| 1 | //===- Localizer.cpp ---------------------- Localize some instrs -*- C++ -*-==// |
| 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 | /// \file |
| 9 | /// This file implements the Localizer class. |
| 10 | //===----------------------------------------------------------------------===// |
| 11 | |
| 12 | #include "llvm/CodeGen/GlobalISel/Localizer.h" |
| 13 | #include "llvm/ADT/DenseMap.h" |
| 14 | #include "llvm/ADT/STLExtras.h" |
| 15 | #include "llvm/ADT/SetVector.h" |
| 16 | #include "llvm/Analysis/TargetTransformInfo.h" |
| 17 | #include "llvm/CodeGen/GlobalISel/GenericMachineInstrs.h" |
| 18 | #include "llvm/CodeGen/GlobalISel/Utils.h" |
| 19 | #include "llvm/CodeGen/MachineFunction.h" |
| 20 | #include "llvm/CodeGen/MachineFunctionAnalysisManager.h" |
| 21 | #include "llvm/CodeGen/MachinePassManager.h" |
| 22 | #include "llvm/CodeGen/MachineRegisterInfo.h" |
| 23 | #include "llvm/CodeGen/TargetLowering.h" |
| 24 | #include "llvm/IR/Analysis.h" |
| 25 | #include "llvm/IR/DebugInfoMetadata.h" |
| 26 | #include "llvm/InitializePasses.h" |
| 27 | #include "llvm/Support/Debug.h" |
| 28 | |
| 29 | #define DEBUG_TYPE "localizer" |
| 30 | |
| 31 | using namespace llvm; |
| 32 | |
| 33 | namespace { |
| 34 | |
| 35 | class LocalizerImpl { |
| 36 | /// MRI contains all the register class/bank information that this |
| 37 | /// pass uses and updates. |
| 38 | MachineRegisterInfo *MRI = nullptr; |
| 39 | /// TTI used for getting remat costs for instructions. |
| 40 | TargetTransformInfo *TTI = nullptr; |
| 41 | |
| 42 | /// Check if \p MOUse is used in the same basic block as \p Def. |
| 43 | /// If the use is in the same block, we say it is local. |
| 44 | /// When the use is not local, \p InsertMBB will contain the basic |
| 45 | /// block when to insert \p Def to have a local use. |
| 46 | static bool isLocalUse(MachineOperand &MOUse, const MachineInstr &Def, |
| 47 | MachineBasicBlock *&InsertMBB); |
| 48 | |
| 49 | /// Initialize the field members using \p MF. |
| 50 | void init(MachineFunction &MF, function_ref<TargetTransformInfo *()> GetTTI); |
| 51 | |
| 52 | typedef SmallSetVector<MachineInstr *, 32> LocalizedSetVecT; |
| 53 | |
| 54 | /// If \p Op is a reg operand of a PHI, return the number of total |
| 55 | /// operands in the PHI that are the same as \p Op, including itself. |
| 56 | unsigned getNumPhiUses(MachineOperand &Op) const; |
| 57 | |
| 58 | /// Do inter-block localization from the entry block. |
| 59 | bool localizeInterBlock(MachineFunction &MF, |
| 60 | LocalizedSetVecT &LocalizedInstrs); |
| 61 | |
| 62 | /// Do intra-block localization of already localized instructions. |
| 63 | bool localizeIntraBlock(LocalizedSetVecT &LocalizedInstrs); |
| 64 | |
| 65 | public: |
| 66 | bool runOnMachineFunction(MachineFunction &MF, |
| 67 | function_ref<TargetTransformInfo *()> GetTTI); |
| 68 | }; |
| 69 | |
| 70 | } // namespace |
| 71 | |
| 72 | char LocalizerLegacy::ID = 0; |
| 73 | INITIALIZE_PASS_BEGIN(LocalizerLegacy, DEBUG_TYPE, |
| 74 | "Move/duplicate certain instructions close to their use" , |
| 75 | false, false) |
| 76 | INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass) |
| 77 | INITIALIZE_PASS_END(LocalizerLegacy, DEBUG_TYPE, |
| 78 | "Move/duplicate certain instructions close to their use" , |
| 79 | false, false) |
| 80 | |
| 81 | LocalizerLegacy::LocalizerLegacy() : MachineFunctionPass(ID) {} |
| 82 | |
| 83 | void LocalizerImpl::init(MachineFunction &MF, |
