1 //===-- GCNSchedStrategy.cpp - GCN Scheduler Strategy ---------------------===// 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 /// This contains a MachineSchedStrategy implementation for maximizing wave 11 /// occupancy on GCN hardware. 12 //===----------------------------------------------------------------------===// 13 14 #include "GCNSchedStrategy.h" 15 #include "AMDGPUSubtarget.h" 16 #include "SIInstrInfo.h" 17 #include "SIMachineFunctionInfo.h" 18 #include "SIRegisterInfo.h" 19 #include "Utils/AMDGPUBaseInfo.h" 20 #include "llvm/CodeGen/RegisterClassInfo.h" 21 #include "llvm/Support/MathExtras.h" 22 23 #define DEBUG_TYPE "machine-scheduler" 24 25 using namespace llvm; 26 27 GCNMaxOccupancySchedStrategy::GCNMaxOccupancySchedStrategy( 28 const MachineSchedContext *C) : 29 GenericScheduler(C), TargetOccupancy(0), MF(nullptr) { } 30 31 void GCNMaxOccupancySchedStrategy::initialize(ScheduleDAGMI *DAG) { 32 GenericScheduler::initialize(DAG); 33 34 const SIRegisterInfo *SRI = static_cast<const SIRegisterInfo*>(TRI); 35 36 MF = &DAG->MF; 37 38 const GCNSubtarget &ST = MF->getSubtarget<GCNSubtarget>(); 39 40 // FIXME: This is also necessary, because some passes that run after 41 // scheduling and before regalloc increase register pressure. 42 const int ErrorMargin = 3; 43 44 SGPRExcessLimit = Context->RegClassInfo 45 ->getNumAllocatableRegs(&AMDGPU::SGPR_32RegClass) - ErrorMargin; 46 VGPRExcessLimit = Context->RegClassInfo 47 ->getNumAllocatableRegs(&AMDGPU::VGPR_32RegClass) - ErrorMargin; 48 if (TargetOccupancy) { 49 SGPRCriticalLimit = ST.getMaxNumSGPRs(TargetOccupancy, true); 50 VGPRCriticalLimit = ST.getMaxNumVGPRs(TargetOccupancy); 51 } else { 52 SGPRCriticalLimit = SRI->getRegPressureSetLimit(DAG->MF, 53 SRI->getSGPRPressureSet()); 54 VGPRCriticalLimit = SRI->getRegPressureSetLimit(DAG->MF, 55 SRI->getVGPRPressureSet()); 56 } 57 58 SGPRCriticalLimit -= ErrorMargin; 59 VGPRCriticalLimit -= ErrorMargin; 60 } 61 62 void GCNMaxOccupancySchedStrategy::initCandidate(SchedCandidate &Cand, SUnit *SU, 63 bool AtTop, const RegPressureTracker &RPTracker, 64 const SIRegisterInfo *SRI, 65 unsigned SGPRPressure, 66 unsigned VGPRPressure) { 67 68 Cand.SU = SU; 69 Cand.AtTop = AtTop; 70 71 // getDownwardPressure() and getUpwardPressure() make temporary changes to 72 // the tracker, so we need to pass those function a non-const copy. 73 RegPressureTracker &TempTracker = const_cast<RegPressureTracker&>(RPTracker); 74 75 Pressure.clear(); 76 MaxPressure.clear(); 77 78 if (AtTop) 79 TempTracker.getDownwardPressure(SU->getInstr(), Pressure, MaxPressure); 80 else { 81 // FIXME: I think for bottom up scheduling, the register pressure is cached 82 // and can be retrieved by DAG->getPressureDif(SU). 83 TempTracker.getUpwardPressure(SU->getInstr(), Pressure, MaxPressure); 84 } 85 86 unsigned NewSGPRPressure = Pressure[SRI->getSGPRPressureSet()]; 87 unsigned NewVGPRPressure = Pressure[SRI->getVGPRPressureSet()]; 88 89 // If two instructions increase the pressure of different register sets 90 // by the same amount, the generic scheduler will prefer to schedule the 91 // instruction that increases the set with the least amount of registers, 92 // which in our case would be SGPRs. This is rarely what we want, so 93 // when we report excess/critical register pressure, we do it either 94 // only for VGPRs or only for SGPRs. 