1//===- ShrinkWrap.cpp - Compute safe point for prolog/epilog insertion ----===//
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 looks for safe point where the prologue and epilogue can be
10// inserted.
11// The safe point for the prologue (resp. epilogue) is called Save
12// (resp. Restore).
13// A point is safe for prologue (resp. epilogue) if and only if
14// it 1) dominates (resp. post-dominates) all the frame related operations and
15// between 2) two executions of the Save (resp. Restore) point there is an
16// execution of the Restore (resp. Save) point.
17//
18// For instance, the following points are safe:
19// for (int i = 0; i < 10; ++i) {
20// Save
21// ...
22// Restore
23// }
24// Indeed, the execution looks like Save -> Restore -> Save -> Restore ...
25// And the following points are not:
26// for (int i = 0; i < 10; ++i) {
27// Save
28// ...
29// }
30// for (int i = 0; i < 10; ++i) {
31// ...
32// Restore
33// }
34// Indeed, the execution looks like Save -> Save -> ... -> Restore -> Restore.
35//
36// This pass also ensures that the safe points are 3) cheaper than the regular
37// entry and exits blocks.
38//
39// Property #1 is ensured via the use of MachineDominatorTree and
40// MachinePostDominatorTree.
41// Property #2 is ensured via property #1 and MachineLoopInfo, i.e., both
42// points must be in the same loop.
43// Property #3 is ensured via the MachineBlockFrequencyInfo.
44//
45// If this pass found points matching all these properties, then
46// MachineFrameInfo is updated with this information.
47//
48//===----------------------------------------------------------------------===//
49
50#include "llvm/CodeGen/ShrinkWrap.h"
51#include "llvm/ADT/BitVector.h"
52#include "llvm/ADT/PostOrderIterator.h"
53#include "llvm/ADT/SetVector.h"
54#include "llvm/ADT/SmallVector.h"
55#include "llvm/ADT/Statistic.h"
56#include "llvm/Analysis/CFG.h"
57#include "llvm/Analysis/ValueTracking.h"
58#include "llvm/CodeGen/MachineBasicBlock.h"
59#include "llvm/CodeGen/MachineBlockFrequencyInfo.h"
60#include "llvm/CodeGen/MachineDominators.h"
61#include "llvm/CodeGen/MachineFrameInfo.h"
62#include "llvm/CodeGen/MachineFunction.h"
63#include "llvm/CodeGen/MachineFunctionPass.h"
64#include "llvm/CodeGen/MachineInstr.h"
65#include "llvm/CodeGen/MachineLoopInfo.h"
66#include "llvm/CodeGen/MachineOperand.h"
67#include "llvm/CodeGen/MachineOptimizationRemarkEmitter.h"
68#include "llvm/CodeGen/MachinePostDominators.h"
69#include "llvm/CodeGen/RegisterClassInfo.h"
70#include "llvm/CodeGen/RegisterScavenging.h"
71#include "llvm/CodeGen/TargetFrameLowering.h"
72#include "llvm/CodeGen/TargetInstrInfo.h"
73#include "llvm/CodeGen/TargetLowering.h"
74#include "llvm/CodeGen/TargetRegisterInfo.h"
75#include "llvm/CodeGen/TargetSubtargetInfo.h"
76#include "llvm/IR/Attributes.h"
77#include "llvm/IR/Function.h"
78#include "llvm/InitializePasses.h"
79#include "llvm/MC/MCAsmInfo.h"
80#include "llvm/Pass.h"
81#include "llvm/Support/CommandLine.h"
82#include "llvm/Support/Debug.h"
83#include "llvm/Support/ErrorHandling.h"
84#include "llvm/Support/raw_ostream.h"
85#include "llvm/Target/TargetMachine.h"
86#include <cassert>
87#include <memory>
88
89using namespace llvm;
90
91#define DEBUG_TYPE "shrink-wrap"
92
93STATISTIC(NumFunc, "Number of functions");
94STATISTIC(NumCandidates, "Number of shrink-wrapping candidates");
95STATISTIC(NumCandidatesDropped,
96 "Number of shrink-wrapping candidates dropped because of frequency");
97
98static cl::opt<cl::boolOrDefault>
99EnableShrinkWrapOpt("enable-shrink-wrap", cl::Hidden,
100 cl::desc("enable the shrink-wrapping pass"));
101static cl::opt<bool> EnablePostShrinkWrapOpt(
102 "enable-shrink-wrap-region-split", cl::init(Val: true), cl::Hidden,
103 cl::desc("enable splitting of the restore block if possible"));
104
105namespace {
106
107/// Class to determine where the safe point to insert the
108/// prologue and epilogue are.
109/// Unlike the paper from Fred C. Chow, PLDI'88, that introduces the
110/// shrink-wrapping term for prologue/epilogue placement, this pass
111/// does not rely on expensive data-flow analysis. Instead we use the
112/// dominance properties and loop information to decide which point
113/// are safe for such insertion.
114class ShrinkWrapImpl {
115 /// Hold callee-saved information.
116 const RegisterClassInfo *RCI = nullptr;
117 MachineDominatorTree *MDT = nullptr;
118 MachinePostDominatorTree *MPDT = nullptr;
119
120 /// Current safe point found for the prologue.
121 /// The prologue will be inserted before the first instruction
122 /// in this basic block.
123 MachineBasicBlock *Save = nullptr;
124
125 /// Current safe point found for the epilogue.
126 /// The epilogue will be inserted before the first terminator instruction
127 /// in this basic block.
128 MachineBasicBlock *Restore = nullptr;
129
130 /// Hold the information of the basic block frequency.
131 /// Use to check the profitability of the new points.
132 MachineBlockFrequencyInfo *MBFI = nullptr;
133
134 /// Hold the loop information. Used to determine if Save and Restore
135 /// are in the same loop.
136 MachineLoopInfo *MLI = nullptr;
137
138 // Emit remarks.
139 MachineOptimizationRemarkEmitter *ORE = nullptr;
140
141 /// Frequency of the Entry block.
142 BlockFrequency EntryFreq;
143
144 /// Current opcode for frame setup.
145 unsigned FrameSetupOpcode = ~0u;
146
147 /// Current opcode for frame destroy.
148 unsigned FrameDestroyOpcode = ~0u;
149
150 /// Stack pointer register, used by llvm.{savestack,restorestack}
151 Register SP;
152
153 /// Entry block.