| 84 | function_ref<TargetTransformInfo *()> GetTTI) { |
| 85 | MRI = &MF.getRegInfo(); |
| 86 | TTI = GetTTI(); |
| 87 | } |
| 88 | |
| 89 | void LocalizerLegacy::getAnalysisUsage(AnalysisUsage &AU) const { |
| 90 | AU.addRequired<TargetTransformInfoWrapperPass>(); |
| 91 | AU.setPreservesCFG(); |
| 92 | getSelectionDAGFallbackAnalysisUsage(AU); |
| 93 | MachineFunctionPass::getAnalysisUsage(AU); |
| 94 | } |
| 95 | |
| 96 | bool LocalizerImpl::isLocalUse(MachineOperand &MOUse, const MachineInstr &Def, |
| 97 | MachineBasicBlock *&InsertMBB) { |
| 98 | MachineInstr &MIUse = *MOUse.getParent(); |
| 99 | InsertMBB = MIUse.getParent(); |
| 100 | if (MIUse.isPHI()) |
| 101 | InsertMBB = MIUse.getOperand(i: MOUse.getOperandNo() + 1).getMBB(); |
| 102 | return InsertMBB == Def.getParent(); |
| 103 | } |
| 104 | |
| 105 | unsigned LocalizerImpl::getNumPhiUses(MachineOperand &Op) const { |
| 106 | auto *MI = dyn_cast<GPhi>(Val: &*Op.getParent()); |
| 107 | if (!MI) |
| 108 | return 0; |
| 109 | |
| 110 | Register SrcReg = Op.getReg(); |
| 111 | unsigned NumUses = 0; |
| 112 | for (unsigned I = 0, NumVals = MI->getNumIncomingValues(); I < NumVals; ++I) { |
| 113 | if (MI->getIncomingValue(I) == SrcReg) |
| 114 | ++NumUses; |
| 115 | } |
| 116 | return NumUses; |
| 117 | } |
| 118 | |
| 119 | bool LocalizerImpl::localizeInterBlock(MachineFunction &MF, |
| 120 | LocalizedSetVecT &LocalizedInstrs) { |
| 121 | bool Changed = false; |
| 122 | DenseMap<std::pair<MachineBasicBlock *, Register>, Register> MBBWithLocalDef; |
| 123 | |
| 124 | // Since the IRTranslator only emits constants into the entry block, and the |
| 125 | // rest of the GISel pipeline generally emits constants close to their users, |
| 126 | // we only localize instructions in the entry block here. This might change if |
| 127 | // we start doing CSE across blocks. |
| 128 | auto &MBB = MF.front(); |
| 129 | auto &TL = *MF.getSubtarget().getTargetLowering(); |
| 130 | for (MachineInstr &MI : llvm::reverse(C&: MBB)) { |
| 131 | if (!TL.shouldLocalize(MI, TTI)) |
| 132 | continue; |
| 133 | LLVM_DEBUG(dbgs() << "Should localize: " << MI); |
| 134 | assert(MI.getDesc().getNumDefs() == 1 && |
| 135 | "More than one definition not supported yet" ); |
| 136 | Register Reg = MI.getOperand(i: 0).getReg(); |
| 137 | // Check if all the users of MI are local. We are going to invalidate the |
| 138 | // list of use operands, so we can't use a range iterator. Debug users stay |
| 139 | // with the original definition. |
| 140 | for (MachineOperand &MOUse : |
| 141 | llvm::make_early_inc_range(Range: MRI->use_nodbg_operands(Reg))) { |
| 142 | // Check if the use is already local. |
| 143 | MachineBasicBlock *InsertMBB; |
| 144 | LLVM_DEBUG(MachineInstr &MIUse = *MOUse.getParent(); |
| 145 | dbgs() << "Checking use: " << MIUse |
| 146 | << " #Opd: " << MOUse.getOperandNo() << '\n'); |
| 147 | if (isLocalUse(MOUse, Def: MI, InsertMBB)) { |
| 148 | // Even if we're in the same block, if the block is very large we could |
| 149 | // still have many long live ranges. Try to do intra-block localization |
| 150 | // too. |
| 151 | LocalizedInstrs.insert(X: &MI); |
| 152 | continue; |
| 153 | } |
| 154 | |
| 155 | // PHIs look like a single user but can use the same register in multiple |
| 156 | // edges, causing remat into each predecessor. Allow this to a certain |
| 157 | // extent. |
| 158 | unsigned NumPhiUses = getNumPhiUses(Op&: MOUse); |
| 159 | const unsigned PhiThreshold = 2; // FIXME: Tune this more. |