95 96 // FIXME: Better heuristics to determine whether to prefer SGPRs or VGPRs. 97 const unsigned MaxVGPRPressureInc = 16; 98 bool ShouldTrackVGPRs = VGPRPressure + MaxVGPRPressureInc >= VGPRExcessLimit; 99 bool ShouldTrackSGPRs = !ShouldTrackVGPRs && SGPRPressure >= SGPRExcessLimit; 100 101 102 // FIXME: We have to enter REG-EXCESS before we reach the actual threshold 103 // to increase the likelihood we don't go over the limits. We should improve 104 // the analysis to look through dependencies to find the path with the least 105 // register pressure. 106 107 // We only need to update the RPDelta for instructions that increase register 108 // pressure. Instructions that decrease or keep reg pressure the same will be 109 // marked as RegExcess in tryCandidate() when they are compared with 110 // instructions that increase the register pressure. 111 if (ShouldTrackVGPRs && NewVGPRPressure >= VGPRExcessLimit) { 112 Cand.RPDelta.Excess = PressureChange(SRI->getVGPRPressureSet()); 113 Cand.RPDelta.Excess.setUnitInc(NewVGPRPressure - VGPRExcessLimit); 114 } 115 116 if (ShouldTrackSGPRs && NewSGPRPressure >= SGPRExcessLimit) { 117 Cand.RPDelta.Excess = PressureChange(SRI->getSGPRPressureSet()); 118 Cand.RPDelta.Excess.setUnitInc(NewSGPRPressure - SGPRExcessLimit); 119 } 120 121 // Register pressure is considered 'CRITICAL' if it is approaching a value 122 // that would reduce the wave occupancy for the execution unit. When 123 // register pressure is 'CRITICAL', increading SGPR and VGPR pressure both 124 // has the same cost, so we don't need to prefer one over the other. 125 126 int SGPRDelta = NewSGPRPressure - SGPRCriticalLimit; 127 int VGPRDelta = NewVGPRPressure - VGPRCriticalLimit; 128 129 if (SGPRDelta >= 0 || VGPRDelta >= 0) { 130 if (SGPRDelta > VGPRDelta) { 131 Cand.RPDelta.CriticalMax = PressureChange(SRI->getSGPRPressureSet()); 132 Cand.RPDelta.CriticalMax.setUnitInc(SGPRDelta); 133 } else { 134 Cand.RPDelta.CriticalMax = PressureChange(SRI->getVGPRPressureSet()); 135 Cand.RPDelta.CriticalMax.setUnitInc(VGPRDelta); 136 } 137 } 138 } 139 140 // This function is mostly cut and pasted from 141 // GenericScheduler::pickNodeFromQueue() 142 void GCNMaxOccupancySchedStrategy::pickNodeFromQueue(SchedBoundary &Zone, 143 const CandPolicy &ZonePolicy, 144 const RegPressureTracker &RPTracker, 145 SchedCandidate &Cand) { 146 const SIRegisterInfo *SRI = static_cast<const SIRegisterInfo*>(TRI); 147 ArrayRef<unsigned> Pressure = RPTracker.getRegSetPressureAtPos(); 148 unsigned SGPRPressure = Pressure[SRI->getSGPRPressureSet()]; 149 unsigned VGPRPressure = Pressure[SRI->getVGPRPressureSet()]; 150 ReadyQueue &Q = Zone.Available; 151 for (SUnit *SU : Q) { 152 153 SchedCandidate TryCand(ZonePolicy); 154 initCandidate(TryCand, SU, Zone.isTop(), RPTracker, SRI, 155 SGPRPressure, VGPRPressure); 156 // Pass SchedBoundary only when comparing nodes from the same boundary. 