File: | src/gnu/usr.bin/clang/libLLVM/../../../llvm/llvm/lib/Target/AMDGPU/SIMachineScheduler.cpp |
Warning: | line 770, column 9 Value stored to 'ProposedColor' is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
1 | //===-- SIMachineScheduler.cpp - SI Scheduler Interface -------------------===// |
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 | /// SI Machine Scheduler interface |
11 | // |
12 | //===----------------------------------------------------------------------===// |
13 | |
14 | #include "SIMachineScheduler.h" |
15 | #include "SIInstrInfo.h" |
16 | #include "MCTargetDesc/AMDGPUMCTargetDesc.h" |
17 | #include "llvm/CodeGen/LiveIntervals.h" |
18 | #include "llvm/CodeGen/MachineRegisterInfo.h" |
19 | |
20 | using namespace llvm; |
21 | |
22 | #define DEBUG_TYPE"machine-scheduler" "machine-scheduler" |
23 | |
24 | // This scheduler implements a different scheduling algorithm than |
25 | // GenericScheduler. |
26 | // |
27 | // There are several specific architecture behaviours that can't be modelled |
28 | // for GenericScheduler: |
29 | // . When accessing the result of an SGPR load instruction, you have to wait |
30 | // for all the SGPR load instructions before your current instruction to |
31 | // have finished. |
32 | // . When accessing the result of an VGPR load instruction, you have to wait |
33 | // for all the VGPR load instructions previous to the VGPR load instruction |
34 | // you are interested in to finish. |
35 | // . The less the register pressure, the best load latencies are hidden |
36 | // |
37 | // Moreover some specifities (like the fact a lot of instructions in the shader |
38 | // have few dependencies) makes the generic scheduler have some unpredictable |
39 | // behaviours. For example when register pressure becomes high, it can either |
40 | // manage to prevent register pressure from going too high, or it can |
41 | // increase register pressure even more than if it hadn't taken register |
42 | // pressure into account. |
43 | // |
44 | // Also some other bad behaviours are generated, like loading at the beginning |
45 | // of the shader a constant in VGPR you won't need until the end of the shader. |
46 | // |
47 | // The scheduling problem for SI can distinguish three main parts: |
48 | // . Hiding high latencies (texture sampling, etc) |
49 | // . Hiding low latencies (SGPR constant loading, etc) |
50 | // . Keeping register usage low for better latency hiding and general |
51 | // performance |
52 | // |
53 | // Some other things can also affect performance, but are hard to predict |
54 | // (cache usage, the fact the HW can issue several instructions from different |
55 | // wavefronts if different types, etc) |
56 | // |
57 | // This scheduler tries to solve the scheduling problem by dividing it into |
58 | // simpler sub-problems. It divides the instructions into blocks, schedules |
59 | // locally inside the blocks where it takes care of low latencies, and then |
60 | // chooses the order of the blocks by taking care of high latencies. |
61 | // Dividing the instructions into blocks helps control keeping register |
62 | // usage low. |
63 | // |
64 | // First the instructions are put into blocks. |
65 | // We want the blocks help control register usage and hide high latencies |
66 | // later. To help control register usage, we typically want all local |
67 | // computations, when for example you create a result that can be comsummed |
68 | // right away, to be contained in a block. Block inputs and outputs would |
69 | // typically be important results that are needed in several locations of |
70 | // the shader. Since we do want blocks to help hide high latencies, we want |
71 | // the instructions inside the block to have a minimal set of dependencies |
72 | // on high latencies. It will make it easy to pick blocks to hide specific |
73 | // high latencies. |
74 | // The block creation algorithm is divided into several steps, and several |
75 | // variants can be tried during the scheduling process. |
76 | // |
77 | // Second the order of the instructions inside the blocks is chosen. |
78 | // At that step we do take into account only register usage and hiding |
79 | // low latency instructions |
80 | // |
81 | // Third the block order is chosen, there we try to hide high latencies |
82 | // and keep register usage low. |
83 | // |
84 | // After the third step, a pass is done to improve the hiding of low |
85 | // latencies. |
86 | // |
87 | // Actually when talking about 'low latency' or 'high latency' it includes |
88 | // both the latency to get the cache (or global mem) data go to the register, |
89 | // and the bandwidth limitations. |
90 | // Increasing the number of active wavefronts helps hide the former, but it |
91 | // doesn't solve the latter, thus why even if wavefront count is high, we have |
92 | // to try have as many instructions hiding high latencies as possible. |
93 | // The OpenCL doc says for example latency of 400 cycles for a global mem access, |
94 | // which is hidden by 10 instructions if the wavefront count is 10. |
95 | |
96 | // Some figures taken from AMD docs: |
97 | // Both texture and constant L1 caches are 4-way associative with 64 bytes |
98 | // lines. |
99 | // Constant cache is shared with 4 CUs. |
100 | // For texture sampling, the address generation unit receives 4 texture |
101 | // addresses per cycle, thus we could expect texture sampling latency to be |
102 | // equivalent to 4 instructions in the very best case (a VGPR is 64 work items, |
103 | // instructions in a wavefront group are executed every 4 cycles), |
104 | // or 16 instructions if the other wavefronts associated to the 3 other VALUs |
105 | // of the CU do texture sampling too. (Don't take these figures too seriously, |
106 | // as I'm not 100% sure of the computation) |
107 | // Data exports should get similar latency. |
108 | // For constant loading, the cache is shader with 4 CUs. |
109 | // The doc says "a throughput of 16B/cycle for each of the 4 Compute Unit" |
110 | // I guess if the other CU don't read the cache, it can go up to 64B/cycle. |
111 | // It means a simple s_buffer_load should take one instruction to hide, as |
112 | // well as a s_buffer_loadx2 and potentially a s_buffer_loadx8 if on the same |
113 | // cache line. |
114 | // |
115 | // As of today the driver doesn't preload the constants in cache, thus the |
116 | // first loads get extra latency. The doc says global memory access can be |
117 | // 300-600 cycles. We do not specially take that into account when scheduling |
118 | // As we expect the driver to be able to preload the constants soon. |
119 | |
120 | // common code // |
121 | |
122 | #ifndef NDEBUG1 |
123 | |
124 | static const char *getReasonStr(SIScheduleCandReason Reason) { |
125 | switch (Reason) { |
126 | case NoCand: return "NOCAND"; |
127 | case RegUsage: return "REGUSAGE"; |
128 | case Latency: return "LATENCY"; |
129 | case Successor: return "SUCCESSOR"; |
130 | case Depth: return "DEPTH"; |
131 | case NodeOrder: return "ORDER"; |
132 | } |
133 | llvm_unreachable("Unknown reason!")__builtin_unreachable(); |
134 | } |
135 | |
136 | #endif |
137 | |
138 | namespace llvm { |
139 | namespace SISched { |
140 | static bool tryLess(int TryVal, int CandVal, |
141 | SISchedulerCandidate &TryCand, |
142 | SISchedulerCandidate &Cand, |
143 | SIScheduleCandReason Reason) { |
144 | if (TryVal < CandVal) { |
145 | TryCand.Reason = Reason; |
146 | return true; |
147 | } |
148 | if (TryVal > CandVal) { |
149 | if (Cand.Reason > Reason) |
150 | Cand.Reason = Reason; |
151 | return true; |
152 | } |
153 | Cand.setRepeat(Reason); |
154 | return false; |
155 | } |
156 | |
157 | static bool tryGreater(int TryVal, int CandVal, |
158 | SISchedulerCandidate &TryCand, |
159 | SISchedulerCandidate &Cand, |
160 | SIScheduleCandReason Reason) { |
161 | if (TryVal > CandVal) { |
162 | TryCand.Reason = Reason; |
163 | return true; |
164 | } |
165 | if (TryVal < CandVal) { |
166 | if (Cand.Reason > Reason) |
167 | Cand.Reason = Reason; |
168 | return true; |
169 | } |
170 | Cand.setRepeat(Reason); |
171 | return false; |
172 | } |
173 | } // end namespace SISched |
174 | } // end namespace llvm |
175 | |
176 | // SIScheduleBlock // |
177 | |
178 | void SIScheduleBlock::addUnit(SUnit *SU) { |
179 | NodeNum2Index[SU->NodeNum] = SUnits.size(); |
180 | SUnits.push_back(SU); |
181 | } |
182 | |
183 | #ifndef NDEBUG1 |
184 | void SIScheduleBlock::traceCandidate(const SISchedCandidate &Cand) { |
185 | |
186 | dbgs() << " SU(" << Cand.SU->NodeNum << ") " << getReasonStr(Cand.Reason); |
187 | dbgs() << '\n'; |
188 | } |
189 | #endif |
190 | |
191 | void SIScheduleBlock::tryCandidateTopDown(SISchedCandidate &Cand, |
192 | SISchedCandidate &TryCand) { |
193 | // Initialize the candidate if needed. |
194 | if (!Cand.isValid()) { |
195 | TryCand.Reason = NodeOrder; |
196 | return; |
197 | } |
198 | |
199 | if (Cand.SGPRUsage > 60 && |
200 | SISched::tryLess(TryCand.SGPRUsage, Cand.SGPRUsage, |
201 | TryCand, Cand, RegUsage)) |
202 | return; |
203 | |
204 | // Schedule low latency instructions as top as possible. |
205 | // Order of priority is: |
206 | // . Low latency instructions which do not depend on other low latency |
207 | // instructions we haven't waited for |
208 | // . Other instructions which do not depend on low latency instructions |
209 | // we haven't waited for |
210 | // . Low latencies |
211 | // . All other instructions |
212 | // Goal is to get: low latency instructions - independent instructions |
213 | // - (eventually some more low latency instructions) |
214 | // - instructions that depend on the first low latency instructions. |
215 | // If in the block there is a lot of constant loads, the SGPR usage |
216 | // could go quite high, thus above the arbitrary limit of 60 will encourage |
217 | // use the already loaded constants (in order to release some SGPRs) before |
218 | // loading more. |
219 | if (SISched::tryLess(TryCand.HasLowLatencyNonWaitedParent, |