154 const MachineBasicBlock *Entry = nullptr;
155
156 using SetOfRegs = SmallSetVector<unsigned, 16>;
157
158 /// Registers that need to be saved for the current function.
159 mutable SetOfRegs CurrentCSRs;
160
161 /// Current MachineFunction.
162 MachineFunction *MachineFunc = nullptr;
163
164 /// Is `true` for the block numbers where we assume possible stack accesses
165 /// or computation of stack-relative addresses on any CFG path including the
166 /// block itself. Is `false` for basic blocks where we can guarantee the
167 /// opposite. False positives won't lead to incorrect analysis results,
168 /// therefore this approach is fair.
169 BitVector StackAddressUsedBlockInfo;
170
171 /// Check if \p MI uses or defines a callee-saved register or
172 /// a frame index. If this is the case, this means \p MI must happen
173 /// after Save and before Restore.
174 bool useOrDefCSROrFI(const MachineInstr &MI, RegScavenger *RS,
175 bool StackAddressUsed) const;
176
177 const SetOfRegs &getCurrentCSRs(RegScavenger *RS) const {
178 if (CurrentCSRs.empty()) {
179 BitVector SavedRegs;
180 const TargetFrameLowering *TFI =
181 MachineFunc->getSubtarget().getFrameLowering();
182
183 TFI->determineCalleeSaves(MF&: *MachineFunc, SavedRegs, RS);
184
185 for (int Reg = SavedRegs.find_first(); Reg != -1;
186 Reg = SavedRegs.find_next(Prev: Reg))
187 CurrentCSRs.insert(X: (unsigned)Reg);
188 }
189 return CurrentCSRs;
190 }
191
192 /// Update the Save and Restore points such that \p MBB is in
193 /// the region that is dominated by Save and post-dominated by Restore
194 /// and Save and Restore still match the safe point definition.
195 /// Such point may not exist and Save and/or Restore may be null after
196 /// this call.
197 void updateSaveRestorePoints(MachineBasicBlock &MBB, RegScavenger *RS);
198
199 // Try to find safe point based on dominance and block frequency without
200 // any change in IR.
201 bool performShrinkWrapping(
202 const ReversePostOrderTraversal<MachineBasicBlock *> &RPOT,
203 RegScavenger *RS);
204
205 /// This function tries to split the restore point if doing so can shrink the
206 /// save point further. \return True if restore point is split.
207 bool postShrinkWrapping(bool HasCandidate, MachineFunction &MF,
208 RegScavenger *RS);
209
210 /// This function analyzes if the restore point can split to create a new
211 /// restore point. This function collects
212 /// 1. Any preds of current restore that are reachable by callee save/FI
213 /// blocks
214 /// - indicated by DirtyPreds
215 /// 2. Any preds of current restore that are not DirtyPreds - indicated by
216 /// CleanPreds
217 /// Both sets should be non-empty for considering restore point split.
218 bool checkIfRestoreSplittable(
219 const MachineBasicBlock *CurRestore,
220 const DenseSet<const MachineBasicBlock *> &ReachableByDirty,
221 SmallVectorImpl<MachineBasicBlock *> &DirtyPreds,
222 SmallVectorImpl<MachineBasicBlock *> &CleanPreds,
223 const TargetInstrInfo *TII, RegScavenger *RS);
224
225 /// Initialize the pass for \p MF.
226 void init(MachineFunction &MF) {
227 Save = nullptr;
228 Restore = nullptr;
229 EntryFreq = MBFI->getEntryFreq();
230 const TargetSubtargetInfo &Subtarget = MF.getSubtarget();
231 const TargetInstrInfo &TII = *Subtarget.getInstrInfo();
232 FrameSetupOpcode = TII.getCallFrameSetupOpcode();
233 FrameDestroyOpcode = TII.getCallFrameDestroyOpcode();
234 SP = Subtarget.getTargetLowering()->getStackPointerRegisterToSaveRestore();
235 Entry = &MF.front();
236 CurrentCSRs.clear();
237 MachineFunc = &MF;
238
239 ++NumFunc;
240 }
241
242 /// Check whether or not Save and Restore points are still interesting for
243 /// shrink-wrapping.
244 bool ArePointsInteresting() const { return Save != Entry && Save && Restore; }
245
246public:
247 ShrinkWrapImpl(const RegisterClassInfo *RCI, MachineDominatorTree *MDT,
248 MachinePostDominatorTree *MPDT,
249 MachineBlockFrequencyInfo *MBFI, MachineLoopInfo *MLI,
250 MachineOptimizationRemarkEmitter *ORE)
251 : RCI(RCI), MDT(MDT), MPDT(MPDT), MBFI(MBFI), MLI(MLI), ORE(ORE) {}
252
253 /// Check if shrink wrapping is enabled for this target and function.
254 static bool isShrinkWrapEnabled(const MachineFunction &MF);
255
256 bool run(MachineFunction &MF);
257};
258
259class ShrinkWrapLegacy : public MachineFunctionPass {
260public:
261 static char ID;
262
263 ShrinkWrapLegacy() : MachineFunctionPass(ID) {}
264
265 void getAnalysisUsage(AnalysisUsage &AU) const override {
266 AU.setPreservesAll();
267 AU.addRequired<MachineBlockFrequencyInfoWrapperPass>();
268 AU.addRequired<MachineDominatorTreeWrapperPass>();
269 AU.addRequired<MachinePostDominatorTreeWrapperPass>();
270 AU.addRequired<MachineLoopInfoWrapperPass>();
271 AU.addRequired<MachineOptimizationRemarkEmitterPass>();
272 AU.addRequired<MachineRegisterClassInfoWrapperPass>();
273 MachineFunctionPass::getAnalysisUsage(AU);
274 }
275
276 MachineFunctionProperties getRequiredProperties() const override {
277 return MachineFunctionProperties().setNoVRegs();
278 }
279
280 StringRef getPassName() const override { return "Shrink Wrapping analysis"; }
281
282 /// Perform the shrink-wrapping analysis and update
283 /// the MachineFrameInfo attached to \p MF with the results.
