1//===- SCCPSolver.cpp - SCCP Utility --------------------------- *- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// \file
10// This file implements the Sparse Conditional Constant Propagation (SCCP)
11// utility.
12//
13//===----------------------------------------------------------------------===//
14
15#include "llvm/Transforms/Utils/SCCPSolver.h"
16#include "llvm/ADT/SetVector.h"
17#include "llvm/Analysis/ConstantFolding.h"
18#include "llvm/Analysis/InstructionSimplify.h"
19#include "llvm/Analysis/Loads.h"
20#include "llvm/Analysis/ValueLattice.h"
21#include "llvm/Analysis/ValueLatticeUtils.h"
22#include "llvm/Analysis/ValueTracking.h"
23#include "llvm/IR/ConstantRange.h"
24#include "llvm/IR/DerivedTypes.h"
25#include "llvm/IR/IRBuilder.h"
26#include "llvm/IR/InstVisitor.h"
27#include "llvm/IR/Instructions.h"
28#include "llvm/IR/NoFolder.h"
29#include "llvm/IR/PatternMatch.h"
30#include "llvm/Support/Casting.h"
31#include "llvm/Support/Debug.h"
32#include "llvm/Support/ErrorHandling.h"
33#include "llvm/Support/raw_ostream.h"
34#include "llvm/Transforms/Utils/Local.h"
35#include <cassert>
36#include <utility>
37#include <vector>
38
39using namespace llvm;
40using namespace PatternMatch;
41
42#define DEBUG_TYPE "sccp"
43
44// The maximum number of range extensions allowed for operations requiring
45// widening.
46static const unsigned MaxNumRangeExtensions = 10;
47
48/// Returns MergeOptions with MaxWidenSteps set to MaxNumRangeExtensions.
49static ValueLatticeElement::MergeOptions getMaxWidenStepsOpts() {
50 return ValueLatticeElement::MergeOptions().setMaxWidenSteps(
51 MaxNumRangeExtensions);
52}
53
54namespace llvm {
55
56bool SCCPSolver::isConstant(const ValueLatticeElement &LV) {
57 return LV.isConstant() ||
58 (LV.isConstantRange() && LV.getConstantRange().isSingleElement());
59}
60
61bool SCCPSolver::isReplaceableConstant(const ValueLatticeElement &LV) {
62 return isConstant(LV) && !LV.mayHaveDifferentProvenance();
63}
64
65bool SCCPSolver::isOverdefined(const ValueLatticeElement &LV) {
66 return !LV.isUnknownOrUndef() && !SCCPSolver::isConstant(LV);
67}
68
69bool SCCPSolver::tryToReplaceWithConstant(Value *V) {
70 Constant *Const = getConstantOrNull(V);
71 if (!Const)
72 return false;
73 // Replacing `musttail` instructions with constant breaks `musttail` invariant
74 // unless the call itself can be removed.
75 // Calls with "clang.arc.attachedcall" implicitly use the return value and
76 // those uses cannot be updated with a constant.
77 CallBase *CB = dyn_cast<CallBase>(Val: V);
78 if (CB && ((CB->isMustTailCall() && !wouldInstructionBeTriviallyDead(I: CB)) ||
79 CB->getOperandBundle(ID: LLVMContext::OB_clang_arc_attachedcall))) {
80 Function *F = CB->getCalledFunction();
81
82 // Don't zap returns of the callee
83 if (F)
84 addToMustPreserveReturnsInFunctions(F);
85
86 LLVM_DEBUG(dbgs() << " Can\'t treat the result of call " << *CB
87 << " as a constant\n");
88 return false;
89 }
90
91 // For pointer constants derived from PredicateInfo, the constant may have
92 // different provenance. Take this into account during constant pointer
93 // propagation.
94 if (V->getType()->isPointerTy()) {
95 const auto &LV = getLatticeValueFor(V);
96 if (LV.mayHaveDifferentProvenance()) {
97 const DataLayout &DL = getDataLayout();
98 bool Changed = V->replaceUsesWithIf(New: Const, ShouldReplace: [&](Use &U) {
99 bool CanReplace = canReplacePointersInUseIfEqual(U, To: Const, DL);
100 if (CanReplace)
101 LLVM_DEBUG(dbgs() << " Constant pointer: " << *Const << " = " << *V
102 << '\n');
103 return CanReplace;
104 });
105 return Changed;
106 }
107 }
108
109 LLVM_DEBUG(dbgs() << " Constant: " << *Const << " = " << *V << '\n');
110
111 // Replaces all of the uses of a variable with uses of the constant.
112 V->replaceAllUsesWith(V: Const);
113 return true;
114}
115
116/// Helper for propagting !implicit.ref metadata from callee to caller before
117/// erasing a call instruction. This ensures that references to global objects
118/// (e.g., copyright strings) are preserved even when calls are optimized away.
119static void propagateImplicitRefFromCall(CallBase *CB) {
120 Function *Callee = CB->getCalledFunction();
121 if (!Callee)
122 return;
123
124 if (!Callee->hasMetadata(KindID: LLVMContext::MD_implicit_ref))
125 return;
126
127 Function *Caller = CB->getParent()->getParent();
128 if (!Caller)
129 return;
130
131 SmallVector<MDNode *> MDs;
132 Callee->getMetadata(KindID: LLVMContext::MD_implicit_ref, MDs);
133 for (MDNode *MD : MDs)
134 Caller->addMetadata(KindID: LLVMContext::MD_implicit_ref, MD&: *MD);
135}
136
137/// Helper for getting ranges from \p Solver. Instructions inserted during
138/// simplification are unavailable in the solver, so we return a full range for
139/// them.
140static ConstantRange getRange(Value *Op, SCCPSolver &Solver,
141 const SmallPtrSetImpl<Value *> &InsertedValues) {
142 if (auto *Const = dyn_cast<Constant>(Val: Op))
143 return Const->toConstantRange();
144 if (InsertedValues.contains(Ptr: Op)) {
145 unsigned Bitwidth = Op->getType()->getScalarSizeInBits();
146 return ConstantRange::getFull(BitWidth: Bitwidth);
147 }
148 return Solver.getLatticeValueFor(V: Op).asConstantRange(Ty: Op->getType(),
149 /*UndefAllowed=*/false);
150}
151
152/// Try to use \p Inst's value range from \p Solver to infer the NUW flag.
153static bool refineInstruction(SCCPSolver &Solver,
154 const SmallPtrSetImpl<Value *> &InsertedValues,
155 Instruction &Inst) {
156 bool Changed = false;
157 auto GetRange = [&Solver, &InsertedValues](Value *Op) {
158 return getRange(Op, Solver, InsertedValues);
159 };
160
161 if (isa<OverflowingBinaryOperator>(Val: Inst)) {
162 if (Inst.hasNoSignedWrap() && Inst.hasNoUnsignedWrap())
163 return false;
164
165 auto RangeA = GetRange(Inst.getOperand(i: 0));
166 auto RangeB = GetRange(Inst.getOperand(i: 1));
167 if (!Inst.hasNoUnsignedWrap()) {
168 auto NUWRange = ConstantRange::makeGuaranteedNoWrapRegion(
169 BinOp: Instruction::BinaryOps(Inst.getOpcode()), Other: RangeB,
170 NoWrapKind: OverflowingBinaryOperator::NoUnsignedWrap);
171 if (NUWRange.contains(CR: RangeA)) {
172 Inst.setHasNoUnsignedWrap();
173 Changed = true;
174 }
175 }
176 if (!Inst.hasNoSignedWrap()) {
177 auto NSWRange = ConstantRange::makeGuaranteedNoWrapRegion(
178 BinOp: Instruction::BinaryOps(Inst.getOpcode()), Other: RangeB,
179 NoWrapKind: OverflowingBinaryOperator::NoSignedWrap);
180 if (NSWRange.contains(CR: RangeA)) {
181 Inst.setHasNoSignedWrap();
182 Changed = true;
183 }
184 }
185 } else if (isa<PossiblyNonNegInst>(Val: Inst) && !Inst.hasNonNeg()) {
186 auto Range = GetRange(Inst.getOperand(i: 0));
187 if (Range.isAllNonNegative()) {
188 Inst.setNonNeg();
189 Changed = true;
190 }
191 } else if (TruncInst *TI = dyn_cast<TruncInst>(Val: &Inst)) {
192 if (TI->hasNoSignedWrap() && TI->hasNoUnsignedWrap())
193 return false;
194
195 auto Range = GetRange(Inst.getOperand(i: 0));
196 uint64_t DestWidth = TI->getDestTy()->getScalarSizeInBits();
197 if (!TI->hasNoUnsignedWrap()) {
198 if (Range.getActiveBits() <= DestWidth) {
199 TI->setHasNoUnsignedWrap(true);
200 Changed = true;
201 }
202 }
203 if (!TI->hasNoSignedWrap()) {
204 if (Range.getMinSignedBits() <= DestWidth) {
205 TI->setHasNoSignedWrap(true);
206 Changed = true;
207 }
208 }
209 } else if (auto *GEP = dyn_cast<GetElementPtrInst>(Val: &Inst)) {
210 if (GEP->hasNoUnsignedWrap() || !GEP->hasNoUnsignedSignedWrap())
211 return false;
212
213 if (all_of(Range: GEP->indices(),
214 P: [&](Value *V) { return GetRange(V).isAllNonNegative(); })) {
215 GEP->setNoWrapFlags(GEP->getNoWrapFlags() |
216 GEPNoWrapFlags::noUnsignedWrap());
217 Changed = true;
218 }
219 }
220
221 return Changed;
222}
223
224/// Try to replace signed instructions with their unsigned equivalent.