220 | Cand.HasLowLatencyNonWaitedParent, |
221 | TryCand, Cand, SIScheduleCandReason::Depth)) |
222 | return; |
223 | |
224 | if (SISched::tryGreater(TryCand.IsLowLatency, Cand.IsLowLatency, |
225 | TryCand, Cand, SIScheduleCandReason::Depth)) |
226 | return; |
227 | |
228 | if (TryCand.IsLowLatency && |
229 | SISched::tryLess(TryCand.LowLatencyOffset, Cand.LowLatencyOffset, |
230 | TryCand, Cand, SIScheduleCandReason::Depth)) |
231 | return; |
232 | |
233 | if (SISched::tryLess(TryCand.VGPRUsage, Cand.VGPRUsage, |
234 | TryCand, Cand, RegUsage)) |
235 | return; |
236 | |
237 | // Fall through to original instruction order. |
238 | if (TryCand.SU->NodeNum < Cand.SU->NodeNum) { |
239 | TryCand.Reason = NodeOrder; |
240 | } |
241 | } |
242 | |
243 | SUnit* SIScheduleBlock::pickNode() { |
244 | SISchedCandidate TopCand; |
245 | |
246 | for (SUnit* SU : TopReadySUs) { |
247 | SISchedCandidate TryCand; |
248 | std::vector<unsigned> pressure; |
249 | std::vector<unsigned> MaxPressure; |
250 | // Predict register usage after this instruction. |
251 | TryCand.SU = SU; |
252 | TopRPTracker.getDownwardPressure(SU->getInstr(), pressure, MaxPressure); |
253 | TryCand.SGPRUsage = pressure[AMDGPU::RegisterPressureSets::SReg_32]; |
254 | TryCand.VGPRUsage = pressure[AMDGPU::RegisterPressureSets::VGPR_32]; |
255 | TryCand.IsLowLatency = DAG->IsLowLatencySU[SU->NodeNum]; |
256 | TryCand.LowLatencyOffset = DAG->LowLatencyOffset[SU->NodeNum]; |
257 | TryCand.HasLowLatencyNonWaitedParent = |
258 | HasLowLatencyNonWaitedParent[NodeNum2Index[SU->NodeNum]]; |
259 | tryCandidateTopDown(TopCand, TryCand); |
260 | if (TryCand.Reason != NoCand) |
261 | TopCand.setBest(TryCand); |
262 | } |
263 | |
264 | return TopCand.SU; |
265 | } |
266 | |
267 | |
268 | // Schedule something valid. |
269 | void SIScheduleBlock::fastSchedule() { |
270 | TopReadySUs.clear(); |
271 | if (Scheduled) |
272 | undoSchedule(); |
273 | |
274 | for (SUnit* SU : SUnits) { |
275 | if (!SU->NumPredsLeft) |
276 | TopReadySUs.push_back(SU); |
277 | } |
278 | |
279 | while (!TopReadySUs.empty()) { |
280 | SUnit *SU = TopReadySUs[0]; |
281 | ScheduledSUnits.push_back(SU); |
282 | nodeScheduled(SU); |
283 | } |
284 | |
285 | Scheduled = true; |
286 | } |
287 | |
288 | // Returns if the register was set between first and last. |
289 | static bool isDefBetween(unsigned Reg, |
290 | SlotIndex First, SlotIndex Last, |
291 | const MachineRegisterInfo *MRI, |
292 | const LiveIntervals *LIS) { |
293 | for (MachineRegisterInfo::def_instr_iterator |
294 | UI = MRI->def_instr_begin(Reg), |
295 | UE = MRI->def_instr_end(); UI != UE; ++UI) { |
296 | const MachineInstr* MI = &*UI; |
297 | if (MI->isDebugValue()) |
298 | continue; |
299 | SlotIndex InstSlot = LIS->getInstructionIndex(*MI).getRegSlot(); |
300 | if (InstSlot >= First && InstSlot <= Last) |
301 | return true; |
302 | } |
303 | return false; |
304 | } |
305 | |
306 | void SIScheduleBlock::initRegPressure(MachineBasicBlock::iterator BeginBlock, |
307 | MachineBasicBlock::iterator EndBlock) { |
308 | IntervalPressure Pressure, BotPressure; |
309 | RegPressureTracker RPTracker(Pressure), BotRPTracker(BotPressure); |
310 | LiveIntervals *LIS = DAG->getLIS(); |
311 | MachineRegisterInfo *MRI = DAG->getMRI(); |
312 | DAG->initRPTracker(TopRPTracker); |
313 | DAG->initRPTracker(BotRPTracker); |
314 | DAG->initRPTracker(RPTracker); |
315 | |
316 | // Goes though all SU. RPTracker captures what had to be alive for the SUs |
317 | // to execute, and what is still alive at the end. |
318 | for (SUnit* SU : ScheduledSUnits) { |
319 | RPTracker.setPos(SU->getInstr()); |
320 | RPTracker.advance(); |
321 | } |
322 | |
323 | // Close the RPTracker to finalize live ins/outs. |
324 | RPTracker.closeRegion(); |
325 | |
326 | // Initialize the live ins and live outs. |
327 | TopRPTracker.addLiveRegs(RPTracker.getPressure().LiveInRegs); |
328 | BotRPTracker.addLiveRegs(RPTracker.getPressure().LiveOutRegs); |
329 | |
330 | // Do not Track Physical Registers, because it messes up. |
331 | for (const auto &RegMaskPair : RPTracker.getPressure().LiveInRegs) { |
332 | if (Register::isVirtualRegister(RegMaskPair.RegUnit)) |
333 | LiveInRegs.insert(RegMaskPair.RegUnit); |
334 | } |
335 | LiveOutRegs.clear(); |
336 | // There is several possibilities to distinguish: |
337 | // 1) Reg is not input to any instruction in the block, but is output of one |
338 | // 2) 1) + read in the block and not needed after it |
339 | // 3) 1) + read in the block but needed in another block |
340 | // 4) Reg is input of an instruction but another block will read it too |
341 | // 5) Reg is input of an instruction and then rewritten in the block. |
342 | // result is not read in the block (implies used in another block) |
343 | // 6) Reg is input of an instruction and then rewritten in the block. |
344 | // result is read in the block and not needed in another block |
345 | // 7) Reg is input of an instruction and then rewritten in the block. |
346 | // result is read in the block but also needed in another block |
347 | // LiveInRegs will contains all the regs in situation 4, 5, 6, 7 |
348 | // We want LiveOutRegs to contain only Regs whose content will be read after |
349 | // in another block, and whose content was written in the current block, |
350 | // that is we want it to get 1, 3, 5, 7 |
351 | // Since we made the MIs of a block to be packed all together before |
352 | // scheduling, then the LiveIntervals were correct, and the RPTracker was |
353 | // able to correctly handle 5 vs 6, 2 vs 3. |
354 | // (Note: This is not sufficient for RPTracker to not do mistakes for case 4) |
355 | // The RPTracker's LiveOutRegs has 1, 3, (some correct or incorrect)4, 5, 7 |
356 | // Comparing to LiveInRegs is not sufficient to differenciate 4 vs 5, 7 |
357 | // The use of findDefBetween removes the case 4. |
358 | for (const auto &RegMaskPair : RPTracker.getPressure().LiveOutRegs) { |
359 | Register Reg = RegMaskPair.RegUnit; |
360 | if (Reg.isVirtual() && |
361 | isDefBetween(Reg, LIS->getInstructionIndex(*BeginBlock).getRegSlot(), |
362 | LIS->getInstructionIndex(*EndBlock).getRegSlot(), MRI, |
363 | LIS)) { |
364 | LiveOutRegs.insert(Reg); |
365 | } |
366 | } |
367 | |
368 | // Pressure = sum_alive_registers register size |
369 | // Internally llvm will represent some registers as big 128 bits registers |
370 | // for example, but they actually correspond to 4 actual 32 bits registers. |
371 | // Thus Pressure is not equal to num_alive_registers * constant. |
372 | LiveInPressure = TopPressure.MaxSetPressure; |
373 | LiveOutPressure = BotPressure.MaxSetPressure; |
374 | |
375 | // Prepares TopRPTracker for top down scheduling. |
376 | TopRPTracker.closeTop(); |
377 | } |
378 | |
379 | void SIScheduleBlock::schedule(MachineBasicBlock::iterator BeginBlock, |
380 | MachineBasicBlock::iterator EndBlock) { |
381 | if (!Scheduled) |
382 | fastSchedule(); |
383 | |
384 | // PreScheduling phase to set LiveIn and LiveOut. |
385 | initRegPressure(BeginBlock, EndBlock); |
386 | undoSchedule(); |
387 | |
388 | // Schedule for real now. |
389 | |
390 | TopReadySUs.clear(); |
391 | |
392 | for (SUnit* SU : SUnits) { |
393 | if (!SU->NumPredsLeft) |
394 | TopReadySUs.push_back(SU); |
395 | } |
396 | |
397 | while (!TopReadySUs.empty()) { |
398 | SUnit *SU = pickNode(); |
399 | ScheduledSUnits.push_back(SU); |
400 | TopRPTracker.setPos(SU->getInstr()); |
401 | TopRPTracker.advance(); |
402 | nodeScheduled(SU); |
403 | } |
404 | |
405 | // TODO: compute InternalAdditionnalPressure. |
406 | InternalAdditionnalPressure.resize(TopPressure.MaxSetPressure.size()); |
407 | |
408 | // Check everything is right. |
409 | #ifndef NDEBUG1 |
410 | assert(SUnits.size() == ScheduledSUnits.size() &&((void)0) |
411 | TopReadySUs.empty())((void)0); |
412 | for (SUnit* SU : SUnits) { |
413 | assert(SU->isScheduled &&((void)0) |
414 | SU->NumPredsLeft == 0)((void)0); |
415 | } |
416 | #endif |
417 | |
418 | Scheduled = true; |
419 | } |
420 | |
421 | void SIScheduleBlock::undoSchedule() { |
422 | for (SUnit* SU : SUnits) { |
423 | SU->isScheduled = false; |
424 | for (SDep& Succ : SU->Succs) { |
425 | if (BC->isSUInBlock(Succ.getSUnit(), ID)) |
426 | undoReleaseSucc(SU, &Succ); |
427 | } |
428 | } |
429 | HasLowLatencyNonWaitedParent.assign(SUnits.size(), 0); |
430 | ScheduledSUnits.clear(); |
431 | Scheduled = false; |
432 | } |
433 | |
434 | void SIScheduleBlock::undoReleaseSucc(SUnit *SU, SDep *SuccEdge) { |
435 | SUnit *SuccSU = SuccEdge->getSUnit(); |
436 | |
437 | if (SuccEdge->isWeak()) { |
438 | ++SuccSU->WeakPredsLeft; |
439 | return; |
440 | } |
441 | ++SuccSU->NumPredsLeft; |
442 | } |
443 | |
444 | void SIScheduleBlock::releaseSucc(SUnit *SU, SDep *SuccEdge) { |
445 | SUnit *SuccSU = SuccEdge->getSUnit(); |
446 | |
447 | if (SuccEdge->isWeak()) { |
448 | --SuccSU->WeakPredsLeft; |
449 | return; |
450 | } |
451 | #ifndef NDEBUG1 |
452 | if (SuccSU->NumPredsLeft == 0) { |
453 | dbgs() << "*** Scheduling failed! ***\n"; |
454 | DAG->dumpNode(*SuccSU); |
455 | dbgs() << " has been released too many times!\n"; |
456 | llvm_unreachable(nullptr)__builtin_unreachable(); |
457 | } |
458 | #endif |
459 | |
460 | --SuccSU->NumPredsLeft; |
461 | } |
462 | |
463 | /// Release Successors of the SU that are in the block or not. |
464 | void SIScheduleBlock::releaseSuccessors(SUnit *SU, bool InOrOutBlock) { |
465 | for (SDep& Succ : SU->Succs) { |
466 | SUnit *SuccSU = Succ.getSUnit(); |
467 | |
468 | if (SuccSU->NodeNum >= DAG->SUnits.size()) |
469 | continue; |
470 | |
471 | if (BC->isSUInBlock(SuccSU, ID) != InOrOutBlock) |
472 | continue; |
473 | |
474 | releaseSucc(SU, &Succ); |
475 | if (SuccSU->NumPredsLeft == 0 && InOrOutBlock) |
476 | TopReadySUs.push_back(SuccSU); |
477 | } |
478 | } |
479 | |
480 | void SIScheduleBlock::nodeScheduled(SUnit *SU) { |
481 | // Is in TopReadySUs |
482 | assert (!SU->NumPredsLeft)((void)0); |
483 | std::vector<SUnit *>::iterator I = llvm::find(TopReadySUs, SU); |
