1 //===- MachineSink.cpp - Sinking for machine instructions -----------------===//
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 // This pass moves instructions into successor blocks when possible, so that
10 // they aren't executed on paths where their results aren't needed.
11 //
12 // This pass is not intended to be a replacement or a complete alternative
13 // for an LLVM-IR-level sinking pass. It is only designed to sink simple
14 // constructs that are not exposed before lowering and instruction selection.
15 //
16 //===----------------------------------------------------------------------===//
17 
18 #include "llvm/ADT/DenseSet.h"
19 #include "llvm/ADT/DepthFirstIterator.h"
20 #include "llvm/ADT/MapVector.h"
21 #include "llvm/ADT/PointerIntPair.h"
22 #include "llvm/ADT/PostOrderIterator.h"
23 #include "llvm/ADT/SetVector.h"
24 #include "llvm/ADT/SmallSet.h"
25 #include "llvm/ADT/SmallVector.h"
26 #include "llvm/ADT/Statistic.h"
27 #include "llvm/Analysis/AliasAnalysis.h"
28 #include "llvm/Analysis/CFG.h"
29 #include "llvm/CodeGen/MachineBasicBlock.h"
30 #include "llvm/CodeGen/MachineBlockFrequencyInfo.h"
31 #include "llvm/CodeGen/MachineBranchProbabilityInfo.h"
32 #include "llvm/CodeGen/MachineCycleAnalysis.h"
33 #include "llvm/CodeGen/MachineDominators.h"
34 #include "llvm/CodeGen/MachineFunction.h"
35 #include "llvm/CodeGen/MachineFunctionPass.h"
36 #include "llvm/CodeGen/MachineInstr.h"
37 #include "llvm/CodeGen/MachineLoopInfo.h"
38 #include "llvm/CodeGen/MachineOperand.h"
39 #include "llvm/CodeGen/MachinePostDominators.h"
40 #include "llvm/CodeGen/MachineRegisterInfo.h"
41 #include "llvm/CodeGen/RegisterClassInfo.h"
42 #include "llvm/CodeGen/RegisterPressure.h"
43 #include "llvm/CodeGen/TargetInstrInfo.h"
44 #include "llvm/CodeGen/TargetRegisterInfo.h"
45 #include "llvm/CodeGen/TargetSubtargetInfo.h"
46 #include "llvm/IR/BasicBlock.h"
47 #include "llvm/IR/DebugInfoMetadata.h"
48 #include "llvm/IR/LLVMContext.h"
49 #include "llvm/InitializePasses.h"
50 #include "llvm/MC/MCRegisterInfo.h"
51 #include "llvm/Pass.h"
52 #include "llvm/Support/BranchProbability.h"
53 #include "llvm/Support/CommandLine.h"
54 #include "llvm/Support/Debug.h"
55 #include "llvm/Support/raw_ostream.h"
56 #include <algorithm>
57 #include <cassert>
58 #include <cstdint>
59 #include <map>
60 #include <utility>
61 #include <vector>
62 
63 using namespace llvm;
64 
65 #define DEBUG_TYPE "machine-sink"
66 
67 static cl::opt<bool>
68 SplitEdges("machine-sink-split",
69            cl::desc("Split critical edges during machine sinking"),
70            cl::init(true), cl::Hidden);
71 
72 static cl::opt<bool>
73 UseBlockFreqInfo("machine-sink-bfi",
74            cl::desc("Use block frequency info to find successors to sink"),
75            cl::init(true), cl::Hidden);
76 
77 static cl::opt<unsigned> SplitEdgeProbabilityThreshold(
78     "machine-sink-split-probability-threshold",
79     cl::desc(
80         "Percentage threshold for splitting single-instruction critical edge. "
81         "If the branch threshold is higher than this threshold, we allow "
82         "speculative execution of up to 1 instruction to avoid branching to "
83         "splitted critical edge"),
84     cl::init(40), cl::Hidden);
85 
86 static cl::opt<unsigned> SinkLoadInstsPerBlockThreshold(
87     "machine-sink-load-instrs-threshold",
88     cl::desc("Do not try to find alias store for a load if there is a in-path "
89              "block whose instruction number is higher than this threshold."),
90     cl::init(2000), cl::Hidden);
91 
92 static cl::opt<unsigned> SinkLoadBlocksThreshold(
93     "machine-sink-load-blocks-threshold",
94     cl::desc("Do not try to find alias store for a load if the block number in "
95              "the straight line is higher than this threshold."),
96     cl::init(20), cl::Hidden);
97 
98 static cl::opt<bool>
99     SinkInstsIntoCycle("sink-insts-to-avoid-spills",
100                        cl::desc("Sink instructions into cycles to avoid "
101                                 "register spills"),
102                        cl::init(false), cl::Hidden);
103 
104 static cl::opt<unsigned> SinkIntoCycleLimit(
105     "machine-sink-cycle-limit",
106     cl::desc("The maximum number of instructions considered for cycle sinking."),
107     cl::init(50), cl::Hidden);
108 
109 STATISTIC(NumSunk,      "Number of machine instructions sunk");
110 STATISTIC(NumCycleSunk,  "Number of machine instructions sunk into a cycle");
111 STATISTIC(NumSplit,     "Number of critical edges split");
112 STATISTIC(NumCoalesces, "Number of copies coalesced");
113 STATISTIC(NumPostRACopySink, "Number of copies sunk after RA");
114 
115 namespace {
116 
117   class MachineSinking : public MachineFunctionPass {
118     const TargetInstrInfo *TII;
119     const TargetRegisterInfo *TRI;
120     MachineRegisterInfo  *MRI;     // Machine register information
121     MachineDominatorTree *DT;      // Machine dominator tree
122     MachinePostDominatorTree *PDT; // Machine post dominator tree
123     MachineCycleInfo *CI;
124     MachineBlockFrequencyInfo *MBFI;
125     const MachineBranchProbabilityInfo *MBPI;
126     AliasAnalysis *AA;
127     RegisterClassInfo RegClassInfo;
128 
129     // Remember which edges have been considered for breaking.
130     SmallSet<std::pair<MachineBasicBlock*, MachineBasicBlock*>, 8>
131     CEBCandidates;
132     // Remember which edges we are about to split.
133     // This is different from CEBCandidates since those edges
134     // will be split.
135     SetVector<std::pair<MachineBasicBlock *, MachineBasicBlock *>> ToSplit;
136 
137     DenseSet<Register> RegsToClearKillFlags;
138 
139     using AllSuccsCache =
140         std::map<MachineBasicBlock *, SmallVector<MachineBasicBlock *, 4>>;
141 
142     /// DBG_VALUE pointer and flag. The flag is true if this DBG_VALUE is
143     /// post-dominated by another DBG_VALUE of the same variable location.
144     /// This is necessary to detect sequences such as:
145     ///     %0 = someinst
146     ///     DBG_VALUE %0, !123, !DIExpression()
147     ///     %1 = anotherinst
148     ///     DBG_VALUE %1, !123, !DIExpression()
149     /// Where if %0 were to sink, the DBG_VAUE should not sink with it, as that
150     /// would re-order assignments.
151     using SeenDbgUser = PointerIntPair<MachineInstr *, 1>;
152 
153     /// Record of DBG_VALUE uses of vregs in a block, so that we can identify
154     /// debug instructions to sink.
155     SmallDenseMap<unsigned, TinyPtrVector<SeenDbgUser>> SeenDbgUsers;
156 
157     /// Record of debug variables that have had their locations set in the
158     /// current block.
159     DenseSet<DebugVariable> SeenDbgVars;
160 
161     std::map<std::pair<MachineBasicBlock *, MachineBasicBlock *>, bool>
162         HasStoreCache;
163     std::map<std::pair<MachineBasicBlock *, MachineBasicBlock *>,
164              std::vector<MachineInstr *>>
165         StoreInstrCache;
166 
167     /// Cached BB's register pressure.
168     std::map<MachineBasicBlock *, std::vector<unsigned>> CachedRegisterPressure;
169 
170   public:
171     static char ID; // Pass identification
172 
173     MachineSinking() : MachineFunctionPass(ID) {
174       initializeMachineSinkingPass(*PassRegistry::getPassRegistry());
175     }
176 
177     bool runOnMachineFunction(MachineFunction &MF) override;
178 
179     void getAnalysisUsage(AnalysisUsage &AU) const override {
180       MachineFunctionPass::getAnalysisUsage(AU);
181       AU.addRequired<AAResultsWrapperPass>();
182       AU.addRequired<MachineDominatorTree>();
183       AU.addRequired<MachinePostDominatorTree>();
184       AU.addRequired<MachineCycleInfoWrapperPass>();
185       AU.addRequired<MachineBranchProbabilityInfo>();
186       AU.addPreserved<MachineCycleInfoWrapperPass>();
187       AU.addPreserved<MachineLoopInfo>();
188       if (UseBlockFreqInfo)
189         AU.addRequired<MachineBlockFrequencyInfo>();
190     }
191 
192     void releaseMemory() override {
193       CEBCandidates.clear();
194     }
195 
196   private:
197     bool ProcessBlock(MachineBasicBlock &MBB);
198     void ProcessDbgInst(MachineInstr &MI);
199     bool isWorthBreakingCriticalEdge(MachineInstr &MI,
200                                      MachineBasicBlock *From,
201                                      MachineBasicBlock *To);
202 
203     bool hasStoreBetween(MachineBasicBlock *From, MachineBasicBlock *To,
204                          MachineInstr &MI);
205 
206     /// Postpone the splitting of the given critical
207     /// edge (\p From, \p To).
208     ///
209     /// We do not split the edges on the fly. Indeed, this invalidates
210     /// the dominance information and thus triggers a lot of updates
211     /// of that information underneath.
212     /// Instead, we postpone all the splits after each iteration of
213     /// the main loop. That way, the information is at least valid
214     /// for the lifetime of an iteration.
215     ///
216     /// \return True if the edge is marked as toSplit, false otherwise.
217     /// False can be returned if, for instance, this is not profitable.
218     bool PostponeSplitCriticalEdge(MachineInstr &MI,
219                                    MachineBasicBlock *From,
220                                    MachineBasicBlock *To,
221                                    bool BreakPHIEdge);
222     bool SinkInstruction(MachineInstr &MI, bool &SawStore,
223                          AllSuccsCache &AllSuccessors);
224 
225     /// If we sink a COPY inst, some debug users of it's destination may no
226     /// longer be dominated by the COPY, and will eventually be dropped.
227     /// This is easily rectified by forwarding the non-dominated debug uses
228     /// to the copy source.