225static bool replaceSignedInst(SCCPSolver &Solver,
226 SmallPtrSetImpl<Value *> &InsertedValues,
227 Instruction &Inst) {
228 // Determine if a signed value is known to be >= 0.
229 auto isNonNegative = [&Solver, &InsertedValues](Value *V) {
230 return getRange(Op: V, Solver, InsertedValues).isAllNonNegative();
231 };
232
233 Instruction *NewInst = nullptr;
234 switch (Inst.getOpcode()) {
235 case Instruction::SIToFP:
236 case Instruction::SExt: {
237 // If the source value is not negative, this is a zext/uitofp.
238 Value *Op0 = Inst.getOperand(i: 0);
239 if (!isNonNegative(Op0))
240 return false;
241 NewInst = CastInst::Create(Inst.getOpcode() == Instruction::SExt
242 ? Instruction::ZExt
243 : Instruction::UIToFP,
244 S: Op0, Ty: Inst.getType(), Name: "", InsertBefore: Inst.getIterator());
245 NewInst->setNonNeg();
246 break;
247 }
248 case Instruction::AShr: {
249 // If the shifted value is not negative, this is a logical shift right.
250 Value *Op0 = Inst.getOperand(i: 0);
251 if (!isNonNegative(Op0))
252 return false;
253 NewInst = BinaryOperator::CreateLShr(V1: Op0, V2: Inst.getOperand(i: 1), Name: "", InsertBefore: Inst.getIterator());
254 NewInst->setIsExact(Inst.isExact());
255 break;
256 }
257 case Instruction::SDiv:
258 case Instruction::SRem: {
259 // If both operands are not negative, this is the same as udiv/urem.
260 Value *Op0 = Inst.getOperand(i: 0), *Op1 = Inst.getOperand(i: 1);
261 if (!isNonNegative(Op0) || !isNonNegative(Op1))
262 return false;
263 auto NewOpcode = Inst.getOpcode() == Instruction::SDiv ? Instruction::UDiv
264 : Instruction::URem;
265 NewInst = BinaryOperator::Create(Op: NewOpcode, S1: Op0, S2: Op1, Name: "", InsertBefore: Inst.getIterator());
266 if (Inst.getOpcode() == Instruction::SDiv)
267 NewInst->setIsExact(Inst.isExact());
268 break;
269 }
270 default:
271 return false;
272 }
273
274 // Wire up the new instruction and update state.
275 assert(NewInst && "Expected replacement instruction");
276 NewInst->takeName(V: &Inst);
277 InsertedValues.insert(Ptr: NewInst);
278 Inst.replaceAllUsesWith(V: NewInst);
279 NewInst->setDebugLoc(Inst.getDebugLoc());
280 Solver.removeLatticeValueFor(V: &Inst);
281 Inst.eraseFromParent();
282 return true;
283}
284
285/// Try to use \p Inst's value range from \p Solver to simplify it.
286static Value *simplifyInstruction(SCCPSolver &Solver,
287 SmallPtrSetImpl<Value *> &InsertedValues,
288 Instruction &Inst) {
289 auto GetRange = [&Solver, &InsertedValues](Value *Op) {
290 return getRange(Op, Solver, InsertedValues);
291 };
292
293 Value *X;
294 const APInt *RHSC;
295 // Remove masking operations.
296 if (match(V: &Inst, P: m_And(L: m_Value(V&: X), R: m_LowBitMask(V&: RHSC)))) {
297 ConstantRange LRange = GetRange(X);
298 if (LRange.getUnsignedMax().ule(RHS: *RHSC))
299 return X;
300 }
301
302 // Check if we can simplify [us]cmp(X, Y) to X - Y.
303 if (auto *Cmp = dyn_cast<CmpIntrinsic>(Val: &Inst)) {
304 Value *LHS = Cmp->getOperand(i_nocapture: 0);
305 Value *RHS = Cmp->getOperand(i_nocapture: 1);
306 unsigned BitWidth = LHS->getType()->getScalarSizeInBits();
307 // Bail out on 1-bit comparisons.
308 if (BitWidth == 1)
309 return nullptr;
310 ConstantRange LRange = GetRange(LHS);
311 if (LRange.isSizeLargerThan(MaxSize: 3))
312 return nullptr;
313 ConstantRange RRange = GetRange(RHS);
314 if (RRange.isSizeLargerThan(MaxSize: 3))
315 return nullptr;
316 ConstantRange RHSLower = RRange.sub(Other: APInt(BitWidth, 1));
317 ConstantRange RHSUpper = RRange.add(Other: APInt(BitWidth, 1));
318 ICmpInst::Predicate Pred =
319 Cmp->isSigned() ? CmpInst::ICMP_SLE : CmpInst::ICMP_ULE;
320 if (!RHSLower.icmp(Pred, Other: LRange) || !LRange.icmp(Pred, Other: RHSUpper))
321 return nullptr;
322
323 IRBuilder<NoFolder> Builder(&Inst);
324 Value *Sub = Builder.CreateSub(LHS, RHS, Name: Inst.getName(), /*HasNUW=*/false,
325 /*HasNSW=*/Cmp->isSigned());
326 InsertedValues.insert(Ptr: Sub);
327 if (Sub->getType() != Inst.getType()) {
328 Sub = Builder.CreateSExtOrTrunc(V: Sub, DestTy: Inst.getType());
329 InsertedValues.insert(Ptr: Sub);
330 }
331 return Sub;
332 }
333
334 // Relax range checks.
335 if (auto *ICmp = dyn_cast<ICmpInst>(Val: &Inst)) {
336 Value *X;
337 auto MatchTwoInstructionExactRangeCheck =
338 [&]() -> std::optional<ConstantRange> {
339 const APInt *RHSC;
340 if (!match(V: ICmp->getOperand(i_nocapture: 1), P: m_APInt(Res&: RHSC)))
341 return std::nullopt;
342
343 Value *LHS = ICmp->getOperand(i_nocapture: 0);
344 ICmpInst::Predicate Pred = ICmp->getPredicate();
345 const APInt *Offset;
346 if (match(V: LHS, P: m_OneUse(SubPattern: m_AddLike(L: m_Value(V&: X), R: m_APInt(Res&: Offset)))))
347 return ConstantRange::makeExactICmpRegion(Pred, Other: *RHSC).sub(Other: *Offset);
348 // Match icmp eq/ne X & NegPow2, C
349 if (ICmp->isEquality()) {
350 const APInt *Mask;
351 if (match(V: LHS, P: m_OneUse(SubPattern: m_And(L: m_Value(V&: X), R: m_NegatedPower2(V&: Mask)))) &&
352 RHSC->countr_zero() >= Mask->countr_zero()) {
353 ConstantRange CR(*RHSC, *RHSC - *Mask);
354 return Pred == ICmpInst::ICMP_EQ ? CR : CR.inverse();
355 }
356 }
357 return std::nullopt;
358 };
359
360 if (auto CR = MatchTwoInstructionExactRangeCheck()) {
361 ConstantRange LRange = GetRange(X);
362 // Early exit if we know nothing about X.
363 if (LRange.isFullSet())
364 return nullptr;
365 auto ConvertCRToICmp =
366 [&](const std::optional<ConstantRange> &NewCR) -> Value * {
367 ICmpInst::Predicate Pred;
368 APInt RHS;
369 // Check if we can represent NewCR as an icmp predicate.
370 if (NewCR && NewCR->getEquivalentICmp(Pred, RHS)) {
371 IRBuilder<NoFolder> Builder(&Inst);
372 Value *NewICmp =
373 Builder.CreateICmp(P: Pred, LHS: X, RHS: ConstantInt::get(Ty: X->getType(), V: RHS));
374 InsertedValues.insert(Ptr: NewICmp);
375 return NewICmp;
376 }
377 return nullptr;
378 };
379 // We are allowed to refine the comparison to either true or false for out
380 // of range inputs.
381 // Here we refine the comparison to false, and check if we can narrow the
382 // range check to a simpler test.