484 | if (I == TopReadySUs.end()) { |
485 | dbgs() << "Data Structure Bug in SI Scheduler\n"; |
486 | llvm_unreachable(nullptr)__builtin_unreachable(); |
487 | } |
488 | TopReadySUs.erase(I); |
489 | |
490 | releaseSuccessors(SU, true); |
491 | // Scheduling this node will trigger a wait, |
492 | // thus propagate to other instructions that they do not need to wait either. |
493 | if (HasLowLatencyNonWaitedParent[NodeNum2Index[SU->NodeNum]]) |
494 | HasLowLatencyNonWaitedParent.assign(SUnits.size(), 0); |
495 | |
496 | if (DAG->IsLowLatencySU[SU->NodeNum]) { |
497 | for (SDep& Succ : SU->Succs) { |
498 | std::map<unsigned, unsigned>::iterator I = |
499 | NodeNum2Index.find(Succ.getSUnit()->NodeNum); |
500 | if (I != NodeNum2Index.end()) |
501 | HasLowLatencyNonWaitedParent[I->second] = 1; |
502 | } |
503 | } |
504 | SU->isScheduled = true; |
505 | } |
506 | |
507 | void SIScheduleBlock::finalizeUnits() { |
508 | // We remove links from outside blocks to enable scheduling inside the block. |
509 | for (SUnit* SU : SUnits) { |
510 | releaseSuccessors(SU, false); |
511 | if (DAG->IsHighLatencySU[SU->NodeNum]) |
512 | HighLatencyBlock = true; |
513 | } |
514 | HasLowLatencyNonWaitedParent.resize(SUnits.size(), 0); |
515 | } |
516 | |
517 | // we maintain ascending order of IDs |
518 | void SIScheduleBlock::addPred(SIScheduleBlock *Pred) { |
519 | unsigned PredID = Pred->getID(); |
520 | |
521 | // Check if not already predecessor. |
522 | for (SIScheduleBlock* P : Preds) { |
523 | if (PredID == P->getID()) |
524 | return; |
525 | } |
526 | Preds.push_back(Pred); |
527 | |
528 | assert(none_of(Succs,((void)0) |
529 | [=](std::pair<SIScheduleBlock*,((void)0) |
530 | SIScheduleBlockLinkKind> S) {((void)0) |
531 | return PredID == S.first->getID();((void)0) |
532 | }) &&((void)0) |
533 | "Loop in the Block Graph!")((void)0); |
534 | } |
535 | |
536 | void SIScheduleBlock::addSucc(SIScheduleBlock *Succ, |
537 | SIScheduleBlockLinkKind Kind) { |
538 | unsigned SuccID = Succ->getID(); |
539 | |
540 | // Check if not already predecessor. |
541 | for (std::pair<SIScheduleBlock*, SIScheduleBlockLinkKind> &S : Succs) { |
542 | if (SuccID == S.first->getID()) { |
543 | if (S.second == SIScheduleBlockLinkKind::NoData && |
544 | Kind == SIScheduleBlockLinkKind::Data) |
545 | S.second = Kind; |
546 | return; |
547 | } |
548 | } |
549 | if (Succ->isHighLatencyBlock()) |
550 | ++NumHighLatencySuccessors; |
551 | Succs.push_back(std::make_pair(Succ, Kind)); |
552 | |
553 | assert(none_of(Preds,((void)0) |
554 | [=](SIScheduleBlock *P) { return SuccID == P->getID(); }) &&((void)0) |
555 | "Loop in the Block Graph!")((void)0); |
556 | } |
557 | |
558 | #ifndef NDEBUG1 |
559 | void SIScheduleBlock::printDebug(bool full) { |
560 | dbgs() << "Block (" << ID << ")\n"; |
561 | if (!full) |
562 | return; |
563 | |
564 | dbgs() << "\nContains High Latency Instruction: " |
565 | << HighLatencyBlock << '\n'; |
566 | dbgs() << "\nDepends On:\n"; |
567 | for (SIScheduleBlock* P : Preds) { |
568 | P->printDebug(false); |
569 | } |
570 | |
571 | dbgs() << "\nSuccessors:\n"; |
572 | for (std::pair<SIScheduleBlock*, SIScheduleBlockLinkKind> S : Succs) { |
573 | if (S.second == SIScheduleBlockLinkKind::Data) |
574 | dbgs() << "(Data Dep) "; |
575 | S.first->printDebug(false); |
576 | } |
577 | |
578 | if (Scheduled) { |
579 | dbgs() << "LiveInPressure " |
580 | << LiveInPressure[AMDGPU::RegisterPressureSets::SReg_32] << ' ' |
581 | << LiveInPressure[AMDGPU::RegisterPressureSets::VGPR_32] << '\n'; |
582 | dbgs() << "LiveOutPressure " |
583 | << LiveOutPressure[AMDGPU::RegisterPressureSets::SReg_32] << ' ' |
584 | << LiveOutPressure[AMDGPU::RegisterPressureSets::VGPR_32] << "\n\n"; |
585 | dbgs() << "LiveIns:\n"; |
586 | for (unsigned Reg : LiveInRegs) |
587 | dbgs() << printVRegOrUnit(Reg, DAG->getTRI()) << ' '; |
588 | |
589 | dbgs() << "\nLiveOuts:\n"; |
590 | for (unsigned Reg : LiveOutRegs) |
591 | dbgs() << printVRegOrUnit(Reg, DAG->getTRI()) << ' '; |
592 | } |
593 | |
594 | dbgs() << "\nInstructions:\n"; |
595 | for (const SUnit* SU : SUnits) |
596 | DAG->dumpNode(*SU); |
597 | |
598 | dbgs() << "///////////////////////\n"; |
599 | } |
600 | #endif |
601 | |
602 | // SIScheduleBlockCreator // |
603 | |
604 | SIScheduleBlockCreator::SIScheduleBlockCreator(SIScheduleDAGMI *DAG) |
605 | : DAG(DAG) {} |
606 | |
607 | SIScheduleBlocks |
608 | SIScheduleBlockCreator::getBlocks(SISchedulerBlockCreatorVariant BlockVariant) { |
609 | std::map<SISchedulerBlockCreatorVariant, SIScheduleBlocks>::iterator B = |
610 | Blocks.find(BlockVariant); |
611 | if (B == Blocks.end()) { |
612 | SIScheduleBlocks Res; |
613 | createBlocksForVariant(BlockVariant); |
614 | topologicalSort(); |
615 | scheduleInsideBlocks(); |
616 | fillStats(); |
617 | Res.Blocks = CurrentBlocks; |
618 | Res.TopDownIndex2Block = TopDownIndex2Block; |
619 | Res.TopDownBlock2Index = TopDownBlock2Index; |
620 | Blocks[BlockVariant] = Res; |
621 | return Res; |
622 | } else { |
623 | return B->second; |
624 | } |
625 | } |
626 | |
627 | bool SIScheduleBlockCreator::isSUInBlock(SUnit *SU, unsigned ID) { |
628 | if (SU->NodeNum >= DAG->SUnits.size()) |
629 | return false; |
630 | return CurrentBlocks[Node2CurrentBlock[SU->NodeNum]]->getID() == ID; |
631 | } |
632 | |
633 | void SIScheduleBlockCreator::colorHighLatenciesAlone() { |
634 | unsigned DAGSize = DAG->SUnits.size(); |
635 | |
636 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
637 | SUnit *SU = &DAG->SUnits[i]; |
638 | if (DAG->IsHighLatencySU[SU->NodeNum]) { |
639 | CurrentColoring[SU->NodeNum] = NextReservedID++; |
640 | } |
641 | } |
642 | } |
643 | |
644 | static bool |
645 | hasDataDependencyPred(const SUnit &SU, const SUnit &FromSU) { |
646 | for (const auto &PredDep : SU.Preds) { |
647 | if (PredDep.getSUnit() == &FromSU && |
648 | PredDep.getKind() == llvm::SDep::Data) |
649 | return true; |
650 | } |
651 | return false; |
652 | } |
653 | |
654 | void SIScheduleBlockCreator::colorHighLatenciesGroups() { |
655 | unsigned DAGSize = DAG->SUnits.size(); |
656 | unsigned NumHighLatencies = 0; |
657 | unsigned GroupSize; |
658 | int Color = NextReservedID; |
659 | unsigned Count = 0; |
660 | std::set<unsigned> FormingGroup; |
661 | |
662 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
663 | SUnit *SU = &DAG->SUnits[i]; |
664 | if (DAG->IsHighLatencySU[SU->NodeNum]) |
665 | ++NumHighLatencies; |
666 | } |
667 | |
668 | if (NumHighLatencies == 0) |
669 | return; |
670 | |
671 | if (NumHighLatencies <= 6) |
672 | GroupSize = 2; |
673 | else if (NumHighLatencies <= 12) |
674 | GroupSize = 3; |
675 | else |
676 | GroupSize = 4; |
677 | |
678 | for (unsigned SUNum : DAG->TopDownIndex2SU) { |
679 | const SUnit &SU = DAG->SUnits[SUNum]; |
680 | if (DAG->IsHighLatencySU[SU.NodeNum]) { |
681 | unsigned CompatibleGroup = true; |
682 | int ProposedColor = Color; |
683 | std::vector<int> AdditionalElements; |
684 | |
685 | // We don't want to put in the same block |
686 | // two high latency instructions that depend |
687 | // on each other. |
688 | // One way would be to check canAddEdge |
689 | // in both directions, but that currently is not |
690 | // enough because there the high latency order is |
691 | // enforced (via links). |
692 | // Instead, look at the dependencies between the |
693 | // high latency instructions and deduce if it is |
694 | // a data dependency or not. |
695 | for (unsigned j : FormingGroup) { |
696 | bool HasSubGraph; |
697 | std::vector<int> SubGraph; |
698 | // By construction (topological order), if SU and |
699 | // DAG->SUnits[j] are linked, DAG->SUnits[j] is neccessary |
700 | // in the parent graph of SU. |
701 | #ifndef NDEBUG1 |
702 | SubGraph = DAG->GetTopo()->GetSubGraph(SU, DAG->SUnits[j], |
703 | HasSubGraph); |
704 | assert(!HasSubGraph)((void)0); |
705 | #endif |
706 | SubGraph = DAG->GetTopo()->GetSubGraph(DAG->SUnits[j], SU, |
707 | HasSubGraph); |
708 | if (!HasSubGraph) |
709 | continue; // No dependencies between each other |
710 | else if (SubGraph.size() > 5) { |
711 | // Too many elements would be required to be added to the block. |
712 | CompatibleGroup = false; |
713 | break; |
714 | } |
715 | else { |
716 | // Check the type of dependency |
717 | for (unsigned k : SubGraph) { |
718 | // If in the path to join the two instructions, |
719 | // there is another high latency instruction, |
720 | // or instructions colored for another block |
721 | // abort the merge. |
722 | if (DAG->IsHighLatencySU[k] || |
723 | (CurrentColoring[k] != ProposedColor && |
724 | CurrentColoring[k] != 0)) { |
725 | CompatibleGroup = false; |
726 | break; |
727 | } |
728 | // If one of the SU in the subgraph depends on the result of SU j, |
729 | // there'll be a data dependency. |
730 | if (hasDataDependencyPred(DAG->SUnits[k], DAG->SUnits[j])) { |
731 | CompatibleGroup = false; |
732 | break; |
733 | } |
734 | } |
735 | if (!CompatibleGroup) |
736 | break; |
737 | // Same check for the SU |
738 | if (hasDataDependencyPred(SU, DAG->SUnits[j])) { |
739 | CompatibleGroup = false; |
740 | break; |
741 | } |
742 | // Add all the required instructions to the block |
743 | // These cannot live in another block (because they |
744 | // depend (order dependency) on one of the |
745 | // instruction in the block, and are required for the |
746 | // high latency instruction we add. |
747 | llvm::append_range(AdditionalElements, SubGraph); |
748 | } |
749 | } |
750 | if (CompatibleGroup) { |
751 | FormingGroup.insert(SU.NodeNum); |
752 | for (unsigned j : AdditionalElements) |
753 | CurrentColoring[j] = ProposedColor; |
754 | CurrentColoring[SU.NodeNum] = ProposedColor; |
755 | ++Count; |
756 | } |
757 | // Found one incompatible instruction, |
758 | // or has filled a big enough group. |
759 | // -> start a new one. |
760 | if (!CompatibleGroup) { |
761 | FormingGroup.clear(); |
762 | Color = ++NextReservedID; |
763 | ProposedColor = Color; |
764 | FormingGroup.insert(SU.NodeNum); |
765 | CurrentColoring[SU.NodeNum] = ProposedColor; |