229     void SalvageUnsunkDebugUsersOfCopy(MachineInstr &,
230                                        MachineBasicBlock *TargetBlock);
231     bool AllUsesDominatedByBlock(Register Reg, MachineBasicBlock *MBB,
232                                  MachineBasicBlock *DefMBB, bool &BreakPHIEdge,
233                                  bool &LocalUse) const;
234     MachineBasicBlock *FindSuccToSinkTo(MachineInstr &MI, MachineBasicBlock *MBB,
235                bool &BreakPHIEdge, AllSuccsCache &AllSuccessors);
236 
237     void FindCycleSinkCandidates(MachineCycle *Cycle, MachineBasicBlock *BB,
238                                  SmallVectorImpl<MachineInstr *> &Candidates);
239     bool SinkIntoCycle(MachineCycle *Cycle, MachineInstr &I);
240 
241     bool isProfitableToSinkTo(Register Reg, MachineInstr &MI,
242                               MachineBasicBlock *MBB,
243                               MachineBasicBlock *SuccToSinkTo,
244                               AllSuccsCache &AllSuccessors);
245 
246     bool PerformTrivialForwardCoalescing(MachineInstr &MI,
247                                          MachineBasicBlock *MBB);
248 
249     SmallVector<MachineBasicBlock *, 4> &
250     GetAllSortedSuccessors(MachineInstr &MI, MachineBasicBlock *MBB,
251                            AllSuccsCache &AllSuccessors) const;
252 
253     std::vector<unsigned> &getBBRegisterPressure(MachineBasicBlock &MBB);
254   };
255 
256 } // end anonymous namespace
257 
258 char MachineSinking::ID = 0;
259 
260 char &llvm::MachineSinkingID = MachineSinking::ID;
261 
262 INITIALIZE_PASS_BEGIN(MachineSinking, DEBUG_TYPE,
263                       "Machine code sinking", false, false)
264 INITIALIZE_PASS_DEPENDENCY(MachineBranchProbabilityInfo)
265 INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
266 INITIALIZE_PASS_DEPENDENCY(MachineCycleInfoWrapperPass)
267 INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)
268 INITIALIZE_PASS_END(MachineSinking, DEBUG_TYPE,
269                     "Machine code sinking", false, false)
270 
271 bool MachineSinking::PerformTrivialForwardCoalescing(MachineInstr &MI,
272                                                      MachineBasicBlock *MBB) {
273   if (!MI.isCopy())
274     return false;
275 
276   Register SrcReg = MI.getOperand(1).getReg();
277   Register DstReg = MI.getOperand(0).getReg();
278   if (!SrcReg.isVirtual() || !DstReg.isVirtual() ||
279       !MRI->hasOneNonDBGUse(SrcReg))
280     return false;
281 
282   const TargetRegisterClass *SRC = MRI->getRegClass(SrcReg);
283   const TargetRegisterClass *DRC = MRI->getRegClass(DstReg);
284   if (SRC != DRC)
285     return false;
286 
287   MachineInstr *DefMI = MRI->getVRegDef(SrcReg);
288   if (DefMI->isCopyLike())
289     return false;
290   LLVM_DEBUG(dbgs() << "Coalescing: " << *DefMI);
291   LLVM_DEBUG(dbgs() << "*** to: " << MI);
292   MRI->replaceRegWith(DstReg, SrcReg);
293   MI.eraseFromParent();
294 
295   // Conservatively, clear any kill flags, since it's possible that they are no
296   // longer correct.
297   MRI->clearKillFlags(SrcReg);
298 
299   ++NumCoalesces;
300   return true;
301 }
302 
303 /// AllUsesDominatedByBlock - Return true if all uses of the specified register
304 /// occur in blocks dominated by the specified block. If any use is in the
305 /// definition block, then return false since it is never legal to move def
306 /// after uses.
307 bool MachineSinking::AllUsesDominatedByBlock(Register Reg,
308                                              MachineBasicBlock *MBB,
309                                              MachineBasicBlock *DefMBB,
310                                              bool &BreakPHIEdge,
311                                              bool &LocalUse) const {
312   assert(Reg.isVirtual() && "Only makes sense for vregs");
313 
314   // Ignore debug uses because debug info doesn't affect the code.
315   if (MRI->use_nodbg_empty(Reg))
316     return true;
317 
318   // BreakPHIEdge is true if all the uses are in the successor MBB being sunken
319   // into and they are all PHI nodes. In this case, machine-sink must break
320   // the critical edge first. e.g.
321   //
322   // %bb.1:
323   //   Predecessors according to CFG: %bb.0
324   //     ...
325   //     %def = DEC64_32r %x, implicit-def dead %eflags
326   //     ...
327   //     JE_4 <%bb.37>, implicit %eflags
328   //   Successors according to CFG: %bb.37 %bb.2
329   //
330   // %bb.2:
331   //     %p = PHI %y, %bb.0, %def, %bb.1
332   if (all_of(MRI->use_nodbg_operands(Reg), [&](MachineOperand &MO) {
333         MachineInstr *UseInst = MO.getParent();
334         unsigned OpNo = UseInst->getOperandNo(&MO);
335         MachineBasicBlock *UseBlock = UseInst->getParent();
336         return UseBlock == MBB && UseInst->isPHI() &&
337                UseInst->getOperand(OpNo + 1).getMBB() == DefMBB;
338       })) {
339     BreakPHIEdge = true;
340     return true;
341   }
342 
343   for (MachineOperand &MO : MRI->use_nodbg_operands(Reg)) {
344     // Determine the block of the use.
345     MachineInstr *UseInst = MO.getParent();
346     unsigned OpNo = &MO - &UseInst->getOperand(0);
347     MachineBasicBlock *UseBlock = UseInst->getParent();
348     if (UseInst->isPHI()) {
349       // PHI nodes use the operand in the predecessor block, not the block with
350       // the PHI.
351       UseBlock = UseInst->getOperand(OpNo+1).getMBB();
352     } else if (UseBlock == DefMBB) {
353       LocalUse = true;
354       return false;
355     }
356 
357     // Check that it dominates.
358     if (!DT->dominates(MBB, UseBlock))
359       return false;
360   }
361 
362   return true;
363 }
364 
365 /// Return true if this machine instruction loads from global offset table or
366 /// constant pool.
367 static bool mayLoadFromGOTOrConstantPool(MachineInstr &MI) {
368   assert(MI.mayLoad() && "Expected MI that loads!");
369 
370   // If we lost memory operands, conservatively assume that the instruction
371   // reads from everything..
372   if (MI.memoperands_empty())
373     return true;
374 
375   for (MachineMemOperand *MemOp : MI.memoperands())
376     if (const PseudoSourceValue *PSV = MemOp->getPseudoValue())
377       if (PSV->isGOT() || PSV->isConstantPool())
378         return true;
379 
380   return false;
381 }
382 
383 void MachineSinking::FindCycleSinkCandidates(
384     MachineCycle *Cycle, MachineBasicBlock *BB,
385     SmallVectorImpl<MachineInstr *> &Candidates) {
386   for (auto &MI : *BB) {
387     LLVM_DEBUG(dbgs() << "CycleSink: Analysing candidate: " << MI);
388     if (!TII->shouldSink(MI)) {
389       LLVM_DEBUG(dbgs() << "CycleSink: Instruction not a candidate for this "
390                            "target\n");
391       continue;
392     }
393     if (!isCycleInvariant(Cycle, MI)) {
394       LLVM_DEBUG(dbgs() << "CycleSink: Instruction is not cycle invariant\n");
395       continue;
396     }
397     bool DontMoveAcrossStore = true;
398     if (!MI.isSafeToMove(AA, DontMoveAcrossStore)) {
399       LLVM_DEBUG(dbgs() << "CycleSink: Instruction not safe to move.\n");
400       continue;
401     }
402     if (MI.mayLoad() && !mayLoadFromGOTOrConstantPool(MI)) {
403       LLVM_DEBUG(dbgs() << "CycleSink: Dont sink GOT or constant pool loads\n");
404       continue;
405     }
406     if (MI.isConvergent())
407       continue;
408 
409     const MachineOperand &MO = MI.getOperand(0);
410     if (!MO.isReg() || !MO.getReg() || !MO.isDef())
411       continue;
412     if (!MRI->hasOneDef(MO.getReg()))
413       continue;
414 
415     LLVM_DEBUG(dbgs() << "CycleSink: Instruction added as candidate.\n");
416     Candidates.push_back(&MI);
417   }
418 }
419 
420 bool MachineSinking::runOnMachineFunction(MachineFunction &MF) {
421   if (skipFunction(MF.getFunction()))
422     return false;
423 
424   LLVM_DEBUG(dbgs() << "******** Machine Sinking ********\n");
425 
426   TII = MF.getSubtarget().getInstrInfo();
427   TRI = MF.getSubtarget().getRegisterInfo();
428   MRI = &MF.getRegInfo();
429   DT = &getAnalysis<MachineDominatorTree>();
430   PDT = &getAnalysis<MachinePostDominatorTree>();
431   CI = &getAnalysis<MachineCycleInfoWrapperPass>().getCycleInfo();
432   MBFI = UseBlockFreqInfo ? &getAnalysis<MachineBlockFrequencyInfo>() : nullptr;
433   MBPI = &getAnalysis<MachineBranchProbabilityInfo>();
434   AA = &getAnalysis<AAResultsWrapperPass>().getAAResults();
435   RegClassInfo.runOnMachineFunction(MF);
436 
437   bool EverMadeChange = false;
438 
439   while (true) {
440     bool MadeChange = false;
441 
442     // Process all basic blocks.
443     CEBCandidates.clear();
444     ToSplit.clear();
445     for (auto &MBB: MF)
446       MadeChange |= ProcessBlock(MBB);
447 
448     // If we have anything we marked as toSplit, split it now.
449     for (const auto &Pair : ToSplit) {
450       auto NewSucc = Pair.first->SplitCriticalEdge(Pair.second, *this);
451       if (NewSucc != nullptr) {
452         LLVM_DEBUG(dbgs() << " *** Splitting critical edge: "
453                           << printMBBReference(*Pair.first) << " -- "
454                           << printMBBReference(*NewSucc) << " -- "
455                           << printMBBReference(*Pair.second) << '\n');
456         if (MBFI)
457           MBFI->onEdgeSplit(*Pair.first, *NewSucc, *MBPI);
458 
459         MadeChange = true;
460         ++NumSplit;
461       } else
462         LLVM_DEBUG(dbgs() << " *** Not legal to break critical edge\n");
463     }
464     // If this iteration over the code changed anything, keep iterating.