157 SchedBoundary *ZoneArg = Cand.AtTop == TryCand.AtTop ? &Zone : nullptr; 158 GenericScheduler::tryCandidate(Cand, TryCand, ZoneArg); 159 if (TryCand.Reason != NoCand) { 160 // Initialize resource delta if needed in case future heuristics query it. 161 if (TryCand.ResDelta == SchedResourceDelta()) 162 TryCand.initResourceDelta(Zone.DAG, SchedModel); 163 Cand.setBest(TryCand); 164 LLVM_DEBUG(traceCandidate(Cand)); 165 } 166 } 167 } 168 169 // This function is mostly cut and pasted from 170 // GenericScheduler::pickNodeBidirectional() 171 SUnit *GCNMaxOccupancySchedStrategy::pickNodeBidirectional(bool &IsTopNode) { 172 // Schedule as far as possible in the direction of no choice. This is most 173 // efficient, but also provides the best heuristics for CriticalPSets. 174 if (SUnit *SU = Bot.pickOnlyChoice()) { 175 IsTopNode = false; 176 return SU; 177 } 178 if (SUnit *SU = Top.pickOnlyChoice()) { 179 IsTopNode = true; 180 return SU; 181 } 182 // Set the bottom-up policy based on the state of the current bottom zone and 183 // the instructions outside the zone, including the top zone. 184 CandPolicy BotPolicy; 185 setPolicy(BotPolicy, /*IsPostRA=*/false, Bot, &Top); 186 // Set the top-down policy based on the state of the current top zone and 187 // the instructions outside the zone, including the bottom zone. 188 CandPolicy TopPolicy; 189 setPolicy(TopPolicy, /*IsPostRA=*/false, Top, &Bot); 190 191 // See if BotCand is still valid (because we previously scheduled from Top). 192 LLVM_DEBUG(dbgs() << "Picking from Bot:\n"); 193 if (!BotCand.isValid() || BotCand.SU->isScheduled || 194 BotCand.Policy != BotPolicy) { 195 BotCand.reset(CandPolicy()); 196 pickNodeFromQueue(Bot, BotPolicy, DAG->getBotRPTracker(), BotCand); 197 assert(BotCand.Reason != NoCand && "failed to find the first candidate"); 198 } else { 199 LLVM_DEBUG(traceCandidate(BotCand)); 200 #ifndef NDEBUG 201 if (VerifyScheduling) { 202 SchedCandidate TCand; 203 TCand.reset(CandPolicy()); 204 pickNodeFromQueue(Bot, BotPolicy, DAG->getBotRPTracker(), TCand); 205 assert(TCand.SU == BotCand.SU && 206 "Last pick result should correspond to re-picking right now"); 207 } 208 #endif 209 } 210 211 // Check if the top Q has a better candidate. 212 LLVM_DEBUG(dbgs() << "Picking from Top:\n"); 213 if (!TopCand.isValid() || TopCand.SU->isScheduled || 214 TopCand.Policy != TopPolicy) { 215 TopCand.reset(CandPolicy()); 216 pickNodeFromQueue(Top, TopPolicy, DAG->getTopRPTracker(), TopCand); 217 assert(TopCand.Reason != NoCand && "failed to find the first candidate"); 218 } else { 219 LLVM_DEBUG(traceCandidate(TopCand)); 220 #ifndef NDEBUG 221 if (VerifyScheduling) { 222 SchedCandidate TCand; 223 TCand.reset(CandPolicy()); 224 pickNodeFromQueue(Top, TopPolicy, DAG->getTopRPTracker(), TCand); 225 assert(TCand.SU == TopCand.SU && 226 "Last pick result should correspond to re-picking right now"); 227 } 228 #endif 229 } 230 231 // Pick best from BotCand and TopCand. 