766 | Count = 0; |
767 | } else if (Count == GroupSize) { |
768 | FormingGroup.clear(); |
769 | Color = ++NextReservedID; |
770 | ProposedColor = Color; |
Value stored to 'ProposedColor' is never read | |
771 | Count = 0; |
772 | } |
773 | } |
774 | } |
775 | } |
776 | |
777 | void SIScheduleBlockCreator::colorComputeReservedDependencies() { |
778 | unsigned DAGSize = DAG->SUnits.size(); |
779 | std::map<std::set<unsigned>, unsigned> ColorCombinations; |
780 | |
781 | CurrentTopDownReservedDependencyColoring.clear(); |
782 | CurrentBottomUpReservedDependencyColoring.clear(); |
783 | |
784 | CurrentTopDownReservedDependencyColoring.resize(DAGSize, 0); |
785 | CurrentBottomUpReservedDependencyColoring.resize(DAGSize, 0); |
786 | |
787 | // Traverse TopDown, and give different colors to SUs depending |
788 | // on which combination of High Latencies they depend on. |
789 | |
790 | for (unsigned SUNum : DAG->TopDownIndex2SU) { |
791 | SUnit *SU = &DAG->SUnits[SUNum]; |
792 | std::set<unsigned> SUColors; |
793 | |
794 | // Already given. |
795 | if (CurrentColoring[SU->NodeNum]) { |
796 | CurrentTopDownReservedDependencyColoring[SU->NodeNum] = |
797 | CurrentColoring[SU->NodeNum]; |
798 | continue; |
799 | } |
800 | |
801 | for (SDep& PredDep : SU->Preds) { |
802 | SUnit *Pred = PredDep.getSUnit(); |
803 | if (PredDep.isWeak() || Pred->NodeNum >= DAGSize) |
804 | continue; |
805 | if (CurrentTopDownReservedDependencyColoring[Pred->NodeNum] > 0) |
806 | SUColors.insert(CurrentTopDownReservedDependencyColoring[Pred->NodeNum]); |
807 | } |
808 | // Color 0 by default. |
809 | if (SUColors.empty()) |
810 | continue; |
811 | // Same color than parents. |
812 | if (SUColors.size() == 1 && *SUColors.begin() > DAGSize) |
813 | CurrentTopDownReservedDependencyColoring[SU->NodeNum] = |
814 | *SUColors.begin(); |
815 | else { |
816 | std::map<std::set<unsigned>, unsigned>::iterator Pos = |
817 | ColorCombinations.find(SUColors); |
818 | if (Pos != ColorCombinations.end()) { |
819 | CurrentTopDownReservedDependencyColoring[SU->NodeNum] = Pos->second; |
820 | } else { |
821 | CurrentTopDownReservedDependencyColoring[SU->NodeNum] = |
822 | NextNonReservedID; |
823 | ColorCombinations[SUColors] = NextNonReservedID++; |
824 | } |
825 | } |
826 | } |
827 | |
828 | ColorCombinations.clear(); |
829 | |
830 | // Same as before, but BottomUp. |
831 | |
832 | for (unsigned SUNum : DAG->BottomUpIndex2SU) { |
833 | SUnit *SU = &DAG->SUnits[SUNum]; |
834 | std::set<unsigned> SUColors; |
835 | |
836 | // Already given. |
837 | if (CurrentColoring[SU->NodeNum]) { |
838 | CurrentBottomUpReservedDependencyColoring[SU->NodeNum] = |
839 | CurrentColoring[SU->NodeNum]; |
840 | continue; |
841 | } |
842 | |
843 | for (SDep& SuccDep : SU->Succs) { |
844 | SUnit *Succ = SuccDep.getSUnit(); |
845 | if (SuccDep.isWeak() || Succ->NodeNum >= DAGSize) |
846 | continue; |
847 | if (CurrentBottomUpReservedDependencyColoring[Succ->NodeNum] > 0) |
848 | SUColors.insert(CurrentBottomUpReservedDependencyColoring[Succ->NodeNum]); |
849 | } |
850 | // Keep color 0. |
851 | if (SUColors.empty()) |
852 | continue; |
853 | // Same color than parents. |
854 | if (SUColors.size() == 1 && *SUColors.begin() > DAGSize) |
855 | CurrentBottomUpReservedDependencyColoring[SU->NodeNum] = |
856 | *SUColors.begin(); |
857 | else { |
858 | std::map<std::set<unsigned>, unsigned>::iterator Pos = |
859 | ColorCombinations.find(SUColors); |
860 | if (Pos != ColorCombinations.end()) { |
861 | CurrentBottomUpReservedDependencyColoring[SU->NodeNum] = Pos->second; |
862 | } else { |
863 | CurrentBottomUpReservedDependencyColoring[SU->NodeNum] = |
864 | NextNonReservedID; |
865 | ColorCombinations[SUColors] = NextNonReservedID++; |
866 | } |
867 | } |
868 | } |
869 | } |
870 | |
871 | void SIScheduleBlockCreator::colorAccordingToReservedDependencies() { |
872 | unsigned DAGSize = DAG->SUnits.size(); |
873 | std::map<std::pair<unsigned, unsigned>, unsigned> ColorCombinations; |
874 | |
875 | // Every combination of colors given by the top down |
876 | // and bottom up Reserved node dependency |
877 | |
878 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
879 | SUnit *SU = &DAG->SUnits[i]; |
880 | std::pair<unsigned, unsigned> SUColors; |
881 | |
882 | // High latency instructions: already given. |
883 | if (CurrentColoring[SU->NodeNum]) |
884 | continue; |
885 | |
886 | SUColors.first = CurrentTopDownReservedDependencyColoring[SU->NodeNum]; |
887 | SUColors.second = CurrentBottomUpReservedDependencyColoring[SU->NodeNum]; |
888 | |
889 | std::map<std::pair<unsigned, unsigned>, unsigned>::iterator Pos = |
890 | ColorCombinations.find(SUColors); |
891 | if (Pos != ColorCombinations.end()) { |
892 | CurrentColoring[SU->NodeNum] = Pos->second; |
893 | } else { |
894 | CurrentColoring[SU->NodeNum] = NextNonReservedID; |
895 | ColorCombinations[SUColors] = NextNonReservedID++; |
896 | } |
897 | } |
898 | } |
899 | |
900 | void SIScheduleBlockCreator::colorEndsAccordingToDependencies() { |
901 | unsigned DAGSize = DAG->SUnits.size(); |
902 | std::vector<int> PendingColoring = CurrentColoring; |
903 | |
904 | assert(DAGSize >= 1 &&((void)0) |
905 | CurrentBottomUpReservedDependencyColoring.size() == DAGSize &&((void)0) |
906 | CurrentTopDownReservedDependencyColoring.size() == DAGSize)((void)0); |
907 | // If there is no reserved block at all, do nothing. We don't want |
908 | // everything in one block. |
909 | if (*std::max_element(CurrentBottomUpReservedDependencyColoring.begin(), |
910 | CurrentBottomUpReservedDependencyColoring.end()) == 0 && |
911 | *std::max_element(CurrentTopDownReservedDependencyColoring.begin(), |
912 | CurrentTopDownReservedDependencyColoring.end()) == 0) |
913 | return; |
914 | |
915 | for (unsigned SUNum : DAG->BottomUpIndex2SU) { |
916 | SUnit *SU = &DAG->SUnits[SUNum]; |
917 | std::set<unsigned> SUColors; |
918 | std::set<unsigned> SUColorsPending; |
919 | |
920 | if (CurrentColoring[SU->NodeNum] <= (int)DAGSize) |
921 | continue; |
922 | |
923 | if (CurrentBottomUpReservedDependencyColoring[SU->NodeNum] > 0 || |
924 | CurrentTopDownReservedDependencyColoring[SU->NodeNum] > 0) |
925 | continue; |
926 | |
927 | for (SDep& SuccDep : SU->Succs) { |
928 | SUnit *Succ = SuccDep.getSUnit(); |
929 | if (SuccDep.isWeak() || Succ->NodeNum >= DAGSize) |
930 | continue; |
931 | if (CurrentBottomUpReservedDependencyColoring[Succ->NodeNum] > 0 || |
932 | CurrentTopDownReservedDependencyColoring[Succ->NodeNum] > 0) |
933 | SUColors.insert(CurrentColoring[Succ->NodeNum]); |
934 | SUColorsPending.insert(PendingColoring[Succ->NodeNum]); |
935 | } |
936 | // If there is only one child/parent block, and that block |
937 | // is not among the ones we are removing in this path, then |
938 | // merge the instruction to that block |
939 | if (SUColors.size() == 1 && SUColorsPending.size() == 1) |
940 | PendingColoring[SU->NodeNum] = *SUColors.begin(); |
941 | else // TODO: Attribute new colors depending on color |
942 | // combination of children. |
943 | PendingColoring[SU->NodeNum] = NextNonReservedID++; |
944 | } |
945 | CurrentColoring = PendingColoring; |
946 | } |
947 | |
948 | |
949 | void SIScheduleBlockCreator::colorForceConsecutiveOrderInGroup() { |
950 | unsigned DAGSize = DAG->SUnits.size(); |
951 | unsigned PreviousColor; |
952 | std::set<unsigned> SeenColors; |
953 | |
954 | if (DAGSize <= 1) |
955 | return; |
956 | |
957 | PreviousColor = CurrentColoring[0]; |
958 | |
959 | for (unsigned i = 1, e = DAGSize; i != e; ++i) { |
960 | SUnit *SU = &DAG->SUnits[i]; |
961 | unsigned CurrentColor = CurrentColoring[i]; |
962 | unsigned PreviousColorSave = PreviousColor; |
963 | assert(i == SU->NodeNum)((void)0); |
964 | |
965 | if (CurrentColor != PreviousColor) |
966 | SeenColors.insert(PreviousColor); |
967 | PreviousColor = CurrentColor; |
968 | |
969 | if (CurrentColoring[SU->NodeNum] <= (int)DAGSize) |
970 | continue; |
971 | |
972 | if (SeenColors.find(CurrentColor) == SeenColors.end()) |
973 | continue; |
974 | |
975 | if (PreviousColorSave != CurrentColor) |
976 | CurrentColoring[i] = NextNonReservedID++; |
977 | else |
978 | CurrentColoring[i] = CurrentColoring[i-1]; |
979 | } |
980 | } |
981 | |
982 | void SIScheduleBlockCreator::colorMergeConstantLoadsNextGroup() { |
983 | unsigned DAGSize = DAG->SUnits.size(); |
984 | |
985 | for (unsigned SUNum : DAG->BottomUpIndex2SU) { |
986 | SUnit *SU = &DAG->SUnits[SUNum]; |
987 | std::set<unsigned> SUColors; |
988 | |
989 | if (CurrentColoring[SU->NodeNum] <= (int)DAGSize) |
990 | continue; |
991 | |
992 | // No predecessor: Vgpr constant loading. |
993 | // Low latency instructions usually have a predecessor (the address) |
994 | if (SU->Preds.size() > 0 && !DAG->IsLowLatencySU[SU->NodeNum]) |
995 | continue; |
996 | |
997 | for (SDep& SuccDep : SU->Succs) { |
998 | SUnit *Succ = SuccDep.getSUnit(); |
999 | if (SuccDep.isWeak() || Succ->NodeNum >= DAGSize) |
1000 | continue; |
1001 | SUColors.insert(CurrentColoring[Succ->NodeNum]); |
1002 | } |
1003 | if (SUColors.size() == 1) |
1004 | CurrentColoring[SU->NodeNum] = *SUColors.begin(); |
1005 | } |
1006 | } |
1007 | |
1008 | void SIScheduleBlockCreator::colorMergeIfPossibleNextGroup() { |
1009 | unsigned DAGSize = DAG->SUnits.size(); |
1010 | |
1011 | for (unsigned SUNum : DAG->BottomUpIndex2SU) { |
1012 | SUnit *SU = &DAG->SUnits[SUNum]; |
1013 | std::set<unsigned> SUColors; |
1014 | |
1015 | if (CurrentColoring[SU->NodeNum] <= (int)DAGSize) |
1016 | continue; |
1017 | |
1018 | for (SDep& SuccDep : SU->Succs) { |
1019 | SUnit *Succ = SuccDep.getSUnit(); |
1020 | if (SuccDep.isWeak() || Succ->NodeNum >= DAGSize) |
1021 | continue; |
1022 | SUColors.insert(CurrentColoring[Succ->NodeNum]); |
1023 | } |
1024 | if (SUColors.size() == 1) |
1025 | CurrentColoring[SU->NodeNum] = *SUColors.begin(); |
1026 | } |
1027 | } |
1028 | |
1029 | void SIScheduleBlockCreator::colorMergeIfPossibleNextGroupOnlyForReserved() { |
1030 | unsigned DAGSize = DAG->SUnits.size(); |
1031 | |
1032 | for (unsigned SUNum : DAG->BottomUpIndex2SU) { |