465     if (!MadeChange) break;
466     EverMadeChange = true;
467   }
468 
469   if (SinkInstsIntoCycle) {
470     SmallVector<MachineCycle *, 8> Cycles(CI->toplevel_begin(),
471                                           CI->toplevel_end());
472     for (auto *Cycle : Cycles) {
473       MachineBasicBlock *Preheader = Cycle->getCyclePreheader();
474       if (!Preheader) {
475         LLVM_DEBUG(dbgs() << "CycleSink: Can't find preheader\n");
476         continue;
477       }
478       SmallVector<MachineInstr *, 8> Candidates;
479       FindCycleSinkCandidates(Cycle, Preheader, Candidates);
480 
481       // Walk the candidates in reverse order so that we start with the use
482       // of a def-use chain, if there is any.
483       // TODO: Sort the candidates using a cost-model.
484       unsigned i = 0;
485       for (MachineInstr *I : llvm::reverse(Candidates)) {
486         if (i++ == SinkIntoCycleLimit) {
487           LLVM_DEBUG(dbgs() << "CycleSink:   Limit reached of instructions to "
488                                "be analysed.");
489           break;
490         }
491 
492         if (!SinkIntoCycle(Cycle, *I))
493           break;
494         EverMadeChange = true;
495         ++NumCycleSunk;
496       }
497     }
498   }
499 
500   HasStoreCache.clear();
501   StoreInstrCache.clear();
502 
503   // Now clear any kill flags for recorded registers.
504   for (auto I : RegsToClearKillFlags)
505     MRI->clearKillFlags(I);
506   RegsToClearKillFlags.clear();
507 
508   return EverMadeChange;
509 }
510 
511 bool MachineSinking::ProcessBlock(MachineBasicBlock &MBB) {
512   // Can't sink anything out of a block that has less than two successors.
513   if (MBB.succ_size() <= 1 || MBB.empty()) return false;
514 
515   // Don't bother sinking code out of unreachable blocks. In addition to being
516   // unprofitable, it can also lead to infinite looping, because in an
517   // unreachable cycle there may be nowhere to stop.
518   if (!DT->isReachableFromEntry(&MBB)) return false;
519 
520   bool MadeChange = false;
521 
522   // Cache all successors, sorted by frequency info and cycle depth.
523   AllSuccsCache AllSuccessors;
524 
525   // Walk the basic block bottom-up.  Remember if we saw a store.
526   MachineBasicBlock::iterator I = MBB.end();
527   --I;
528   bool ProcessedBegin, SawStore = false;
529   do {
530     MachineInstr &MI = *I;  // The instruction to sink.
531 
532     // Predecrement I (if it's not begin) so that it isn't invalidated by
533     // sinking.
534     ProcessedBegin = I == MBB.begin();
535     if (!ProcessedBegin)
536       --I;
537 
538     if (MI.isDebugOrPseudoInstr()) {
539       if (MI.isDebugValue())
540         ProcessDbgInst(MI);
541       continue;
542     }
543 
544     bool Joined = PerformTrivialForwardCoalescing(MI, &MBB);
545     if (Joined) {
546       MadeChange = true;
547       continue;
548     }
549 
550     if (SinkInstruction(MI, SawStore, AllSuccessors)) {
551       ++NumSunk;
552       MadeChange = true;
553     }
554 
555     // If we just processed the first instruction in the block, we're done.
556   } while (!ProcessedBegin);
557 
558   SeenDbgUsers.clear();
559   SeenDbgVars.clear();
560   // recalculate the bb register pressure after sinking one BB.
561   CachedRegisterPressure.clear();
562 
563   return MadeChange;
564 }
565 
566 void MachineSinking::ProcessDbgInst(MachineInstr &MI) {
567   // When we see DBG_VALUEs for registers, record any vreg it reads, so that
568   // we know what to sink if the vreg def sinks.
569   assert(MI.isDebugValue() && "Expected DBG_VALUE for processing");
570 
571   DebugVariable Var(MI.getDebugVariable(), MI.getDebugExpression(),
572                     MI.getDebugLoc()->getInlinedAt());
573   bool SeenBefore = SeenDbgVars.contains(Var);
574 
575   for (MachineOperand &MO : MI.debug_operands()) {
576     if (MO.isReg() && MO.getReg().isVirtual())
577       SeenDbgUsers[MO.getReg()].push_back(SeenDbgUser(&MI, SeenBefore));
578   }
579 
580   // Record the variable for any DBG_VALUE, to avoid re-ordering any of them.
581   SeenDbgVars.insert(Var);
582 }
583 
584 bool MachineSinking::isWorthBreakingCriticalEdge(MachineInstr &MI,
585                                                  MachineBasicBlock *From,
586                                                  MachineBasicBlock *To) {
587   // FIXME: Need much better heuristics.
588 
589   // If the pass has already considered breaking this edge (during this pass
590   // through the function), then let's go ahead and break it. This means
591   // sinking multiple "cheap" instructions into the same block.
592   if (!CEBCandidates.insert(std::make_pair(From, To)).second)
593     return true;
594 
595   if (!MI.isCopy() && !TII->isAsCheapAsAMove(MI))
596     return true;
597 
598   if (From->isSuccessor(To) && MBPI->getEdgeProbability(From, To) <=
599       BranchProbability(SplitEdgeProbabilityThreshold, 100))
600     return true;
601 
602   // MI is cheap, we probably don't want to break the critical edge for it.
603   // However, if this would allow some definitions of its source operands
604   // to be sunk then it's probably worth it.
605   for (const MachineOperand &MO : MI.operands()) {
606     if (!MO.isReg() || !MO.isUse())
607       continue;
608     Register Reg = MO.getReg();
609     if (Reg == 0)
610       continue;
611 
612     // We don't move live definitions of physical registers,
613     // so sinking their uses won't enable any opportunities.
614     if (Reg.isPhysical())
615       continue;
616 
617     // If this instruction is the only user of a virtual register,
618     // check if breaking the edge will enable sinking
619     // both this instruction and the defining instruction.
620     if (MRI->hasOneNonDBGUse(Reg)) {
621       // If the definition resides in same MBB,
622       // claim it's likely we can sink these together.
623       // If definition resides elsewhere, we aren't
624       // blocking it from being sunk so don't break the edge.
625       MachineInstr *DefMI = MRI->getVRegDef(Reg);
626       if (DefMI->getParent() == MI.getParent())
627         return true;
628     }
629   }
630 
631   return false;
632 }
633 
634 bool MachineSinking::PostponeSplitCriticalEdge(MachineInstr &MI,
635                                                MachineBasicBlock *FromBB,
636                                                MachineBasicBlock *ToBB,
637                                                bool BreakPHIEdge) {
638   if (!isWorthBreakingCriticalEdge(MI, FromBB, ToBB))
639     return false;
640 
641   // Avoid breaking back edge. From == To means backedge for single BB cycle.
642   if (!SplitEdges || FromBB == ToBB)
643     return false;
644 
645   MachineCycle *FromCycle = CI->getCycle(FromBB);
646   MachineCycle *ToCycle = CI->getCycle(ToBB);
647 
648   // Check for backedges of more "complex" cycles.
649   if (FromCycle == ToCycle && FromCycle &&
650       (!FromCycle->isReducible() || FromCycle->getHeader() == ToBB))
651     return false;
652 
653   // It's not always legal to break critical edges and sink the computation
654   // to the edge.
655   //
656   // %bb.1:
657   // v1024
658   // Beq %bb.3
659   // <fallthrough>
660   // %bb.2:
661   // ... no uses of v1024
662   // <fallthrough>
663   // %bb.3:
664   // ...
665   //       = v1024
666   //
667   // If %bb.1 -> %bb.3 edge is broken and computation of v1024 is inserted:
668   //
669   // %bb.1:
670   // ...
671   // Bne %bb.2
672   // %bb.4:
673   // v1024 =
674   // B %bb.3
675   // %bb.2:
676   // ... no uses of v1024
677   // <fallthrough>
678   // %bb.3:
679   // ...
680   //       = v1024
681   //
682   // This is incorrect since v1024 is not computed along the %bb.1->%bb.2->%bb.3
683   // flow. We need to ensure the new basic block where the computation is
684   // sunk to dominates all the uses.
685   // It's only legal to break critical edge and sink the computation to the
686   // new block if all the predecessors of "To", except for "From", are
687   // not dominated by "From". Given SSA property, this means these
688   // predecessors are dominated by "To".
689   //
690   // There is no need to do this check if all the uses are PHI nodes. PHI
691   // sources are only defined on the specific predecessor edges.
692   if (!BreakPHIEdge) {
693     for (MachineBasicBlock *Pred : ToBB->predecessors())
694       if (Pred != FromBB && !DT->dominates(ToBB, Pred))
695         return false;
696   }
697 
698   ToSplit.insert(std::make_pair(FromBB, ToBB));
699 
700   return true;
701 }
702 
703 std::vector<unsigned> &
704 MachineSinking::getBBRegisterPressure(MachineBasicBlock &MBB) {
705   // Currently to save compiling time, MBB's register pressure will not change
706   // in one ProcessBlock iteration because of CachedRegisterPressure. but MBB's
707   // register pressure is changed after sinking any instructions into it.
708   // FIXME: need a accurate and cheap register pressure estiminate model here.
709   auto RP = CachedRegisterPressure.find(&MBB);
710   if (RP != CachedRegisterPressure.end())
711     return RP->second;
712 
713   RegionPressure Pressure;
714   RegPressureTracker RPTracker(Pressure);
715 
716   // Initialize the register pressure tracker.
717   RPTracker.init(MBB.getParent(), &RegClassInfo, nullptr, &MBB, MBB.end(),
718                  /*TrackLaneMasks*/ false, /*TrackUntiedDefs=*/true);
719 
720   for (MachineBasicBlock::iterator MII = MBB.instr_end(),
721                                    MIE = MBB.instr_begin();
722        MII != MIE; --MII) {
723     MachineInstr &MI = *std::prev(MII);
724     if (MI.isDebugInstr() || MI.isPseudoProbe())
725       continue;
726     RegisterOperands RegOpers;
727     RegOpers.collect(MI, *TRI, *MRI, false, false);
728     RPTracker.recedeSkipDebugValues();
729     assert(&*RPTracker.getPos() == &MI && "RPTracker sync error!");
730     RPTracker.recede(RegOpers);
731   }
732 
733   RPTracker.closeRegion();
734   auto It = CachedRegisterPressure.insert(
735       std::make_pair(&MBB, RPTracker.getPressure().MaxSetPressure));
736   return It.first->second;
737 }
738 
739 /// isProfitableToSinkTo - Return true if it is profitable to sink MI.