232 LLVM_DEBUG(dbgs() << "Top Cand: "; traceCandidate(TopCand); 233 dbgs() << "Bot Cand: "; traceCandidate(BotCand);); 234 SchedCandidate Cand; 235 if (TopCand.Reason == BotCand.Reason) { 236 Cand = BotCand; 237 GenericSchedulerBase::CandReason TopReason = TopCand.Reason; 238 TopCand.Reason = NoCand; 239 GenericScheduler::tryCandidate(Cand, TopCand, nullptr); 240 if (TopCand.Reason != NoCand) { 241 Cand.setBest(TopCand); 242 } else { 243 TopCand.Reason = TopReason; 244 } 245 } else { 246 if (TopCand.Reason == RegExcess && TopCand.RPDelta.Excess.getUnitInc() <= 0) { 247 Cand = TopCand; 248 } else if (BotCand.Reason == RegExcess && BotCand.RPDelta.Excess.getUnitInc() <= 0) { 249 Cand = BotCand; 250 } else if (TopCand.Reason == RegCritical && TopCand.RPDelta.CriticalMax.getUnitInc() <= 0) { 251 Cand = TopCand; 252 } else if (BotCand.Reason == RegCritical && BotCand.RPDelta.CriticalMax.getUnitInc() <= 0) { 253 Cand = BotCand; 254 } else { 255 if (BotCand.Reason > TopCand.Reason) { 256 Cand = TopCand; 257 } else { 258 Cand = BotCand; 259 } 260 } 261 } 262 LLVM_DEBUG(dbgs() << "Picking: "; traceCandidate(Cand);); 263 264 IsTopNode = Cand.AtTop; 265 return Cand.SU; 266 } 267 268 // This function is mostly cut and pasted from 269 // GenericScheduler::pickNode() 270 SUnit *GCNMaxOccupancySchedStrategy::pickNode(bool &IsTopNode) { 271 if (DAG->top() == DAG->bottom()) { 272 assert(Top.Available.empty() && Top.Pending.empty() && 273 Bot.Available.empty() && Bot.Pending.empty() && "ReadyQ garbage"); 274 return nullptr; 275 } 276 SUnit *SU; 277 do { 278 if (RegionPolicy.OnlyTopDown) { 279 SU = Top.pickOnlyChoice(); 280 if (!SU) { 281 CandPolicy NoPolicy; 282 TopCand.reset(NoPolicy); 283 pickNodeFromQueue(Top, NoPolicy, DAG->getTopRPTracker(), TopCand); 284 assert(TopCand.Reason != NoCand && "failed to find a candidate"); 285 SU = TopCand.SU; 286 } 287 IsTopNode = true; 288 } else if (RegionPolicy.OnlyBottomUp) { 289 SU = Bot.pickOnlyChoice(); 290 if (!SU) { 291 CandPolicy NoPolicy; 292 BotCand.reset(NoPolicy); 293 pickNodeFromQueue(Bot, NoPolicy, DAG->getBotRPTracker(), BotCand); 294 assert(BotCand.Reason != NoCand && "failed to find a candidate"); 295 SU = BotCand.SU; 296 } 297 IsTopNode = false; 298 } else { 299 SU = pickNodeBidirectional(IsTopNode); 300 } 301 } while (SU->isScheduled); 302 303 if (SU->isTopReady()) 304 Top.removeReady(SU); 305 if (SU->isBottomReady()) 306 Bot.removeReady(SU); 307 308 LLVM_DEBUG(dbgs() << "Scheduling SU(" << SU->NodeNum << ") " 309 << *SU->getInstr()); 310 return SU; 311 } 312 313 GCNScheduleDAGMILive::GCNScheduleDAGMILive(MachineSchedContext *C, 314 std::unique_ptr<MachineSchedStrategy> S) : 315 ScheduleDAGMILive(C, std::move(S)), 316 ST(MF.getSubtarget<GCNSubtarget>()), 317 MFI(*MF.getInfo<SIMachineFunctionInfo>()), 318 StartingOccupancy(MFI.getOccupancy()), 319 MinOccupancy(StartingOccupancy), Stage(0), RegionIdx(0) { 320 321 LLVM_DEBUG(dbgs() << "Starting occupancy is " << StartingOccupancy << ".