284 bool runOnMachineFunction(MachineFunction &MF) override;
285};
286
287} // end anonymous namespace
288
289char ShrinkWrapLegacy::ID = 0;
290
291char &llvm::ShrinkWrapID = ShrinkWrapLegacy::ID;
292
293INITIALIZE_PASS_BEGIN(ShrinkWrapLegacy, DEBUG_TYPE, "Shrink Wrap Pass", false,
294 false)
295INITIALIZE_PASS_DEPENDENCY(MachineBlockFrequencyInfoWrapperPass)
296INITIALIZE_PASS_DEPENDENCY(MachineDominatorTreeWrapperPass)
297INITIALIZE_PASS_DEPENDENCY(MachinePostDominatorTreeWrapperPass)
298INITIALIZE_PASS_DEPENDENCY(MachineLoopInfoWrapperPass)
299INITIALIZE_PASS_DEPENDENCY(MachineOptimizationRemarkEmitterPass)
300INITIALIZE_PASS_DEPENDENCY(MachineRegisterClassInfoWrapperPass)
301INITIALIZE_PASS_END(ShrinkWrapLegacy, DEBUG_TYPE, "Shrink Wrap Pass", false,
302 false)
303
304bool ShrinkWrapImpl::useOrDefCSROrFI(const MachineInstr &MI, RegScavenger *RS,
305 bool StackAddressUsed) const {
306 /// Check if \p Op is known to access an address not on the function's stack .
307 /// At the moment, accesses where the underlying object is a global, function
308 /// argument, or jump table are considered non-stack accesses. Note that the
309 /// caller's stack may get accessed when passing an argument via the stack,
310 /// but not the stack of the current function.
311 ///
312 auto IsKnownNonStackPtr = [](MachineMemOperand *Op) {
313 if (Op->getValue()) {
314 const Value *UO = getUnderlyingObject(V: Op->getValue());
315 if (!UO)
316 return false;
317 if (auto *Arg = dyn_cast<Argument>(Val: UO))
318 return !Arg->hasPassPointeeByValueCopyAttr();
319 return isa<GlobalValue>(Val: UO);
320 }
321 if (const PseudoSourceValue *PSV = Op->getPseudoValue())
322 return PSV->isJumpTable() || PSV->isConstantPool();
323 return false;
324 };
325 // Load/store operations may access the stack indirectly when we previously
326 // computed an address to a stack location.
327 if (StackAddressUsed && MI.mayLoadOrStore() &&
328 (MI.isCall() || MI.hasUnmodeledSideEffects() || MI.memoperands_empty() ||
329 !all_of(Range: MI.memoperands(), P: IsKnownNonStackPtr)))
330 return true;
331
332 if (MI.getOpcode() == FrameSetupOpcode ||
333 MI.getOpcode() == FrameDestroyOpcode) {
334 LLVM_DEBUG(dbgs() << "Frame instruction: " << MI << '\n');
335 return true;
336 }
337 const MachineFunction *MF = MI.getParent()->getParent();
338 const TargetRegisterInfo *TRI = MF->getSubtarget().getRegisterInfo();
339 for (const MachineOperand &MO : MI.operands()) {
340 bool UseOrDefCSR = false;
341 if (MO.isReg()) {
342 // Ignore instructions like DBG_VALUE which don't read/def the register.
343 if (!MO.isDef() && !MO.readsReg())
344 continue;
345 Register PhysReg = MO.getReg();
346 if (!PhysReg)
347 continue;
348 assert(PhysReg.isPhysical() && "Unallocated register?!");
349 // The stack pointer is not normally described as a callee-saved register
350 // in calling convention definitions, so we need to watch for it
351 // separately. An SP mentioned by a call instruction, we can ignore,
352 // though, as it's harmless and we do not want to effectively disable tail
353 // calls by forcing the restore point to post-dominate them.
354 // PPC's LR is also not normally described as a callee-saved register in
355 // calling convention definitions, so we need to watch for it, too. An LR
356 // mentioned implicitly by a return (or "branch to link register")
357 // instruction we can ignore, otherwise we may pessimize shrinkwrapping.
358 // PPC's Frame pointer (FP) is also not described as a callee-saved
359 // register. Until the FP is assigned a Physical Register PPC's FP needs
360 // to be checked separately.
361 UseOrDefCSR = (!MI.isCall() && PhysReg == SP) ||
362 RCI->getLastCalleeSavedAlias(PhysReg) ||
363 (!MI.isReturn() &&
364 TRI->isNonallocatableRegisterCalleeSave(Reg: PhysReg)) ||
365 TRI->isVirtualFrameRegister(Reg: PhysReg);
366 } else if (MO.isRegMask()) {
367 // Check if this regmask clobbers any of the CSRs.
368 for (unsigned Reg : getCurrentCSRs(RS)) {
369 if (MO.clobbersPhysReg(PhysReg: Reg)) {
370 UseOrDefCSR = true;
371 break;
372 }
373 }
374 }
375 // Skip FrameIndex operands in DBG_VALUE instructions.
376 if (UseOrDefCSR || (MO.isFI() && !MI.isDebugValue())) {
377 LLVM_DEBUG(dbgs() << "Use or define CSR(" << UseOrDefCSR << ") or FI("
378 << MO.isFI() << "): " << MI << '\n');
379 return true;
380 }
381 }
382 return false;
383}
384
385/// Helper function to find the immediate (post) dominator.
386template <typename ListOfBBs, typename DominanceAnalysis>
387static MachineBasicBlock *FindIDom(MachineBasicBlock &Block, ListOfBBs BBs,
388 DominanceAnalysis &Dom, bool Strict = true) {
389 if (BBs.begin() == BBs.end())
390 return Strict ? nullptr : &Block;
391 MachineBasicBlock *IDom = Dom.findNearestCommonDominator(iterator_range(BBs));
392 if (Strict && IDom == &Block)
393 return nullptr;
394 return IDom;
395}
396
397static bool isAnalyzableBB(const TargetInstrInfo &TII,
398 MachineBasicBlock &Entry) {
399 // Check if the block is analyzable.
400 MachineBasicBlock *TBB = nullptr, *FBB = nullptr;
401 SmallVector<MachineOperand, 4> Cond;
402 return !TII.analyzeBranch(MBB&: Entry, TBB, FBB, Cond);
403}
404
405/// Determines if any predecessor of MBB is on the path from block that has use
406/// or def of CSRs/FI to MBB.
407/// ReachableByDirty: All blocks reachable from block that has use or def of
408/// CSR/FI.