740 bool MachineSinking::isProfitableToSinkTo(Register Reg, MachineInstr &MI,
741                                           MachineBasicBlock *MBB,
742                                           MachineBasicBlock *SuccToSinkTo,
743                                           AllSuccsCache &AllSuccessors) {
744   assert (SuccToSinkTo && "Invalid SinkTo Candidate BB");
745 
746   if (MBB == SuccToSinkTo)
747     return false;
748 
749   // It is profitable if SuccToSinkTo does not post dominate current block.
750   if (!PDT->dominates(SuccToSinkTo, MBB))
751     return true;
752 
753   // It is profitable to sink an instruction from a deeper cycle to a shallower
754   // cycle, even if the latter post-dominates the former (PR21115).
755   if (CI->getCycleDepth(MBB) > CI->getCycleDepth(SuccToSinkTo))
756     return true;
757 
758   // Check if only use in post dominated block is PHI instruction.
759   bool NonPHIUse = false;
760   for (MachineInstr &UseInst : MRI->use_nodbg_instructions(Reg)) {
761     MachineBasicBlock *UseBlock = UseInst.getParent();
762     if (UseBlock == SuccToSinkTo && !UseInst.isPHI())
763       NonPHIUse = true;
764   }
765   if (!NonPHIUse)
766     return true;
767 
768   // If SuccToSinkTo post dominates then also it may be profitable if MI
769   // can further profitably sinked into another block in next round.
770   bool BreakPHIEdge = false;
771   // FIXME - If finding successor is compile time expensive then cache results.
772   if (MachineBasicBlock *MBB2 =
773           FindSuccToSinkTo(MI, SuccToSinkTo, BreakPHIEdge, AllSuccessors))
774     return isProfitableToSinkTo(Reg, MI, SuccToSinkTo, MBB2, AllSuccessors);
775 
776   MachineCycle *MCycle = CI->getCycle(MBB);
777 
778   // If the instruction is not inside a cycle, it is not profitable to sink MI to
779   // a post dominate block SuccToSinkTo.
780   if (!MCycle)
781     return false;
782 
783   auto isRegisterPressureSetExceedLimit = [&](const TargetRegisterClass *RC) {
784     unsigned Weight = TRI->getRegClassWeight(RC).RegWeight;
785     const int *PS = TRI->getRegClassPressureSets(RC);
786     // Get register pressure for block SuccToSinkTo.
787     std::vector<unsigned> BBRegisterPressure =
788         getBBRegisterPressure(*SuccToSinkTo);
789     for (; *PS != -1; PS++)
790       // check if any register pressure set exceeds limit in block SuccToSinkTo
791       // after sinking.
792       if (Weight + BBRegisterPressure[*PS] >=
793           TRI->getRegPressureSetLimit(*MBB->getParent(), *PS))
794         return true;
795     return false;
796   };
797 
798   // If this instruction is inside a Cycle and sinking this instruction can make
799   // more registers live range shorten, it is still prifitable.
800   for (const MachineOperand &MO : MI.operands()) {
801     // Ignore non-register operands.
802     if (!MO.isReg())
803       continue;
804     Register Reg = MO.getReg();
805     if (Reg == 0)
806       continue;
807 
808     if (Reg.isPhysical()) {
809       if (MO.isUse() &&
810           (MRI->isConstantPhysReg(Reg) || TII->isIgnorableUse(MO)))
811         continue;
812 
813       // Don't handle non-constant and non-ignorable physical register.
814       return false;
815     }
816 
817     // Users for the defs are all dominated by SuccToSinkTo.
818     if (MO.isDef()) {
819       // This def register's live range is shortened after sinking.
820       bool LocalUse = false;
821       if (!AllUsesDominatedByBlock(Reg, SuccToSinkTo, MBB, BreakPHIEdge,
822                                    LocalUse))
823         return false;
824     } else {
825       MachineInstr *DefMI = MRI->getVRegDef(Reg);
826       if (!DefMI)
827         continue;
828       MachineCycle *Cycle = CI->getCycle(DefMI->getParent());
829       // DefMI is defined outside of cycle. There should be no live range
830       // impact for this operand. Defination outside of cycle means:
831       // 1: defination is outside of cycle.
832       // 2: defination is in this cycle, but it is a PHI in the cycle header.
833       if (Cycle != MCycle || (DefMI->isPHI() && Cycle && Cycle->isReducible() &&
834                               Cycle->getHeader() == DefMI->getParent()))
835         continue;
836       // The DefMI is defined inside the cycle.
837       // If sinking this operand makes some register pressure set exceed limit,
838       // it is not profitable.
839       if (isRegisterPressureSetExceedLimit(MRI->getRegClass(Reg))) {
840         LLVM_DEBUG(dbgs() << "register pressure exceed limit, not profitable.");
841         return false;
842       }
843     }
844   }
845 
846   // If MI is in cycle and all its operands are alive across the whole cycle or
847   // if no operand sinking make register pressure set exceed limit, it is
848   // profitable to sink MI.
849   return true;
850 }
851 
852 /// Get the sorted sequence of successors for this MachineBasicBlock, possibly
853 /// computing it if it was not already cached.
854 SmallVector<MachineBasicBlock *, 4> &
855 MachineSinking::GetAllSortedSuccessors(MachineInstr &MI, MachineBasicBlock *MBB,
856                                        AllSuccsCache &AllSuccessors) const {
857   // Do we have the sorted successors in cache ?
858   auto Succs = AllSuccessors.find(MBB);
859   if (Succs != AllSuccessors.end())
860     return Succs->second;
861 
862   SmallVector<MachineBasicBlock *, 4> AllSuccs(MBB->successors());
863 
864   // Handle cases where sinking can happen but where the sink point isn't a
865   // successor. For example:
866   //
867   //   x = computation
868   //   if () {} else {}
869   //   use x
870   //
871   for (MachineDomTreeNode *DTChild : DT->getNode(MBB)->children()) {
872     // DomTree children of MBB that have MBB as immediate dominator are added.
873     if (DTChild->getIDom()->getBlock() == MI.getParent() &&
874         // Skip MBBs already added to the AllSuccs vector above.
875         !MBB->isSuccessor(DTChild->getBlock()))
876       AllSuccs.push_back(DTChild->getBlock());
877   }
878 
879   // Sort Successors according to their cycle depth or block frequency info.
880   llvm::stable_sort(
881       AllSuccs, [this](const MachineBasicBlock *L, const MachineBasicBlock *R) {
882         uint64_t LHSFreq = MBFI ? MBFI->getBlockFreq(L).getFrequency() : 0;
883         uint64_t RHSFreq = MBFI ? MBFI->getBlockFreq(R).getFrequency() : 0;
884         bool HasBlockFreq = LHSFreq != 0 && RHSFreq != 0;
885         return HasBlockFreq ? LHSFreq < RHSFreq
886                             : CI->getCycleDepth(L) < CI->getCycleDepth(R);
887       });
888 
889   auto it = AllSuccessors.insert(std::make_pair(MBB, AllSuccs));
890 
891   return it.first->second;
892 }
893 
894 /// FindSuccToSinkTo - Find a successor to sink this instruction to.
895 MachineBasicBlock *
896 MachineSinking::FindSuccToSinkTo(MachineInstr &MI, MachineBasicBlock *MBB,
897                                  bool &BreakPHIEdge,
898                                  AllSuccsCache &AllSuccessors) {
899   assert (MBB && "Invalid MachineBasicBlock!");
900 
901   // loop over all the operands of the specified instruction.  If there is
902   // anything we can't handle, bail out.
903 
904   // SuccToSinkTo - This is the successor to sink this instruction to, once we
905   // decide.
906   MachineBasicBlock *SuccToSinkTo = nullptr;
907   for (const MachineOperand &MO : MI.operands()) {
908     if (!MO.isReg()) continue;  // Ignore non-register operands.
909 
910     Register Reg = MO.getReg();
911     if (Reg == 0) continue;
912 
913     if (Reg.isPhysical()) {
914       if (MO.isUse()) {
915         // If the physreg has no defs anywhere, it's just an ambient register
916         // and we can freely move its uses. Alternatively, if it's allocatable,
917         // it could get allocated to something with a def during allocation.
918         if (!MRI->isConstantPhysReg(Reg) && !TII->isIgnorableUse(MO))
919           return nullptr;
920       } else if (!MO.isDead()) {
921         // A def that isn't dead. We can't move it.
922         return nullptr;
923       }
924     } else {
925       // Virtual register uses are always safe to sink.
926       if (MO.isUse()) continue;
927 
928       // If it's not safe to move defs of the register class, then abort.
929       if (!TII->isSafeToMoveRegClassDefs(MRI->getRegClass(Reg)))
930         return nullptr;
931 
932       // Virtual register defs can only be sunk if all their uses are in blocks
933       // dominated by one of the successors.
934       if (SuccToSinkTo) {
935         // If a previous operand picked a block to sink to, then this operand
936         // must be sinkable to the same block.
937         bool LocalUse = false;
938         if (!AllUsesDominatedByBlock(Reg, SuccToSinkTo, MBB,
939                                      BreakPHIEdge, LocalUse))
940           return nullptr;
941 
942         continue;
943       }
944 
945       // Otherwise, we should look at all the successors and decide which one
946       // we should sink to. If we have reliable block frequency information
947       // (frequency != 0) available, give successors with smaller frequencies
948       // higher priority, otherwise prioritize smaller cycle depths.
949       for (MachineBasicBlock *SuccBlock :
950            GetAllSortedSuccessors(MI, MBB, AllSuccessors)) {
951         bool LocalUse = false;
952         if (AllUsesDominatedByBlock(Reg, SuccBlock, MBB,
953                                     BreakPHIEdge, LocalUse)) {
954           SuccToSinkTo = SuccBlock;
955           break;
956         }
957         if (LocalUse)
958           // Def is used locally, it's never safe to move this def.
959           return nullptr;
960       }
961 
962       // If we couldn't find a block to sink to, ignore this instruction.
963       if (!SuccToSinkTo)
964         return nullptr;
965       if (!isProfitableToSinkTo(Reg, MI, MBB, SuccToSinkTo, AllSuccessors))
966         return nullptr;
967     }
968   }
969 
970   // It is not possible to sink an instruction into its own block.  This can
971   // happen with cycles.
972   if (MBB == SuccToSinkTo)
973     return nullptr;
974 
975   // It's not safe to sink instructions to EH landing pad. Control flow into
976   // landing pad is implicitly defined.
977   if (SuccToSinkTo && SuccToSinkTo->isEHPad())
978     return nullptr;
979 
980   // It ought to be okay to sink instructions into an INLINEASM_BR target, but
981   // only if we make sure that MI occurs _before_ an INLINEASM_BR instruction in
982   // the source block (which this code does not yet do). So for now, forbid
983   // doing so.