383 if (auto *V = ConvertCRToICmp(CR->exactIntersectWith(CR: LRange)))
384 return V;
385 // Here we refine the comparison to true, i.e. we relax the range check.
386 if (auto *V = ConvertCRToICmp(CR->exactUnionWith(CR: LRange.inverse())))
387 return V;
388 }
389 }
390
391 return nullptr;
392}
393
394bool SCCPSolver::simplifyInstsInBlock(BasicBlock &BB,
395 SmallPtrSetImpl<Value *> &InsertedValues,
396 Statistic &InstRemovedStat,
397 Statistic &InstReplacedStat) {
398 bool MadeChanges = false;
399 for (Instruction &Inst : make_early_inc_range(Range&: BB)) {
400 if (Inst.getType()->isVoidTy())
401 continue;
402 if (tryToReplaceWithConstant(V: &Inst)) {
403 if (isInstructionTriviallyDead(I: &Inst)) {
404 // Propagate !implicit.ref before erasing the call.
405 if (auto *CB = dyn_cast<CallBase>(Val: &Inst))
406 propagateImplicitRefFromCall(CB);
407
408 Inst.eraseFromParent();
409 ++InstRemovedStat;
410 }
411 MadeChanges = true;
412 } else if (replaceSignedInst(Solver&: *this, InsertedValues, Inst)) {
413 MadeChanges = true;
414 ++InstReplacedStat;
415 } else if (refineInstruction(Solver&: *this, InsertedValues, Inst)) {
416 MadeChanges = true;
417 } else if (auto *V = simplifyInstruction(Solver&: *this, InsertedValues, Inst)) {
418 Inst.replaceAllUsesWith(V);
419 Inst.eraseFromParent();
420 ++InstRemovedStat;
421 MadeChanges = true;
422 }
423 }
424 return MadeChanges;
425}
426
427bool SCCPSolver::removeNonFeasibleEdges(BasicBlock *BB, DomTreeUpdater &DTU,
428 BasicBlock *&NewUnreachableBB) const {
429 SmallPtrSet<BasicBlock *, 8> FeasibleSuccessors;
430 bool HasNonFeasibleEdges = false;
431 for (BasicBlock *Succ : successors(BB)) {
432 if (isEdgeFeasible(From: BB, To: Succ))
433 FeasibleSuccessors.insert(Ptr: Succ);
434 else
435 HasNonFeasibleEdges = true;
436 }
437
438 // All edges feasible, nothing to do.
439 if (!HasNonFeasibleEdges)
440 return false;
441
442 // SCCP can only determine non-feasible edges for br, switch and indirectbr.
443 Instruction *TI = BB->getTerminator();
444 assert((isa<UncondBrInst, CondBrInst, SwitchInst, IndirectBrInst>(TI)) &&
445 "Terminator must be a br, switch or indirectbr");
446
447 if (FeasibleSuccessors.size() == 0) {
448 // Branch on undef/poison, replace with unreachable.
449 SmallPtrSet<BasicBlock *, 8> SeenSuccs;
450 SmallVector<DominatorTree::UpdateType, 8> Updates;
451 for (BasicBlock *Succ : successors(BB)) {
452 Succ->removePredecessor(Pred: BB);
453 if (SeenSuccs.insert(Ptr: Succ).second)
454 Updates.push_back(Elt: {DominatorTree::Delete, BB, Succ});
455 }
456 TI->eraseFromParent();
457 new UnreachableInst(BB->getContext(), BB);
458 DTU.applyUpdatesPermissive(Updates);
459 } else if (FeasibleSuccessors.size() == 1) {
460 // Replace with an unconditional branch to the only feasible successor.
461 BasicBlock *OnlyFeasibleSuccessor = *FeasibleSuccessors.begin();
462 SmallVector<DominatorTree::UpdateType, 8> Updates;
463 bool HaveSeenOnlyFeasibleSuccessor = false;
464 for (BasicBlock *Succ : successors(BB)) {
465 if (Succ == OnlyFeasibleSuccessor && !HaveSeenOnlyFeasibleSuccessor) {
466 // Don't remove the edge to the only feasible successor the first time
467 // we see it. We still do need to remove any multi-edges to it though.
468 HaveSeenOnlyFeasibleSuccessor = true;
469 continue;
470 }
471
472 Succ->removePredecessor(Pred: BB);
473 Updates.push_back(Elt: {DominatorTree::Delete, BB, Succ});
474 }
475
476 Instruction *BI = UncondBrInst::Create(Target: OnlyFeasibleSuccessor, InsertBefore: BB);
477 BI->setDebugLoc(TI->getDebugLoc());
478 TI->eraseFromParent();
479 DTU.applyUpdatesPermissive(Updates);
480 } else if (FeasibleSuccessors.size() > 1) {
481 SwitchInstProfUpdateWrapper SI(*cast<SwitchInst>(Val: TI));
482 SmallVector<DominatorTree::UpdateType, 8> Updates;
483
484 // If the default destination is unfeasible it will never be taken. Replace
485 // it with a new block with a single Unreachable instruction.
486 BasicBlock *DefaultDest = SI->getDefaultDest();
487 if (!FeasibleSuccessors.contains(Ptr: DefaultDest)) {
488 if (!NewUnreachableBB) {
489 NewUnreachableBB =
490 BasicBlock::Create(Context&: DefaultDest->getContext(), Name: "default.unreachable",
491 Parent: DefaultDest->getParent(), InsertBefore: DefaultDest);
492 auto *UI =
493 new UnreachableInst(DefaultDest->getContext(), NewUnreachableBB);
494 UI->setDebugLoc(DebugLoc::getTemporary());
495 }
496
497 DefaultDest->removePredecessor(Pred: BB);
498 SI->setDefaultDest(NewUnreachableBB);
499 Updates.push_back(Elt: {DominatorTree::Delete, BB, DefaultDest});
500 Updates.push_back(Elt: {DominatorTree::Insert, BB, NewUnreachableBB});
501 }
502
503 for (auto CI = SI->case_begin(); CI != SI->case_end();) {
504 if (FeasibleSuccessors.contains(Ptr: CI->getCaseSuccessor())) {
505 ++CI;
506 continue;
507 }
508
509 BasicBlock *Succ = CI->getCaseSuccessor();
510 Succ->removePredecessor(Pred: BB);
511 Updates.push_back(Elt: {DominatorTree::Delete, BB, Succ});
512 SI.removeCase(I: CI);
513 // Don't increment CI, as we removed a case.
514 }
515
516 DTU.applyUpdatesPermissive(Updates);
517 } else {
518 llvm_unreachable("Must have at least one feasible successor");
519 }
520 return true;
521}
522
523static void inferAttribute(Function *F, unsigned AttrIndex,
524 const ValueLatticeElement &Val) {
525 // If there is a known constant range for the value, add range attribute.
526 if (Val.isConstantRange() && !Val.getConstantRange().isSingleElement()) {
527 // Do not add range attribute if the value may include undef.
528 if (Val.isConstantRangeIncludingUndef())
529 return;
530
531 // Take the intersection of the existing attribute and the inferred range.
532 Attribute OldAttr = F->getAttributeAtIndex(i: AttrIndex, Kind: Attribute::Range);
533 ConstantRange CR = Val.getConstantRange();
534 if (OldAttr.isValid())
535 CR = CR.intersectWith(CR: OldAttr.getRange());
536 F->addAttributeAtIndex(
537 i: AttrIndex, Attr: Attribute::get(Context&: F->getContext(), Kind: Attribute::Range, CR));
538 return;
539 }
540 // Infer nonnull attribute.
541 if (Val.isNotConstant() && Val.getNotConstant()->getType()->isPointerTy() &&
542 Val.getNotConstant()->isNullValue() &&
543 !F->hasAttributeAtIndex(Idx: AttrIndex, Kind: Attribute::NonNull)) {
544 F->addAttributeAtIndex(i: AttrIndex,
545 Attr: Attribute::get(Context&: F->getContext(), Kind: Attribute::NonNull));
546 }
547}
548
549void SCCPSolver::inferReturnAttributes() const {
550 for (const auto &[F, ReturnValue] : getTrackedRetVals())
551 inferAttribute(F, AttrIndex: AttributeList::ReturnIndex, Val: ReturnValue);
552}
553
554void SCCPSolver::inferArgAttributes() const {
555 for (Function *F : getArgumentTrackedFunctions()) {
556 if (!isBlockExecutable(BB: &F->front()))
557 continue;
558 for (Argument &A : F->args())
559 if (!A.getType()->isStructTy())
560 inferAttribute(F, AttrIndex: AttributeList::FirstArgIndex + A.getArgNo(),
561 Val: getLatticeValueFor(V: &A));
562 }
563}
564
565/// Helper class for SCCPSolver. This implements the instruction visitor and
566/// holds all the state.