\n"); 322 } 323 324 void GCNScheduleDAGMILive::schedule() { 325 if (Stage == 0) { 326 // Just record regions at the first pass. 327 Regions.push_back(std::make_pair(RegionBegin, RegionEnd)); 328 return; 329 } 330 331 std::vector<MachineInstr*> Unsched; 332 Unsched.reserve(NumRegionInstrs); 333 for (auto &I : *this) { 334 Unsched.push_back(&I); 335 } 336 337 GCNRegPressure PressureBefore; 338 if (LIS) { 339 PressureBefore = Pressure[RegionIdx]; 340 341 LLVM_DEBUG(dbgs() << "Pressure before scheduling:\nRegion live-ins:"; 342 GCNRPTracker::printLiveRegs(dbgs(), LiveIns[RegionIdx], MRI); 343 dbgs() << "Region live-in pressure: "; 344 llvm::getRegPressure(MRI, LiveIns[RegionIdx]).print(dbgs()); 345 dbgs() << "Region register pressure: "; 346 PressureBefore.print(dbgs())); 347 } 348 349 ScheduleDAGMILive::schedule(); 350 Regions[RegionIdx] = std::make_pair(RegionBegin, RegionEnd); 351 352 if (!LIS) 353 return; 354 355 // Check the results of scheduling. 356 GCNMaxOccupancySchedStrategy &S = (GCNMaxOccupancySchedStrategy&)*SchedImpl; 357 auto PressureAfter = getRealRegPressure(); 358 359 LLVM_DEBUG(dbgs() << "Pressure after scheduling: "; 360 PressureAfter.print(dbgs())); 361 362 if (PressureAfter.getSGPRNum() <= S.SGPRCriticalLimit && 363 PressureAfter.getVGPRNum() <= S.VGPRCriticalLimit) { 364 Pressure[RegionIdx] = PressureAfter; 365 LLVM_DEBUG(dbgs() << "Pressure in desired limits, done.\n"); 366 return; 367 } 368 unsigned Occ = MFI.getOccupancy(); 369 unsigned WavesAfter = std::min(Occ, PressureAfter.getOccupancy(ST)); 370 unsigned WavesBefore = std::min(Occ, PressureBefore.getOccupancy(ST)); 371 LLVM_DEBUG(dbgs() << "Occupancy before scheduling: " << WavesBefore 372 << ", after " << WavesAfter << ".\n"); 373 374 // We could not keep current target occupancy because of the just scheduled 375 // region. Record new occupancy for next scheduling cycle. 376 unsigned NewOccupancy = std::max(WavesAfter, WavesBefore); 377 // Allow memory bound functions to drop to 4 waves if not limited by an 378 // attribute. 379 if (WavesAfter < WavesBefore && WavesAfter < MinOccupancy && 380 WavesAfter >= MFI.getMinAllowedOccupancy()) { 381 LLVM_DEBUG(dbgs() << "Function is memory bound, allow occupancy drop up to " 382 << MFI.getMinAllowedOccupancy() << " waves\n"); 383 NewOccupancy = WavesAfter; 384 } 385 if (NewOccupancy < MinOccupancy) { 386 MinOccupancy = NewOccupancy; 387 MFI.limitOccupancy(MinOccupancy); 388 LLVM_DEBUG(dbgs() << "Occupancy lowered for the function to " 389 << MinOccupancy << ".\n"); 390 } 391 392 if (WavesAfter >= MinOccupancy) { 393 unsigned TotalVGPRs = AMDGPU::IsaInfo::getAddressableNumVGPRs(&ST); 394 unsigned TotalSGPRs = AMDGPU::IsaInfo::getAddressableNumSGPRs(&ST); 395 if (WavesAfter > MFI.getMinWavesPerEU() || 396 PressureAfter.less(ST, PressureBefore) || 397 (TotalVGPRs >= PressureAfter.getVGPRNum() && 398 TotalSGPRs >= PressureAfter.getSGPRNum())) { 399 Pressure[RegionIdx] = PressureAfter; 400 return; 401 } 402 LLVM_DEBUG(dbgs() << "New pressure will result in more spilling.\n"); 403 } 404 405 LLVM_DEBUG(dbgs() << "Attempting to revert scheduling.