984   if (SuccToSinkTo && SuccToSinkTo->isInlineAsmBrIndirectTarget())
985     return nullptr;
986 
987   return SuccToSinkTo;
988 }
989 
990 /// Return true if MI is likely to be usable as a memory operation by the
991 /// implicit null check optimization.
992 ///
993 /// This is a "best effort" heuristic, and should not be relied upon for
994 /// correctness.  This returning true does not guarantee that the implicit null
995 /// check optimization is legal over MI, and this returning false does not
996 /// guarantee MI cannot possibly be used to do a null check.
997 static bool SinkingPreventsImplicitNullCheck(MachineInstr &MI,
998                                              const TargetInstrInfo *TII,
999                                              const TargetRegisterInfo *TRI) {
1000   using MachineBranchPredicate = TargetInstrInfo::MachineBranchPredicate;
1001 
1002   auto *MBB = MI.getParent();
1003   if (MBB->pred_size() != 1)
1004     return false;
1005 
1006   auto *PredMBB = *MBB->pred_begin();
1007   auto *PredBB = PredMBB->getBasicBlock();
1008 
1009   // Frontends that don't use implicit null checks have no reason to emit
1010   // branches with make.implicit metadata, and this function should always
1011   // return false for them.
1012   if (!PredBB ||
1013       !PredBB->getTerminator()->getMetadata(LLVMContext::MD_make_implicit))
1014     return false;
1015 
1016   const MachineOperand *BaseOp;
1017   int64_t Offset;
1018   bool OffsetIsScalable;
1019   if (!TII->getMemOperandWithOffset(MI, BaseOp, Offset, OffsetIsScalable, TRI))
1020     return false;
1021 
1022   if (!BaseOp->isReg())
1023     return false;
1024 
1025   if (!(MI.mayLoad() && !MI.isPredicable()))
1026     return false;
1027 
1028   MachineBranchPredicate MBP;
1029   if (TII->analyzeBranchPredicate(*PredMBB, MBP, false))
1030     return false;
1031 
1032   return MBP.LHS.isReg() && MBP.RHS.isImm() && MBP.RHS.getImm() == 0 &&
1033          (MBP.Predicate == MachineBranchPredicate::PRED_NE ||
1034           MBP.Predicate == MachineBranchPredicate::PRED_EQ) &&
1035          MBP.LHS.getReg() == BaseOp->getReg();
1036 }
1037 
1038 /// If the sunk instruction is a copy, try to forward the copy instead of
1039 /// leaving an 'undef' DBG_VALUE in the original location. Don't do this if
1040 /// there's any subregister weirdness involved. Returns true if copy
1041 /// propagation occurred.
1042 static bool attemptDebugCopyProp(MachineInstr &SinkInst, MachineInstr &DbgMI,
1043                                  Register Reg) {
1044   const MachineRegisterInfo &MRI = SinkInst.getMF()->getRegInfo();
1045   const TargetInstrInfo &TII = *SinkInst.getMF()->getSubtarget().getInstrInfo();
1046 
1047   // Copy DBG_VALUE operand and set the original to undef. We then check to
1048   // see whether this is something that can be copy-forwarded. If it isn't,
1049   // continue around the loop.
1050 
1051   const MachineOperand *SrcMO = nullptr, *DstMO = nullptr;
1052   auto CopyOperands = TII.isCopyInstr(SinkInst);
1053   if (!CopyOperands)
1054     return false;
1055   SrcMO = CopyOperands->Source;
1056   DstMO = CopyOperands->Destination;
1057 
1058   // Check validity of forwarding this copy.
1059   bool PostRA = MRI.getNumVirtRegs() == 0;
1060 
1061   // Trying to forward between physical and virtual registers is too hard.
1062   if (Reg.isVirtual() != SrcMO->getReg().isVirtual())
1063     return false;
1064 
1065   // Only try virtual register copy-forwarding before regalloc, and physical
1066   // register copy-forwarding after regalloc.
1067   bool arePhysRegs = !Reg.isVirtual();
1068   if (arePhysRegs != PostRA)
1069     return false;
1070 
1071   // Pre-regalloc, only forward if all subregisters agree (or there are no
1072   // subregs at all). More analysis might recover some forwardable copies.
1073   if (!PostRA)
1074     for (auto &DbgMO : DbgMI.getDebugOperandsForReg(Reg))
1075       if (DbgMO.getSubReg() != SrcMO->getSubReg() ||
1076           DbgMO.getSubReg() != DstMO->getSubReg())
1077         return false;
1078 
1079   // Post-regalloc, we may be sinking a DBG_VALUE of a sub or super-register
1080   // of this copy. Only forward the copy if the DBG_VALUE operand exactly
1081   // matches the copy destination.
1082   if (PostRA && Reg != DstMO->getReg())
1083     return false;
1084 
1085   for (auto &DbgMO : DbgMI.getDebugOperandsForReg(Reg)) {
1086     DbgMO.setReg(SrcMO->getReg());
1087     DbgMO.setSubReg(SrcMO->getSubReg());
1088   }
1089   return true;
1090 }
1091 
1092 using MIRegs = std::pair<MachineInstr *, SmallVector<unsigned, 2>>;
1093 /// Sink an instruction and its associated debug instructions.
1094 static void performSink(MachineInstr &MI, MachineBasicBlock &SuccToSinkTo,
1095                         MachineBasicBlock::iterator InsertPos,
1096                         ArrayRef<MIRegs> DbgValuesToSink) {
1097   // If we cannot find a location to use (merge with), then we erase the debug
1098   // location to prevent debug-info driven tools from potentially reporting
1099   // wrong location information.
1100   if (!SuccToSinkTo.empty() && InsertPos != SuccToSinkTo.end())
1101     MI.setDebugLoc(DILocation::getMergedLocation(MI.getDebugLoc(),
1102                                                  InsertPos->getDebugLoc()));
1103   else
1104     MI.setDebugLoc(DebugLoc());
1105 
1106   // Move the instruction.
1107   MachineBasicBlock *ParentBlock = MI.getParent();
1108   SuccToSinkTo.splice(InsertPos, ParentBlock, MI,
1109                       ++MachineBasicBlock::iterator(MI));
1110 
1111   // Sink a copy of debug users to the insert position. Mark the original
1112   // DBG_VALUE location as 'undef', indicating that any earlier variable
1113   // location should be terminated as we've optimised away the value at this
1114   // point.
1115   for (const auto &DbgValueToSink : DbgValuesToSink) {
1116     MachineInstr *DbgMI = DbgValueToSink.first;
1117     MachineInstr *NewDbgMI = DbgMI->getMF()->CloneMachineInstr(DbgMI);
1118     SuccToSinkTo.insert(InsertPos, NewDbgMI);
1119 
1120     bool PropagatedAllSunkOps = true;
1121     for (unsigned Reg : DbgValueToSink.second) {
1122       if (DbgMI->hasDebugOperandForReg(Reg)) {
1123         if (!attemptDebugCopyProp(MI, *DbgMI, Reg)) {
1124           PropagatedAllSunkOps = false;
1125           break;
1126         }
1127       }
1128     }
1129     if (!PropagatedAllSunkOps)
1130       DbgMI->setDebugValueUndef();
1131   }
1132 }
1133 
1134 /// hasStoreBetween - check if there is store betweeen straight line blocks From
1135 /// and To.
1136 bool MachineSinking::hasStoreBetween(MachineBasicBlock *From,
1137                                      MachineBasicBlock *To, MachineInstr &MI) {
1138   // Make sure From and To are in straight line which means From dominates To
1139   // and To post dominates From.
1140   if (!DT->dominates(From, To) || !PDT->dominates(To, From))
1141     return true;
1142 
1143   auto BlockPair = std::make_pair(From, To);
1144 
1145   // Does these two blocks pair be queried before and have a definite cached
1146   // result?
1147   if (HasStoreCache.find(BlockPair) != HasStoreCache.end())
1148     return HasStoreCache[BlockPair];
1149 
1150   if (StoreInstrCache.find(BlockPair) != StoreInstrCache.end())
1151     return llvm::any_of(StoreInstrCache[BlockPair], [&](MachineInstr *I) {
1152       return I->mayAlias(AA, MI, false);
1153     });
1154 
1155   bool SawStore = false;
1156   bool HasAliasedStore = false;
1157   DenseSet<MachineBasicBlock *> HandledBlocks;
1158   DenseSet<MachineBasicBlock *> HandledDomBlocks;
1159   // Go through all reachable blocks from From.
1160   for (MachineBasicBlock *BB : depth_first(From)) {
1161     // We insert the instruction at the start of block To, so no need to worry
1162     // about stores inside To.
1163     // Store in block From should be already considered when just enter function
1164     // SinkInstruction.
1165     if (BB == To || BB == From)
1166       continue;
1167 
1168     // We already handle this BB in previous iteration.
1169     if (HandledBlocks.count(BB))
1170       continue;
1171 
1172     HandledBlocks.insert(BB);
1173     // To post dominates BB, it must be a path from block From.
1174     if (PDT->dominates(To, BB)) {
1175       if (!HandledDomBlocks.count(BB))
1176         HandledDomBlocks.insert(BB);
1177 
1178       // If this BB is too big or the block number in straight line between From
1179       // and To is too big, stop searching to save compiling time.
1180       if (BB->sizeWithoutDebugLargerThan(SinkLoadInstsPerBlockThreshold) ||
1181           HandledDomBlocks.size() > SinkLoadBlocksThreshold) {
1182         for (auto *DomBB : HandledDomBlocks) {
1183           if (DomBB != BB && DT->dominates(DomBB, BB))
1184             HasStoreCache[std::make_pair(DomBB, To)] = true;
1185           else if(DomBB != BB && DT->dominates(BB, DomBB))
1186             HasStoreCache[std::make_pair(From, DomBB)] = true;
1187         }
1188         HasStoreCache[BlockPair] = true;
1189         return true;
1190       }
1191 
1192       for (MachineInstr &I : *BB) {
1193         // Treat as alias conservatively for a call or an ordered memory
1194         // operation.
1195         if (I.isCall() || I.hasOrderedMemoryRef()) {
1196           for (auto *DomBB : HandledDomBlocks) {
1197             if (DomBB != BB && DT->dominates(DomBB, BB))
1198               HasStoreCache[std::make_pair(DomBB, To)] = true;
1199             else if(DomBB != BB && DT->dominates(BB, DomBB))
1200               HasStoreCache[std::make_pair(From, DomBB)] = true;
1201           }
1202           HasStoreCache[BlockPair] = true;
1203           return true;
1204         }
1205 
1206         if (I.mayStore()) {
1207           SawStore = true;
1208           // We still have chance to sink MI if all stores between are not
1209           // aliased to MI.