1033 | SUnit *SU = &DAG->SUnits[SUNum]; |
1034 | std::set<unsigned> SUColors; |
1035 | |
1036 | if (CurrentColoring[SU->NodeNum] <= (int)DAGSize) |
1037 | continue; |
1038 | |
1039 | for (SDep& SuccDep : SU->Succs) { |
1040 | SUnit *Succ = SuccDep.getSUnit(); |
1041 | if (SuccDep.isWeak() || Succ->NodeNum >= DAGSize) |
1042 | continue; |
1043 | SUColors.insert(CurrentColoring[Succ->NodeNum]); |
1044 | } |
1045 | if (SUColors.size() == 1 && *SUColors.begin() <= DAGSize) |
1046 | CurrentColoring[SU->NodeNum] = *SUColors.begin(); |
1047 | } |
1048 | } |
1049 | |
1050 | void SIScheduleBlockCreator::colorMergeIfPossibleSmallGroupsToNextGroup() { |
1051 | unsigned DAGSize = DAG->SUnits.size(); |
1052 | std::map<unsigned, unsigned> ColorCount; |
1053 | |
1054 | for (unsigned SUNum : DAG->BottomUpIndex2SU) { |
1055 | SUnit *SU = &DAG->SUnits[SUNum]; |
1056 | unsigned color = CurrentColoring[SU->NodeNum]; |
1057 | ++ColorCount[color]; |
1058 | } |
1059 | |
1060 | for (unsigned SUNum : DAG->BottomUpIndex2SU) { |
1061 | SUnit *SU = &DAG->SUnits[SUNum]; |
1062 | unsigned color = CurrentColoring[SU->NodeNum]; |
1063 | std::set<unsigned> SUColors; |
1064 | |
1065 | if (CurrentColoring[SU->NodeNum] <= (int)DAGSize) |
1066 | continue; |
1067 | |
1068 | if (ColorCount[color] > 1) |
1069 | continue; |
1070 | |
1071 | for (SDep& SuccDep : SU->Succs) { |
1072 | SUnit *Succ = SuccDep.getSUnit(); |
1073 | if (SuccDep.isWeak() || Succ->NodeNum >= DAGSize) |
1074 | continue; |
1075 | SUColors.insert(CurrentColoring[Succ->NodeNum]); |
1076 | } |
1077 | if (SUColors.size() == 1 && *SUColors.begin() != color) { |
1078 | --ColorCount[color]; |
1079 | CurrentColoring[SU->NodeNum] = *SUColors.begin(); |
1080 | ++ColorCount[*SUColors.begin()]; |
1081 | } |
1082 | } |
1083 | } |
1084 | |
1085 | void SIScheduleBlockCreator::cutHugeBlocks() { |
1086 | // TODO |
1087 | } |
1088 | |
1089 | void SIScheduleBlockCreator::regroupNoUserInstructions() { |
1090 | unsigned DAGSize = DAG->SUnits.size(); |
1091 | int GroupID = NextNonReservedID++; |
1092 | |
1093 | for (unsigned SUNum : DAG->BottomUpIndex2SU) { |
1094 | SUnit *SU = &DAG->SUnits[SUNum]; |
1095 | bool hasSuccessor = false; |
1096 | |
1097 | if (CurrentColoring[SU->NodeNum] <= (int)DAGSize) |
1098 | continue; |
1099 | |
1100 | for (SDep& SuccDep : SU->Succs) { |
1101 | SUnit *Succ = SuccDep.getSUnit(); |
1102 | if (SuccDep.isWeak() || Succ->NodeNum >= DAGSize) |
1103 | continue; |
1104 | hasSuccessor = true; |
1105 | } |
1106 | if (!hasSuccessor) |
1107 | CurrentColoring[SU->NodeNum] = GroupID; |
1108 | } |
1109 | } |
1110 | |
1111 | void SIScheduleBlockCreator::colorExports() { |
1112 | unsigned ExportColor = NextNonReservedID++; |
1113 | SmallVector<unsigned, 8> ExpGroup; |
1114 | |
1115 | // Put all exports together in a block. |
1116 | // The block will naturally end up being scheduled last, |
1117 | // thus putting exports at the end of the schedule, which |
1118 | // is better for performance. |
1119 | // However we must ensure, for safety, the exports can be put |
1120 | // together in the same block without any other instruction. |
1121 | // This could happen, for example, when scheduling after regalloc |
1122 | // if reloading a spilled register from memory using the same |
1123 | // register than used in a previous export. |
1124 | // If that happens, do not regroup the exports. |
1125 | for (unsigned SUNum : DAG->TopDownIndex2SU) { |
1126 | const SUnit &SU = DAG->SUnits[SUNum]; |
1127 | if (SIInstrInfo::isEXP(*SU.getInstr())) { |
1128 | // Check the EXP can be added to the group safely, |
1129 | // ie without needing any other instruction. |
1130 | // The EXP is allowed to depend on other EXP |
1131 | // (they will be in the same group). |
1132 | for (unsigned j : ExpGroup) { |
1133 | bool HasSubGraph; |
1134 | std::vector<int> SubGraph; |
1135 | // By construction (topological order), if SU and |
1136 | // DAG->SUnits[j] are linked, DAG->SUnits[j] is neccessary |
1137 | // in the parent graph of SU. |
1138 | #ifndef NDEBUG1 |
1139 | SubGraph = DAG->GetTopo()->GetSubGraph(SU, DAG->SUnits[j], |
1140 | HasSubGraph); |
1141 | assert(!HasSubGraph)((void)0); |
1142 | #endif |
1143 | SubGraph = DAG->GetTopo()->GetSubGraph(DAG->SUnits[j], SU, |
1144 | HasSubGraph); |
1145 | if (!HasSubGraph) |
1146 | continue; // No dependencies between each other |
1147 | |
1148 | // SubGraph contains all the instructions required |
1149 | // between EXP SUnits[j] and EXP SU. |
1150 | for (unsigned k : SubGraph) { |
1151 | if (!SIInstrInfo::isEXP(*DAG->SUnits[k].getInstr())) |
1152 | // Other instructions than EXP would be required in the group. |
1153 | // Abort the groupping. |
1154 | return; |
1155 | } |
1156 | } |
1157 | |
1158 | ExpGroup.push_back(SUNum); |
1159 | } |
1160 | } |
1161 | |
1162 | // The group can be formed. Give the color. |
1163 | for (unsigned j : ExpGroup) |
1164 | CurrentColoring[j] = ExportColor; |
1165 | } |
1166 | |
1167 | void SIScheduleBlockCreator::createBlocksForVariant(SISchedulerBlockCreatorVariant BlockVariant) { |
1168 | unsigned DAGSize = DAG->SUnits.size(); |
1169 | std::map<unsigned,unsigned> RealID; |
1170 | |
1171 | CurrentBlocks.clear(); |
1172 | CurrentColoring.clear(); |
1173 | CurrentColoring.resize(DAGSize, 0); |
1174 | Node2CurrentBlock.clear(); |
1175 | |
1176 | // Restore links previous scheduling variant has overridden. |
1177 | DAG->restoreSULinksLeft(); |
1178 | |
1179 | NextReservedID = 1; |
1180 | NextNonReservedID = DAGSize + 1; |
1181 | |
1182 | LLVM_DEBUG(dbgs() << "Coloring the graph\n")do { } while (false); |
1183 | |
1184 | if (BlockVariant == SISchedulerBlockCreatorVariant::LatenciesGrouped) |
1185 | colorHighLatenciesGroups(); |
1186 | else |
1187 | colorHighLatenciesAlone(); |
1188 | colorComputeReservedDependencies(); |
1189 | colorAccordingToReservedDependencies(); |
1190 | colorEndsAccordingToDependencies(); |
1191 | if (BlockVariant == SISchedulerBlockCreatorVariant::LatenciesAlonePlusConsecutive) |
1192 | colorForceConsecutiveOrderInGroup(); |
1193 | regroupNoUserInstructions(); |
1194 | colorMergeConstantLoadsNextGroup(); |
1195 | colorMergeIfPossibleNextGroupOnlyForReserved(); |
1196 | colorExports(); |
1197 | |
1198 | // Put SUs of same color into same block |
1199 | Node2CurrentBlock.resize(DAGSize, -1); |
1200 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
1201 | SUnit *SU = &DAG->SUnits[i]; |
1202 | unsigned Color = CurrentColoring[SU->NodeNum]; |
1203 | if (RealID.find(Color) == RealID.end()) { |
1204 | int ID = CurrentBlocks.size(); |
1205 | BlockPtrs.push_back(std::make_unique<SIScheduleBlock>(DAG, this, ID)); |
1206 | CurrentBlocks.push_back(BlockPtrs.rbegin()->get()); |
1207 | RealID[Color] = ID; |
1208 | } |
1209 | CurrentBlocks[RealID[Color]]->addUnit(SU); |
1210 | Node2CurrentBlock[SU->NodeNum] = RealID[Color]; |
1211 | } |
1212 | |
1213 | // Build dependencies between blocks. |
1214 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
1215 | SUnit *SU = &DAG->SUnits[i]; |
1216 | int SUID = Node2CurrentBlock[i]; |
1217 | for (SDep& SuccDep : SU->Succs) { |
1218 | SUnit *Succ = SuccDep.getSUnit(); |
1219 | if (SuccDep.isWeak() || Succ->NodeNum >= DAGSize) |
1220 | continue; |
1221 | if (Node2CurrentBlock[Succ->NodeNum] != SUID) |
1222 | CurrentBlocks[SUID]->addSucc(CurrentBlocks[Node2CurrentBlock[Succ->NodeNum]], |
1223 | SuccDep.isCtrl() ? NoData : Data); |
1224 | } |
1225 | for (SDep& PredDep : SU->Preds) { |
1226 | SUnit *Pred = PredDep.getSUnit(); |
1227 | if (PredDep.isWeak() || Pred->NodeNum >= DAGSize) |
1228 | continue; |
1229 | if (Node2CurrentBlock[Pred->NodeNum] != SUID) |
1230 | CurrentBlocks[SUID]->addPred(CurrentBlocks[Node2CurrentBlock[Pred->NodeNum]]); |
1231 | } |
1232 | } |
1233 | |
1234 | // Free root and leafs of all blocks to enable scheduling inside them. |
1235 | for (unsigned i = 0, e = CurrentBlocks.size(); i != e; ++i) { |
1236 | SIScheduleBlock *Block = CurrentBlocks[i]; |
1237 | Block->finalizeUnits(); |
1238 | } |
1239 | LLVM_DEBUG(dbgs() << "Blocks created:\n\n";do { } while (false) |
1240 | for (unsigned i = 0, e = CurrentBlocks.size(); i != e; ++i) {do { } while (false) |
1241 | SIScheduleBlock *Block = CurrentBlocks[i];do { } while (false) |
1242 | Block->printDebug(true);do { } while (false) |
1243 | })do { } while (false); |
1244 | } |
1245 | |
1246 | // Two functions taken from Codegen/MachineScheduler.cpp |
1247 | |
1248 | /// Non-const version. |
1249 | static MachineBasicBlock::iterator |
1250 | nextIfDebug(MachineBasicBlock::iterator I, |
1251 | MachineBasicBlock::const_iterator End) { |
1252 | for (; I != End; ++I) { |
1253 | if (!I->isDebugInstr()) |
1254 | break; |
1255 | } |
1256 | return I; |
1257 | } |
1258 | |
1259 | void SIScheduleBlockCreator::topologicalSort() { |
1260 | unsigned DAGSize = CurrentBlocks.size(); |
1261 | std::vector<int> WorkList; |
1262 | |
1263 | LLVM_DEBUG(dbgs() << "Topological Sort\n")do { } while (false); |
1264 | |
1265 | WorkList.reserve(DAGSize); |
1266 | TopDownIndex2Block.resize(DAGSize); |
1267 | TopDownBlock2Index.resize(DAGSize); |
1268 | BottomUpIndex2Block.resize(DAGSize); |
1269 | |
1270 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
1271 | SIScheduleBlock *Block = CurrentBlocks[i]; |
1272 | unsigned Degree = Block->getSuccs().size(); |
1273 | TopDownBlock2Index[i] = Degree; |
1274 | if (Degree == 0) { |
1275 | WorkList.push_back(i); |
1276 | } |
1277 | } |
1278 | |
1279 | int Id = DAGSize; |
1280 | while (!WorkList.empty()) { |
1281 | int i = WorkList.back(); |
1282 | SIScheduleBlock *Block = CurrentBlocks[i]; |
1283 | WorkList.pop_back(); |
1284 | TopDownBlock2Index[i] = --Id; |
1285 | TopDownIndex2Block[Id] = i; |
1286 | for (SIScheduleBlock* Pred : Block->getPreds()) { |
1287 | if (!--TopDownBlock2Index[Pred->getID()]) |
1288 | WorkList.push_back(Pred->getID()); |
1289 | } |
1290 | } |
1291 | |
1292 | #ifndef NDEBUG1 |
1293 | // Check correctness of the ordering. |
1294 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
