1 //===-- MachineSink.cpp - Sinking for machine instructions ----------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This pass moves instructions into successor blocks when possible, so that
11 // they aren't executed on paths where their results aren't needed.
13 // This pass is not intended to be a replacement or a complete alternative
14 // for an LLVM-IR-level sinking pass. It is only designed to sink simple
15 // constructs that are not exposed before lowering and instruction selection.
17 //===----------------------------------------------------------------------===//
19 #define DEBUG_TYPE "machine-sink"
20 #include "llvm/CodeGen/Passes.h"
21 #include "llvm/CodeGen/MachineRegisterInfo.h"
22 #include "llvm/CodeGen/MachineDominators.h"
23 #include "llvm/CodeGen/MachineLoopInfo.h"
24 #include "llvm/Analysis/AliasAnalysis.h"
25 #include "llvm/Target/TargetRegisterInfo.h"
26 #include "llvm/Target/TargetInstrInfo.h"
27 #include "llvm/Target/TargetMachine.h"
28 #include "llvm/ADT/SmallSet.h"
29 #include "llvm/ADT/Statistic.h"
30 #include "llvm/Support/CommandLine.h"
31 #include "llvm/Support/Debug.h"
32 #include "llvm/Support/raw_ostream.h"
36 SplitEdges("machine-sink-split",
37 cl::desc("Split critical edges during machine sinking"),
38 cl::init(true), cl::Hidden);
40 STATISTIC(NumSunk, "Number of machine instructions sunk");
41 STATISTIC(NumSplit, "Number of critical edges split");
42 STATISTIC(NumCoalesces, "Number of copies coalesced");
45 class MachineSinking : public MachineFunctionPass {
46 const TargetInstrInfo *TII;
47 const TargetRegisterInfo *TRI;
48 MachineRegisterInfo *MRI; // Machine register information
49 MachineDominatorTree *DT; // Machine dominator tree
52 BitVector AllocatableSet; // Which physregs are allocatable?
54 // Remember which edges have been considered for breaking.
55 SmallSet<std::pair<MachineBasicBlock*,MachineBasicBlock*>, 8>
59 static char ID; // Pass identification
60 MachineSinking() : MachineFunctionPass(ID) {
61 initializeMachineSinkingPass(*PassRegistry::getPassRegistry());
64 virtual bool runOnMachineFunction(MachineFunction &MF);
66 virtual void getAnalysisUsage(AnalysisUsage &AU) const {
68 MachineFunctionPass::getAnalysisUsage(AU);
69 AU.addRequired<AliasAnalysis>();
70 AU.addRequired<MachineDominatorTree>();
71 AU.addRequired<MachineLoopInfo>();
72 AU.addPreserved<MachineDominatorTree>();
73 AU.addPreserved<MachineLoopInfo>();
76 virtual void releaseMemory() {
77 CEBCandidates.clear();
81 bool ProcessBlock(MachineBasicBlock &MBB);
82 bool isWorthBreakingCriticalEdge(MachineInstr *MI,
83 MachineBasicBlock *From,
84 MachineBasicBlock *To);
85 MachineBasicBlock *SplitCriticalEdge(MachineInstr *MI,
86 MachineBasicBlock *From,
87 MachineBasicBlock *To,
89 bool SinkInstruction(MachineInstr *MI, bool &SawStore);
90 bool AllUsesDominatedByBlock(unsigned Reg, MachineBasicBlock *MBB,
91 MachineBasicBlock *DefMBB,
92 bool &BreakPHIEdge, bool &LocalUse) const;
93 MachineBasicBlock *FindSuccToSinkTo(MachineInstr *MI, MachineBasicBlock *MBB,
95 bool isProfitableToSinkTo(unsigned Reg, MachineInstr *MI,
96 MachineBasicBlock *MBB,
97 MachineBasicBlock *SuccToSinkTo);
99 bool PerformTrivialForwardCoalescing(MachineInstr *MI,
100 MachineBasicBlock *MBB);
103 // SuccessorSorter - Sort Successors according to their loop depth.
