1e8d8bef9SDimitry Andric //===-- MachineFunctionSplitter.cpp - Split machine functions //-----------===// 2e8d8bef9SDimitry Andric // 3e8d8bef9SDimitry Andric // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4e8d8bef9SDimitry Andric // See https://llvm.org/LICENSE.txt for license information. 5e8d8bef9SDimitry Andric // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6e8d8bef9SDimitry Andric // 7e8d8bef9SDimitry Andric //===----------------------------------------------------------------------===// 8e8d8bef9SDimitry Andric // 9e8d8bef9SDimitry Andric // \file 10e8d8bef9SDimitry Andric // Uses profile information to split out cold blocks. 11e8d8bef9SDimitry Andric // 12e8d8bef9SDimitry Andric // This pass splits out cold machine basic blocks from the parent function. This 13e8d8bef9SDimitry Andric // implementation leverages the basic block section framework. Blocks marked 14e8d8bef9SDimitry Andric // cold by this pass are grouped together in a separate section prefixed with 15e8d8bef9SDimitry Andric // ".text.unlikely.*". The linker can then group these together as a cold 16e8d8bef9SDimitry Andric // section. The split part of the function is a contiguous region identified by 17e8d8bef9SDimitry Andric // the symbol "foo.cold". Grouping all cold blocks across functions together 18e8d8bef9SDimitry Andric // decreases fragmentation and improves icache and itlb utilization. Note that 19e8d8bef9SDimitry Andric // the overall changes to the binary size are negligible; only a small number of 20e8d8bef9SDimitry Andric // additional jump instructions may be introduced. 21e8d8bef9SDimitry Andric // 22e8d8bef9SDimitry Andric // For the original RFC of this pass please see 23e8d8bef9SDimitry Andric // https://groups.google.com/d/msg/llvm-dev/RUegaMg-iqc/wFAVxa6fCgAJ 24e8d8bef9SDimitry Andric //===----------------------------------------------------------------------===// 25e8d8bef9SDimitry Andric 26e8d8bef9SDimitry Andric #include "llvm/ADT/Statistic.h" 27e8d8bef9SDimitry Andric #include "llvm/Analysis/ProfileSummaryInfo.h" 28e8d8bef9SDimitry Andric #include "llvm/CodeGen/BasicBlockSectionUtils.h" 29e8d8bef9SDimitry Andric #include "llvm/CodeGen/MachineBasicBlock.h" 30e8d8bef9SDimitry Andric #include "llvm/CodeGen/MachineBlockFrequencyInfo.h" 31e8d8bef9SDimitry Andric #include "llvm/CodeGen/MachineFunction.h" 32e8d8bef9SDimitry Andric #include "llvm/CodeGen/MachineFunctionPass.h" 33e8d8bef9SDimitry Andric #include "llvm/CodeGen/MachineModuleInfo.h" 34e8d8bef9SDimitry Andric #include "llvm/CodeGen/Passes.h" 35e8d8bef9SDimitry Andric #include "llvm/IR/Function.h" 36e8d8bef9SDimitry Andric #include "llvm/IR/Module.h" 37e8d8bef9SDimitry Andric #include "llvm/InitializePasses.h" 38e8d8bef9SDimitry Andric #include "llvm/Support/CommandLine.h" 39e8d8bef9SDimitry Andric 40e8d8bef9SDimitry Andric using namespace llvm; 41e8d8bef9SDimitry Andric 42e8d8bef9SDimitry Andric // FIXME: This cutoff value is CPU dependent and should be moved to 43e8d8bef9SDimitry Andric // TargetTransformInfo once we consider enabling this on other platforms. 44e8d8bef9SDimitry Andric // The value is expressed as a ProfileSummaryInfo integer percentile cutoff. 45e8d8bef9SDimitry Andric // Defaults to 999950, i.e. all blocks colder than 99.995 percentile are split. 