409static bool
410hasDirtyPred(const DenseSet<const MachineBasicBlock *> &ReachableByDirty,
411 const MachineBasicBlock &MBB) {
412 for (const MachineBasicBlock *PredBB : MBB.predecessors())
413 if (ReachableByDirty.count(V: PredBB))
414 return true;
415 return false;
416}
417
418/// Derives the list of all the basic blocks reachable from MBB.
419static void markAllReachable(DenseSet<const MachineBasicBlock *> &Visited,
420 const MachineBasicBlock &MBB) {
421 SmallVector<MachineBasicBlock *, 4> Worklist(MBB.successors());
422 Visited.insert(V: &MBB);
423 while (!Worklist.empty()) {
424 MachineBasicBlock *SuccMBB = Worklist.pop_back_val();
425 if (!Visited.insert(V: SuccMBB).second)
426 continue;
427 Worklist.append(in_start: SuccMBB->succ_begin(), in_end: SuccMBB->succ_end());
428 }
429}
430
431/// Collect blocks reachable by use or def of CSRs/FI.
432static void collectBlocksReachableByDirty(
433 const DenseSet<const MachineBasicBlock *> &DirtyBBs,
434 DenseSet<const MachineBasicBlock *> &ReachableByDirty) {
435 for (const MachineBasicBlock *MBB : DirtyBBs) {
436 if (ReachableByDirty.count(V: MBB))
437 continue;
438 // Mark all offsprings as reachable.
439 markAllReachable(Visited&: ReachableByDirty, MBB: *MBB);
440 }
441}
442
443/// \return true if there is a clean path from SavePoint to the original
444/// Restore.
445static bool
446isSaveReachableThroughClean(const MachineBasicBlock *SavePoint,
447 ArrayRef<MachineBasicBlock *> CleanPreds) {
448 DenseSet<const MachineBasicBlock *> Visited;
449 SmallVector<MachineBasicBlock *, 4> Worklist(CleanPreds);
450 while (!Worklist.empty()) {
451 MachineBasicBlock *CleanBB = Worklist.pop_back_val();
452 if (CleanBB == SavePoint)
453 return true;
454 if (!Visited.insert(V: CleanBB).second || !CleanBB->pred_size())
455 continue;
456 Worklist.append(in_start: CleanBB->pred_begin(), in_end: CleanBB->pred_end());
457 }
458 return false;
459}
460
461/// This function updates the branches post restore point split.
462///
463/// Restore point has been split.
464/// Old restore point: MBB
465/// New restore point: NMBB
466/// Any basic block(say BBToUpdate) which had a fallthrough to MBB
467/// previously should
468/// 1. Fallthrough to NMBB iff NMBB is inserted immediately above MBB in the
469/// block layout OR
470/// 2. Branch unconditionally to NMBB iff NMBB is inserted at any other place.
471static void updateTerminator(MachineBasicBlock *BBToUpdate,
472 MachineBasicBlock *NMBB,
473 const TargetInstrInfo *TII) {
474 DebugLoc DL = BBToUpdate->findBranchDebugLoc();
475 // if NMBB isn't the new layout successor for BBToUpdate, insert unconditional
476 // branch to it
477 if (!BBToUpdate->isLayoutSuccessor(MBB: NMBB))
478 TII->insertUnconditionalBranch(MBB&: *BBToUpdate, DestBB: NMBB, DL);
479}
480
481/// This function splits the restore point and returns new restore point/BB.
482///
483/// DirtyPreds: Predessors of \p MBB that are ReachableByDirty
484///
485/// Decision has been made to split the restore point.
486/// old restore point: \p MBB
487/// new restore point: \p NMBB
488/// This function makes the necessary block layout changes so that
489/// 1. \p NMBB points to \p MBB unconditionally
490/// 2. All dirtyPreds that previously pointed to \p MBB point to \p NMBB
491static MachineBasicBlock *
492tryToSplitRestore(MachineBasicBlock *MBB,
493 ArrayRef<MachineBasicBlock *> DirtyPreds,
494 const TargetInstrInfo *TII) {
495 MachineFunction *MF = MBB->getParent();
496
497 // get the list of DirtyPreds who have a fallthrough to MBB
498 // before the block layout change. This is just to ensure that if the NMBB is
499 // inserted after MBB, then we create unconditional branch from
500 // DirtyPred/CleanPred to NMBB
501 SmallPtrSet<MachineBasicBlock *, 8> MBBFallthrough;
502 for (MachineBasicBlock *BB : DirtyPreds)
503 if (BB->getFallThrough(JumpToFallThrough: false) == MBB)
504 MBBFallthrough.insert(Ptr: BB);
505
506 MachineBasicBlock *NMBB = MF->CreateMachineBasicBlock();
507 // Insert this block at the end of the function. Inserting in between may
508 // interfere with control flow optimizer decisions.
509 MF->insert(MBBI: MF->end(), MBB: NMBB);
510
511 for (const MachineBasicBlock::RegisterMaskPair &LI : MBB->liveins())
512 NMBB->addLiveIn(PhysReg: LI.PhysReg);
513
514 TII->insertUnconditionalBranch(MBB&: *NMBB, DestBB: MBB, DL: DebugLoc());
515
516 // After splitting, all predecessors of the restore point should be dirty
517 // blocks.
518 for (MachineBasicBlock *SuccBB : DirtyPreds)
519 SuccBB->ReplaceUsesOfBlockWith(Old: MBB, New: NMBB);
520
521 NMBB->addSuccessor(Succ: MBB);
522
523 for (MachineBasicBlock *BBToUpdate : MBBFallthrough)
524 updateTerminator(BBToUpdate, NMBB, TII);
525
526 return NMBB;
527}
528
529/// This function undoes the restore point split done earlier.
530///
531/// DirtyPreds: All predecessors of \p NMBB that are ReachableByDirty.
532///
533/// Restore point was split and the change needs to be unrolled. Make necessary
534/// changes to reset restore point from \p NMBB to \p MBB.
