBlockSectionUtils.h" 29 #include "llvm/CodeGen/MachineBasicBlock.h" 30 #include "llvm/CodeGen/MachineBlockFrequencyInfo.h" 31 #include "llvm/CodeGen/MachineFunction.h" 32 #include "llvm/CodeGen/MachineFunctionPass.h" 33 #include "llvm/CodeGen/MachineModuleInfo.h" 34 #include "llvm/CodeGen/Passes.h" 35 #include "llvm/IR/Function.h" 36 #include "llvm/InitializePasses.h" 37 #include "llvm/Support/CommandLine.h" 38 39 using namespace llvm; 40 41 // FIXME: This cutoff value is CPU dependent and should be moved to 42 // TargetTransformInfo once we consider enabling this on other platforms. 43 // The value is expressed as a ProfileSummaryInfo integer percentile cutoff. 44 // Defaults to 999950, i.e. all blocks colder than 99.995 percentile are split. 45 // The default was empirically determined to be optimal when considering cutoff 46 // values between 99%-ile to 100%-ile with respect to iTLB and icache metrics on 47 // Intel CPUs. 48 static cl::opt<unsigned> 49 PercentileCutoff("mfs-psi-cutoff", 50 cl::desc("Percentile profile summary cutoff used to " 51 "determine cold blocks. Unused if set to zero."), 52 cl::init(999950), cl::Hidden); 53 54 static cl::opt<unsigned> ColdCountThreshold( 55 "mfs-count-threshold", 56 cl::desc( 57 "Minimum number of times a block must be executed to be retained."), 58 cl::init(1), cl::Hidden); 59 60 namespace { 61 62 class MachineFunctionSplitter : public MachineFunctionPass { 63 public: 64 static char ID; 65 MachineFunctionSplitter() : MachineFunctionPass(ID) { 66 initializeMachineFunctionSplitterPass(*PassRegistry::getPassRegistry()); 67 } 68 69 StringRef getPassName() const override { 70 return "Machine Function Splitter Transformation"; 71 } 72 73 void getAnalysisUsage(AnalysisUsage &AU) const override; 74 75 bool runOnMachineFunction(MachineFunction &F) override; 76 }; 77 } // end anonymous namespace 78 79 static bool isColdBlock(const MachineBasicBlock &MBB, 80 const MachineBlockFrequencyInfo *MBFI, 81 ProfileSummaryInfo *PSI) { 82 Optional<uint64_t> Count = MBFI->getBlockProfileCount(&MBB); 83 if (!Count.hasValue()) 84 return true; 85 86 if (PercentileCutoff > 0) { 87 return PSI->isColdCountNthPercentile(PercentileCutoff, *Count); 88 } 89 return (*Count < ColdCountThreshold); 90 } 91 92 bool MachineFunctionSplitter::runOnMachineFunction(MachineFunction &MF) { 93 // TODO: We only target functions with profile data. Static information may 94 // also be considered but we don't see performance improvements yet. 95 if (!MF.getFunction().hasProfileData()) 96 return false; 97 98 // TODO: We don't split functions where a section attribute has been set 99 // since the split part may not be placed in a contiguous region. It may also 100 // be more beneficial to augment the linker to ensure contiguous layout of 101 // split functions within the same section as specified by the attribute. 102 if (MF.getFunction().hasSection() || 103 MF.getFunction().hasFnAttribute("implicit-section-name")) 104 return false; 105 106 // We don't want to proceed further for cold functions 107 // or functions of unknown hotness. Lukewarm functions have no prefix. 108 Optional<StringRef> SectionPrefix = MF.getFunction().getSectionPrefix(); 109 if (SectionPrefix.hasValue() && 110 (SectionPrefix.getValue().equals("unlikely") || 111 SectionPrefix.getValue().equals("unknown"))) { 112 return false; 113 } 114 115 // Renumbering blocks here preserves the order of the blocks as 116 // sortBasicBlocksAndUpdateBranches uses the numeric identifier to sort 117 // blocks. Preserving the order of blocks is essential to retaining decisions 118 // made by prior passes such as MachineBlockPlacement. 119 MF.RenumberBlocks(); 120 MF.setBBSectionsType(BasicBlockSection::Preset); 121 auto *MBFI = &getAnalysis<MachineBlockFrequencyInfo>(); 122 auto *PSI = &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI(); 123 124 SmallVector<MachineBasicBlock *, 2> LandingPads; 125 for (auto &MBB : MF) { 126 if (MBB.isEntryBlock()) 127 continue; 128 129 if (MBB.isEHPad()) 130 LandingPads.push_back(&MBB); 131 else if (isColdBlock(MBB, MBFI, PSI)) 132 MBB.setSectionID(MBBSectionID::ColdSectionID); 133 } 134 135 // We only split out eh pads if all of them are cold. 136 bool HasHotLandingPads = false; 137 for (const MachineBasicBlock *LP : LandingPads) { 138 if (!isColdBlock(*LP, MBFI, PSI)) 139 HasHotLandingPads = true; 140 } 141 if (!HasHotLandingPads) { 142 for (MachineBasicBlock *LP : LandingPads) 143 LP->setSectionID(MBBSectionID::ColdSectionID); 144 } 145 146 auto Comparator = [](const MachineBasicBlock &X, const MachineBasicBlock &Y) { 147 return X.getSectionID().Type < Y.getSectionID().Type; 148 }; 149 llvm::sortBasicBlocksAndUpdateBranches(MF, Comparator); 150 151 return true; 152 } 153 154 void MachineFunctionSplitter::getAnalysisUsage(AnalysisUsage &AU) const { 155 AU.addRequired<MachineModuleInfoWrapperPass>(); 156 AU.addRequired<MachineBlockFrequencyInfo>(); 157 AU.addRequired<ProfileSummaryInfoWrapperPass>(); 158 } 159 160 char MachineFunctionSplitter::ID = 0; 161 INITIALIZE_PASS(MachineFunctionSplitter, "machine-function-splitter", 162 "Split machine functions using profile information", false, 163 false) 164 165 MachineFunctionPass *llvm::createMachineFunctionSplitterPass() { 166 return new MachineFunctionSplitter(); 167 } 168