567class SCCPInstVisitor : public InstVisitor<SCCPInstVisitor> {
568 const DataLayout &DL;
569 std::function<const TargetLibraryInfo &(Function &)> GetTLI;
570 /// Basic blocks that are executable (but may not have been visited yet).
571 SmallPtrSet<BasicBlock *, 8> BBExecutable;
572 /// Basic blocks that are executable and have been visited at least once.
573 SmallPtrSet<BasicBlock *, 8> BBVisited;
574 DenseMap<Value *, ValueLatticeElement>
575 ValueState; // The state each value is in.
576
577 /// StructValueState - This maintains ValueState for values that have
578 /// StructType, for example for formal arguments, calls, insertelement, etc.
579 DenseMap<std::pair<Value *, unsigned>, ValueLatticeElement> StructValueState;
580
581 /// GlobalValue - If we are tracking any values for the contents of a global
582 /// variable, we keep a mapping from the constant accessor to the element of
583 /// the global, to the currently known value. If the value becomes
584 /// overdefined, it's entry is simply removed from this map.
585 DenseMap<GlobalVariable *, ValueLatticeElement> TrackedGlobals;
586
587 /// TrackedRetVals - If we are tracking arguments into and the return
588 /// value out of a function, it will have an entry in this map, indicating
589 /// what the known return value for the function is.
590 MapVector<Function *, ValueLatticeElement> TrackedRetVals;
591
592 /// TrackedMultipleRetVals - Same as TrackedRetVals, but used for functions
593 /// that return multiple values.
594 MapVector<std::pair<Function *, unsigned>, ValueLatticeElement>
595 TrackedMultipleRetVals;
596
597 /// The set of values whose lattice has been invalidated.
598 /// Populated by resetLatticeValueFor(), cleared after resolving undefs.
599 DenseSet<Value *> Invalidated;
600
601 /// MRVFunctionsTracked - Each function in TrackedMultipleRetVals is
602 /// represented here for efficient lookup.
603 SmallPtrSet<Function *, 16> MRVFunctionsTracked;
604
605 /// A list of functions whose return cannot be modified.
606 SmallPtrSet<Function *, 16> MustPreserveReturnsInFunctions;
607
608 /// TrackingIncomingArguments - This is the set of functions for whose
609 /// arguments we make optimistic assumptions about and try to prove as
610 /// constants.
611 SmallPtrSet<Function *, 16> TrackingIncomingArguments;
612
613 /// Worklist of instructions to re-visit. This only includes instructions
614 /// in blocks that have already been visited at least once.
615 SmallSetVector<Instruction *, 16> InstWorkList;
616
617 /// Current instruction while visiting a block for the first time, used to
618 /// avoid unnecessary instruction worklist insertions. Null if an instruction
619 /// is visited outside a whole-block visitation.
620 Instruction *CurI = nullptr;
621
622 // The BasicBlock work list
623 SmallVector<BasicBlock *, 64> BBWorkList;
624
625 /// KnownFeasibleEdges - Entries in this set are edges which have already had
626 /// PHI nodes retriggered.
627 using Edge = std::pair<BasicBlock *, BasicBlock *>;
628 DenseSet<Edge> KnownFeasibleEdges;
629
630 DenseMap<Function *, std::unique_ptr<PredicateInfo>> FnPredicateInfo;
631
632 DenseMap<Value *, SmallSetVector<User *, 2>> AdditionalUsers;
633
634 LLVMContext &Ctx;
635
636 BumpPtrAllocator PredicateInfoAllocator;
637
638private:
639 ConstantInt *getConstantInt(const ValueLatticeElement &IV, Type *Ty) const {
640 return dyn_cast_or_null<ConstantInt>(Val: getConstant(LV: IV, Ty));
641 }
642
643 /// Push instruction \p I to the worklist.
644 void pushToWorkList(Instruction *I);
645
646 /// Push users of value \p V to the worklist.
647 void pushUsersToWorkList(Value *V);
648
649 /// Like pushUsersToWorkList(), but also prints a debug message with the
650 /// updated value.
651 void pushUsersToWorkListMsg(ValueLatticeElement &IV, Value *V);
652
653 // markConstant - Make a value be marked as "constant". If the value
654 // is not already a constant, add it to the instruction work list so that
655 // the users of the instruction are updated later.
656 bool markConstant(ValueLatticeElement &IV, Value *V, Constant *C,
657 bool MayIncludeUndef = false);
658
659 bool markConstant(Value *V, Constant *C) {
660 assert(!V->getType()->isStructTy() && "structs should use mergeInValue");
661 return markConstant(IV&: ValueState[V], V, C);
662 }
663
664 bool markNotConstant(ValueLatticeElement &IV, Value *V, Constant *C);
665
666 bool markNotNull(ValueLatticeElement &IV, Value *V) {
667 return markNotConstant(IV, V, C: Constant::getNullValue(Ty: V->getType()));
668 }
669
670 /// markConstantRange - Mark the object as constant range with \p CR. If the
671 /// object is not a constant range with the range \p CR, add it to the
672 /// instruction work list so that the users of the instruction are updated
673 /// later.
674 bool markConstantRange(ValueLatticeElement &IV, Value *V,
675 const ConstantRange &CR);
676
677 // markOverdefined - Make a value be marked as "overdefined". If the
678 // value is not already overdefined, add it to the overdefined instruction
679 // work list so that the users of the instruction are updated later.
680 bool markOverdefined(ValueLatticeElement &IV, Value *V);
681
682 /// Merge \p MergeWithV into \p IV and push \p V to the worklist, if \p IV
683 /// changes.
684 bool mergeInValue(ValueLatticeElement &IV, Value *V,
685 const ValueLatticeElement &MergeWithV,
686 ValueLatticeElement::MergeOptions Opts = {
687 /*MayIncludeUndef=*/false, /*CheckWiden=*/false});
688
689 /// getValueState - Return the ValueLatticeElement object that corresponds to
690 /// the value. This function handles the case when the value hasn't been seen
691 /// yet by properly seeding constants etc.
692 ValueLatticeElement &getValueState(Value *V) {
693 assert(!V->getType()->isStructTy() && "Should use getStructValueState");
694
695 auto I = ValueState.try_emplace(Key: V);
696 ValueLatticeElement &LV = I.first->second;
697
698 if (!I.second)
699 return LV; // Common case, already in the map.
700
701 if (auto *C = dyn_cast<Constant>(Val: V))
702 LV.markConstant(V: C); // Constants are constant
703
704 // All others are unknown by default.
705 return LV;
706 }
707
708 /// getStructValueState - Return the ValueLatticeElement object that
709 /// corresponds to the value/field pair. This function handles the case when
710 /// the value hasn't been seen yet by properly seeding constants etc.
711 ValueLatticeElement &getStructValueState(Value *V, unsigned i) {
712 assert(V->getType()->isStructTy() && "Should use getValueState");
713 assert(i < cast<StructType>(V->getType())->getNumElements() &&
714 "Invalid element #");
715
716 auto I = StructValueState.insert(
717 KV: std::make_pair(x: std::make_pair(x&: V, y&: i), y: ValueLatticeElement()));
718 ValueLatticeElement &LV = I.first->second;
719
720 if (!I.second)
721 return LV; // Common case, already in the map.
722
723 if (auto *C = dyn_cast<Constant>(Val: V)) {
724 Constant *Elt = C->getAggregateElement(Elt: i);
725
726 if (!Elt)
727 LV.markOverdefined(); // Unknown sort of constant.
728 else
729 LV.markConstant(V: Elt); // Constants are constant.
730 }
731
732 // All others are underdefined by default.
733 return LV;
734 }
735
736 /// Traverse the use-def chain of \p Call, marking itself and its users as
737 /// "unknown" on the way.
738 void invalidate(CallBase *Call) {
739 SmallVector<Instruction *, 64> ToInvalidate;
740 ToInvalidate.push_back(Elt: Call);
741
742 while (!ToInvalidate.empty()) {
743 Instruction *Inst = ToInvalidate.pop_back_val();
744
745 if (!Invalidated.insert(V: Inst).second)
746 continue;
747
748 if (!BBExecutable.count(Ptr: Inst->getParent()))
749 continue;
750
751 Value *V = nullptr;
752 // For return instructions we need to invalidate the tracked returns map.
753 // Anything else has its lattice in the value map.