1295 | SIScheduleBlock *Block = CurrentBlocks[i]; |
1296 | for (SIScheduleBlock* Pred : Block->getPreds()) { |
1297 | assert(TopDownBlock2Index[i] > TopDownBlock2Index[Pred->getID()] &&((void)0) |
1298 | "Wrong Top Down topological sorting")((void)0); |
1299 | } |
1300 | } |
1301 | #endif |
1302 | |
1303 | BottomUpIndex2Block = std::vector<int>(TopDownIndex2Block.rbegin(), |
1304 | TopDownIndex2Block.rend()); |
1305 | } |
1306 | |
1307 | void SIScheduleBlockCreator::scheduleInsideBlocks() { |
1308 | unsigned DAGSize = CurrentBlocks.size(); |
1309 | |
1310 | LLVM_DEBUG(dbgs() << "\nScheduling Blocks\n\n")do { } while (false); |
1311 | |
1312 | // We do schedule a valid scheduling such that a Block corresponds |
1313 | // to a range of instructions. |
1314 | LLVM_DEBUG(dbgs() << "First phase: Fast scheduling for Reg Liveness\n")do { } while (false); |
1315 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
1316 | SIScheduleBlock *Block = CurrentBlocks[i]; |
1317 | Block->fastSchedule(); |
1318 | } |
1319 | |
1320 | // Note: the following code, and the part restoring previous position |
1321 | // is by far the most expensive operation of the Scheduler. |
1322 | |
1323 | // Do not update CurrentTop. |
1324 | MachineBasicBlock::iterator CurrentTopFastSched = DAG->getCurrentTop(); |
1325 | std::vector<MachineBasicBlock::iterator> PosOld; |
1326 | std::vector<MachineBasicBlock::iterator> PosNew; |
1327 | PosOld.reserve(DAG->SUnits.size()); |
1328 | PosNew.reserve(DAG->SUnits.size()); |
1329 | |
1330 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
1331 | int BlockIndice = TopDownIndex2Block[i]; |
1332 | SIScheduleBlock *Block = CurrentBlocks[BlockIndice]; |
1333 | std::vector<SUnit*> SUs = Block->getScheduledUnits(); |
1334 | |
1335 | for (SUnit* SU : SUs) { |
1336 | MachineInstr *MI = SU->getInstr(); |
1337 | MachineBasicBlock::iterator Pos = MI; |
1338 | PosOld.push_back(Pos); |
1339 | if (&*CurrentTopFastSched == MI) { |
1340 | PosNew.push_back(Pos); |
1341 | CurrentTopFastSched = nextIfDebug(++CurrentTopFastSched, |
1342 | DAG->getCurrentBottom()); |
1343 | } else { |
1344 | // Update the instruction stream. |
1345 | DAG->getBB()->splice(CurrentTopFastSched, DAG->getBB(), MI); |
1346 | |
1347 | // Update LiveIntervals. |
1348 | // Note: Moving all instructions and calling handleMove every time |
1349 | // is the most cpu intensive operation of the scheduler. |
1350 | // It would gain a lot if there was a way to recompute the |
1351 | // LiveIntervals for the entire scheduling region. |
1352 | DAG->getLIS()->handleMove(*MI, /*UpdateFlags=*/true); |
1353 | PosNew.push_back(CurrentTopFastSched); |
1354 | } |
1355 | } |
1356 | } |
1357 | |
1358 | // Now we have Block of SUs == Block of MI. |
1359 | // We do the final schedule for the instructions inside the block. |
1360 | // The property that all the SUs of the Block are grouped together as MI |
1361 | // is used for correct reg usage tracking. |
1362 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
1363 | SIScheduleBlock *Block = CurrentBlocks[i]; |
1364 | std::vector<SUnit*> SUs = Block->getScheduledUnits(); |
1365 | Block->schedule((*SUs.begin())->getInstr(), (*SUs.rbegin())->getInstr()); |
1366 | } |
1367 | |
1368 | LLVM_DEBUG(dbgs() << "Restoring MI Pos\n")do { } while (false); |
1369 | // Restore old ordering (which prevents a LIS->handleMove bug). |
1370 | for (unsigned i = PosOld.size(), e = 0; i != e; --i) { |
1371 | MachineBasicBlock::iterator POld = PosOld[i-1]; |
1372 | MachineBasicBlock::iterator PNew = PosNew[i-1]; |
1373 | if (PNew != POld) { |
1374 | // Update the instruction stream. |
1375 | DAG->getBB()->splice(POld, DAG->getBB(), PNew); |
1376 | |
1377 | // Update LiveIntervals. |
1378 | DAG->getLIS()->handleMove(*POld, /*UpdateFlags=*/true); |
1379 | } |
1380 | } |
1381 | |
1382 | LLVM_DEBUG(for (unsigned i = 0, e = CurrentBlocks.size(); i != e; ++i) {do { } while (false) |
1383 | SIScheduleBlock *Block = CurrentBlocks[i];do { } while (false) |
1384 | Block->printDebug(true);do { } while (false) |
1385 | })do { } while (false); |
1386 | } |
1387 | |
1388 | void SIScheduleBlockCreator::fillStats() { |
1389 | unsigned DAGSize = CurrentBlocks.size(); |
1390 | |
1391 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
1392 | int BlockIndice = TopDownIndex2Block[i]; |
1393 | SIScheduleBlock *Block = CurrentBlocks[BlockIndice]; |
1394 | if (Block->getPreds().empty()) |
1395 | Block->Depth = 0; |
1396 | else { |
1397 | unsigned Depth = 0; |
1398 | for (SIScheduleBlock *Pred : Block->getPreds()) { |
1399 | if (Depth < Pred->Depth + Pred->getCost()) |
1400 | Depth = Pred->Depth + Pred->getCost(); |
1401 | } |
1402 | Block->Depth = Depth; |
1403 | } |
1404 | } |
1405 | |
1406 | for (unsigned i = 0, e = DAGSize; i != e; ++i) { |
1407 | int BlockIndice = BottomUpIndex2Block[i]; |
1408 | SIScheduleBlock *Block = CurrentBlocks[BlockIndice]; |
1409 | if (Block->getSuccs().empty()) |
1410 | Block->Height = 0; |
1411 | else { |
1412 | unsigned Height = 0; |
1413 | for (const auto &Succ : Block->getSuccs()) |
1414 | Height = std::max(Height, Succ.first->Height + Succ.first->getCost()); |
1415 | Block->Height = Height; |
1416 | } |
1417 | } |
1418 | } |
1419 | |
1420 | // SIScheduleBlockScheduler // |
1421 | |
1422 | SIScheduleBlockScheduler::SIScheduleBlockScheduler(SIScheduleDAGMI *DAG, |
1423 | SISchedulerBlockSchedulerVariant Variant, |
1424 | SIScheduleBlocks BlocksStruct) : |
1425 | DAG(DAG), Variant(Variant), Blocks(BlocksStruct.Blocks), |
1426 | LastPosWaitedHighLatency(0), NumBlockScheduled(0), VregCurrentUsage(0), |
1427 | SregCurrentUsage(0), maxVregUsage(0), maxSregUsage(0) { |
1428 | |
1429 | // Fill the usage of every output |
1430 | // Warning: while by construction we always have a link between two blocks |
1431 | // when one needs a result from the other, the number of users of an output |
1432 | // is not the sum of child blocks having as input the same virtual register. |
1433 | // Here is an example. A produces x and y. B eats x and produces x'. |
1434 | // C eats x' and y. The register coalescer may have attributed the same |
1435 | // virtual register to x and x'. |
1436 | // To count accurately, we do a topological sort. In case the register is |
1437 | // found for several parents, we increment the usage of the one with the |
1438 | // highest topological index. |
1439 | LiveOutRegsNumUsages.resize(Blocks.size()); |
1440 | for (unsigned i = 0, e = Blocks.size(); i != e; ++i) { |
1441 | SIScheduleBlock *Block = Blocks[i]; |
1442 | for (unsigned Reg : Block->getInRegs()) { |
1443 | bool Found = false; |
1444 | int topoInd = -1; |
1445 | for (SIScheduleBlock* Pred: Block->getPreds()) { |
1446 | std::set<unsigned> PredOutRegs = Pred->getOutRegs(); |
1447 | std::set<unsigned>::iterator RegPos = PredOutRegs.find(Reg); |
1448 | |
1449 | if (RegPos != PredOutRegs.end()) { |
1450 | Found = true; |
1451 | if (topoInd < BlocksStruct.TopDownBlock2Index[Pred->getID()]) { |
1452 | topoInd = BlocksStruct.TopDownBlock2Index[Pred->getID()]; |
1453 | } |
1454 | } |
1455 | } |
1456 | |
1457 | if (!Found) |
1458 | continue; |
1459 | |
1460 | int PredID = BlocksStruct.TopDownIndex2Block[topoInd]; |
1461 | ++LiveOutRegsNumUsages[PredID][Reg]; |
1462 | } |
1463 | } |
1464 | |
1465 | LastPosHighLatencyParentScheduled.resize(Blocks.size(), 0); |
1466 | BlockNumPredsLeft.resize(Blocks.size()); |
1467 | BlockNumSuccsLeft.resize(Blocks.size()); |
1468 | |
1469 | for (unsigned i = 0, e = Blocks.size(); i != e; ++i) { |
1470 | SIScheduleBlock *Block = Blocks[i]; |
1471 | BlockNumPredsLeft[i] = Block->getPreds().size(); |
1472 | BlockNumSuccsLeft[i] = Block->getSuccs().size(); |
1473 | } |
1474 | |
1475 | #ifndef NDEBUG1 |
1476 | for (unsigned i = 0, e = Blocks.size(); i != e; ++i) { |
1477 | SIScheduleBlock *Block = Blocks[i]; |
1478 | assert(Block->getID() == i)((void)0); |
1479 | } |
1480 | #endif |
1481 | |
1482 | std::set<unsigned> InRegs = DAG->getInRegs(); |
1483 | addLiveRegs(InRegs); |
1484 | |
1485 | // Increase LiveOutRegsNumUsages for blocks |
1486 | // producing registers consumed in another |
1487 | // scheduling region. |
1488 | for (unsigned Reg : DAG->getOutRegs()) { |
1489 | for (unsigned i = 0, e = Blocks.size(); i != e; ++i) { |
1490 | // Do reverse traversal |
1491 | int ID = BlocksStruct.TopDownIndex2Block[Blocks.size()-1-i]; |
1492 | SIScheduleBlock *Block = Blocks[ID]; |
1493 | const std::set<unsigned> &OutRegs = Block->getOutRegs(); |
1494 | |
1495 | if (OutRegs.find(Reg) == OutRegs.end()) |
1496 | continue; |
1497 | |
1498 | ++LiveOutRegsNumUsages[ID][Reg]; |
1499 | break; |
1500 | } |
1501 | } |
1502 | |
1503 | // Fill LiveRegsConsumers for regs that were already |
1504 | // defined before scheduling. |
1505 | for (unsigned i = 0, e = Blocks.size(); i != e; ++i) { |
1506 | SIScheduleBlock *Block = Blocks[i]; |
1507 | for (unsigned Reg : Block->getInRegs()) { |
1508 | bool Found = false; |
1509 | for (SIScheduleBlock* Pred: Block->getPreds()) { |
1510 | std::set<unsigned> PredOutRegs = Pred->getOutRegs(); |
1511 | std::set<unsigned>::iterator RegPos = PredOutRegs.find(Reg); |
1512 | |
1513 | if (RegPos != PredOutRegs.end()) { |
1514 | Found = true; |
1515 | break; |
1516 | } |
1517 | } |
1518 | |
1519 | if (!Found) |
1520 | ++LiveRegsConsumers[Reg]; |
1521 | } |
1522 | } |
1523 | |
1524 | for (unsigned i = 0, e = Blocks.size(); i != e; ++i) { |
1525 | SIScheduleBlock *Block = Blocks[i]; |
1526 | if (BlockNumPredsLeft[i] == 0) { |
1527 | ReadyBlocks.push_back(Block); |
1528 | } |
1529 | } |
1530 | |
1531 | while (SIScheduleBlock *Block = pickBlock()) { |
1532 | BlocksScheduled.push_back(Block); |
1533 | blockScheduled(Block); |
1534 | } |
1535 | |
1536 | LLVM_DEBUG(dbgs() << "Block Order:"; for (SIScheduleBlock *Blockdo { } while (false) |
1537 | : BlocksScheduled) {do { } while (false) |
1538 | dbgs() << ' ' << Block->getID();do { } while (false) |
1539 | } dbgs() << '\n';)do { } while (false); |
1540 | } |
1541 | |
1542 | bool SIScheduleBlockScheduler::tryCandidateLatency(SIBlockSchedCandidate &Cand, |