1210           // Cache all store instructions, so that we don't need to go through
1211           // all From reachable blocks for next load instruction.
1212           if (I.mayAlias(AA, MI, false))
1213             HasAliasedStore = true;
1214           StoreInstrCache[BlockPair].push_back(&I);
1215         }
1216       }
1217     }
1218   }
1219   // If there is no store at all, cache the result.
1220   if (!SawStore)
1221     HasStoreCache[BlockPair] = false;
1222   return HasAliasedStore;
1223 }
1224 
1225 /// Sink instructions into cycles if profitable. This especially tries to
1226 /// prevent register spills caused by register pressure if there is little to no
1227 /// overhead moving instructions into cycles.
1228 bool MachineSinking::SinkIntoCycle(MachineCycle *Cycle, MachineInstr &I) {
1229   LLVM_DEBUG(dbgs() << "CycleSink: Finding sink block for: " << I);
1230   MachineBasicBlock *Preheader = Cycle->getCyclePreheader();
1231   assert(Preheader && "Cycle sink needs a preheader block");
1232   MachineBasicBlock *SinkBlock = nullptr;
1233   bool CanSink = true;
1234   const MachineOperand &MO = I.getOperand(0);
1235 
1236   for (MachineInstr &MI : MRI->use_instructions(MO.getReg())) {
1237     LLVM_DEBUG(dbgs() << "CycleSink:   Analysing use: " << MI);
1238     if (!Cycle->contains(MI.getParent())) {
1239       LLVM_DEBUG(dbgs() << "CycleSink:   Use not in cycle, can't sink.\n");
1240       CanSink = false;
1241       break;
1242     }
1243 
1244     // FIXME: Come up with a proper cost model that estimates whether sinking
1245     // the instruction (and thus possibly executing it on every cycle
1246     // iteration) is more expensive than a register.
1247     // For now assumes that copies are cheap and thus almost always worth it.
1248     if (!MI.isCopy()) {
1249       LLVM_DEBUG(dbgs() << "CycleSink:   Use is not a copy\n");
1250       CanSink = false;
1251       break;
1252     }
1253     if (!SinkBlock) {
1254       SinkBlock = MI.getParent();
1255       LLVM_DEBUG(dbgs() << "CycleSink:   Setting sink block to: "
1256                         << printMBBReference(*SinkBlock) << "\n");
1257       continue;
1258     }
1259     SinkBlock = DT->findNearestCommonDominator(SinkBlock, MI.getParent());
1260     if (!SinkBlock) {
1261       LLVM_DEBUG(dbgs() << "CycleSink:   Can't find nearest dominator\n");
1262       CanSink = false;
1263       break;
1264     }
1265     LLVM_DEBUG(dbgs() << "CycleSink:   Setting nearest common dom block: " <<
1266                printMBBReference(*SinkBlock) << "\n");
1267   }
1268 
1269   if (!CanSink) {
1270     LLVM_DEBUG(dbgs() << "CycleSink: Can't sink instruction.\n");
1271     return false;
1272   }
1273   if (!SinkBlock) {
1274     LLVM_DEBUG(dbgs() << "CycleSink: Not sinking, can't find sink block.\n");
1275     return false;
1276   }
1277   if (SinkBlock == Preheader) {
1278     LLVM_DEBUG(
1279         dbgs() << "CycleSink: Not sinking, sink block is the preheader\n");
1280     return false;
1281   }
1282   if (SinkBlock->sizeWithoutDebugLargerThan(SinkLoadInstsPerBlockThreshold)) {
1283     LLVM_DEBUG(
1284         dbgs() << "CycleSink: Not Sinking, block too large to analyse.\n");
1285     return false;
1286   }
1287 
1288   LLVM_DEBUG(dbgs() << "CycleSink: Sinking instruction!\n");
1289   SinkBlock->splice(SinkBlock->SkipPHIsAndLabels(SinkBlock->begin()), Preheader,
1290                     I);
1291 
1292   // Conservatively clear any kill flags on uses of sunk instruction
1293   for (MachineOperand &MO : I.operands()) {
1294     if (MO.isReg() && MO.readsReg())
1295       RegsToClearKillFlags.insert(MO.getReg());
1296   }
1297 
1298   // The instruction is moved from its basic block, so do not retain the
1299   // debug information.
1300   assert(!I.isDebugInstr() && "Should not sink debug inst");
1301   I.setDebugLoc(DebugLoc());
1302   return true;
1303 }
1304 
1305 /// Return true if a target defined block prologue instruction interferes
1306 /// with a sink candidate.
1307 static bool blockPrologueInterferes(MachineBasicBlock *BB,
1308                                     MachineBasicBlock::iterator End,
1309                                     MachineInstr &MI,
1310                                     const TargetRegisterInfo *TRI,
1311                                     const TargetInstrInfo *TII,
1312                                     const MachineRegisterInfo *MRI) {
1313   if (BB->begin() == End)
1314     return false; // no prologue
1315   for (MachineBasicBlock::iterator PI = BB->getFirstNonPHI(); PI != End; ++PI) {
1316     // Only check target defined prologue instructions
1317     if (!TII->isBasicBlockPrologue(*PI))
1318       continue;
1319     for (auto &MO : MI.operands()) {
1320       if (!MO.isReg())
1321         continue;
1322       Register Reg = MO.getReg();
1323       if (!Reg)
1324         continue;
1325       if (MO.isUse()) {
1326         if (Reg.isPhysical() &&
1327             (TII->isIgnorableUse(MO) || (MRI && MRI->isConstantPhysReg(Reg))))
1328           continue;
1329         if (PI->modifiesRegister(Reg, TRI))
1330           return true;
1331       } else {
1332         if (PI->readsRegister(Reg, TRI))
1333           return true;
1334         // Check for interference with non-dead defs
1335         auto *DefOp = PI->findRegisterDefOperand(Reg, false, true, TRI);
1336         if (DefOp && !DefOp->isDead())
1337           return true;
1338       }
1339     }
1340   }
1341   return false;
1342 }
1343 
1344 /// SinkInstruction - Determine whether it is safe to sink the specified machine
1345 /// instruction out of its current block into a successor.
1346 bool MachineSinking::SinkInstruction(MachineInstr &MI, bool &SawStore,
1347                                      AllSuccsCache &AllSuccessors) {
1348   // Don't sink instructions that the target prefers not to sink.
1349   if (!TII->shouldSink(MI))
1350     return false;
1351 
1352   // Check if it's safe to move the instruction.
1353   if (!MI.isSafeToMove(AA, SawStore))
1354     return false;
1355 
1356   // Convergent operations may not be made control-dependent on additional
1357   // values.
1358   if (MI.isConvergent())
1359     return false;
1360 
1361   // Don't break implicit null checks.  This is a performance heuristic, and not
1362   // required for correctness.
1363   if (SinkingPreventsImplicitNullCheck(MI, TII, TRI))
1364     return false;
1365 
1366   // FIXME: This should include support for sinking instructions within the
1367   // block they are currently in to shorten the live ranges.  We often get
1368   // instructions sunk into the top of a large block, but it would be better to
1369   // also sink them down before their first use in the block.  This xform has to
1370   // be careful not to *increase* register pressure though, e.g. sinking
1371   // "x = y + z" down if it kills y and z would increase the live ranges of y
1372   // and z and only shrink the live range of x.
1373 
1374   bool BreakPHIEdge = false;
1375   MachineBasicBlock *ParentBlock = MI.getParent();
1376   MachineBasicBlock *SuccToSinkTo =
1377       FindSuccToSinkTo(MI, ParentBlock, BreakPHIEdge, AllSuccessors);
1378 
1379   // If there are no outputs, it must have side-effects.
1380   if (!SuccToSinkTo)
1381     return false;
1382 
1383   // If the instruction to move defines a dead physical register which is live
1384   // when leaving the basic block, don't move it because it could turn into a
1385   // "zombie" define of that preg. E.g., EFLAGS. (<rdar://problem/8030636>)
1386   for (const MachineOperand &MO : MI.operands()) {
1387     if (!MO.isReg() || MO.isUse())
1388       continue;
1389     Register Reg = MO.getReg();
1390     if (Reg == 0 || !Reg.isPhysical())
1391       continue;
1392     if (SuccToSinkTo->isLiveIn(Reg))
1393       return false;
1394   }
1395 
1396   LLVM_DEBUG(dbgs() << "Sink instr " << MI << "\tinto block " << *SuccToSinkTo);
1397 
1398   // If the block has multiple predecessors, this is a critical edge.
1399   // Decide if we can sink along it or need to break the edge.
1400   if (SuccToSinkTo->pred_size() > 1) {
1401     // We cannot sink a load across a critical edge - there may be stores in
1402     // other code paths.
1403     bool TryBreak = false;
1404     bool Store =
1405         MI.mayLoad() ? hasStoreBetween(ParentBlock, SuccToSinkTo, MI) : true;
1406     if (!MI.isSafeToMove(AA, Store)) {
1407       LLVM_DEBUG(dbgs() << " *** NOTE: Won't sink load along critical edge.\n");
1408       TryBreak = true;
1409     }
1410 
1411     // We don't want to sink across a critical edge if we don't dominate the
1412     // successor. We could be introducing calculations to new code paths.
1413     if (!TryBreak && !DT->dominates(ParentBlock, SuccToSinkTo)) {
1414       LLVM_DEBUG(dbgs() << " *** NOTE: Critical edge found\n");
1415       TryBreak = true;
1416     }
1417 
1418     // Don't sink instructions into a cycle.
1419     if (!TryBreak && CI->getCycle(SuccToSinkTo) &&
1420         (!CI->getCycle(SuccToSinkTo)->isReducible() ||
1421          CI->getCycle(SuccToSinkTo)->getHeader() == SuccToSinkTo)) {
1422       LLVM_DEBUG(dbgs() << " *** NOTE: cycle header found\n");
1423       TryBreak = true;
1424     }
1425 
1426     // Otherwise we are OK with sinking along a critical edge.
1427     if (!TryBreak)
1428       LLVM_DEBUG(dbgs() << "Sinking along critical edge.\n");
1429     else {
1430       // Mark this edge as to be split.
1431       // If the edge can actually be split, the next iteration of the main loop
1432       // will sink MI in the newly created block.
1433       bool Status =
1434         PostponeSplitCriticalEdge(MI, ParentBlock, SuccToSinkTo, BreakPHIEdge);
1435       if (!Status)
1436         LLVM_DEBUG(dbgs() << " *** PUNTING: Not legal or profitable to "
1437                              "break critical edge\n");
1438       // The instruction will not be sunk this time.