754 if (auto *RetInst = dyn_cast<ReturnInst>(Val: Inst)) {
755 Function *F = RetInst->getParent()->getParent();
756 if (auto It = TrackedRetVals.find(Key: F); It != TrackedRetVals.end()) {
757 It->second = ValueLatticeElement();
758 V = F;
759 } else if (MRVFunctionsTracked.count(Ptr: F)) {
760 auto *STy = cast<StructType>(Val: F->getReturnType());
761 for (unsigned I = 0, E = STy->getNumElements(); I != E; ++I)
762 TrackedMultipleRetVals[{F, I}] = ValueLatticeElement();
763 V = F;
764 }
765 } else if (auto *STy = dyn_cast<StructType>(Val: Inst->getType())) {
766 for (unsigned I = 0, E = STy->getNumElements(); I != E; ++I) {
767 if (auto It = StructValueState.find(Val: {Inst, I});
768 It != StructValueState.end()) {
769 It->second = ValueLatticeElement();
770 V = Inst;
771 }
772 }
773 } else if (auto It = ValueState.find(Val: Inst); It != ValueState.end()) {
774 It->second = ValueLatticeElement();
775 V = Inst;
776 }
777
778 if (V) {
779 LLVM_DEBUG(dbgs() << "Invalidated lattice for " << *V << "\n");
780
781 for (User *U : V->users())
782 if (auto *UI = dyn_cast<Instruction>(Val: U))
783 ToInvalidate.push_back(Elt: UI);
784
785 auto It = AdditionalUsers.find(Val: V);
786 if (It != AdditionalUsers.end())
787 for (User *U : It->second)
788 if (auto *UI = dyn_cast<Instruction>(Val: U))
789 ToInvalidate.push_back(Elt: UI);
790 }
791 }
792 }
793
794 /// markEdgeExecutable - Mark a basic block as executable, adding it to the BB
795 /// work list if it is not already executable.
796 bool markEdgeExecutable(BasicBlock *Source, BasicBlock *Dest);
797
798 // getFeasibleSuccessors - Return a vector of booleans to indicate which
799 // successors are reachable from a given terminator instruction.
800 void getFeasibleSuccessors(Instruction &TI, SmallVectorImpl<bool> &Succs);
801
802 // Add U as additional user of V.
803 void addAdditionalUser(Value *V, User *U) { AdditionalUsers[V].insert(X: U); }
804
805 void handlePredicate(Instruction *I, Value *CopyOf, const PredicateBase *PI);
806 void handleCallOverdefined(CallBase &CB);
807 void handleCallResult(CallBase &CB);
808 void handleCallArguments(CallBase &CB);
809 void handleExtractOfWithOverflow(ExtractValueInst &EVI,
810 const WithOverflowInst *WO, unsigned Idx);
811 bool isInstFullyOverDefined(Instruction &Inst);
812
813private:
814 friend class InstVisitor<SCCPInstVisitor>;
815
816 // visit implementations - Something changed in this instruction. Either an
817 // operand made a transition, or the instruction is newly executable. Change
818 // the value type of I to reflect these changes if appropriate.
819 void visitPHINode(PHINode &I);
820
821 // Terminators
822
823 void visitReturnInst(ReturnInst &I);
824 void visitTerminator(Instruction &TI);
825
826 void visitCastInst(CastInst &I);
827 void visitSelectInst(SelectInst &I);
828 void visitUnaryOperator(Instruction &I);
829 void visitFreezeInst(FreezeInst &I);
830 void visitBinaryOperator(Instruction &I);
831 void visitCmpInst(CmpInst &I);
832 void visitExtractValueInst(ExtractValueInst &EVI);
833 void visitInsertValueInst(InsertValueInst &IVI);
834
835 void visitCatchSwitchInst(CatchSwitchInst &CPI) {
836 markOverdefined(V: &CPI);
837 visitTerminator(TI&: CPI);
838 }
839
840 // Instructions that cannot be folded away.
841
842 void visitStoreInst(StoreInst &I);
843 void visitLoadInst(LoadInst &I);
844 void visitGetElementPtrInst(GetElementPtrInst &I);
845 void visitAllocaInst(AllocaInst &AI);
846
847 void visitInvokeInst(InvokeInst &II) {
848 visitCallBase(CB&: II);
849 visitTerminator(TI&: II);
850 }
851
852 void visitCallBrInst(CallBrInst &CBI) {
853 visitCallBase(CB&: CBI);
854 visitTerminator(TI&: CBI);
855 }
856
857 void visitCallBase(CallBase &CB);
858 void visitResumeInst(ResumeInst &I) { /*returns void*/
859 }
860 void visitUnreachableInst(UnreachableInst &I) { /*returns void*/
861 }
862 void visitFenceInst(FenceInst &I) { /*returns void*/
863 }
864
865 void visitInstruction(Instruction &I);
866
867public:
868 const DataLayout &getDataLayout() const { return DL; }
869
870 void addPredicateInfo(Function &F, DominatorTree &DT, AssumptionCache &AC) {
871 FnPredicateInfo.insert(KV: {&F, std::make_unique<PredicateInfo>(
872 args&: F, args&: DT, args&: AC, args&: PredicateInfoAllocator)});
873 }
874
875 void removeSSACopies(Function &F) {
876 auto It = FnPredicateInfo.find(Val: &F);
877 if (It == FnPredicateInfo.end())
878 return;
879
880 for (BasicBlock &BB : F) {
881 for (Instruction &Inst : llvm::make_early_inc_range(Range&: BB)) {
882 if (auto *BC = dyn_cast<BitCastInst>(Val: &Inst)) {
883 if (BC->getType() == BC->getOperand(i_nocapture: 0)->getType()) {
884 if (It->second->getPredicateInfoFor(V: &Inst)) {
885 Value *Op = BC->getOperand(i_nocapture: 0);
886 Inst.replaceAllUsesWith(V: Op);
887 Inst.eraseFromParent();
888 }
889 }
890 }
891 }
892 }
893 }
894
895 void visitCallInst(CallInst &I) { visitCallBase(CB&: I); }
896
897 bool markBlockExecutable(BasicBlock *BB);
898
899 const PredicateBase *getPredicateInfoFor(Instruction *I) {
900 auto It = FnPredicateInfo.find(Val: I->getParent()->getParent());
901 if (It == FnPredicateInfo.end())
902 return nullptr;
903 return It->second->getPredicateInfoFor(V: I);
904 }
905
906 SCCPInstVisitor(const DataLayout &DL,
907 std::function<const TargetLibraryInfo &(Function &)> GetTLI,
908 LLVMContext &Ctx)
909 : DL(DL), GetTLI(GetTLI), Ctx(Ctx) {}
910
911 void trackValueOfGlobalVariable(GlobalVariable *GV) {
912 // We only track the contents of scalar globals.
913 if (GV->getValueType()->isSingleValueType()) {
914 ValueLatticeElement &IV = TrackedGlobals[GV];
915 IV.markConstant(V: GV->getInitializer());
916 }
917 }
918
919 void addTrackedFunction(Function *F) {
920 // Add an entry, F -> undef.
921 if (auto *STy = dyn_cast<StructType>(Val: F->getReturnType())) {
922 MRVFunctionsTracked.insert(Ptr: F);
923 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
924 TrackedMultipleRetVals.try_emplace(Key: std::make_pair(x&: F, y&: i));
925 } else if (!F->getReturnType()->isVoidTy())
926 TrackedRetVals.try_emplace(Key: F);
927 }
928
929 void addToMustPreserveReturnsInFunctions(Function *F) {
930 MustPreserveReturnsInFunctions.insert(Ptr: F);
931 }
932
933 bool mustPreserveReturn(Function *F) {
934 return MustPreserveReturnsInFunctions.count(Ptr: F);
935 }
936
937 void addArgumentTrackedFunction(Function *F) {
938 TrackingIncomingArguments.insert(Ptr: F);
939 }
940
941 bool isArgumentTrackedFunction(Function *F) {
942 return TrackingIncomingArguments.count(Ptr: F);
943 }
944
945 const SmallPtrSetImpl<Function *> &getArgumentTrackedFunctions() const {
946 return TrackingIncomingArguments;
947 }
948
949 void solve();
950
951 bool resolvedUndef(Instruction &I);
952
953 bool resolvedUndefsIn(Function &F);
954
955 bool isBlockExecutable(BasicBlock *BB) const {
956 return BBExecutable.count(Ptr: BB);
957 }
958
959 bool isEdgeFeasible(BasicBlock *From, BasicBlock *To) const;
960
961 std::vector<ValueLatticeElement> getStructLatticeValueFor(Value *V) const {
962 std::vector<ValueLatticeElement> StructValues;
963 auto *STy = dyn_cast<StructType>(Val: V->getType());
964 assert(STy && "getStructLatticeValueFor() can be called only on structs");
965 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
966 auto I = StructValueState.find(Val: std::make_pair(x&: V, y&: i));
967 assert(I != StructValueState.end() && "Value not in valuemap!");
968 StructValues.push_back(x: I->second);
969 }
970 return StructValues;
971 }
972
973 void removeLatticeValueFor(Value *V) { ValueState.erase(Val: V); }
974
975 /// Invalidate the Lattice Value of \p Call and its users after specializing
976 /// the call. Then recompute it.
977 void resetLatticeValueFor(CallBase *Call) {
978 // Calls to void returning functions do not need invalidation.