104 struct SuccessorSorter {
105 SuccessorSorter(MachineLoopInfo *LoopInfo) : LI(LoopInfo) {}
106 bool operator()(const MachineBasicBlock *LHS,
107 const MachineBasicBlock *RHS) const {
108 return LI->getLoopDepth(LHS) < LI->getLoopDepth(RHS);
112 } // end anonymous namespace
114 char MachineSinking::ID = 0;
115 char &llvm::MachineSinkingID = MachineSinking::ID;
116 INITIALIZE_PASS_BEGIN(MachineSinking, "machine-sink",
117 "Machine code sinking", false, false)
118 INITIALIZE_PASS_DEPENDENCY(MachineDominatorTree)
119 INITIALIZE_PASS_DEPENDENCY(MachineLoopInfo)
120 INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
121 INITIALIZE_PASS_END(MachineSinking, "machine-sink",
122 "Machine code sinking", false, false)
124 bool MachineSinking::PerformTrivialForwardCoalescing(MachineInstr *MI,
125 MachineBasicBlock *MBB) {
129 unsigned SrcReg = MI->getOperand(1).getReg();
130 unsigned DstReg = MI->getOperand(0).getReg();
131 if (!TargetRegisterInfo::isVirtualRegister(SrcReg) ||
132 !TargetRegisterInfo::isVirtualRegister(DstReg) ||
133 !MRI->hasOneNonDBGUse(SrcReg))
136 const TargetRegisterClass *SRC = MRI->getRegClass(SrcReg);
137 const TargetRegisterClass *DRC = MRI->getRegClass(DstReg);
141 MachineInstr *DefMI = MRI->getVRegDef(SrcReg);
142 if (DefMI->isCopyLike())
144 DEBUG(dbgs() << "Coalescing: " << *DefMI);
145 DEBUG(dbgs() << "*** to: " << *MI);
146 MRI->replaceRegWith(DstReg, SrcReg);
147 MI->eraseFromParent();
152 /// AllUsesDominatedByBlock - Return true if all uses of the specified register
153 /// occur in blocks dominated by the specified block. If any use is in the
154 /// definition block, then return false since it is never legal to move def
157 MachineSinking::AllUsesDominatedByBlock(unsigned Reg,
158 MachineBasicBlock *MBB,
159 MachineBasicBlock *DefMBB,
161 bool &LocalUse) const {
162 assert(TargetRegisterInfo::isVirtualRegister(Reg) &&
163 "Only makes sense for vregs");
165 // Ignore debug uses because debug info doesn't affect the code.
166 if (MRI->use_nodbg_empty(Reg))
169 // BreakPHIEdge is true if all the uses are in the successor MBB being sunken
170 // into and they are all PHI nodes. In this case, machine-sink must break
171 // the critical edge first. e.g.
173 // BB#1: derived from LLVM BB %bb4.preheader
174 // Predecessors according to CFG: BB#0
176 // %reg16385<def> = DEC64_32r %reg16437, %EFLAGS<imp-def,dead>
178 // JE_4 <BB#37>, %EFLAGS<imp-use>
179 // Successors according to CFG: BB#37 BB#2
181 // BB#2: derived from LLVM BB %bb.nph
182 // Predecessors according to CFG: BB#0 BB#1
183 // %reg16386<def> = PHI %reg16434, <BB#0>, %reg16385, <BB#1>
185 for (MachineRegisterInfo::use_nodbg_iterator
186 I = MRI->use_nodbg_begin(Reg), E = MRI->use_nodbg_end();
188 MachineInstr *UseInst = &*I;
189 MachineBasicBlock *UseBlock = UseInst->getParent();
190 if (!(UseBlock == MBB && UseInst->isPHI() &&
191 UseInst->getOperand(I.getOperandNo()+1).getMBB() == DefMBB)) {
192 BreakPHIEdge = false;
199 for (MachineRegisterInfo::use_nodbg_iterator
200 I = MRI->use_nodbg_begin(Reg), E = MRI->use_nodbg_end();
202 // Determine the block of the use.
203 MachineInstr *UseInst = &*I;
204 MachineBasicBlock *UseBlock = UseInst->getParent();
205 if (UseInst->isPHI()) {
206 // PHI nodes use the operand in the predecessor block, not the block with
208 UseBlock = UseInst->getOperand(I.getOperandNo()+1).getMBB();
209 } else if (UseBlock == DefMBB) {
214 // Check that it dominates.
