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/BranchProbabilityInfo.h" 28 #include "llvm/Analysis/EHUtils.h" 29 #include "llvm/Analysis/ProfileSummaryInfo.h" 30 #include "llvm/CodeGen/BasicBlockSectionUtils.h" 31 #include "llvm/CodeGen/MachineBasicBlock.h" 32 #include "llvm/CodeGen/MachineBlockFrequencyInfo.h" 33 #include "llvm/CodeGen/MachineFunction.h" 34 #include "llvm/CodeGen/MachineFunctionPass.h" 35 #include "llvm/CodeGen/MachineModuleInfo.h" 36 #include "llvm/CodeGen/Passes.h" 37 #include "llvm/CodeGen/TargetInstrInfo.h" 38 #include "llvm/IR/Function.h" 39 #include "llvm/InitializePasses.h" 40 #include "llvm/Support/CommandLine.h" 41 #include <optional> 42 43 using namespace llvm; 44 45 // FIXME: This cutoff value is CPU dependent and should be moved to 46 // TargetTransformInfo once we consider enabling this on other platforms. 47 // The value is expressed as a ProfileSummaryInfo integer percentile cutoff. 48 // Defaults to 999950, i.e. all blocks colder than 99.995 percentile are split. 49 // The default was empirically determined to be optimal when considering cutoff 50 // values between 99%-ile to 100%-ile with respect to iTLB and icache metrics on 51 // Intel CPUs. 52 static cl::opt<unsigned> 53 PercentileCutoff("mfs-psi-cutoff", 54 cl::desc("Percentile profile summary cutoff used to " 55 "determine cold blocks. Unused if set to zero."), 56 cl::init(999950), cl::Hidden); 57 58 static cl::opt<unsigned> ColdCountThreshold( 59 "mfs-count-threshold", 60 cl::desc( 61 "Minimum number of times a block must be executed to be retained."), 62 cl::init(1), cl::Hidden); 63 64 static cl::opt<bool> SplitAllEHCode( 65 "mfs-split-ehcode", 66 cl::desc("Splits all EH code and it's descendants by default."), 67 cl::init(false), cl::Hidden); 68 69 namespace { 70 71 class MachineFunctionSplitter : public MachineFunctionPass { 72 public: 73 static char ID; 74 MachineFunctionSplitter() : MachineFunctionPass(ID) { 75 initializeMachineFunctionSplitterPass(*PassRegistry::getPassRegistry()); 76 } 77 78 StringRef getPassName() const override { 79 return "Machine Function Splitter Transformation"; 80 } 81 82 void getAnalysisUsage(AnalysisUsage &AU) const override; 83 84 bool runOnMachineFunction(MachineFunction &F) override; 85 }; 86 } // end anonymous namespace 87 88 /// setDescendantEHBlocksCold - This splits all EH pads and blocks reachable 89 /// only by EH pad as cold. This will help mark EH pads statically cold 90 /// instead of relying on profile data. 91 static void setDescendantEHBlocksCold(MachineFunction &MF) { 92 DenseSet<MachineBasicBlock *> EHBlocks; 93 computeEHOnlyBlocks(MF, EHBlocks); 94 for (auto Block : EHBlocks) { 95 Block->setSectionID(MBBSectionID::ColdSectionID); 96 } 97 } 98 99 static void finishAdjustingBasicBlocksAndLandingPads(MachineFunction &MF) { 100 auto Comparator = [](const MachineBasicBlock &X, const MachineBasicBlock &Y) { 101 return X.getSectionID().Type < Y.getSectionID().Type; 102 }; 103 llvm::sortBasicBlocksAndUpdateBranches(MF, Comparator); 104 llvm::avoidZeroOffsetLandingPad(MF); 105 } 106 107 static bool isColdBlock(const MachineBasicBlock &MBB, 108 const MachineBlockFrequencyInfo *MBFI, 109 ProfileSummaryInfo *PSI) { 110 std::optional<uint64_t> Count = MBFI->getBlockProfileCount(&MBB); 111 // For instrumentation profiles and sample profiles, we use different ways 112 // to judge whether a block is cold and should be split. 113 if (PSI->hasInstrumentationProfile() || PSI->hasCSInstrumentationProfile()) { 114 // If using instrument profile, which is deemed "accurate", no count means 115 // cold. 116 if (!Count) 117 return true; 118 if (PercentileCutoff > 0) 119 return PSI->isColdCountNthPercentile(PercentileCutoff, *Count); 120 // Fallthrough to end of function. 