535static void rollbackRestoreSplit(MachineFunction &MF, MachineBasicBlock *NMBB,
536 MachineBasicBlock *MBB,
537 ArrayRef<MachineBasicBlock *> DirtyPreds,
538 const TargetInstrInfo *TII) {
539 // For a BB, if NMBB is fallthrough in the current layout, then in the new
540 // layout a. BB should fallthrough to MBB OR b. BB should undconditionally
541 // branch to MBB
542 SmallPtrSet<MachineBasicBlock *, 8> NMBBFallthrough;
543 for (MachineBasicBlock *BB : DirtyPreds)
544 if (BB->getFallThrough(JumpToFallThrough: false) == NMBB)
545 NMBBFallthrough.insert(Ptr: BB);
546
547 NMBB->removeSuccessor(Succ: MBB);
548 for (MachineBasicBlock *SuccBB : DirtyPreds)
549 SuccBB->ReplaceUsesOfBlockWith(Old: NMBB, New: MBB);
550
551 NMBB->erase(I: NMBB->begin(), E: NMBB->end());
552 NMBB->eraseFromParent();
553
554 for (MachineBasicBlock *BBToUpdate : NMBBFallthrough)
555 updateTerminator(BBToUpdate, NMBB: MBB, TII);
556}
557
558// A block is deemed fit for restore point split iff there exist
559// 1. DirtyPreds - preds of CurRestore reachable from use or def of CSR/FI
560// 2. CleanPreds - preds of CurRestore that arent DirtyPreds
561bool ShrinkWrapImpl::checkIfRestoreSplittable(
562 const MachineBasicBlock *CurRestore,
563 const DenseSet<const MachineBasicBlock *> &ReachableByDirty,
564 SmallVectorImpl<MachineBasicBlock *> &DirtyPreds,
565 SmallVectorImpl<MachineBasicBlock *> &CleanPreds,
566 const TargetInstrInfo *TII, RegScavenger *RS) {
567 for (const MachineInstr &MI : *CurRestore)
568 if (useOrDefCSROrFI(MI, RS, /*StackAddressUsed=*/true))
569 return false;
570
571 for (MachineBasicBlock *PredBB : CurRestore->predecessors()) {
572 if (!isAnalyzableBB(TII: *TII, Entry&: *PredBB))
573 return false;
574
575 if (ReachableByDirty.count(V: PredBB))
576 DirtyPreds.push_back(Elt: PredBB);
577 else
578 CleanPreds.push_back(Elt: PredBB);
579 }
580
581 return !(CleanPreds.empty() || DirtyPreds.empty());
582}
583
584bool ShrinkWrapImpl::postShrinkWrapping(bool HasCandidate, MachineFunction &MF,
585 RegScavenger *RS) {
586 if (!EnablePostShrinkWrapOpt)
587 return false;
588
589 MachineBasicBlock *InitSave = nullptr;
590 MachineBasicBlock *InitRestore = nullptr;
591
592 if (HasCandidate) {
593 InitSave = Save;
594 InitRestore = Restore;
595 } else {
596 InitRestore = nullptr;
597 InitSave = &MF.front();
598 for (MachineBasicBlock &MBB : MF) {
599 if (MBB.isEHFuncletEntry())
600 return false;
601 if (MBB.isReturnBlock()) {
602 // Do not support multiple restore points.
603 if (InitRestore)
604 return false;
605 InitRestore = &MBB;
606 }
607 }
608 }
609
610 if (!InitSave || !InitRestore || InitRestore == InitSave ||
611 !MDT->dominates(A: InitSave, B: InitRestore) ||
612 !MPDT->dominates(A: InitRestore, B: InitSave))
613 return false;
614
615 // Bail out of the optimization if any of the basic block is target of
616 // INLINEASM_BR instruction
617 for (MachineBasicBlock &MBB : MF)
618 if (MBB.isInlineAsmBrIndirectTarget())
619 return false;
620
621 DenseSet<const MachineBasicBlock *> DirtyBBs;
622 for (MachineBasicBlock &MBB : MF) {
623 if (!MDT->isReachableFromEntry(A: &MBB))
624 continue;
625 if (MBB.isEHPad()) {
626 DirtyBBs.insert(V: &MBB);
627 continue;
628 }
629 for (const MachineInstr &MI : MBB)
630 if (useOrDefCSROrFI(MI, RS, /*StackAddressUsed=*/true)) {
631 DirtyBBs.insert(V: &MBB);
632 break;
633 }
634 }
635
636 // Find blocks reachable from the use or def of CSRs/FI.
637 DenseSet<const MachineBasicBlock *> ReachableByDirty;
638 collectBlocksReachableByDirty(DirtyBBs, ReachableByDirty);
639
640 const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
641 SmallVector<MachineBasicBlock *, 2> DirtyPreds;
642 SmallVector<MachineBasicBlock *, 2> CleanPreds;
643 if (!checkIfRestoreSplittable(CurRestore: InitRestore, ReachableByDirty, DirtyPreds,
644 CleanPreds, TII, RS))
645 return false;
646
647 // Trying to reach out to the new save point which dominates all dirty blocks.
648 MachineBasicBlock *NewSave =
649 FindIDom<>(Block&: **DirtyPreds.begin(), BBs: DirtyPreds, Dom&: *MDT, Strict: false);
650
651 while (NewSave && (hasDirtyPred(ReachableByDirty, MBB: *NewSave) ||
652 EntryFreq < MBFI->getBlockFreq(MBB: NewSave) ||
653 /*Entry freq has been observed more than a loop block in
654 some cases*/
655 MLI->getLoopFor(BB: NewSave))) {
656 SmallVector<MachineBasicBlock*> ReachablePreds;
657 for (auto BB: NewSave->predecessors())
658 if (MDT->isReachableFromEntry(A: BB))
659 ReachablePreds.push_back(Elt: BB);
660 if (ReachablePreds.empty())
661 break;
662
663 NewSave = FindIDom<>(Block&: **ReachablePreds.begin(), BBs: ReachablePreds, Dom&: *MDT,
664 Strict: false);
665 }
666
667 const TargetFrameLowering *TFI = MF.getSubtarget().getFrameLowering();
668 if (!NewSave || NewSave == InitSave ||
669 isSaveReachableThroughClean(SavePoint: NewSave, CleanPreds) ||
670 !TFI->canUseAsPrologue(MBB: *NewSave))
671 return false;
672
673 // Now we know that splitting a restore point can isolate the restore point
674 // from clean blocks and doing so can shrink the save point.
675 MachineBasicBlock *NewRestore =
676 tryToSplitRestore(MBB: InitRestore, DirtyPreds, TII);
677
678 // Make sure if the new restore point is valid as an epilogue, depending on
679 // targets.