979 Function *F = Call->getCalledFunction();
980 (void)F;
981 assert(!F->getReturnType()->isVoidTy() &&
982 (TrackedRetVals.count(F) || MRVFunctionsTracked.count(F)) &&
983 "All non void specializations should be tracked");
984 invalidate(Call);
985 handleCallResult(CB&: *Call);
986 }
987
988 const ValueLatticeElement &getLatticeValueFor(Value *V) const {
989 assert(!V->getType()->isStructTy() &&
990 "Should use getStructLatticeValueFor");
991 auto I = ValueState.find(Val: V);
992 assert(I != ValueState.end() &&
993 "V not found in ValueState nor Paramstate map!");
994 return I->second;
995 }
996
997 const MapVector<Function *, ValueLatticeElement> &getTrackedRetVals() const {
998 return TrackedRetVals;
999 }
1000
1001 const DenseMap<GlobalVariable *, ValueLatticeElement> &
1002 getTrackedGlobals() const {
1003 return TrackedGlobals;
1004 }
1005
1006 const SmallPtrSet<Function *, 16> &getMRVFunctionsTracked() const {
1007 return MRVFunctionsTracked;
1008 }
1009
1010 void markOverdefined(Value *V) {
1011 if (auto *STy = dyn_cast<StructType>(Val: V->getType()))
1012 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
1013 markOverdefined(IV&: getStructValueState(V, i), V);
1014 else
1015 markOverdefined(IV&: ValueState[V], V);
1016 }
1017
1018 ValueLatticeElement getArgAttributeVL(Argument *A) {
1019 if (A->getType()->isIntOrIntVectorTy()) {
1020 if (std::optional<ConstantRange> Range = A->getRange())
1021 return ValueLatticeElement::getRange(CR: *Range);
1022 }
1023 if (A->hasNonNullAttr())
1024 return ValueLatticeElement::getNot(C: Constant::getNullValue(Ty: A->getType()));
1025 // Assume nothing about the incoming arguments without attributes.
1026 return ValueLatticeElement::getOverdefined();
1027 }
1028
1029 void trackValueOfArgument(Argument *A) {
1030 if (A->getType()->isStructTy())
1031 return (void)markOverdefined(V: A);
1032 mergeInValue(IV&: ValueState[A], V: A, MergeWithV: getArgAttributeVL(A));
1033 }
1034
1035 bool isStructLatticeConstant(Function *F, StructType *STy);
1036
1037 Constant *getConstant(const ValueLatticeElement &LV, Type *Ty) const;
1038
1039 Constant *getConstantOrNull(Value *V) const;
1040
1041 void setLatticeValueForSpecializationArguments(Function *F,
1042 const SmallVectorImpl<ArgInfo> &Args);
1043
1044 void markFunctionUnreachable(Function *F) {
1045 for (auto &BB : *F)
1046 BBExecutable.erase(Ptr: &BB);
1047 }
1048
1049 void solveWhileResolvedUndefsIn(Module &M) {
1050 bool ResolvedUndefs = true;
1051 while (ResolvedUndefs) {
1052 solve();
1053 ResolvedUndefs = false;
1054 for (Function &F : M)
1055 ResolvedUndefs |= resolvedUndefsIn(F);
1056 }
1057 }
1058
1059 void solveWhileResolvedUndefsIn(SmallVectorImpl<Function *> &WorkList) {
1060 bool ResolvedUndefs = true;
1061 while (ResolvedUndefs) {
1062 solve();
1063 ResolvedUndefs = false;
1064 for (Function *F : WorkList)
1065 ResolvedUndefs |= resolvedUndefsIn(F&: *F);
1066 }
1067 }
1068
1069 void solveWhileResolvedUndefs() {
1070 bool ResolvedUndefs = true;
1071 while (ResolvedUndefs) {
1072 solve();
1073 ResolvedUndefs = false;
1074 for (Value *V : Invalidated)
1075 if (auto *I = dyn_cast<Instruction>(Val: V))
1076 ResolvedUndefs |= resolvedUndef(I&: *I);
1077 }
1078 Invalidated.clear();
1079 }
1080};
1081
1082} // namespace llvm
1083
1084bool SCCPInstVisitor::markBlockExecutable(BasicBlock *BB) {
1085 if (!BBExecutable.insert(Ptr: BB).second)
1086 return false;
1087 LLVM_DEBUG(dbgs() << "Marking Block Executable: " << BB->getName() << '\n');
1088 BBWorkList.push_back(Elt: BB); // Add the block to the work list!
1089 return true;
1090}
1091
1092void SCCPInstVisitor::pushToWorkList(Instruction *I) {
1093 // If we're currently visiting a block, do not push any instructions in the
1094 // same blocks that are after the current one, as they will be visited
1095 // anyway. We do have to push updates to earlier instructions (e.g. phi
1096 // nodes or loads of tracked globals).
1097 if (CurI && I->getParent() == CurI->getParent() && CurI->comesBefore(Other: I))
1098 return;
1099 // Only push instructions in already visited blocks. Otherwise we'll handle
1100 // it when we visit the block for the first time.
1101 if (BBVisited.contains(Ptr: I->getParent()))
1102 InstWorkList.insert(X: I);
1103}
1104
1105void SCCPInstVisitor::pushUsersToWorkList(Value *V) {
1106 for (User *U : V->users())
1107 if (auto *UI = dyn_cast<Instruction>(Val: U))
1108 pushToWorkList(I: UI);
1109
1110 auto Iter = AdditionalUsers.find(Val: V);
1111 if (Iter != AdditionalUsers.end()) {
1112 // Copy additional users before notifying them of changes, because new
1113 // users may be added, potentially invalidating the iterator.
1114 SmallVector<Instruction *, 2> ToNotify;
1115 for (User *U : Iter->second)
1116 if (auto *UI = dyn_cast<Instruction>(Val: U))
1117 ToNotify.push_back(Elt: UI);
1118 for (Instruction *UI : ToNotify)
1119 pushToWorkList(I: UI);
1120 }
1121}
1122
1123void SCCPInstVisitor::pushUsersToWorkListMsg(ValueLatticeElement &IV,
1124 Value *V) {
1125 LLVM_DEBUG(dbgs() << "updated " << IV << ": " << *V << '\n');
1126 pushUsersToWorkList(V);
1127}
1128
1129bool SCCPInstVisitor::markConstant(ValueLatticeElement &IV, Value *V,
1130 Constant *C, bool MayIncludeUndef) {
1131 if (!IV.markConstant(V: C, MayIncludeUndef))
1132 return false;
1133 LLVM_DEBUG(dbgs() << "markConstant: " << *C << ": " << *V << '\n');
1134 pushUsersToWorkList(V);
1135 return true;
1136}
1137
1138bool SCCPInstVisitor::markNotConstant(ValueLatticeElement &IV, Value *V,
1139 Constant *C) {
1140 if (!IV.markNotConstant(V: C))
1141 return false;
1142 LLVM_DEBUG(dbgs() << "markNotConstant: " << *C << ": " << *V << '\n');
1143 pushUsersToWorkList(V);
1144 return true;
1145}
1146
1147bool SCCPInstVisitor::markConstantRange(ValueLatticeElement &IV, Value *V,
1148 const ConstantRange &CR) {
1149 if (!IV.markConstantRange(NewR: CR))
1150 return false;
1151 LLVM_DEBUG(dbgs() << "markConstantRange: " << CR << ": " << *V << '\n');
1152 pushUsersToWorkList(V);
1153 return true;
1154}
1155
1156bool SCCPInstVisitor::markOverdefined(ValueLatticeElement &IV, Value *V) {
1157 if (!IV.markOverdefined())
1158 return false;
1159
1160 LLVM_DEBUG(dbgs() << "markOverdefined: ";
1161 if (auto *F = dyn_cast<Function>(V)) dbgs()
1162 << "Function '" << F->getName() << "'\n";
1163 else dbgs() << *V << '\n');
1164 // Only instructions go on the work list
1165 pushUsersToWorkList(V);
1166 return true;
1167}
1168
1169bool SCCPInstVisitor::isStructLatticeConstant(Function *F, StructType *STy) {
1170 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
1171 const auto &It = TrackedMultipleRetVals.find(Key: std::make_pair(x&: F, y&: i));
1172 assert(It != TrackedMultipleRetVals.end());
1173 if (!SCCPSolver::isReplaceableConstant(LV: It->second))
1174 return false;
1175 }
1176 return true;
1177}
1178
1179Constant *SCCPInstVisitor::getConstant(const ValueLatticeElement &LV,
1180 Type *Ty) const {
1181 if (LV.isConstant()) {
1182 Constant *C = LV.getConstant();
1183 assert(C->getType() == Ty && "Type mismatch");
1184 return C;
1185 }
1186
1187 if (LV.isConstantRange()) {
1188 const auto &CR = LV.getConstantRange();
1189 if (CR.getSingleElement())
1190 return ConstantInt::get(Ty, V: *CR.getSingleElement());
1191 }
1192 return nullptr;
1193}
1194
1195Constant *SCCPInstVisitor::getConstantOrNull(Value *V) const {
1196 Constant *Const = nullptr;
1197 if (V->getType()->isStructTy()) {
1198 std::vector<ValueLatticeElement> LVs = getStructLatticeValueFor(V);
1199 if (any_of(Range&: LVs, P: SCCPSolver::isOverdefined))
1200 return nullptr;
1201 std::vector<Constant *> ConstVals;
1202 auto *ST = cast<StructType>(Val: V->getType());
1203 for (unsigned I = 0, E = ST->getNumElements(); I != E; ++I) {
1204 const ValueLatticeElement &LV = LVs[I];
1205 ConstVals.push_back(x: SCCPSolver::isConstant(LV)
1206 ? getConstant(LV, Ty: ST->getElementType(N: I))
1207 : UndefValue::get(T: ST->getElementType(N: I)));
1208 }
1209 Const = ConstantStruct::get(T: ST, V: ConstVals);
1210 } else {
1211 const ValueLatticeElement &LV = getLatticeValueFor(V);
1212 if (SCCPSolver::isOverdefined(LV))
1213 return nullptr;
1214 Const = SCCPSolver::isConstant(LV) ? getConstant(LV, Ty: V->getType())
1215 : UndefValue::get(T: V->getType());
1216 }
1217 assert(Const && "Constant is nullptr here!");
1218 return Const;
1219}
1220
1221void SCCPInstVisitor::setLatticeValueForSpecializationArguments(Function *F,
1222 const SmallVectorImpl<ArgInfo> &Args) {
1223 assert(!Args.empty() && "Specialization without arguments");
1224 assert(F->arg_size() == Args[0].Formal->getParent()->arg_size() &&
1225 "Functions should have the same number of arguments");
1226
1227 auto Iter = Args.begin();
1228 Function::arg_iterator NewArg = F->arg_begin();
1229 Function::arg_iterator OldArg = Args[0].Formal->getParent()->arg_begin();
1230 for (auto End = F->arg_end(); NewArg != End; ++NewArg, ++OldArg) {
1231
1232 LLVM_DEBUG(dbgs() << "SCCP: Marking argument "
1233 << NewArg->getNameOrAsOperand() << "\n");
1234
1235 // Mark the argument constants in the new function
1236 // or copy the lattice state over from the old function.