1543 | SIBlockSchedCandidate &TryCand) { |
1544 | if (!Cand.isValid()) { |
1545 | TryCand.Reason = NodeOrder; |
1546 | return true; |
1547 | } |
1548 | |
1549 | // Try to hide high latencies. |
1550 | if (SISched::tryLess(TryCand.LastPosHighLatParentScheduled, |
1551 | Cand.LastPosHighLatParentScheduled, TryCand, Cand, Latency)) |
1552 | return true; |
1553 | // Schedule high latencies early so you can hide them better. |
1554 | if (SISched::tryGreater(TryCand.IsHighLatency, Cand.IsHighLatency, |
1555 | TryCand, Cand, Latency)) |
1556 | return true; |
1557 | if (TryCand.IsHighLatency && SISched::tryGreater(TryCand.Height, Cand.Height, |
1558 | TryCand, Cand, Depth)) |
1559 | return true; |
1560 | if (SISched::tryGreater(TryCand.NumHighLatencySuccessors, |
1561 | Cand.NumHighLatencySuccessors, |
1562 | TryCand, Cand, Successor)) |
1563 | return true; |
1564 | return false; |
1565 | } |
1566 | |
1567 | bool SIScheduleBlockScheduler::tryCandidateRegUsage(SIBlockSchedCandidate &Cand, |
1568 | SIBlockSchedCandidate &TryCand) { |
1569 | if (!Cand.isValid()) { |
1570 | TryCand.Reason = NodeOrder; |
1571 | return true; |
1572 | } |
1573 | |
1574 | if (SISched::tryLess(TryCand.VGPRUsageDiff > 0, Cand.VGPRUsageDiff > 0, |
1575 | TryCand, Cand, RegUsage)) |
1576 | return true; |
1577 | if (SISched::tryGreater(TryCand.NumSuccessors > 0, |
1578 | Cand.NumSuccessors > 0, |
1579 | TryCand, Cand, Successor)) |
1580 | return true; |
1581 | if (SISched::tryGreater(TryCand.Height, Cand.Height, TryCand, Cand, Depth)) |
1582 | return true; |
1583 | if (SISched::tryLess(TryCand.VGPRUsageDiff, Cand.VGPRUsageDiff, |
1584 | TryCand, Cand, RegUsage)) |
1585 | return true; |
1586 | return false; |
1587 | } |
1588 | |
1589 | SIScheduleBlock *SIScheduleBlockScheduler::pickBlock() { |
1590 | SIBlockSchedCandidate Cand; |
1591 | std::vector<SIScheduleBlock*>::iterator Best; |
1592 | SIScheduleBlock *Block; |
1593 | if (ReadyBlocks.empty()) |
1594 | return nullptr; |
1595 | |
1596 | DAG->fillVgprSgprCost(LiveRegs.begin(), LiveRegs.end(), |
1597 | VregCurrentUsage, SregCurrentUsage); |
1598 | if (VregCurrentUsage > maxVregUsage) |
1599 | maxVregUsage = VregCurrentUsage; |
1600 | if (SregCurrentUsage > maxSregUsage) |
1601 | maxSregUsage = SregCurrentUsage; |
1602 | LLVM_DEBUG(dbgs() << "Picking New Blocks\n"; dbgs() << "Available: ";do { } while (false) |
1603 | for (SIScheduleBlock *Blockdo { } while (false) |
1604 | : ReadyBlocks) dbgs()do { } while (false) |
1605 | << Block->getID() << ' ';do { } while (false) |
1606 | dbgs() << "\nCurrent Live:\n";do { } while (false) |
1607 | for (unsigned Regdo { } while (false) |
1608 | : LiveRegs) dbgs()do { } while (false) |
1609 | << printVRegOrUnit(Reg, DAG->getTRI()) << ' ';do { } while (false) |
1610 | dbgs() << '\n';do { } while (false) |
1611 | dbgs() << "Current VGPRs: " << VregCurrentUsage << '\n';do { } while (false) |
1612 | dbgs() << "Current SGPRs: " << SregCurrentUsage << '\n';)do { } while (false); |
1613 | |
1614 | Cand.Block = nullptr; |
1615 | for (std::vector<SIScheduleBlock*>::iterator I = ReadyBlocks.begin(), |
1616 | E = ReadyBlocks.end(); I != E; ++I) { |
1617 | SIBlockSchedCandidate TryCand; |
1618 | TryCand.Block = *I; |
1619 | TryCand.IsHighLatency = TryCand.Block->isHighLatencyBlock(); |
1620 | TryCand.VGPRUsageDiff = |
1621 | checkRegUsageImpact(TryCand.Block->getInRegs(), |
1622 | TryCand.Block->getOutRegs())[AMDGPU::RegisterPressureSets::VGPR_32]; |
1623 | TryCand.NumSuccessors = TryCand.Block->getSuccs().size(); |
1624 | TryCand.NumHighLatencySuccessors = |
1625 | TryCand.Block->getNumHighLatencySuccessors(); |
1626 | TryCand.LastPosHighLatParentScheduled = |
1627 | (unsigned int) std::max<int> (0, |
1628 | LastPosHighLatencyParentScheduled[TryCand.Block->getID()] - |
1629 | LastPosWaitedHighLatency); |
1630 | TryCand.Height = TryCand.Block->Height; |
1631 | // Try not to increase VGPR usage too much, else we may spill. |
1632 | if (VregCurrentUsage > 120 || |
1633 | Variant != SISchedulerBlockSchedulerVariant::BlockLatencyRegUsage) { |
1634 | if (!tryCandidateRegUsage(Cand, TryCand) && |
1635 | Variant != SISchedulerBlockSchedulerVariant::BlockRegUsage) |
1636 | tryCandidateLatency(Cand, TryCand); |
1637 | } else { |
1638 | if (!tryCandidateLatency(Cand, TryCand)) |
1639 | tryCandidateRegUsage(Cand, TryCand); |
1640 | } |
1641 | if (TryCand.Reason != NoCand) { |
1642 | Cand.setBest(TryCand); |
1643 | Best = I; |
1644 | LLVM_DEBUG(dbgs() << "Best Current Choice: " << Cand.Block->getID() << ' 'do { } while (false) |
1645 | << getReasonStr(Cand.Reason) << '\n')do { } while (false); |
1646 | } |
1647 | } |
1648 | |
1649 | LLVM_DEBUG(dbgs() << "Picking: " << Cand.Block->getID() << '\n';do { } while (false) |
1650 | dbgs() << "Is a block with high latency instruction: "do { } while (false) |
1651 | << (Cand.IsHighLatency ? "yes\n" : "no\n");do { } while (false) |
1652 | dbgs() << "Position of last high latency dependency: "do { } while (false) |
1653 | << Cand.LastPosHighLatParentScheduled << '\n';do { } while (false) |
1654 | dbgs() << "VGPRUsageDiff: " << Cand.VGPRUsageDiff << '\n';do { } while (false) |
1655 | dbgs() << '\n';)do { } while (false); |
1656 | |
1657 | Block = Cand.Block; |
1658 | ReadyBlocks.erase(Best); |
1659 | return Block; |
1660 | } |
1661 | |
1662 | // Tracking of currently alive registers to determine VGPR Usage. |
1663 | |
1664 | void SIScheduleBlockScheduler::addLiveRegs(std::set<unsigned> &Regs) { |
1665 | for (Register Reg : Regs) { |
1666 | // For now only track virtual registers. |
1667 | if (!Reg.isVirtual()) |
1668 | continue; |
1669 | // If not already in the live set, then add it. |
1670 | (void) LiveRegs.insert(Reg); |
1671 | } |
1672 | } |
1673 | |
1674 | void SIScheduleBlockScheduler::decreaseLiveRegs(SIScheduleBlock *Block, |
1675 | std::set<unsigned> &Regs) { |
1676 | for (unsigned Reg : Regs) { |
1677 | // For now only track virtual registers. |
1678 | std::set<unsigned>::iterator Pos = LiveRegs.find(Reg); |
1679 | assert (Pos != LiveRegs.end() && // Reg must be live.((void)0) |
1680 | LiveRegsConsumers.find(Reg) != LiveRegsConsumers.end() &&((void)0) |
1681 | LiveRegsConsumers[Reg] >= 1)((void)0); |
1682 | --LiveRegsConsumers[Reg]; |
1683 | if (LiveRegsConsumers[Reg] == 0) |
1684 | LiveRegs.erase(Pos); |
1685 | } |
1686 | } |
1687 | |
1688 | void SIScheduleBlockScheduler::releaseBlockSuccs(SIScheduleBlock *Parent) { |
1689 | for (const auto &Block : Parent->getSuccs()) { |
1690 | if (--BlockNumPredsLeft[Block.first->getID()] == 0) |
1691 | ReadyBlocks.push_back(Block.first); |
1692 | |
1693 | if (Parent->isHighLatencyBlock() && |
1694 | Block.second == SIScheduleBlockLinkKind::Data) |
1695 | LastPosHighLatencyParentScheduled[Block.first->getID()] = NumBlockScheduled; |
1696 | } |
1697 | } |
1698 | |
1699 | void SIScheduleBlockScheduler::blockScheduled(SIScheduleBlock *Block) { |
1700 | decreaseLiveRegs(Block, Block->getInRegs()); |
1701 | addLiveRegs(Block->getOutRegs()); |
1702 | releaseBlockSuccs(Block); |
1703 | for (std::map<unsigned, unsigned>::iterator RegI = |
1704 | LiveOutRegsNumUsages[Block->getID()].begin(), |
1705 | E = LiveOutRegsNumUsages[Block->getID()].end(); RegI != E; ++RegI) { |
1706 | std::pair<unsigned, unsigned> RegP = *RegI; |
1707 | // We produce this register, thus it must not be previously alive. |
1708 | assert(LiveRegsConsumers.find(RegP.first) == LiveRegsConsumers.end() ||((void)0) |
1709 | LiveRegsConsumers[RegP.first] == 0)((void)0); |
1710 | LiveRegsConsumers[RegP.first] += RegP.second; |
1711 | } |
1712 | if (LastPosHighLatencyParentScheduled[Block->getID()] > |
1713 | (unsigned)LastPosWaitedHighLatency) |
1714 | LastPosWaitedHighLatency = |
1715 | LastPosHighLatencyParentScheduled[Block->getID()]; |
1716 | ++NumBlockScheduled; |
1717 | } |
1718 | |
1719 | std::vector<int> |
1720 | SIScheduleBlockScheduler::checkRegUsageImpact(std::set<unsigned> &InRegs, |
1721 | std::set<unsigned> &OutRegs) { |
1722 | std::vector<int> DiffSetPressure; |
1723 | DiffSetPressure.assign(DAG->getTRI()->getNumRegPressureSets(), 0); |
1724 | |
1725 | for (Register Reg : InRegs) { |
1726 | // For now only track virtual registers. |
1727 | if (!Reg.isVirtual()) |
1728 | continue; |
1729 | if (LiveRegsConsumers[Reg] > 1) |
1730 | continue; |
1731 | PSetIterator PSetI = DAG->getMRI()->getPressureSets(Reg); |
1732 | for (; PSetI.isValid(); ++PSetI) { |
1733 | DiffSetPressure[*PSetI] -= PSetI.getWeight(); |
1734 | } |
1735 | } |
1736 | |
1737 | for (Register Reg : OutRegs) { |
1738 | // For now only track virtual registers. |
1739 | if (!Reg.isVirtual()) |
1740 | continue; |
1741 | PSetIterator PSetI = DAG->getMRI()->getPressureSets(Reg); |
1742 | for (; PSetI.isValid(); ++PSetI) { |
1743 | DiffSetPressure[*PSetI] += PSetI.getWeight(); |
1744 | } |
1745 | } |
1746 | |
1747 | return DiffSetPressure; |
1748 | } |
1749 | |
1750 | // SIScheduler // |
1751 | |
1752 | struct SIScheduleBlockResult |
1753 | SIScheduler::scheduleVariant(SISchedulerBlockCreatorVariant BlockVariant, |
1754 | SISchedulerBlockSchedulerVariant ScheduleVariant) { |
1755 | SIScheduleBlocks Blocks = BlockCreator.getBlocks(BlockVariant); |
1756 | SIScheduleBlockScheduler Scheduler(DAG, ScheduleVariant, Blocks); |
1757 | std::vector<SIScheduleBlock*> ScheduledBlocks; |
1758 | struct SIScheduleBlockResult Res; |
1759 | |
1760 | ScheduledBlocks = Scheduler.getBlocks(); |
1761 | |
1762 | for (unsigned b = 0; b < ScheduledBlocks.size(); ++b) { |
1763 | SIScheduleBlock *Block = ScheduledBlocks[b]; |
1764 | std::vector<SUnit*> SUs = Block->getScheduledUnits(); |
1765 | |
1766 | for (SUnit* SU : SUs) |
1767 | Res.SUs.push_back(SU->NodeNum); |
1768 | } |
1769 | |
1770 | Res.MaxSGPRUsage = Scheduler.getSGPRUsage(); |
1771 | Res.MaxVGPRUsage = Scheduler.getVGPRUsage(); |
1772 | return Res; |
1773 | } |
1774 | |
1775 | // SIScheduleDAGMI // |
1776 | |
1777 | SIScheduleDAGMI::SIScheduleDAGMI(MachineSchedContext *C) : |
1778 | ScheduleDAGMILive(C, std::make_unique<GenericScheduler>(C)) { |