680 if (!TFI->canUseAsEpilogue(MBB: *NewRestore)) {
681 rollbackRestoreSplit(MF, NMBB: NewRestore, MBB: InitRestore, DirtyPreds, TII);
682 return false;
683 }
684
685 Save = NewSave;
686 Restore = NewRestore;
687
688 MDT->recalculate(Func&: MF);
689 MPDT->recalculate(Func&: MF);
690
691 assert((MDT->dominates(Save, Restore) && MPDT->dominates(Restore, Save)) &&
692 "Incorrect save or restore point due to dominance relations");
693 assert((!MLI->getLoopFor(Save) && !MLI->getLoopFor(Restore)) &&
694 "Unexpected save or restore point in a loop");
695 assert((EntryFreq >= MBFI->getBlockFreq(Save) &&
696 EntryFreq >= MBFI->getBlockFreq(Restore)) &&
697 "Incorrect save or restore point based on block frequency");
698 return true;
699}
700
701void ShrinkWrapImpl::updateSaveRestorePoints(MachineBasicBlock &MBB,
702 RegScavenger *RS) {
703 // Get rid of the easy cases first.
704 if (!Save)
705 Save = &MBB;
706 else
707 Save = MDT->findNearestCommonDominator(A: Save, B: &MBB);
708 assert(Save);
709
710 if (!Restore)
711 Restore = &MBB;
712 else if (MPDT->getNode(BB: &MBB)) // If the block is not in the post dom tree, it
713 // means the block never returns. If that's the
714 // case, we don't want to call
715 // `findNearestCommonDominator`, which will
716 // return `Restore`.
717 Restore = MPDT->findNearestCommonDominator(A: Restore, B: &MBB);
718 else
719 Restore = nullptr; // Abort, we can't find a restore point in this case.
720
721 // Make sure we would be able to insert the restore code before the
722 // terminator.
723 if (Restore == &MBB) {
724 for (const MachineInstr &Terminator : MBB.terminators()) {
725 if (!useOrDefCSROrFI(MI: Terminator, RS, /*StackAddressUsed=*/true))
726 continue;
727 // One of the terminator needs to happen before the restore point.
728 if (MBB.succ_empty()) {
729 Restore = nullptr; // Abort, we can't find a restore point in this case.
730 break;
731 }
732 // Look for a restore point that post-dominates all the successors.
733 // The immediate post-dominator is what we are looking for.
734 Restore = FindIDom<>(Block&: *Restore, BBs: Restore->successors(), Dom&: *MPDT);
735 break;
736 }
737 }
738
739 if (!Restore) {
740 LLVM_DEBUG(
741 dbgs() << "Restore point needs to be spanned on several blocks\n");
742 return;
743 }
744
745 // Make sure Save and Restore are suitable for shrink-wrapping:
746 // 1. all path from Save needs to lead to Restore before exiting.
747 // 2. all path to Restore needs to go through Save from Entry.
748 // We achieve that by making sure that:
749 // A. Save dominates Restore.
750 // B. Restore post-dominates Save.
751 // C. Save and Restore are in the same loop.
752 bool SaveDominatesRestore = false;
753 bool RestorePostDominatesSave = false;
754 while (Restore &&
755 (!(SaveDominatesRestore = MDT->dominates(A: Save, B: Restore)) ||
756 !(RestorePostDominatesSave = MPDT->dominates(A: Restore, B: Save)) ||
757 // Post-dominance is not enough in loops to ensure that all uses/defs
758 // are after the prologue and before the epilogue at runtime.
759 // E.g.,
760 // while(1) {
761 // Save
762 // Restore
763 // if (...)
764 // break;
765 // use/def CSRs
766 // }
767 // All the uses/defs of CSRs are dominated by Save and post-dominated
768 // by Restore. However, the CSRs uses are still reachable after
769 // Restore and before Save are executed.
770 //
771 // For now, just push the restore/save points outside of loops.
772 // FIXME: Refine the criteria to still find interesting cases
773 // for loops.
774 MLI->getLoopFor(BB: Save) || MLI->getLoopFor(BB: Restore))) {
775 // Fix (A).
776 if (!SaveDominatesRestore) {
777 Save = MDT->findNearestCommonDominator(A: Save, B: Restore);
778 continue;
779 }
780 // Fix (B).
781 if (!RestorePostDominatesSave)
782 Restore = MPDT->findNearestCommonDominator(A: Restore, B: Save);
783
784 // Fix (C).
785 if (Restore && (MLI->getLoopFor(BB: Save) || MLI->getLoopFor(BB: Restore))) {
786 if (MLI->getLoopDepth(BB: Save) > MLI->getLoopDepth(BB: Restore)) {
787 // Push Save outside of this loop if immediate dominator is different
788 // from save block. If immediate dominator is not different, bail out.
789 SmallVector<MachineBasicBlock *> Preds;
790 for (auto *PBB : Save->predecessors())
791 if (MDT->isReachableFromEntry(A: PBB))
792 Preds.push_back(Elt: PBB);
793 Save = FindIDom<>(Block&: *Save, BBs: Preds, Dom&: *MDT);
794 if (!Save)
795 break;
796 } else {
797 // If the loop does not exit, there is no point in looking
798 // for a post-dominator outside the loop.
799 SmallVector<MachineBasicBlock*, 4> ExitBlocks;
800 MLI->getLoopFor(BB: Restore)->getExitingBlocks(ExitingBlocks&: ExitBlocks);
801 // Push Restore outside of this loop.
802 // Look for the immediate post-dominator of the loop exits.
803 MachineBasicBlock *IPdom = Restore;
804 for (MachineBasicBlock *LoopExitBB: ExitBlocks) {
805 IPdom = FindIDom<>(Block&: *IPdom, BBs: LoopExitBB->successors(), Dom&: *MPDT);
806 if (!IPdom)
807 break;
808 }
809 // If the immediate post-dominator is not in a less nested loop,
810 // then we are stuck in a program with an infinite loop.
811 // In that case, we will not find a safe point, hence, bail out.