\n"); 406 RegionEnd = RegionBegin; 407 for (MachineInstr *MI : Unsched) { 408 if (MI->isDebugInstr()) 409 continue; 410 411 if (MI->getIterator() != RegionEnd) { 412 BB->remove(MI); 413 BB->insert(RegionEnd, MI); 414 if (!MI->isDebugInstr()) 415 LIS->handleMove(*MI, true); 416 } 417 // Reset read-undef flags and update them later. 418 for (auto &Op : MI->operands()) 419 if (Op.isReg() && Op.isDef()) 420 Op.setIsUndef(false); 421 RegisterOperands RegOpers; 422 RegOpers.collect(*MI, *TRI, MRI, ShouldTrackLaneMasks, false); 423 if (!MI->isDebugInstr()) { 424 if (ShouldTrackLaneMasks) { 425 // Adjust liveness and add missing dead+read-undef flags. 426 SlotIndex SlotIdx = LIS->getInstructionIndex(*MI).getRegSlot(); 427 RegOpers.adjustLaneLiveness(*LIS, MRI, SlotIdx, MI); 428 } else { 429 // Adjust for missing dead-def flags. 430 RegOpers.detectDeadDefs(*MI, *LIS); 431 } 432 } 433 RegionEnd = MI->getIterator(); 434 ++RegionEnd; 435 LLVM_DEBUG(dbgs() << "Scheduling " << *MI); 436 } 437 RegionBegin = Unsched.front()->getIterator(); 438 Regions[RegionIdx] = std::make_pair(RegionBegin, RegionEnd); 439 440 placeDebugValues(); 441 } 442 443 GCNRegPressure GCNScheduleDAGMILive::getRealRegPressure() const { 444 GCNDownwardRPTracker RPTracker(*LIS); 445 RPTracker.advance(begin(), end(), &LiveIns[RegionIdx]); 446 return RPTracker.moveMaxPressure(); 447 } 448 449 void GCNScheduleDAGMILive::computeBlockPressure(const MachineBasicBlock *MBB) { 450 GCNDownwardRPTracker RPTracker(*LIS); 451 452 // If the block has the only successor then live-ins of that successor are 453 // live-outs of the current block. We can reuse calculated live set if the 454 // successor will be sent to scheduling past current block. 455 const MachineBasicBlock *OnlySucc = nullptr; 456 if (MBB->succ_size() == 1 && !(*MBB->succ_begin())->empty()) { 457 SlotIndexes *Ind = LIS->getSlotIndexes(); 458 if (Ind->getMBBStartIdx(MBB) < Ind->getMBBStartIdx(*MBB->succ_begin())) 459 OnlySucc = *MBB->succ_begin(); 460 } 461 462 // Scheduler sends regions from the end of the block upwards. 463 size_t CurRegion = RegionIdx; 464 for (size_t E = Regions.size(); CurRegion != E; ++CurRegion) 465 if (Regions[CurRegion].first->getParent() != MBB) 466 break; 467 --CurRegion; 468 469 auto I = MBB->begin(); 470 auto LiveInIt = MBBLiveIns.find(MBB); 471 if (LiveInIt != MBBLiveIns.end()) { 472 auto LiveIn = std::move(LiveInIt->second); 473 RPTracker.reset(*MBB->begin(), &LiveIn); 474 MBBLiveIns.erase(LiveInIt); 475 } else { 476 auto &Rgn = Regions[CurRegion]; 477 I = Rgn.first; 478 auto *NonDbgMI = &*skipDebugInstructionsForward(Rgn.first, Rgn.second); 479 auto LRS = BBLiveInMap.lookup(NonDbgMI); 480 assert(isEqual(getLiveRegsBefore(*NonDbgMI, *LIS), LRS)); 481 RPTracker.reset(*I, &LRS); 482 } 483 484 for ( ; ; ) { 485 I = RPTracker.getNext(); 486 487 if (Regions[CurRegion].first == I) { 488 LiveIns[CurRegion] = RPTracker.getLiveRegs(); 489 RPTracker.clearMaxPressure(); 490 } 491 492 if (Regions[CurRegion].second == I) { 493 Pressure[CurRegion] = RPTracker.moveMaxPressure(); 