215 if (!DT->dominates(MBB, UseBlock))
222 bool MachineSinking::runOnMachineFunction(MachineFunction &MF) {
223 DEBUG(dbgs() << "******** Machine Sinking ********\n");
225 const TargetMachine &TM = MF.getTarget();
226 TII = TM.getInstrInfo();
227 TRI = TM.getRegisterInfo();
228 MRI = &MF.getRegInfo();
229 DT = &getAnalysis<MachineDominatorTree>();
230 LI = &getAnalysis<MachineLoopInfo>();
231 AA = &getAnalysis<AliasAnalysis>();
232 AllocatableSet = TRI->getAllocatableSet(MF);
234 bool EverMadeChange = false;
237 bool MadeChange = false;
239 // Process all basic blocks.
240 CEBCandidates.clear();
241 for (MachineFunction::iterator I = MF.begin(), E = MF.end();
243 MadeChange |= ProcessBlock(*I);
245 // If this iteration over the code changed anything, keep iterating.
246 if (!MadeChange) break;
247 EverMadeChange = true;
249 return EverMadeChange;
252 bool MachineSinking::ProcessBlock(MachineBasicBlock &MBB) {
253 // Can't sink anything out of a block that has less than two successors.
254 if (MBB.succ_size() <= 1 || MBB.empty()) return false;
256 // Don't bother sinking code out of unreachable blocks. In addition to being
257 // unprofitable, it can also lead to infinite looping, because in an
258 // unreachable loop there may be nowhere to stop.
259 if (!DT->isReachableFromEntry(&MBB)) return false;
261 bool MadeChange = false;
263 // Walk the basic block bottom-up. Remember if we saw a store.
264 MachineBasicBlock::iterator I = MBB.end();
266 bool ProcessedBegin, SawStore = false;
268 MachineInstr *MI = I; // The instruction to sink.
270 // Predecrement I (if it's not begin) so that it isn't invalidated by
272 ProcessedBegin = I == MBB.begin();
276 if (MI->isDebugValue())
279 bool Joined = PerformTrivialForwardCoalescing(MI, &MBB);
285 if (SinkInstruction(MI, SawStore))
286 ++NumSunk, MadeChange = true;
288 // If we just processed the first instruction in the block, we're done.
289 } while (!ProcessedBegin);
294 bool MachineSinking::isWorthBreakingCriticalEdge(MachineInstr *MI,
295 MachineBasicBlock *From,
296 MachineBasicBlock *To) {
297 // FIXME: Need much better heuristics.
299 // If the pass has already considered breaking this edge (during this pass
300 // through the function), then let's go ahead and break it. This means
301 // sinking multiple "cheap" instructions into the same block.
302 if (!CEBCandidates.insert(std::make_pair(From, To)))
305 if (!MI->isCopy() && !MI->isAsCheapAsAMove())
308 // MI is cheap, we probably don't want to break the critical edge for it.
309 // However, if this would allow some definitions of its source operands
310 // to be sunk then it's probably worth it.
311 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
312 const MachineOperand &MO = MI->getOperand(i);
313 if (!MO.isReg()) continue;
314 unsigned Reg = MO.getReg();
315 if (Reg == 0 || !TargetRegisterInfo::isPhysicalRegister(Reg))
317 if (MRI->hasOneNonDBGUse(Reg))
324 MachineBasicBlock *MachineSinking::SplitCriticalEdge(MachineInstr *MI,
325 MachineBasicBlock *FromBB,
326 MachineBasicBlock *ToBB,
328 if (!isWorthBreakingCriticalEdge(MI, FromBB, ToBB))
331 // Avoid breaking back edge. From == To means backedge for single BB loop.
332 if (!SplitEdges || FromBB == ToBB)
335 // Check for backedges of more "complex" loops.
336 if (LI->getLoopFor(FromBB) == LI->getLoopFor(ToBB) &&
337 LI->isLoopHeader(ToBB))
340 // It's not always legal to break critical edges and sink the computation
348 // ... no uses of v1024
354 // If BB#1 -> BB#3 edge is broken and computation of v1024 is inserted:
363 // ... no uses of v1024
369 // This is incorrect since v1024 is not computed along the BB#1->BB#2->BB#3
370 // flow. We need to ensure the new basic block where the computation is
371 // sunk to dominates all the uses.