| 160 | if (NumPhiUses > PhiThreshold) |
| 161 | continue; |
| 162 | |
| 163 | LLVM_DEBUG(dbgs() << "Fixing non-local use\n" ); |
| 164 | Changed = true; |
| 165 | auto MBBAndReg = std::make_pair(x&: InsertMBB, y&: Reg); |
| 166 | auto NewVRegIt = MBBWithLocalDef.find(Val: MBBAndReg); |
| 167 | if (NewVRegIt == MBBWithLocalDef.end()) { |
| 168 | // Create the localized instruction. |
| 169 | MachineInstr *LocalizedMI = MF.CloneMachineInstr(Orig: &MI); |
| 170 | LocalizedInstrs.insert(X: LocalizedMI); |
| 171 | MachineInstr &UseMI = *MOUse.getParent(); |
| 172 | if (MRI->hasOneNonDBGUse(RegNo: Reg) && !UseMI.isPHI()) |
| 173 | InsertMBB->insert(I: UseMI, MI: LocalizedMI); |
| 174 | else |
| 175 | InsertMBB->insert(I: InsertMBB->SkipPHIsAndLabels(I: InsertMBB->begin()), |
| 176 | MI: LocalizedMI); |
| 177 | |
| 178 | // Set a new register for the definition. |
| 179 | Register NewReg = MRI->cloneVirtualRegister(VReg: Reg); |
| 180 | LocalizedMI->getOperand(i: 0).setReg(NewReg); |
| 181 | NewVRegIt = |
| 182 | MBBWithLocalDef.try_emplace(Key: MBBAndReg, Args&: NewReg).first; |
| 183 | LLVM_DEBUG(dbgs() << "Inserted: " << *LocalizedMI); |
| 184 | } |
| 185 | LLVM_DEBUG(dbgs() << "Update use with: " << printReg(NewVRegIt->second) |
| 186 | << '\n'); |
| 187 | // Update the user reg. |
| 188 | MOUse.setReg(NewVRegIt->second); |
| 189 | } |
| 190 | } |
| 191 | return Changed; |
| 192 | } |
| 193 | |
| 194 | bool LocalizerImpl::localizeIntraBlock(LocalizedSetVecT &LocalizedInstrs) { |
| 195 | bool Changed = false; |
| 196 | |
| 197 | // For each already-localized instruction which has multiple users, then we |
| 198 | // scan the block top down from the current position until we hit one of them. |
| 199 | |
| 200 | // FIXME: Consider doing inst duplication if live ranges are very long due to |
| 201 | // many users, but this case may be better served by regalloc improvements. |
| 202 | |
| 203 | for (MachineInstr *MI : LocalizedInstrs) { |
| 204 | Register Reg = MI->getOperand(i: 0).getReg(); |
| 205 | MachineBasicBlock &MBB = *MI->getParent(); |
| 206 | // All of the user MIs of this reg. |
| 207 | SmallPtrSet<MachineInstr *, 32> Users; |
| 208 | for (MachineInstr &UseMI : MRI->use_nodbg_instructions(Reg)) { |
| 209 | if (!UseMI.isPHI()) |
| 210 | Users.insert(Ptr: &UseMI); |
| 211 | } |
| 212 | MachineBasicBlock::iterator II(MI); |
| 213 | // If all the users were PHIs then they're not going to be in our block, we |
| 214 | // may still benefit from sinking, especially since the value might be live |
| 215 | // across a call. |
| 216 | if (Users.empty()) { |
| 217 | // Make sure we don't sink in between two terminator sequences by scanning |
| 218 | // forward, not backward. |
| 219 | II = MBB.getFirstTerminatorForward(); |
| 220 | LLVM_DEBUG(dbgs() << "Only phi users: moving inst to end: " << *MI); |
| 221 | } else { |
| 222 | ++II; |
| 223 | while (II != MBB.end() && !Users.count(Ptr: &*II)) |
| 224 | ++II; |
| 225 | assert(II != MBB.end() && "Didn't find the user in the MBB" ); |
| 226 | LLVM_DEBUG(dbgs() << "Intra-block: moving " << *MI << " before " << *II); |
| 227 | } |
| 228 | |
| 229 | // Leave each DBG_VALUE above MI that uses one of its results behind as |
| 230 | // undef, so that it still ends any earlier location of the variable, and |
| 231 | // put a copy right after MI, like MachineSink does. No copy is needed if a |
| 232 | // later DBG_VALUE of the same variable replaces it before that. |