812 if (IPdom && MLI->getLoopDepth(BB: IPdom) < MLI->getLoopDepth(BB: Restore))
813 Restore = IPdom;
814 else {
815 Restore = nullptr;
816 break;
817 }
818 }
819 }
820 }
821}
822
823static bool giveUpWithRemarks(MachineOptimizationRemarkEmitter *ORE,
824 StringRef RemarkName, StringRef RemarkMessage,
825 const DiagnosticLocation &Loc,
826 const MachineBasicBlock *MBB) {
827 ORE->emit(RemarkBuilder: [&]() {
828 return MachineOptimizationRemarkMissed(DEBUG_TYPE, RemarkName, Loc, MBB)
829 << RemarkMessage;
830 });
831
832 LLVM_DEBUG(dbgs() << RemarkMessage << '\n');
833 return false;
834}
835
836bool ShrinkWrapImpl::performShrinkWrapping(
837 const ReversePostOrderTraversal<MachineBasicBlock *> &RPOT,
838 RegScavenger *RS) {
839 for (MachineBasicBlock *MBB : RPOT) {
840 LLVM_DEBUG(dbgs() << "Look into: " << printMBBReference(*MBB) << '\n');
841
842 if (MBB->isEHFuncletEntry())
843 return giveUpWithRemarks(ORE, RemarkName: "UnsupportedEHFunclets",
844 RemarkMessage: "EH Funclets are not supported yet.",
845 Loc: MBB->front().getDebugLoc(), MBB);
846
847 if (MBB->isEHPad() || MBB->isInlineAsmBrIndirectTarget()) {
848 // Push the prologue and epilogue outside of the region that may throw (or
849 // jump out via inlineasm_br), by making sure that all the landing pads
850 // are at least at the boundary of the save and restore points. The
851 // problem is that a basic block can jump out from the middle in these
852 // cases, which we do not handle.
853 updateSaveRestorePoints(MBB&: *MBB, RS);
854 if (!ArePointsInteresting()) {
855 LLVM_DEBUG(dbgs() << "EHPad/inlineasm_br prevents shrink-wrapping\n");
856 return false;
857 }
858 continue;
859 }
860
861 bool StackAddressUsed = false;
862 // Check if we found any stack accesses in the predecessors. We are not
863 // doing a full dataflow analysis here to keep things simple but just
864 // rely on a reverse portorder traversal (RPOT) to guarantee predecessors
865 // are already processed except for loops (and accept the conservative
866 // result for loops).
867 for (const MachineBasicBlock *Pred : MBB->predecessors()) {
868 if (StackAddressUsedBlockInfo.test(Idx: Pred->getNumber())) {
869 StackAddressUsed = true;
870 break;
871 }
872 }
873
874 for (const MachineInstr &MI : *MBB) {
875 if (useOrDefCSROrFI(MI, RS, StackAddressUsed)) {
876 // Save (resp. restore) point must dominate (resp. post dominate)
877 // MI. Look for the proper basic block for those.
878 updateSaveRestorePoints(MBB&: *MBB, RS);
879 // If we are at a point where we cannot improve the placement of
880 // save/restore instructions, just give up.
881 if (!ArePointsInteresting()) {
882 LLVM_DEBUG(dbgs() << "No Shrink wrap candidate found\n");
883 return false;
884 }
885 // No need to look for other instructions, this basic block
886 // will already be part of the handled region.
887 StackAddressUsed = true;
888 break;
889 }
890 }
891 StackAddressUsedBlockInfo[MBB->getNumber()] = StackAddressUsed;
892 }
893 if (!ArePointsInteresting()) {
894 // If the points are not interesting at this point, then they must be null
895 // because it means we did not encounter any frame/CSR related code.
896 // Otherwise, we would have returned from the previous loop.
897 assert(!Save && !Restore && "We miss a shrink-wrap opportunity?!");
898 LLVM_DEBUG(dbgs() << "Nothing to shrink-wrap\n");
899 return false;
900 }
901
902 LLVM_DEBUG(dbgs() << "\n ** Results **\nFrequency of the Entry: "
903 << EntryFreq.getFrequency() << '\n');
904
905 const TargetFrameLowering *TFI =
906 MachineFunc->getSubtarget().getFrameLowering();
907 do {
908 LLVM_DEBUG(dbgs() << "Shrink wrap candidates (#, Name, Freq):\nSave: "
909 << printMBBReference(*Save) << ' '
910 << printBlockFreq(*MBFI, *Save)
911 << "\nRestore: " << printMBBReference(*Restore) << ' '
912 << printBlockFreq(*MBFI, *Restore) << '\n');
913
914 bool IsSaveCheap, TargetCanUseSaveAsPrologue = false;
915 if (((IsSaveCheap = EntryFreq >= MBFI->getBlockFreq(MBB: Save)) &&
916 EntryFreq >= MBFI->getBlockFreq(MBB: Restore)) &&
917 ((TargetCanUseSaveAsPrologue = TFI->canUseAsPrologue(MBB: *Save)) &&
918 TFI->canUseAsEpilogue(MBB: *Restore)))
919 break;
920 LLVM_DEBUG(
921 dbgs() << "New points are too expensive or invalid for the target\n");
922 MachineBasicBlock *NewBB;
923 if (!IsSaveCheap || !TargetCanUseSaveAsPrologue) {
924 Save = FindIDom<>(Block&: *Save, BBs: Save->predecessors(), Dom&: *MDT);
925 if (!Save)
926 break;
927 NewBB = Save;
928 } else {
929 // Restore is expensive.
930 Restore = FindIDom<>(Block&: *Restore, BBs: Restore->successors(), Dom&: *MPDT);
931 if (!Restore)
932 break;
933 NewBB = Restore;
934 }
935 updateSaveRestorePoints(MBB&: *NewBB, RS);
936 } while (Save && Restore);
937
938 if (!ArePointsInteresting()) {
939 ++NumCandidatesDropped;
940 return false;
941 }
942 return true;
943}
944
945bool ShrinkWrapImpl::run(MachineFunction &MF) {
946 LLVM_DEBUG(dbgs() << "**** Analysing " << MF.getName() << '\n');
947
948 init(MF);
949
950 ReversePostOrderTraversal<MachineBasicBlock *> RPOT(&*MF.begin());
951 if (containsIrreducibleCFG<MachineBasicBlock *>(RPOTraversal&: RPOT, LI: *MLI)) {
952 // If MF is irreducible, a block may be in a loop without
953 // MachineLoopInfo reporting it. I.e., we may use the
954 // post-dominance property in loops, which lead to incorrect
955 // results. Moreover, we may miss that the prologue and
956 // epilogue are not in the same loop, leading to unbalanced
957 // construction/deconstruction of the stack frame.