372 // It's only legal to break critical edge and sink the computation to the
373 // new block if all the predecessors of "To", except for "From", are
374 // not dominated by "From". Given SSA property, this means these
375 // predecessors are dominated by "To".
377 // There is no need to do this check if all the uses are PHI nodes. PHI
378 // sources are only defined on the specific predecessor edges.
380 for (MachineBasicBlock::pred_iterator PI = ToBB->pred_begin(),
381 E = ToBB->pred_end(); PI != E; ++PI) {
384 if (!DT->dominates(ToBB, *PI))
389 return FromBB->SplitCriticalEdge(ToBB, this);
392 static bool AvoidsSinking(MachineInstr *MI, MachineRegisterInfo *MRI) {
393 return MI->isInsertSubreg() || MI->isSubregToReg() || MI->isRegSequence();
396 /// collectDebgValues - Scan instructions following MI and collect any
397 /// matching DBG_VALUEs.
398 static void collectDebugValues(MachineInstr *MI,
399 SmallVector<MachineInstr *, 2> & DbgValues) {
401 if (!MI->getOperand(0).isReg())
404 MachineBasicBlock::iterator DI = MI; ++DI;
405 for (MachineBasicBlock::iterator DE = MI->getParent()->end();
407 if (!DI->isDebugValue())
409 if (DI->getOperand(0).isReg() &&
410 DI->getOperand(0).getReg() == MI->getOperand(0).getReg())
411 DbgValues.push_back(DI);
415 /// isPostDominatedBy - Return true if A is post dominated by B.
416 static bool isPostDominatedBy(MachineBasicBlock *A, MachineBasicBlock *B) {
418 // FIXME - Use real post dominator.
419 if (A->succ_size() != 2)
421 MachineBasicBlock::succ_iterator I = A->succ_begin();
424 MachineBasicBlock *OtherSuccBlock = *I;
425 if (OtherSuccBlock->succ_size() != 1 ||
426 *(OtherSuccBlock->succ_begin()) != B)
432 /// isProfitableToSinkTo - Return true if it is profitable to sink MI.
433 bool MachineSinking::isProfitableToSinkTo(unsigned Reg, MachineInstr *MI,
434 MachineBasicBlock *MBB,
435 MachineBasicBlock *SuccToSinkTo) {
436 assert (MI && "Invalid MachineInstr!");
437 assert (SuccToSinkTo && "Invalid SinkTo Candidate BB");
439 if (MBB == SuccToSinkTo)
442 // It is profitable if SuccToSinkTo does not post dominate current block.
443 if (!isPostDominatedBy(MBB, SuccToSinkTo))
446 // Check if only use in post dominated block is PHI instruction.
447 bool NonPHIUse = false;
448 for (MachineRegisterInfo::use_nodbg_iterator
449 I = MRI->use_nodbg_begin(Reg), E = MRI->use_nodbg_end();
451 MachineInstr *UseInst = &*I;
452 MachineBasicBlock *UseBlock = UseInst->getParent();
453 if (UseBlock == SuccToSinkTo && !UseInst->isPHI())
459 // If SuccToSinkTo post dominates then also it may be profitable if MI
460 // can further profitably sinked into another block in next round.
461 bool BreakPHIEdge = false;
462 // FIXME - If finding successor is compile time expensive then catch results.
463 if (MachineBasicBlock *MBB2 = FindSuccToSinkTo(MI, SuccToSinkTo, BreakPHIEdge))
464 return isProfitableToSinkTo(Reg, MI, SuccToSinkTo, MBB2);
466 // If SuccToSinkTo is final destination and it is a post dominator of current
467 // block then it is not profitable to sink MI into SuccToSinkTo block.
471 /// FindSuccToSinkTo - Find a successor to sink this instruction to.
472 MachineBasicBlock *MachineSinking::FindSuccToSinkTo(MachineInstr *MI,
473 MachineBasicBlock *MBB,
474 bool &BreakPHIEdge) {
476 assert (MI && "Invalid MachineInstr!");
477 assert (MBB && "Invalid MachineBasicBlock!");
479 // Loop over all the operands of the specified instruction. If there is
480 // anything we can't handle, bail out.
482 // SuccToSinkTo - This is the successor to sink this instruction to, once we
484 MachineBasicBlock *SuccToSinkTo = 0;
485 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
486 const MachineOperand &MO = MI->getOperand(i);
487 if (!MO.isReg()) continue; // Ignore non-register operands.