1439       return false;
1440     }
1441   }
1442 
1443   if (BreakPHIEdge) {
1444     // BreakPHIEdge is true if all the uses are in the successor MBB being
1445     // sunken into and they are all PHI nodes. In this case, machine-sink must
1446     // break the critical edge first.
1447     bool Status = PostponeSplitCriticalEdge(MI, ParentBlock,
1448                                             SuccToSinkTo, BreakPHIEdge);
1449     if (!Status)
1450       LLVM_DEBUG(dbgs() << " *** PUNTING: Not legal or profitable to "
1451                            "break critical edge\n");
1452     // The instruction will not be sunk this time.
1453     return false;
1454   }
1455 
1456   // Determine where to insert into. Skip phi nodes.
1457   MachineBasicBlock::iterator InsertPos =
1458       SuccToSinkTo->SkipPHIsAndLabels(SuccToSinkTo->begin());
1459   if (blockPrologueInterferes(SuccToSinkTo, InsertPos, MI, TRI, TII, MRI)) {
1460     LLVM_DEBUG(dbgs() << " *** Not sinking: prologue interference\n");
1461     return false;
1462   }
1463 
1464   // Collect debug users of any vreg that this inst defines.
1465   SmallVector<MIRegs, 4> DbgUsersToSink;
1466   for (auto &MO : MI.operands()) {
1467     if (!MO.isReg() || !MO.isDef() || !MO.getReg().isVirtual())
1468       continue;
1469     if (!SeenDbgUsers.count(MO.getReg()))
1470       continue;
1471 
1472     // Sink any users that don't pass any other DBG_VALUEs for this variable.
1473     auto &Users = SeenDbgUsers[MO.getReg()];
1474     for (auto &User : Users) {
1475       MachineInstr *DbgMI = User.getPointer();
1476       if (User.getInt()) {
1477         // This DBG_VALUE would re-order assignments. If we can't copy-propagate
1478         // it, it can't be recovered. Set it undef.
1479         if (!attemptDebugCopyProp(MI, *DbgMI, MO.getReg()))
1480           DbgMI->setDebugValueUndef();
1481       } else {
1482         DbgUsersToSink.push_back(
1483             {DbgMI, SmallVector<unsigned, 2>(1, MO.getReg())});
1484       }
1485     }
1486   }
1487 
1488   // After sinking, some debug users may not be dominated any more. If possible,
1489   // copy-propagate their operands. As it's expensive, don't do this if there's
1490   // no debuginfo in the program.
1491   if (MI.getMF()->getFunction().getSubprogram() && MI.isCopy())
1492     SalvageUnsunkDebugUsersOfCopy(MI, SuccToSinkTo);
1493 
1494   performSink(MI, *SuccToSinkTo, InsertPos, DbgUsersToSink);
1495 
1496   // Conservatively, clear any kill flags, since it's possible that they are no
1497   // longer correct.
1498   // Note that we have to clear the kill flags for any register this instruction
1499   // uses as we may sink over another instruction which currently kills the
1500   // used registers.
1501   for (MachineOperand &MO : MI.operands()) {
1502     if (MO.isReg() && MO.isUse())
1503       RegsToClearKillFlags.insert(MO.getReg()); // Remember to clear kill flags.
1504   }
1505 
1506   return true;
1507 }
1508 
1509 void MachineSinking::SalvageUnsunkDebugUsersOfCopy(
1510     MachineInstr &MI, MachineBasicBlock *TargetBlock) {
1511   assert(MI.isCopy());
1512   assert(MI.getOperand(1).isReg());
1513 
1514   // Enumerate all users of vreg operands that are def'd. Skip those that will
1515   // be sunk. For the rest, if they are not dominated by the block we will sink
1516   // MI into, propagate the copy source to them.
1517   SmallVector<MachineInstr *, 4> DbgDefUsers;
1518   SmallVector<Register, 4> DbgUseRegs;
1519   const MachineRegisterInfo &MRI = MI.getMF()->getRegInfo();
1520   for (auto &MO : MI.operands()) {
1521     if (!MO.isReg() || !MO.isDef() || !MO.getReg().isVirtual())
1522       continue;
1523     DbgUseRegs.push_back(MO.getReg());
1524     for (auto &User : MRI.use_instructions(MO.getReg())) {
1525       if (!User.isDebugValue() || DT->dominates(TargetBlock, User.getParent()))
1526         continue;
1527 
1528       // If is in same block, will either sink or be use-before-def.
1529       if (User.getParent() == MI.getParent())
1530         continue;
1531 
1532       assert(User.hasDebugOperandForReg(MO.getReg()) &&
1533              "DBG_VALUE user of vreg, but has no operand for it?");
1534       DbgDefUsers.push_back(&User);
1535     }
1536   }
1537 
1538   // Point the users of this copy that are no longer dominated, at the source
1539   // of the copy.
1540   for (auto *User : DbgDefUsers) {
1541     for (auto &Reg : DbgUseRegs) {
1542       for (auto &DbgOp : User->getDebugOperandsForReg(Reg)) {
1543         DbgOp.setReg(MI.getOperand(1).getReg());
1544         DbgOp.setSubReg(MI.getOperand(1).getSubReg());
1545       }
1546     }
1547   }
1548 }
1549 
1550 //===----------------------------------------------------------------------===//
1551 // This pass is not intended to be a replacement or a complete alternative
1552 // for the pre-ra machine sink pass. It is only designed to sink COPY
1553 // instructions which should be handled after RA.
1554 //
1555 // This pass sinks COPY instructions into a successor block, if the COPY is not
1556 // used in the current block and the COPY is live-in to a single successor
1557 // (i.e., doesn't require the COPY to be duplicated).  This avoids executing the
1558 // copy on paths where their results aren't needed.  This also exposes
1559 // additional opportunites for dead copy elimination and shrink wrapping.
1560 //
1561 // These copies were either not handled by or are inserted after the MachineSink
1562 // pass. As an example of the former case, the MachineSink pass cannot sink
1563 // COPY instructions with allocatable source registers; for AArch64 these type
1564 // of copy instructions are frequently used to move function parameters (PhyReg)
1565 // into virtual registers in the entry block.
1566 //
1567 // For the machine IR below, this pass will sink %w19 in the entry into its
1568 // successor (%bb.1) because %w19 is only live-in in %bb.1.
1569 // %bb.0:
1570 //   %wzr = SUBSWri %w1, 1
1571 //   %w19 = COPY %w0
1572 //   Bcc 11, %bb.2
1573 // %bb.1:
1574 //   Live Ins: %w19
1575 //   BL @fun
1576 //   %w0 = ADDWrr %w0, %w19
1577 //   RET %w0
1578 // %bb.2:
1579 //   %w0 = COPY %wzr
1580 //   RET %w0
1581 // As we sink %w19 (CSR in AArch64) into %bb.1, the shrink-wrapping pass will be
1582 // able to see %bb.0 as a candidate.
1583 //===----------------------------------------------------------------------===//
1584 namespace {
1585 
1586 class PostRAMachineSinking : public MachineFunctionPass {
1587 public:
1588   bool runOnMachineFunction(MachineFunction &MF) override;
1589 
1590   static char ID;
1591   PostRAMachineSinking() : MachineFunctionPass(ID) {}
1592   StringRef getPassName() const override { return "PostRA Machine Sink"; }
1593 
1594   void getAnalysisUsage(AnalysisUsage &AU) const override {
1595     AU.setPreservesCFG();
1596     MachineFunctionPass::getAnalysisUsage(AU);
1597   }
1598 
1599   MachineFunctionProperties getRequiredProperties() const override {
1600     return MachineFunctionProperties().set(
1601         MachineFunctionProperties::Property::NoVRegs);
1602   }
1603 
1604 private:
1605   /// Track which register units have been modified and used.
1606   LiveRegUnits ModifiedRegUnits, UsedRegUnits;
1607 
1608   /// Track DBG_VALUEs of (unmodified) register units. Each DBG_VALUE has an
1609   /// entry in this map for each unit it touches. The DBG_VALUE's entry
1610   /// consists of a pointer to the instruction itself, and a vector of registers
1611   /// referred to by the instruction that overlap the key register unit.
1612   DenseMap<unsigned, SmallVector<MIRegs, 2>> SeenDbgInstrs;
1613 
1614   /// Sink Copy instructions unused in the same block close to their uses in
1615   /// successors.
1616   bool tryToSinkCopy(MachineBasicBlock &BB, MachineFunction &MF,
1617                      const TargetRegisterInfo *TRI, const TargetInstrInfo *TII);
1618 };
1619 } // namespace
1620 
1621 char PostRAMachineSinking::ID = 0;
1622 char &llvm::PostRAMachineSinkingID = PostRAMachineSinking::ID;
1623 
1624 INITIALIZE_PASS(PostRAMachineSinking, "postra-machine-sink",
1625                 "PostRA Machine Sink", false, false)
1626 
1627 static bool aliasWithRegsInLiveIn(MachineBasicBlock &MBB, unsigned Reg,
1628                                   const TargetRegisterInfo *TRI) {
1629   LiveRegUnits LiveInRegUnits(*TRI);
1630   LiveInRegUnits.addLiveIns(MBB);
1631   return !LiveInRegUnits.available(Reg);
1632 }
1633 
1634 static MachineBasicBlock *
1635 getSingleLiveInSuccBB(MachineBasicBlock &CurBB,
1636                       const SmallPtrSetImpl<MachineBasicBlock *> &SinkableBBs,
1637                       unsigned Reg, const TargetRegisterInfo *TRI) {
1638   // Try to find a single sinkable successor in which Reg is live-in.
1639   MachineBasicBlock *BB = nullptr;
1640   for (auto *SI : SinkableBBs) {
1641     if (aliasWithRegsInLiveIn(*SI, Reg, TRI)) {
1642       // If BB is set here, Reg is live-in to at least two sinkable successors,
1643       // so quit.
1644       if (BB)
1645         return nullptr;
1646       BB = SI;
1647     }
1648   }
1649   // Reg is not live-in to any sinkable successors.
1650   if (!BB)
1651     return nullptr;
1652 
1653   // Check if any register aliased with Reg is live-in in other successors.