46e8d8bef9SDimitry Andric // The default was empirically determined to be optimal when considering cutoff 47e8d8bef9SDimitry Andric // values between 99%-ile to 100%-ile with respect to iTLB and icache metrics on 48e8d8bef9SDimitry Andric // Intel CPUs. 49e8d8bef9SDimitry Andric static cl::opt<unsigned> 50e8d8bef9SDimitry Andric PercentileCutoff("mfs-psi-cutoff", 51e8d8bef9SDimitry Andric cl::desc("Percentile profile summary cutoff used to " 52e8d8bef9SDimitry Andric "determine cold blocks. Unused if set to zero."), 53e8d8bef9SDimitry Andric cl::init(999950), cl::Hidden); 54e8d8bef9SDimitry Andric 55e8d8bef9SDimitry Andric static cl::opt<unsigned> ColdCountThreshold( 56e8d8bef9SDimitry Andric "mfs-count-threshold", 57e8d8bef9SDimitry Andric cl::desc( 58e8d8bef9SDimitry Andric "Minimum number of times a block must be executed to be retained."), 59e8d8bef9SDimitry Andric cl::init(1), cl::Hidden); 60e8d8bef9SDimitry Andric 61e8d8bef9SDimitry Andric namespace { 62e8d8bef9SDimitry Andric 63e8d8bef9SDimitry Andric class MachineFunctionSplitter : public MachineFunctionPass { 64e8d8bef9SDimitry Andric public: 65e8d8bef9SDimitry Andric static char ID; 66e8d8bef9SDimitry Andric MachineFunctionSplitter() : MachineFunctionPass(ID) { 67e8d8bef9SDimitry Andric initializeMachineFunctionSplitterPass(*PassRegistry::getPassRegistry()); 68e8d8bef9SDimitry Andric } 69e8d8bef9SDimitry Andric 70e8d8bef9SDimitry Andric StringRef getPassName() const override { 71e8d8bef9SDimitry Andric return "Machine Function Splitter Transformation"; 72e8d8bef9SDimitry Andric } 73e8d8bef9SDimitry Andric 74e8d8bef9SDimitry Andric void getAnalysisUsage(AnalysisUsage &AU) const override; 75e8d8bef9SDimitry Andric 76e8d8bef9SDimitry Andric bool runOnMachineFunction(MachineFunction &F) override; 77e8d8bef9SDimitry Andric }; 78e8d8bef9SDimitry Andric } // end anonymous namespace 79e8d8bef9SDimitry Andric 80e8d8bef9SDimitry Andric static bool isColdBlock(MachineBasicBlock &MBB, 81e8d8bef9SDimitry Andric const MachineBlockFrequencyInfo *MBFI, 82e8d8bef9SDimitry Andric ProfileSummaryInfo *PSI) { 83e8d8bef9SDimitry Andric Optional<uint64_t> Count = MBFI->getBlockProfileCount(&MBB); 84e8d8bef9SDimitry Andric if (!Count.hasValue()) 85e8d8bef9SDimitry Andric return true; 86e8d8bef9SDimitry Andric 87e8d8bef9SDimitry Andric if (PercentileCutoff > 0) { 88e8d8bef9SDimitry Andric return PSI->isColdCountNthPercentile(PercentileCutoff, *Count); 89e8d8bef9SDimitry Andric } 90e8d8bef9SDimitry Andric return (*Count < ColdCountThreshold); 91e8d8bef9SDimitry Andric } 92e8d8bef9SDimitry Andric 93e8d8bef9SDimitry Andric bool MachineFunctionSplitter::runOnMachineFunction(MachineFunction &MF) { 94e8d8bef9SDimitry Andric // TODO: We only target functions with profile data. Static information may 95e8d8bef9SDimitry Andric // also be considered but we don't see performance improvements yet. 96e8d8bef9SDimitry Andric if (!MF.getFunction().hasProfileData()) 97e8d8bef9SDimitry Andric return false; 98e8d8bef9SDimitry Andric 99e8d8bef9SDimitry Andric // TODO: We don't split functions where a section attribute has been set 100e8d8bef9SDimitry Andric // since the split part may not be placed in a contiguous region. It may also 101e8d8bef9SDimitry Andric // be more beneficial to augment the linker to ensure contiguous layout of 102e8d8bef9SDimitry Andric // split functions within the same section as specified by the attribute. 