1779 | SITII = static_cast<const SIInstrInfo*>(TII); |
1780 | SITRI = static_cast<const SIRegisterInfo*>(TRI); |
1781 | } |
1782 | |
1783 | SIScheduleDAGMI::~SIScheduleDAGMI() = default; |
1784 | |
1785 | // Code adapted from scheduleDAG.cpp |
1786 | // Does a topological sort over the SUs. |
1787 | // Both TopDown and BottomUp |
1788 | void SIScheduleDAGMI::topologicalSort() { |
1789 | Topo.InitDAGTopologicalSorting(); |
1790 | |
1791 | TopDownIndex2SU = std::vector<int>(Topo.begin(), Topo.end()); |
1792 | BottomUpIndex2SU = std::vector<int>(Topo.rbegin(), Topo.rend()); |
1793 | } |
1794 | |
1795 | // Move low latencies further from their user without |
1796 | // increasing SGPR usage (in general) |
1797 | // This is to be replaced by a better pass that would |
1798 | // take into account SGPR usage (based on VGPR Usage |
1799 | // and the corresponding wavefront count), that would |
1800 | // try to merge groups of loads if it make sense, etc |
1801 | void SIScheduleDAGMI::moveLowLatencies() { |
1802 | unsigned DAGSize = SUnits.size(); |
1803 | int LastLowLatencyUser = -1; |
1804 | int LastLowLatencyPos = -1; |
1805 | |
1806 | for (unsigned i = 0, e = ScheduledSUnits.size(); i != e; ++i) { |
1807 | SUnit *SU = &SUnits[ScheduledSUnits[i]]; |
1808 | bool IsLowLatencyUser = false; |
1809 | unsigned MinPos = 0; |
1810 | |
1811 | for (SDep& PredDep : SU->Preds) { |
1812 | SUnit *Pred = PredDep.getSUnit(); |
1813 | if (SITII->isLowLatencyInstruction(*Pred->getInstr())) { |
1814 | IsLowLatencyUser = true; |
1815 | } |
1816 | if (Pred->NodeNum >= DAGSize) |
1817 | continue; |
1818 | unsigned PredPos = ScheduledSUnitsInv[Pred->NodeNum]; |
1819 | if (PredPos >= MinPos) |
1820 | MinPos = PredPos + 1; |
1821 | } |
1822 | |
1823 | if (SITII->isLowLatencyInstruction(*SU->getInstr())) { |
1824 | unsigned BestPos = LastLowLatencyUser + 1; |
1825 | if ((int)BestPos <= LastLowLatencyPos) |
1826 | BestPos = LastLowLatencyPos + 1; |
1827 | if (BestPos < MinPos) |
1828 | BestPos = MinPos; |
1829 | if (BestPos < i) { |
1830 | for (unsigned u = i; u > BestPos; --u) { |
1831 | ++ScheduledSUnitsInv[ScheduledSUnits[u-1]]; |
1832 | ScheduledSUnits[u] = ScheduledSUnits[u-1]; |
1833 | } |
1834 | ScheduledSUnits[BestPos] = SU->NodeNum; |
1835 | ScheduledSUnitsInv[SU->NodeNum] = BestPos; |
1836 | } |
1837 | LastLowLatencyPos = BestPos; |
1838 | if (IsLowLatencyUser) |
1839 | LastLowLatencyUser = BestPos; |
1840 | } else if (IsLowLatencyUser) { |
1841 | LastLowLatencyUser = i; |
1842 | // Moves COPY instructions on which depends |
1843 | // the low latency instructions too. |
1844 | } else if (SU->getInstr()->getOpcode() == AMDGPU::COPY) { |
1845 | bool CopyForLowLat = false; |
1846 | for (SDep& SuccDep : SU->Succs) { |
1847 | SUnit *Succ = SuccDep.getSUnit(); |
1848 | if (SuccDep.isWeak() || Succ->NodeNum >= DAGSize) |
1849 | continue; |
1850 | if (SITII->isLowLatencyInstruction(*Succ->getInstr())) { |
1851 | CopyForLowLat = true; |
1852 | } |
1853 | } |
1854 | if (!CopyForLowLat) |
1855 | continue; |
1856 | if (MinPos < i) { |
1857 | for (unsigned u = i; u > MinPos; --u) { |
1858 | ++ScheduledSUnitsInv[ScheduledSUnits[u-1]]; |
1859 | ScheduledSUnits[u] = ScheduledSUnits[u-1]; |
1860 | } |
1861 | ScheduledSUnits[MinPos] = SU->NodeNum; |
1862 | ScheduledSUnitsInv[SU->NodeNum] = MinPos; |
1863 | } |
1864 | } |
1865 | } |
1866 | } |
1867 | |
1868 | void SIScheduleDAGMI::restoreSULinksLeft() { |
1869 | for (unsigned i = 0, e = SUnits.size(); i != e; ++i) { |
1870 | SUnits[i].isScheduled = false; |
1871 | SUnits[i].WeakPredsLeft = SUnitsLinksBackup[i].WeakPredsLeft; |
1872 | SUnits[i].NumPredsLeft = SUnitsLinksBackup[i].NumPredsLeft; |
1873 | SUnits[i].WeakSuccsLeft = SUnitsLinksBackup[i].WeakSuccsLeft; |
1874 | SUnits[i].NumSuccsLeft = SUnitsLinksBackup[i].NumSuccsLeft; |
1875 | } |
1876 | } |
1877 | |
1878 | // Return the Vgpr and Sgpr usage corresponding to some virtual registers. |
1879 | template<typename _Iterator> void |
1880 | SIScheduleDAGMI::fillVgprSgprCost(_Iterator First, _Iterator End, |
1881 | unsigned &VgprUsage, unsigned &SgprUsage) { |
1882 | VgprUsage = 0; |
1883 | SgprUsage = 0; |
1884 | for (_Iterator RegI = First; RegI != End; ++RegI) { |
1885 | Register Reg = *RegI; |
1886 | // For now only track virtual registers |
1887 | if (!Reg.isVirtual()) |
1888 | continue; |
1889 | PSetIterator PSetI = MRI.getPressureSets(Reg); |
1890 | for (; PSetI.isValid(); ++PSetI) { |
1891 | if (*PSetI == AMDGPU::RegisterPressureSets::VGPR_32) |
1892 | VgprUsage += PSetI.getWeight(); |
1893 | else if (*PSetI == AMDGPU::RegisterPressureSets::SReg_32) |
1894 | SgprUsage += PSetI.getWeight(); |
1895 | } |
1896 | } |
1897 | } |
1898 | |
1899 | void SIScheduleDAGMI::schedule() |
1900 | { |
1901 | SmallVector<SUnit*, 8> TopRoots, BotRoots; |
1902 | SIScheduleBlockResult Best, Temp; |
1903 | LLVM_DEBUG(dbgs() << "Preparing Scheduling\n")do { } while (false); |
1904 | |
1905 | buildDAGWithRegPressure(); |
1906 | LLVM_DEBUG(dump())do { } while (false); |
1907 | |
1908 | topologicalSort(); |
1909 | findRootsAndBiasEdges(TopRoots, BotRoots); |
1910 | // We reuse several ScheduleDAGMI and ScheduleDAGMILive |
1911 | // functions, but to make them happy we must initialize |
1912 | // the default Scheduler implementation (even if we do not |
1913 | // run it) |
1914 | SchedImpl->initialize(this); |
1915 | initQueues(TopRoots, BotRoots); |
1916 | |
1917 | // Fill some stats to help scheduling. |
1918 | |
1919 | SUnitsLinksBackup = SUnits; |
1920 | IsLowLatencySU.clear(); |
1921 | LowLatencyOffset.clear(); |
1922 | IsHighLatencySU.clear(); |
1923 | |
1924 | IsLowLatencySU.resize(SUnits.size(), 0); |
1925 | LowLatencyOffset.resize(SUnits.size(), 0); |
1926 | IsHighLatencySU.resize(SUnits.size(), 0); |
1927 | |
1928 | for (unsigned i = 0, e = (unsigned)SUnits.size(); i != e; ++i) { |
1929 | SUnit *SU = &SUnits[i]; |
1930 | const MachineOperand *BaseLatOp; |
1931 | int64_t OffLatReg; |
1932 | if (SITII->isLowLatencyInstruction(*SU->getInstr())) { |
1933 | IsLowLatencySU[i] = 1; |
1934 | bool OffsetIsScalable; |
1935 | if (SITII->getMemOperandWithOffset(*SU->getInstr(), BaseLatOp, OffLatReg, |
1936 | OffsetIsScalable, TRI)) |
1937 | LowLatencyOffset[i] = OffLatReg; |
1938 | } else if (SITII->isHighLatencyDef(SU->getInstr()->getOpcode())) |
1939 | IsHighLatencySU[i] = 1; |
1940 | } |
1941 | |
1942 | SIScheduler Scheduler(this); |
1943 | Best = Scheduler.scheduleVariant(SISchedulerBlockCreatorVariant::LatenciesAlone, |
1944 | SISchedulerBlockSchedulerVariant::BlockLatencyRegUsage); |
1945 | |
1946 | // if VGPR usage is extremely high, try other good performing variants |
1947 | // which could lead to lower VGPR usage |
1948 | if (Best.MaxVGPRUsage > 180) { |
1949 | static const std::pair<SISchedulerBlockCreatorVariant, |
1950 | SISchedulerBlockSchedulerVariant> |
1951 | Variants[] = { |
1952 | { LatenciesAlone, BlockRegUsageLatency }, |
1953 | // { LatenciesAlone, BlockRegUsage }, |
1954 | { LatenciesGrouped, BlockLatencyRegUsage }, |
1955 | // { LatenciesGrouped, BlockRegUsageLatency }, |
1956 | // { LatenciesGrouped, BlockRegUsage }, |
1957 | { LatenciesAlonePlusConsecutive, BlockLatencyRegUsage }, |
1958 | // { LatenciesAlonePlusConsecutive, BlockRegUsageLatency }, |
1959 | // { LatenciesAlonePlusConsecutive, BlockRegUsage } |
1960 | }; |
1961 | for (std::pair<SISchedulerBlockCreatorVariant, SISchedulerBlockSchedulerVariant> v : Variants) { |
1962 | Temp = Scheduler.scheduleVariant(v.first, v.second); |
1963 | if (Temp.MaxVGPRUsage < Best.MaxVGPRUsage) |
1964 | Best = Temp; |
1965 | } |
1966 | } |
1967 | // if VGPR usage is still extremely high, we may spill. Try other variants |
1968 | // which are less performing, but that could lead to lower VGPR usage. |
1969 | if (Best.MaxVGPRUsage > 200) { |
1970 | static const std::pair<SISchedulerBlockCreatorVariant, |
1971 | SISchedulerBlockSchedulerVariant> |
1972 | Variants[] = { |
1973 | // { LatenciesAlone, BlockRegUsageLatency }, |
1974 | { LatenciesAlone, BlockRegUsage }, |
1975 | // { LatenciesGrouped, BlockLatencyRegUsage }, |
1976 | { LatenciesGrouped, BlockRegUsageLatency }, |
1977 | { LatenciesGrouped, BlockRegUsage }, |
1978 | // { LatenciesAlonePlusConsecutive, BlockLatencyRegUsage }, |
1979 | { LatenciesAlonePlusConsecutive, BlockRegUsageLatency }, |
1980 | { LatenciesAlonePlusConsecutive, BlockRegUsage } |
1981 | }; |
1982 | for (std::pair<SISchedulerBlockCreatorVariant, SISchedulerBlockSchedulerVariant> v : Variants) { |
1983 | Temp = Scheduler.scheduleVariant(v.first, v.second); |
1984 | if (Temp.MaxVGPRUsage < Best.MaxVGPRUsage) |
1985 | Best = Temp; |
1986 | } |
1987 | } |
1988 | |
1989 | ScheduledSUnits = Best.SUs; |
1990 | ScheduledSUnitsInv.resize(SUnits.size()); |
1991 | |
1992 | for (unsigned i = 0, e = (unsigned)SUnits.size(); i != e; ++i) { |
1993 | ScheduledSUnitsInv[ScheduledSUnits[i]] = i; |
1994 | } |
1995 | |
1996 | moveLowLatencies(); |
1997 | |
1998 | // Tell the outside world about the result of the scheduling. |
1999 | |
2000 | assert(TopRPTracker.getPos() == RegionBegin && "bad initial Top tracker")((void)0); |
2001 | TopRPTracker.setPos(CurrentTop); |
2002 | |
2003 | for (std::vector<unsigned>::iterator I = ScheduledSUnits.begin(), |
2004 | E = ScheduledSUnits.end(); I != E; ++I) { |
2005 | SUnit *SU = &SUnits[*I]; |
2006 | |
2007 | scheduleMI(SU, true); |
2008 | |
2009 | LLVM_DEBUG(dbgs() << "Scheduling SU(" << SU->NodeNum << ") "do { } while (false) |
2010 | << *SU->getInstr())do { } while (false); |
2011 | } |
2012 | |
2013 | assert(CurrentTop == CurrentBottom && "Nonempty unscheduled zone.")((void)0); |
2014 | |
2015 | placeDebugValues(); |
2016 | |
2017 | LLVM_DEBUG({do { } while (false) |
2018 | dbgs() << "*** Final schedule for "do { } while (false) |
2019 | << printMBBReference(*begin()->getParent()) << " ***\n";do { } while (false) |
2020 | dumpSchedule();do { } while (false) |
2021 | dbgs() << '\n';do { } while (false) |
2022 | })do { } while (false); |
2023 | } |