958 return giveUpWithRemarks(ORE, RemarkName: "UnsupportedIrreducibleCFG",
959 RemarkMessage: "Irreducible CFGs are not supported yet.",
960 Loc: MF.getFunction().getSubprogram(), MBB: &MF.front());
961 }
962
963 const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
964 std::unique_ptr<RegScavenger> RS(
965 TRI->requiresRegisterScavenging(MF) ? new RegScavenger() : nullptr);
966
967 bool Changed = false;
968
969 // Initially, conservatively assume that stack addresses can be used in each
970 // basic block and change the state only for those basic blocks for which we
971 // were able to prove the opposite.
972 StackAddressUsedBlockInfo.resize(N: MF.getNumBlockIDs(), t: true);
973 bool HasCandidate = performShrinkWrapping(RPOT, RS: RS.get());
974 StackAddressUsedBlockInfo.clear();
975 Changed = postShrinkWrapping(HasCandidate, MF, RS: RS.get());
976 if (!HasCandidate && !Changed)
977 return false;
978 if (!ArePointsInteresting())
979 return Changed;
980
981 LLVM_DEBUG(dbgs() << "Final shrink wrap candidates:\nSave: "
982 << printMBBReference(*Save) << ' '
983 << "\nRestore: " << printMBBReference(*Restore) << '\n');
984
985 MachineFrameInfo &MFI = MF.getFrameInfo();
986
987 // List of CalleeSavedInfo for registers will be added during prologepilog
988 // pass
989 SaveRestorePoints SavePoints({{Save, {}}});
990 SaveRestorePoints RestorePoints({{Restore, {}}});
991
992 MFI.setSavePoints(SavePoints);
993 MFI.setRestorePoints(RestorePoints);
994 ++NumCandidates;
995 return Changed;
996}
997
998bool ShrinkWrapLegacy::runOnMachineFunction(MachineFunction &MF) {
999 if (skipFunction(F: MF.getFunction()) || MF.empty() ||
1000 !ShrinkWrapImpl::isShrinkWrapEnabled(MF))
1001 return false;
1002
1003 const RegisterClassInfo *RCI =
1004 &getAnalysis<MachineRegisterClassInfoWrapperPass>().getRCI();
1005 MachineDominatorTree *MDT =
1006 &getAnalysis<MachineDominatorTreeWrapperPass>().getDomTree();
1007 MachinePostDominatorTree *MPDT =
1008 &getAnalysis<MachinePostDominatorTreeWrapperPass>().getPostDomTree();
1009 MachineBlockFrequencyInfo *MBFI =
1010 &getAnalysis<MachineBlockFrequencyInfoWrapperPass>().getMBFI();
1011 MachineLoopInfo *MLI = &getAnalysis<MachineLoopInfoWrapperPass>().getLI();
1012 MachineOptimizationRemarkEmitter *ORE =
1013 &getAnalysis<MachineOptimizationRemarkEmitterPass>().getORE();
1014
1015 return ShrinkWrapImpl(RCI, MDT, MPDT, MBFI, MLI, ORE).run(MF);
1016}
1017
1018PreservedAnalyses ShrinkWrapPass::run(MachineFunction &MF,
1019 MachineFunctionAnalysisManager &MFAM) {
1020 MFPropsModifier _(*this, MF);
1021 if (MF.empty() || !ShrinkWrapImpl::isShrinkWrapEnabled(MF))
1022 return PreservedAnalyses::all();
1023
1024 const RegisterClassInfo &RCI =
1025 MFAM.getResult<MachineRegisterClassAnalysis>(IR&: MF);
1026 MachineDominatorTree &MDT = MFAM.getResult<MachineDominatorTreeAnalysis>(IR&: MF);
1027 MachinePostDominatorTree &MPDT =
1028 MFAM.getResult<MachinePostDominatorTreeAnalysis>(IR&: MF);
1029 MachineBlockFrequencyInfo &MBFI =
1030 MFAM.getResult<MachineBlockFrequencyAnalysis>(IR&: MF);
1031 MachineLoopInfo &MLI = MFAM.getResult<MachineLoopAnalysis>(IR&: MF);
1032 MachineOptimizationRemarkEmitter &ORE =
1033 MFAM.getResult<MachineOptimizationRemarkEmitterAnalysis>(IR&: MF);
1034
1035 ShrinkWrapImpl(&RCI, &MDT, &MPDT, &MBFI, &MLI, &ORE).run(MF);
1036 return PreservedAnalyses::all();
1037}
1038
1039bool ShrinkWrapImpl::isShrinkWrapEnabled(const MachineFunction &MF) {
1040 const TargetFrameLowering *TFI = MF.getSubtarget().getFrameLowering();
1041
1042 switch (EnableShrinkWrapOpt) {
1043 case cl::boolOrDefault::BOU_UNSET:
1044 return TFI->enableShrinkWrapping(MF) &&
1045 // Windows with CFI has some limitations that make it impossible
1046 // to use shrink-wrapping.
1047 !MF.getTarget().getMCAsmInfo().usesWindowsCFI() &&
1048 // Sanitizers look at the value of the stack at the location
1049 // of the crash. Since a crash can happen anywhere, the
1050 // frame must be lowered before anything else happen for the
1051 // sanitizers to be able to get a correct stack frame.
1052 !(MF.getFunction().hasFnAttribute(Kind: Attribute::SanitizeAddress) ||
1053 MF.getFunction().hasFnAttribute(Kind: Attribute::SanitizeThread) ||
1054 MF.getFunction().hasFnAttribute(Kind: Attribute::SanitizeMemory) ||
1055 MF.getFunction().hasFnAttribute(Kind: Attribute::SanitizeType) ||
1056 MF.getFunction().hasFnAttribute(Kind: Attribute::SanitizeHWAddress));
1057 // If EnableShrinkWrap is set, it takes precedence on whatever the
1058 // target sets. The rational is that we assume we want to test
1059 // something related to shrink-wrapping.
1060 case cl::boolOrDefault::BOU_TRUE:
1061 return true;
1062 case cl::boolOrDefault::BOU_FALSE:
1063 return false;
1064 }
1065 llvm_unreachable("Invalid shrink-wrapping state");
1066}
1067