489 unsigned Reg = MO.getReg();
490 if (Reg == 0) continue;
492 if (TargetRegisterInfo::isPhysicalRegister(Reg)) {
494 // If the physreg has no defs anywhere, it's just an ambient register
495 // and we can freely move its uses. Alternatively, if it's allocatable,
496 // it could get allocated to something with a def during allocation.
497 if (!MRI->isConstantPhysReg(Reg, *MBB->getParent()))
499 } else if (!MO.isDead()) {
500 // A def that isn't dead. We can't move it.
504 // Virtual register uses are always safe to sink.
505 if (MO.isUse()) continue;
507 // If it's not safe to move defs of the register class, then abort.
508 if (!TII->isSafeToMoveRegClassDefs(MRI->getRegClass(Reg)))
511 // FIXME: This picks a successor to sink into based on having one
512 // successor that dominates all the uses. However, there are cases where
513 // sinking can happen but where the sink point isn't a successor. For
520 // the instruction could be sunk over the whole diamond for the
521 // if/then/else (or loop, etc), allowing it to be sunk into other blocks
524 // Virtual register defs can only be sunk if all their uses are in blocks
525 // dominated by one of the successors.
527 // If a previous operand picked a block to sink to, then this operand
528 // must be sinkable to the same block.
529 bool LocalUse = false;
530 if (!AllUsesDominatedByBlock(Reg, SuccToSinkTo, MBB,
531 BreakPHIEdge, LocalUse))
537 // Otherwise, we should look at all the successors and decide which one
538 // we should sink to.
539 // We give successors with smaller loop depth higher priority.
540 SmallVector<MachineBasicBlock*, 4> Succs(MBB->succ_begin(), MBB->succ_end());
541 std::stable_sort(Succs.begin(), Succs.end(), SuccessorSorter(LI));
542 for (SmallVector<MachineBasicBlock*, 4>::iterator SI = Succs.begin(),
543 E = Succs.end(); SI != E; ++SI) {
544 MachineBasicBlock *SuccBlock = *SI;
545 bool LocalUse = false;
546 if (AllUsesDominatedByBlock(Reg, SuccBlock, MBB,
547 BreakPHIEdge, LocalUse)) {
548 SuccToSinkTo = SuccBlock;
552 // Def is used locally, it's never safe to move this def.
556 // If we couldn't find a block to sink to, ignore this instruction.
557 if (SuccToSinkTo == 0)
559 else if (!isProfitableToSinkTo(Reg, MI, MBB, SuccToSinkTo))
564 // It is not possible to sink an instruction into its own block. This can
565 // happen with loops.
566 if (MBB == SuccToSinkTo)
569 // It's not safe to sink instructions to EH landing pad. Control flow into
570 // landing pad is implicitly defined.
571 if (SuccToSinkTo && SuccToSinkTo->isLandingPad())
577 /// SinkInstruction - Determine whether it is safe to sink the specified machine
578 /// instruction out of its current block into a successor.
579 bool MachineSinking::SinkInstruction(MachineInstr *MI, bool &SawStore) {
580 // Don't sink insert_subreg, subreg_to_reg, reg_sequence. These are meant to
581 // be close to the source to make it easier to coalesce.
582 if (AvoidsSinking(MI, MRI))
585 // Check if it's safe to move the instruction.
586 if (!MI->isSafeToMove(TII, AA, SawStore))
589 // FIXME: This should include support for sinking instructions within the
590 // block they are currently in to shorten the live ranges. We often get
591 // instructions sunk into the top of a large block, but it would be better to
592 // also sink them down before their first use in the block. This xform has to
593 // be careful not to *increase* register pressure though, e.g. sinking
594 // "x = y + z" down if it kills y and z would increase the live ranges of y
595 // and z and only shrink the live range of x.
597 bool BreakPHIEdge = false;
598 MachineBasicBlock *ParentBlock = MI->getParent();
599 MachineBasicBlock *SuccToSinkTo = FindSuccToSinkTo(MI, ParentBlock, BreakPHIEdge);
601 // If there are no outputs, it must have side-effects.