| 233 | SmallVector<MachineInstr *, 4> DbgUsers; |
| 234 | SmallVector<MachineInstr *, 4> DbgUsersToCopy; |
| 235 | for (MachineInstr &DbgMI : |
| 236 | make_range(x: std::next(x: MachineBasicBlock::iterator(MI)), y: II)) { |
| 237 | if (!DbgMI.isDebugValue()) |
| 238 | continue; |
| 239 | erase_if(C&: DbgUsersToCopy, P: [&](const MachineInstr *Earlier) { |
| 240 | return Earlier->getDebugVariable() == DbgMI.getDebugVariable() && |
| 241 | Earlier->getDebugLoc().getInlinedAt() == |
| 242 | DbgMI.getDebugLoc().getInlinedAt() && |
| 243 | Earlier->getDebugExpression()->fragmentsOverlap( |
| 244 | Other: DbgMI.getDebugExpression()); |
| 245 | }); |
| 246 | if (DbgMI.hasDebugOperandForReg(Reg)) { |
| 247 | DbgUsers.push_back(Elt: &DbgMI); |
| 248 | DbgUsersToCopy.push_back(Elt: &DbgMI); |
| 249 | } |
| 250 | } |
| 251 | |
| 252 | MI->removeFromParent(); |
| 253 | MBB.insert(I: II, MI); |
| 254 | Changed = true; |
| 255 | |
| 256 | for (MachineInstr *DbgMI : DbgUsersToCopy) |
| 257 | MBB.insert(I: II, MI: MBB.getParent()->CloneMachineInstr(Orig: DbgMI)); |
| 258 | for (MachineInstr *DbgMI : DbgUsers) |
| 259 | DbgMI->setDebugValueUndef(); |
| 260 | |
| 261 | // If the instruction (constant) being localized has single user, we can |
| 262 | // propagate debug location from user. |
| 263 | if (Users.size() == 1) { |
| 264 | const auto &DefDL = MI->getDebugLoc(); |
| 265 | const auto &UserDL = (*Users.begin())->getDebugLoc(); |
| 266 | |
| 267 | if ((!DefDL || DefDL.getLine() == 0) && UserDL && UserDL.getLine() != 0) { |
| 268 | MI->setDebugLoc(UserDL); |
| 269 | } |
| 270 | } |
| 271 | } |
| 272 | return Changed; |
| 273 | } |
| 274 | |
| 275 | bool LocalizerImpl::runOnMachineFunction( |
| 276 | MachineFunction &MF, function_ref<TargetTransformInfo *()> GetTTI) { |
| 277 | // If the ISel pipeline failed, do not bother running that pass. |
| 278 | if (MF.getProperties().hasFailedISel()) |
| 279 | return false; |
| 280 | |
| 281 | LLVM_DEBUG(dbgs() << "Localize instructions for: " << MF.getName() << '\n'); |
| 282 | |
| 283 | init(MF, GetTTI); |
| 284 | |
| 285 | // Keep track of the instructions we localized. We'll do a second pass of |
| 286 | // intra-block localization to further reduce live ranges. |
| 287 | LocalizedSetVecT LocalizedInstrs; |
| 288 | |
| 289 | bool Changed = localizeInterBlock(MF, LocalizedInstrs); |
| 290 | Changed |= localizeIntraBlock(LocalizedInstrs); |
| 291 | return Changed; |
| 292 | } |
| 293 | |
| 294 | bool LocalizerLegacy::runOnMachineFunction(MachineFunction &MF) { |
| 295 | LocalizerImpl Impl; |
| 296 | return Impl.runOnMachineFunction(MF, GetTTI: [&]() { |
| 297 | return &getAnalysis<TargetTransformInfoWrapperPass>().getTTI( |
| 298 | F: MF.getFunction()); |
| 299 | }); |
| 300 | } |
| 301 | |
| 302 | PreservedAnalyses LocalizerPass::run(MachineFunction &MF, |
| 303 | MachineFunctionAnalysisManager &MFAM) { |
| 304 | MFPropsModifier<LocalizerPass> _(*this, MF); |
| 305 | LocalizerImpl Impl; |
| 306 | bool Changed = Impl.runOnMachineFunction(MF, GetTTI: [&]() { |
| 307 | Function &F = MF.getFunction(); |
| 308 | FunctionAnalysisManager &FAM = |
| 309 | MFAM.getResult<FunctionAnalysisManagerMachineFunctionProxy>(IR&: MF) |
| 310 | .getManager(); |
| 311 | return &FAM.getResult<TargetIRAnalysis>(IR&: F); |
| 312 | }); |
| 313 | return Changed ? getMachineFunctionPassPreservedAnalyses() |
| 314 | .preserveSet<CFGAnalyses>() |
| 315 | : PreservedAnalyses::all(); |
| 316 | } |
| 317 | |