494 if (CurRegion-- == RegionIdx) 495 break; 496 } 497 RPTracker.advanceToNext(); 498 RPTracker.advanceBeforeNext(); 499 } 500 501 if (OnlySucc) { 502 if (I != MBB->end()) { 503 RPTracker.advanceToNext(); 504 RPTracker.advance(MBB->end()); 505 } 506 RPTracker.reset(*OnlySucc->begin(), &RPTracker.getLiveRegs()); 507 RPTracker.advanceBeforeNext(); 508 MBBLiveIns[OnlySucc] = RPTracker.moveLiveRegs(); 509 } 510 } 511 512 DenseMap<MachineInstr *, GCNRPTracker::LiveRegSet> 513 GCNScheduleDAGMILive::getBBLiveInMap() const { 514 assert(!Regions.empty()); 515 std::vector<MachineInstr *> BBStarters; 516 BBStarters.reserve(Regions.size()); 517 auto I = Regions.rbegin(), E = Regions.rend(); 518 auto *BB = I->first->getParent(); 519 do { 520 auto *MI = &*skipDebugInstructionsForward(I->first, I->second); 521 BBStarters.push_back(MI); 522 do { 523 ++I; 524 } while (I != E && I->first->getParent() == BB); 525 } while (I != E); 526 return getLiveRegMap(BBStarters, false /*After*/, *LIS); 527 } 528 529 void GCNScheduleDAGMILive::finalizeSchedule() { 530 GCNMaxOccupancySchedStrategy &S = (GCNMaxOccupancySchedStrategy&)*SchedImpl; 531 LLVM_DEBUG(dbgs() << "All regions recorded, starting actual scheduling.\n"); 532 533 LiveIns.resize(Regions.size()); 534 Pressure.resize(Regions.size()); 535 536 if (!Regions.empty()) 537 BBLiveInMap = getBBLiveInMap(); 538 539 do { 540 Stage++; 541 RegionIdx = 0; 542 MachineBasicBlock *MBB = nullptr; 543 544 if (Stage > 1) { 545 // Retry function scheduling if we found resulting occupancy and it is 546 // lower than used for first pass scheduling. This will give more freedom 547 // to schedule low register pressure blocks. 548 // Code is partially copied from MachineSchedulerBase::scheduleRegions(). 549 550 if (!LIS || StartingOccupancy <= MinOccupancy) 551 break; 552 553 LLVM_DEBUG( 554 dbgs() 555 << "Retrying function scheduling with lowest recorded occupancy " 556 << MinOccupancy << ".\n"); 557 558 S.setTargetOccupancy(MinOccupancy); 559 } 560 561 for (auto Region : Regions) { 562 RegionBegin = Region.first; 563 RegionEnd = Region.second; 564 565 if (RegionBegin->getParent() != MBB) { 566 if (MBB) finishBlock(); 567 MBB = RegionBegin->getParent(); 568 startBlock(MBB); 569 if (Stage == 1) 570 computeBlockPressure(MBB); 571 } 572 573 unsigned NumRegionInstrs = std::distance(begin(), end()); 574 enterRegion(MBB, begin(), end(), NumRegionInstrs); 575 576 // Skip empty scheduling regions (0 or 1 schedulable instructions). 577 if (begin() == end() || begin() == std::prev(end())) { 578 exitRegion(); 579 continue; 580 } 581 582 LLVM_DEBUG(dbgs() << "********** MI Scheduling **********\n"); 583 LLVM_DEBUG(dbgs() << MF.getName() << ":" << printMBBReference(*MBB) << " " 584 << MBB->getName() << "\n From: " << *begin() 585 << " To: "; 586 if (RegionEnd != MBB->end()) dbgs() << *RegionEnd; 587 else dbgs() << "End"; 588 dbgs() << " RegionInstrs: " << NumRegionInstrs << '\n'); 589 590 schedule(); 591 592 exitRegion(); 593 ++RegionIdx; 594 } 595 finishBlock(); 596 597 } while (Stage < 2); 598 } 599