602 if (SuccToSinkTo == 0)
606 // If the instruction to move defines a dead physical register which is live
607 // when leaving the basic block, don't move it because it could turn into a
608 // "zombie" define of that preg. E.g., EFLAGS. (<rdar://problem/8030636>)
609 for (unsigned I = 0, E = MI->getNumOperands(); I != E; ++I) {
610 const MachineOperand &MO = MI->getOperand(I);
611 if (!MO.isReg()) continue;
612 unsigned Reg = MO.getReg();
613 if (Reg == 0 || !TargetRegisterInfo::isPhysicalRegister(Reg)) continue;
614 if (SuccToSinkTo->isLiveIn(Reg))
618 DEBUG(dbgs() << "Sink instr " << *MI << "\tinto block " << *SuccToSinkTo);
620 // If the block has multiple predecessors, this would introduce computation on
621 // a path that it doesn't already exist. We could split the critical edge,
622 // but for now we just punt.
623 if (SuccToSinkTo->pred_size() > 1) {
624 // We cannot sink a load across a critical edge - there may be stores in
626 bool TryBreak = false;
628 if (!MI->isSafeToMove(TII, AA, store)) {
629 DEBUG(dbgs() << " *** NOTE: Won't sink load along critical edge.\n");
633 // We don't want to sink across a critical edge if we don't dominate the
634 // successor. We could be introducing calculations to new code paths.
635 if (!TryBreak && !DT->dominates(ParentBlock, SuccToSinkTo)) {
636 DEBUG(dbgs() << " *** NOTE: Critical edge found\n");
640 // Don't sink instructions into a loop.
641 if (!TryBreak && LI->isLoopHeader(SuccToSinkTo)) {
642 DEBUG(dbgs() << " *** NOTE: Loop header found\n");
646 // Otherwise we are OK with sinking along a critical edge.
648 DEBUG(dbgs() << "Sinking along critical edge.\n");
650 MachineBasicBlock *NewSucc =
651 SplitCriticalEdge(MI, ParentBlock, SuccToSinkTo, BreakPHIEdge);
653 DEBUG(dbgs() << " *** PUNTING: Not legal or profitable to "
654 "break critical edge\n");
657 DEBUG(dbgs() << " *** Splitting critical edge:"
658 " BB#" << ParentBlock->getNumber()
659 << " -- BB#" << NewSucc->getNumber()
660 << " -- BB#" << SuccToSinkTo->getNumber() << '\n');
661 SuccToSinkTo = NewSucc;
663 BreakPHIEdge = false;
669 // BreakPHIEdge is true if all the uses are in the successor MBB being
670 // sunken into and they are all PHI nodes. In this case, machine-sink must
671 // break the critical edge first.
672 MachineBasicBlock *NewSucc = SplitCriticalEdge(MI, ParentBlock,
673 SuccToSinkTo, BreakPHIEdge);
675 DEBUG(dbgs() << " *** PUNTING: Not legal or profitable to "
676 "break critical edge\n");
680 DEBUG(dbgs() << " *** Splitting critical edge:"
681 " BB#" << ParentBlock->getNumber()
682 << " -- BB#" << NewSucc->getNumber()
683 << " -- BB#" << SuccToSinkTo->getNumber() << '\n');
684 SuccToSinkTo = NewSucc;
688 // Determine where to insert into. Skip phi nodes.
689 MachineBasicBlock::iterator InsertPos = SuccToSinkTo->begin();
690 while (InsertPos != SuccToSinkTo->end() && InsertPos->isPHI())
693 // collect matching debug values.
694 SmallVector<MachineInstr *, 2> DbgValuesToSink;
695 collectDebugValues(MI, DbgValuesToSink);
697 // Move the instruction.
698 SuccToSinkTo->splice(InsertPos, ParentBlock, MI,
699 ++MachineBasicBlock::iterator(MI));
701 // Move debug values.
702 for (SmallVector<MachineInstr *, 2>::iterator DBI = DbgValuesToSink.begin(),
703 DBE = DbgValuesToSink.end(); DBI != DBE; ++DBI) {
704 MachineInstr *DbgMI = *DBI;
705 SuccToSinkTo->splice(InsertPos, ParentBlock, DbgMI,
706 ++MachineBasicBlock::iterator(DbgMI));
709 // Conservatively, clear any kill flags, since it's possible that they are no