121 } else if (PSI->hasSampleProfile()) { 122 // For sample profile, no count means "do not judege coldness". 123 if (!Count) 124 return false; 125 } 126 127 return (*Count < ColdCountThreshold); 128 } 129 130 bool MachineFunctionSplitter::runOnMachineFunction(MachineFunction &MF) { 131 // We target functions with profile data. Static information in the form 132 // of exception handling code may be split to cold if user passes the 133 // mfs-split-ehcode flag. 134 bool UseProfileData = MF.getFunction().hasProfileData(); 135 if (!UseProfileData && !SplitAllEHCode) 136 return false; 137 138 const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo(); 139 if (!TII.isFunctionSafeToSplit(MF)) 140 return false; 141 142 // Renumbering blocks here preserves the order of the blocks as 143 // sortBasicBlocksAndUpdateBranches uses the numeric identifier to sort 144 // blocks. Preserving the order of blocks is essential to retaining decisions 145 // made by prior passes such as MachineBlockPlacement. 146 MF.RenumberBlocks(); 147 MF.setBBSectionsType(BasicBlockSection::Preset); 148 149 MachineBlockFrequencyInfo *MBFI = nullptr; 150 ProfileSummaryInfo *PSI = nullptr; 151 if (UseProfileData) { 152 MBFI = &getAnalysis<MachineBlockFrequencyInfoWrapperPass>().getMBFI(); 153 PSI = &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI(); 154 // If we don't have a good profile (sample profile is not deemed 155 // as a "good profile") and the function is not hot, then early 156 // return. (Because we can only trust hot functions when profile 157 // quality is not good.) 158 if (PSI->hasSampleProfile() && !PSI->isFunctionHotInCallGraph(&MF, *MBFI)) { 159 // Split all EH code and it's descendant statically by default. 160 if (SplitAllEHCode) 161 setDescendantEHBlocksCold(MF); 162 finishAdjustingBasicBlocksAndLandingPads(MF); 163 return true; 164 } 165 } 166 167 SmallVector<MachineBasicBlock *, 2> LandingPads; 168 for (auto &MBB : MF) { 169 if (MBB.isEntryBlock()) 170 continue; 171 172 if (MBB.isEHPad()) 173 LandingPads.push_back(&MBB); 174 else if (UseProfileData && isColdBlock(MBB, MBFI, PSI) && 175 TII.isMBBSafeToSplitToCold(MBB) && !SplitAllEHCode) 176 MBB.setSectionID(MBBSectionID::ColdSectionID); 177 } 178 179 // Split all EH code and it's descendant statically by default. 180 if (SplitAllEHCode) 181 setDescendantEHBlocksCold(MF); 182 // We only split out eh pads if all of them are cold. 183 else { 184 // Here we have UseProfileData == true. 185 bool HasHotLandingPads = false; 186 for (const MachineBasicBlock *LP : LandingPads) { 187 if (!isColdBlock(*LP, MBFI, PSI) || !TII.isMBBSafeToSplitToCold(*LP)) 188 HasHotLandingPads = true; 189 } 190 if (!HasHotLandingPads) { 191 for (MachineBasicBlock *LP : LandingPads) 192 LP->setSectionID(MBBSectionID::ColdSectionID); 193 } 194 } 195 196 finishAdjustingBasicBlocksAndLandingPads(MF); 197 return true; 198 } 199 200 void MachineFunctionSplitter::getAnalysisUsage(AnalysisUsage &AU) const { 201 AU.addRequired<MachineModuleInfoWrapperPass>(); 202 AU.addRequired<MachineBlockFrequencyInfoWrapperPass>(); 203 AU.addRequired<ProfileSummaryInfoWrapperPass>(); 204 } 205 206 char MachineFunctionSplitter::ID = 0; 207 INITIALIZE_PASS(MachineFunctionSplitter, "machine-function-splitter", 208 "Split machine functions using profile information", false, 209 false) 210 211 MachineFunctionPass *llvm::createMachineFunctionSplitterPass() { 212 return new MachineFunctionSplitter(); 213 } 214