1237 if (Iter != Args.end() && Iter->Formal == &*OldArg) {
1238 if (auto *STy = dyn_cast<StructType>(Val: NewArg->getType())) {
1239 for (unsigned I = 0, E = STy->getNumElements(); I != E; ++I) {
1240 ValueLatticeElement &NewValue = StructValueState[{&*NewArg, I}];
1241 NewValue.markConstant(V: Iter->Actual->getAggregateElement(Elt: I));
1242 }
1243 } else {
1244 ValueState[&*NewArg].markConstant(V: Iter->Actual);
1245 }
1246 ++Iter;
1247 } else {
1248 if (auto *STy = dyn_cast<StructType>(Val: NewArg->getType())) {
1249 for (unsigned I = 0, E = STy->getNumElements(); I != E; ++I) {
1250 ValueLatticeElement &NewValue = StructValueState[{&*NewArg, I}];
1251 NewValue = StructValueState[{&*OldArg, I}];
1252 }
1253 } else {
1254 ValueLatticeElement &NewValue = ValueState[&*NewArg];
1255 NewValue = ValueState[&*OldArg];
1256 }
1257 }
1258 }
1259}
1260
1261void SCCPInstVisitor::visitInstruction(Instruction &I) {
1262 // All the instructions we don't do any special handling for just
1263 // go to overdefined.
1264 LLVM_DEBUG(dbgs() << "SCCP: Don't know how to handle: " << I << '\n');
1265 markOverdefined(V: &I);
1266}
1267
1268bool SCCPInstVisitor::mergeInValue(ValueLatticeElement &IV, Value *V,
1269 const ValueLatticeElement &MergeWithV,
1270 ValueLatticeElement::MergeOptions Opts) {
1271 if (IV.mergeIn(RHS: MergeWithV, Opts)) {
1272 pushUsersToWorkList(V);
1273 LLVM_DEBUG(dbgs() << "Merged " << MergeWithV << " into " << *V << " : "
1274 << IV << "\n");
1275 return true;
1276 }
1277 return false;
1278}
1279
1280bool SCCPInstVisitor::markEdgeExecutable(BasicBlock *Source, BasicBlock *Dest) {
1281 if (!KnownFeasibleEdges.insert(V: Edge(Source, Dest)).second)
1282 return false; // This edge is already known to be executable!
1283
1284 if (!markBlockExecutable(BB: Dest)) {
1285 // If the destination is already executable, we just made an *edge*
1286 // feasible that wasn't before. Revisit the PHI nodes in the block
1287 // because they have potentially new operands.
1288 LLVM_DEBUG(dbgs() << "Marking Edge Executable: " << Source->getName()
1289 << " -> " << Dest->getName() << '\n');
1290
1291 for (PHINode &PN : Dest->phis())
1292 pushToWorkList(I: &PN);
1293 }
1294 return true;
1295}
1296
1297// getFeasibleSuccessors - Return a vector of booleans to indicate which
1298// successors are reachable from a given terminator instruction.
1299void SCCPInstVisitor::getFeasibleSuccessors(Instruction &TI,
1300 SmallVectorImpl<bool> &Succs) {
1301 Succs.resize(N: TI.getNumSuccessors());
1302 if (isa<UncondBrInst>(Val: TI)) {
1303 Succs[0] = true;
1304 return;
1305 }
1306
1307 if (auto *BI = dyn_cast<CondBrInst>(Val: &TI)) {
1308 const ValueLatticeElement &BCValue = getValueState(V: BI->getCondition());
1309 ConstantInt *CI = getConstantInt(IV: BCValue, Ty: BI->getCondition()->getType());
1310 if (!CI) {
1311 // Overdefined condition variables, and branches on unfoldable constant
1312 // conditions, mean the branch could go either way.
1313 if (!BCValue.isUnknownOrUndef())
1314 Succs[0] = Succs[1] = true;
1315 return;
1316 }
1317
1318 // Constant condition variables mean the branch can only go a single way.
1319 Succs[CI->isZero()] = true;
1320 return;
1321 }
1322
1323 // We cannot analyze special terminators, so consider all successors
1324 // executable.
1325 if (TI.isSpecialTerminator()) {
1326 Succs.assign(NumElts: TI.getNumSuccessors(), Elt: true);
1327 return;
1328 }
1329
1330 if (auto *SI = dyn_cast<SwitchInst>(Val: &TI)) {
1331 if (!SI->getNumCases()) {
1332 Succs[0] = true;
1333 return;
1334 }
1335 const ValueLatticeElement &SCValue = getValueState(V: SI->getCondition());
1336 if (ConstantInt *CI =
1337 getConstantInt(IV: SCValue, Ty: SI->getCondition()->getType())) {
1338 Succs[SI->findCaseValue(C: CI)->getSuccessorIndex()] = true;
1339 return;
1340 }
1341
1342 // TODO: Switch on undef is UB. Stop passing false once the rest of LLVM
1343 // is ready.
1344 if (SCValue.isConstantRange(/*UndefAllowed=*/false)) {
1345 const ConstantRange &Range = SCValue.getConstantRange();
1346 unsigned ReachableCaseCount = 0;
1347 for (const auto &Case : SI->cases()) {
1348 const APInt &CaseValue = Case.getCaseValue()->getValue();
1349 if (Range.contains(Val: CaseValue)) {
1350 Succs[Case.getSuccessorIndex()] = true;
1351 ++ReachableCaseCount;
1352 }
1353 }
1354
1355 Succs[SI->case_default()->getSuccessorIndex()] =
1356 Range.isSizeLargerThan(MaxSize: ReachableCaseCount);
1357 return;
1358 }
1359
1360 // Overdefined or unknown condition? All destinations are executable!
1361 if (!SCValue.isUnknownOrUndef())
1362 Succs.assign(NumElts: TI.getNumSuccessors(), Elt: true);
1363 return;
1364 }
1365
1366 // In case of indirect branch and its address is a blockaddress, we mark
1367 // the target as executable.
1368 if (auto *IBR = dyn_cast<IndirectBrInst>(Val: &TI)) {
1369 // Casts are folded by visitCastInst.
1370 const ValueLatticeElement &IBRValue = getValueState(V: IBR->getAddress());
1371 BlockAddress *Addr = dyn_cast_or_null<BlockAddress>(
1372 Val: getConstant(LV: IBRValue, Ty: IBR->getAddress()->getType()));
1373 if (!Addr) { // Overdefined or unknown condition?
1374 // All destinations are executable!