1654   for (auto *SI : CurBB.successors()) {
1655     if (!SinkableBBs.count(SI) && aliasWithRegsInLiveIn(*SI, Reg, TRI))
1656       return nullptr;
1657   }
1658   return BB;
1659 }
1660 
1661 static MachineBasicBlock *
1662 getSingleLiveInSuccBB(MachineBasicBlock &CurBB,
1663                       const SmallPtrSetImpl<MachineBasicBlock *> &SinkableBBs,
1664                       ArrayRef<unsigned> DefedRegsInCopy,
1665                       const TargetRegisterInfo *TRI) {
1666   MachineBasicBlock *SingleBB = nullptr;
1667   for (auto DefReg : DefedRegsInCopy) {
1668     MachineBasicBlock *BB =
1669         getSingleLiveInSuccBB(CurBB, SinkableBBs, DefReg, TRI);
1670     if (!BB || (SingleBB && SingleBB != BB))
1671       return nullptr;
1672     SingleBB = BB;
1673   }
1674   return SingleBB;
1675 }
1676 
1677 static void clearKillFlags(MachineInstr *MI, MachineBasicBlock &CurBB,
1678                            SmallVectorImpl<unsigned> &UsedOpsInCopy,
1679                            LiveRegUnits &UsedRegUnits,
1680                            const TargetRegisterInfo *TRI) {
1681   for (auto U : UsedOpsInCopy) {
1682     MachineOperand &MO = MI->getOperand(U);
1683     Register SrcReg = MO.getReg();
1684     if (!UsedRegUnits.available(SrcReg)) {
1685       MachineBasicBlock::iterator NI = std::next(MI->getIterator());
1686       for (MachineInstr &UI : make_range(NI, CurBB.end())) {
1687         if (UI.killsRegister(SrcReg, TRI)) {
1688           UI.clearRegisterKills(SrcReg, TRI);
1689           MO.setIsKill(true);
1690           break;
1691         }
1692       }
1693     }
1694   }
1695 }
1696 
1697 static void updateLiveIn(MachineInstr *MI, MachineBasicBlock *SuccBB,
1698                          SmallVectorImpl<unsigned> &UsedOpsInCopy,
1699                          SmallVectorImpl<unsigned> &DefedRegsInCopy) {
1700   MachineFunction &MF = *SuccBB->getParent();
1701   const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
1702   for (unsigned DefReg : DefedRegsInCopy)
1703     for (MCSubRegIterator S(DefReg, TRI, true); S.isValid(); ++S)
1704       SuccBB->removeLiveIn(*S);
1705   for (auto U : UsedOpsInCopy) {
1706     Register SrcReg = MI->getOperand(U).getReg();
1707     LaneBitmask Mask;
1708     for (MCRegUnitMaskIterator S(SrcReg, TRI); S.isValid(); ++S) {
1709       Mask |= (*S).second;
1710     }
1711     SuccBB->addLiveIn(SrcReg, Mask.any() ? Mask : LaneBitmask::getAll());
1712   }
1713   SuccBB->sortUniqueLiveIns();
1714 }
1715 
1716 static bool hasRegisterDependency(MachineInstr *MI,
1717                                   SmallVectorImpl<unsigned> &UsedOpsInCopy,
1718                                   SmallVectorImpl<unsigned> &DefedRegsInCopy,
1719                                   LiveRegUnits &ModifiedRegUnits,
1720                                   LiveRegUnits &UsedRegUnits) {
1721   bool HasRegDependency = false;
1722   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
1723     MachineOperand &MO = MI->getOperand(i);
1724     if (!MO.isReg())
1725       continue;
1726     Register Reg = MO.getReg();
1727     if (!Reg)
1728       continue;
1729     if (MO.isDef()) {
1730       if (!ModifiedRegUnits.available(Reg) || !UsedRegUnits.available(Reg)) {
1731         HasRegDependency = true;
1732         break;
1733       }
1734       DefedRegsInCopy.push_back(Reg);
1735 
1736       // FIXME: instead of isUse(), readsReg() would be a better fix here,
1737       // For example, we can ignore modifications in reg with undef. However,
1738       // it's not perfectly clear if skipping the internal read is safe in all
1739       // other targets.
1740     } else if (MO.isUse()) {
1741       if (!ModifiedRegUnits.available(Reg)) {
1742         HasRegDependency = true;
1743         break;
1744       }
1745       UsedOpsInCopy.push_back(i);
1746     }
1747   }
1748   return HasRegDependency;
1749 }
1750 
1751 bool PostRAMachineSinking::tryToSinkCopy(MachineBasicBlock &CurBB,
1752                                          MachineFunction &MF,
1753                                          const TargetRegisterInfo *TRI,
1754                                          const TargetInstrInfo *TII) {
1755   SmallPtrSet<MachineBasicBlock *, 2> SinkableBBs;
1756   // FIXME: For now, we sink only to a successor which has a single predecessor
1757   // so that we can directly sink COPY instructions to the successor without
1758   // adding any new block or branch instruction.
1759   for (MachineBasicBlock *SI : CurBB.successors())
1760     if (!SI->livein_empty() && SI->pred_size() == 1)
1761       SinkableBBs.insert(SI);
1762 
1763   if (SinkableBBs.empty())
1764     return false;
1765 
1766   bool Changed = false;
1767 
1768   // Track which registers have been modified and used between the end of the
1769   // block and the current instruction.
1770   ModifiedRegUnits.clear();
1771   UsedRegUnits.clear();
1772   SeenDbgInstrs.clear();
1773 
1774   for (MachineInstr &MI : llvm::make_early_inc_range(llvm::reverse(CurBB))) {
1775     // Track the operand index for use in Copy.
1776     SmallVector<unsigned, 2> UsedOpsInCopy;
1777     // Track the register number defed in Copy.
1778     SmallVector<unsigned, 2> DefedRegsInCopy;
1779 
1780     // We must sink this DBG_VALUE if its operand is sunk. To avoid searching
1781     // for DBG_VALUEs later, record them when they're encountered.
1782     if (MI.isDebugValue() && !MI.isDebugRef()) {
1783       SmallDenseMap<MCRegister, SmallVector<unsigned, 2>, 4> MIUnits;
1784       bool IsValid = true;
1785       for (MachineOperand &MO : MI.debug_operands()) {
1786         if (MO.isReg() && MO.getReg().isPhysical()) {
1787           // Bail if we can already tell the sink would be rejected, rather
1788           // than needlessly accumulating lots of DBG_VALUEs.
1789           if (hasRegisterDependency(&MI, UsedOpsInCopy, DefedRegsInCopy,
1790                                     ModifiedRegUnits, UsedRegUnits)) {
1791             IsValid = false;
1792             break;
1793           }
1794 
1795           // Record debug use of each reg unit.
1796           for (auto RI = MCRegUnitIterator(MO.getReg(), TRI); RI.isValid();
1797                ++RI)
1798             MIUnits[*RI].push_back(MO.getReg());
1799         }
1800       }
1801       if (IsValid) {
1802         for (auto &RegOps : MIUnits)
1803           SeenDbgInstrs[RegOps.first].emplace_back(&MI,
1804                                                    std::move(RegOps.second));
1805       }
1806       continue;
1807     }
1808 
1809     if (MI.isDebugOrPseudoInstr())
1810       continue;
1811 
1812     // Do not move any instruction across function call.
1813     if (MI.isCall())
1814       return false;
1815 
1816     if (!MI.isCopy() || !MI.getOperand(0).isRenamable()) {
1817       LiveRegUnits::accumulateUsedDefed(MI, ModifiedRegUnits, UsedRegUnits,
1818                                         TRI);
1819       continue;
1820     }
1821 
1822     // Don't sink the COPY if it would violate a register dependency.
1823     if (hasRegisterDependency(&MI, UsedOpsInCopy, DefedRegsInCopy,
1824                               ModifiedRegUnits, UsedRegUnits)) {
1825       LiveRegUnits::accumulateUsedDefed(MI, ModifiedRegUnits, UsedRegUnits,
1826                                         TRI);
1827       continue;
1828     }
1829     assert((!UsedOpsInCopy.empty() && !DefedRegsInCopy.empty()) &&
1830            "Unexpect SrcReg or DefReg");
1831     MachineBasicBlock *SuccBB =
1832         getSingleLiveInSuccBB(CurBB, SinkableBBs, DefedRegsInCopy, TRI);
1833     // Don't sink if we cannot find a single sinkable successor in which Reg
1834     // is live-in.
1835     if (!SuccBB) {
1836       LiveRegUnits::accumulateUsedDefed(MI, ModifiedRegUnits, UsedRegUnits,
1837                                         TRI);
1838       continue;
1839     }
1840     assert((SuccBB->pred_size() == 1 && *SuccBB->pred_begin() == &CurBB) &&
1841            "Unexpected predecessor");
1842 
1843     // Collect DBG_VALUEs that must sink with this copy. We've previously
1844     // recorded which reg units that DBG_VALUEs read, if this instruction
1845     // writes any of those units then the corresponding DBG_VALUEs must sink.
1846     MapVector<MachineInstr *, MIRegs::second_type> DbgValsToSinkMap;
1847     for (auto &MO : MI.operands()) {
1848       if (!MO.isReg() || !MO.isDef())
1849         continue;
1850 
1851       for (auto RI = MCRegUnitIterator(MO.getReg(), TRI); RI.isValid(); ++RI) {
1852         for (const auto &MIRegs : SeenDbgInstrs.lookup(*RI)) {
1853           auto &Regs = DbgValsToSinkMap[MIRegs.first];
1854           for (unsigned Reg : MIRegs.second)
1855             Regs.push_back(Reg);
1856         }
1857       }
1858     }
1859     auto DbgValsToSink = DbgValsToSinkMap.takeVector();
1860 
1861     LLVM_DEBUG(dbgs() << "Sink instr " << MI << "\tinto block " << *SuccBB);
1862 
1863     MachineBasicBlock::iterator InsertPos =
1864         SuccBB->SkipPHIsAndLabels(SuccBB->begin());
1865     if (blockPrologueInterferes(SuccBB, InsertPos, MI, TRI, TII, nullptr)) {
1866       LLVM_DEBUG(
1867           dbgs() << " *** Not sinking: prologue interference\n");
1868       continue;
1869     }
1870 
1871     // Clear the kill flag if SrcReg is killed between MI and the end of the
1872     // block.
1873     clearKillFlags(&MI, CurBB, UsedOpsInCopy, UsedRegUnits, TRI);
1874     performSink(MI, *SuccBB, InsertPos, DbgValsToSink);
1875     updateLiveIn(&MI, SuccBB, UsedOpsInCopy, DefedRegsInCopy);
1876 
1877     Changed = true;
1878     ++NumPostRACopySink;
1879   }
1880   return Changed;
1881 }
1882 
1883 bool PostRAMachineSinking::runOnMachineFunction(MachineFunction &MF) {
1884   if (skipFunction(MF.getFunction()))
1885     return false;
1886 
1887   bool Changed = false;
1888   const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
1889   const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
1890 
1891   ModifiedRegUnits.init(*TRI);
1892   UsedRegUnits.init(*TRI);
1893   for (auto &BB : MF)
1894     Changed |= tryToSinkCopy(BB, MF, TRI, TII);
1895 
1896   return Changed;
1897 }
1898