103e8d8bef9SDimitry Andric if (!MF.getFunction().getSection().empty()) 104e8d8bef9SDimitry Andric return false; 105e8d8bef9SDimitry Andric 106e8d8bef9SDimitry Andric // We don't want to proceed further for cold functions 107e8d8bef9SDimitry Andric // or functions of unknown hotness. Lukewarm functions have no prefix. 108e8d8bef9SDimitry Andric Optional<StringRef> SectionPrefix = MF.getFunction().getSectionPrefix(); 109e8d8bef9SDimitry Andric if (SectionPrefix.hasValue() && 110e8d8bef9SDimitry Andric (SectionPrefix.getValue().equals("unlikely") || 111e8d8bef9SDimitry Andric SectionPrefix.getValue().equals("unknown"))) { 112e8d8bef9SDimitry Andric return false; 113e8d8bef9SDimitry Andric } 114e8d8bef9SDimitry Andric 115e8d8bef9SDimitry Andric // Renumbering blocks here preserves the order of the blocks as 116e8d8bef9SDimitry Andric // sortBasicBlocksAndUpdateBranches uses the numeric identifier to sort 117e8d8bef9SDimitry Andric // blocks. Preserving the order of blocks is essential to retaining decisions 118e8d8bef9SDimitry Andric // made by prior passes such as MachineBlockPlacement. 119e8d8bef9SDimitry Andric MF.RenumberBlocks(); 120e8d8bef9SDimitry Andric MF.setBBSectionsType(BasicBlockSection::Preset); 121e8d8bef9SDimitry Andric auto *MBFI = &getAnalysis<MachineBlockFrequencyInfo>(); 122e8d8bef9SDimitry Andric auto *PSI = &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI(); 123e8d8bef9SDimitry Andric 124e8d8bef9SDimitry Andric for (auto &MBB : MF) { 125e8d8bef9SDimitry Andric // FIXME: We retain the entry block and conservatively keep all landing pad 126e8d8bef9SDimitry Andric // blocks as part of the original function. Once D73739 is submitted, we can 127e8d8bef9SDimitry Andric // improve the handling of ehpads. 128e8d8bef9SDimitry Andric if ((MBB.pred_empty() || MBB.isEHPad())) 129e8d8bef9SDimitry Andric continue; 130e8d8bef9SDimitry Andric if (isColdBlock(MBB, MBFI, PSI)) 131e8d8bef9SDimitry Andric MBB.setSectionID(MBBSectionID::ColdSectionID); 132e8d8bef9SDimitry Andric } 133e8d8bef9SDimitry Andric 134e8d8bef9SDimitry Andric auto Comparator = [](const MachineBasicBlock &X, const MachineBasicBlock &Y) { 135e8d8bef9SDimitry Andric return X.getSectionID().Type < Y.getSectionID().Type; 136e8d8bef9SDimitry Andric }; 137e8d8bef9SDimitry Andric llvm::sortBasicBlocksAndUpdateBranches(MF, Comparator); 138e8d8bef9SDimitry Andric 139e8d8bef9SDimitry Andric return true; 140e8d8bef9SDimitry Andric } 141e8d8bef9SDimitry Andric 142e8d8bef9SDimitry Andric void MachineFunctionSplitter::getAnalysisUsage(AnalysisUsage &AU) const { 143e8d8bef9SDimitry Andric AU.addRequired<MachineModuleInfoWrapperPass>(); 144e8d8bef9SDimitry Andric AU.addRequired<MachineBlockFrequencyInfo>(); 145e8d8bef9SDimitry Andric AU.addRequired<ProfileSummaryInfoWrapperPass>(); 146e8d8bef9SDimitry Andric } 147e8d8bef9SDimitry Andric 148e8d8bef9SDimitry Andric char MachineFunctionSplitter::ID = 0; 149e8d8bef9SDimitry Andric INITIALIZE_PASS(MachineFunctionSplitter, "machine-function-splitter", 150e8d8bef9SDimitry Andric "Split machine functions using profile information", false, 151e8d8bef9SDimitry Andric false) 152e8d8bef9SDimitry Andric 153e8d8bef9SDimitry Andric MachineFunctionPass *llvm::createMachineFunctionSplitterPass() { 154e8d8bef9SDimitry Andric return new MachineFunctionSplitter(); 155e8d8bef9SDimitry Andric } 156