1375 if (!IBRValue.isUnknownOrUndef())
1376 Succs.assign(NumElts: TI.getNumSuccessors(), Elt: true);
1377 return;
1378 }
1379
1380 BasicBlock *T = Addr->getBasicBlock();
1381 assert(Addr->getFunction() == T->getParent() &&
1382 "Block address of a different function ?");
1383 for (unsigned i = 0; i < IBR->getNumSuccessors(); ++i) {
1384 // This is the target.
1385 if (IBR->getDestination(i) == T) {
1386 Succs[i] = true;
1387 return;
1388 }
1389 }
1390
1391 // If we didn't find our destination in the IBR successor list, then we
1392 // have undefined behavior. Its ok to assume no successor is executable.
1393 return;
1394 }
1395
1396 LLVM_DEBUG(dbgs() << "Unknown terminator instruction: " << TI << '\n');
1397 llvm_unreachable("SCCP: Don't know how to handle this terminator!");
1398}
1399
1400// isEdgeFeasible - Return true if the control flow edge from the 'From' basic
1401// block to the 'To' basic block is currently feasible.
1402bool SCCPInstVisitor::isEdgeFeasible(BasicBlock *From, BasicBlock *To) const {
1403 // Check if we've called markEdgeExecutable on the edge yet. (We could
1404 // be more aggressive and try to consider edges which haven't been marked
1405 // yet, but there isn't any need.)
1406 return KnownFeasibleEdges.count(V: Edge(From, To));
1407}
1408
1409// visit Implementations - Something changed in this instruction, either an
1410// operand made a transition, or the instruction is newly executable. Change
1411// the value type of I to reflect these changes if appropriate. This method
1412// makes sure to do the following actions:
1413//
1414// 1. If a phi node merges two constants in, and has conflicting value coming
1415// from different branches, or if the PHI node merges in an overdefined
1416// value, then the PHI node becomes overdefined.
1417// 2. If a phi node merges only constants in, and they all agree on value, the
1418// PHI node becomes a constant value equal to that.
1419// 3. If V <- x (op) y && isConstant(x) && isConstant(y) V = Constant
1420// 4. If V <- x (op) y && (isOverdefined(x) || isOverdefined(y)) V = Overdefined
1421// 5. If V <- MEM or V <- CALL or V <- (unknown) then V = Overdefined
1422// 6. If a conditional branch has a value that is constant, make the selected
1423// destination executable
1424// 7. If a conditional branch has a value that is overdefined, make all
1425// successors executable.
1426void SCCPInstVisitor::visitPHINode(PHINode &PN) {
1427 // Super-extra-high-degree PHI nodes are unlikely to ever be marked constant,
1428 // and slow us down a lot. Just mark them overdefined.
1429 if (PN.getNumIncomingValues() > 64)
1430 return (void)markOverdefined(V: &PN);
1431
1432 if (isInstFullyOverDefined(Inst&: PN))
1433 return;
1434 SmallVector<unsigned> FeasibleIncomingIndices;
1435 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) {
1436 if (!isEdgeFeasible(From: PN.getIncomingBlock(i), To: PN.getParent()))
1437 continue;
1438 FeasibleIncomingIndices.push_back(Elt: i);
1439 }
1440
1441 // Look at all of the executable operands of the PHI node. If any of them
1442 // are overdefined, the PHI becomes overdefined as well. If they are all
1443 // constant, and they agree with each other, the PHI becomes the identical
1444 // constant. If they are constant and don't agree, the PHI is a constant
1445 // range. If there are no executable operands, the PHI remains unknown.
1446 if (StructType *STy = dyn_cast<StructType>(Val: PN.getType())) {
1447 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
1448 ValueLatticeElement PhiState = getStructValueState(V: &PN, i);
1449 if (PhiState.isOverdefined())
1450 continue;
1451 for (unsigned j : FeasibleIncomingIndices) {
1452 const ValueLatticeElement &IV =
1453 getStructValueState(V: PN.getIncomingValue(i: j), i);
1454 PhiState.mergeIn(RHS: IV);
1455 if (PhiState.isOverdefined())
1456 break;
1457 }
1458 ValueLatticeElement &PhiStateRef = getStructValueState(V: &PN, i);
1459 mergeInValue(IV&: PhiStateRef, V: &PN, MergeWithV: PhiState,
1460 Opts: ValueLatticeElement::MergeOptions().setMaxWidenSteps(
1461 FeasibleIncomingIndices.size() + 1));
1462 PhiStateRef.setNumRangeExtensions(
1463 std::max(a: (unsigned)FeasibleIncomingIndices.size(),
1464 b: PhiStateRef.getNumRangeExtensions()));
1465 }
1466 } else {
1467 ValueLatticeElement PhiState = getValueState(V: &PN);
1468 for (unsigned i : FeasibleIncomingIndices) {
1469 const ValueLatticeElement &IV = getValueState(V: PN.getIncomingValue(i));
1470 PhiState.mergeIn(RHS: IV);
1471 if (PhiState.isOverdefined())
1472 break;
1473 }
1474 // We allow up to 1 range extension per active incoming value and one
1475 // additional extension. Note that we manually adjust the number of range
1476 // extensions to match the number of active incoming values. This helps to
1477 // limit multiple extensions caused by the same incoming value, if other
1478 // incoming values are equal.
1479 ValueLatticeElement &PhiStateRef = ValueState[&PN];
1480 mergeInValue(IV&: PhiStateRef, V: &PN, MergeWithV: PhiState,
1481 Opts: ValueLatticeElement::MergeOptions().setMaxWidenSteps(
1482 FeasibleIncomingIndices.size() + 1));
1483 PhiStateRef.setNumRangeExtensions(
1484 std::max(a: (unsigned)FeasibleIncomingIndices.size(),
1485 b: PhiStateRef.getNumRangeExtensions()));
1486 }
1487}
1488
1489void SCCPInstVisitor::visitReturnInst(ReturnInst &I) {
1490 if (I.getNumOperands() == 0)
1491 return; // ret void
1492
1493 Function *F = I.getParent()->getParent();
1494 Value *ResultOp = I.getOperand(i_nocapture: 0);
1495
1496 // If we are tracking the return value of this function, merge it in.
1497 if (!TrackedRetVals.empty() && !ResultOp->getType()->isStructTy()) {
1498 auto TFRVI = TrackedRetVals.find(Key: F);
1499 if (TFRVI != TrackedRetVals.end()) {
1500 mergeInValue(IV&: TFRVI->second, V: F, MergeWithV: getValueState(V: ResultOp));
1501 return;
1502 }
1503 }
1504
1505 // Handle functions that return multiple values.
1506 if (!TrackedMultipleRetVals.empty()) {
1507 if (auto *STy = dyn_cast<StructType>(Val: ResultOp->getType()))
1508 if (MRVFunctionsTracked.count(Ptr: F))
1509 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
1510 mergeInValue(IV&: TrackedMultipleRetVals[std::make_pair(x&: F, y&: i)], V: F,
1511 MergeWithV: getStructValueState(V: ResultOp, i));
1512 }
1513}
1514
1515void SCCPInstVisitor::visitTerminator(Instruction &TI) {
1516 SmallVector<bool, 16> SuccFeasible;
1517 getFeasibleSuccessors(TI, Succs&: SuccFeasible);
1518
1519 BasicBlock *BB = TI.getParent();
1520
1521 // Mark all feasible successors executable.
1522 for (unsigned i = 0, e = SuccFeasible.size(); i != e; ++i)
1523 if (SuccFeasible[i])
1524 markEdgeExecutable(Source: BB, Dest: TI.getSuccessor(Idx: i));
1525}
1526
1527void SCCPInstVisitor::visitCastInst(CastInst &I) {
1528 // ResolvedUndefsIn might mark I as overdefined. Bail out, even if we would
1529 // discover a concrete value later.
1530 if (ValueState[&I].isOverdefined())
1531 return;
1532
1533 if (auto *BC = dyn_cast<BitCastInst>(Val: &I)) {
1534 if (BC->getType() == BC->getOperand(i_nocapture: 0)->getType()) {
1535 if (const PredicateBase *PI = getPredicateInfoFor(I: &I)) {
1536 handlePredicate(I: &I, CopyOf: I.getOperand(i_nocapture: 0), PI);
1537 return;
1538 }
1539 }
1540 }
1541
1542 const ValueLatticeElement &OpSt = getValueState(V: I.getOperand(i_nocapture: 0));
1543 if (OpSt.isUnknownOrUndef())
1544 return;
1545
1546 if (Constant *OpC = getConstant(LV: OpSt, Ty: I.getOperand(i_nocapture: 0)->getType())) {
1547 // Fold the constant as we build.
1548 if (Constant *C =
1549 ConstantFoldCastOperand(Opcode: I.getOpcode(), C: OpC, DestTy: I.getType(), DL)) {
1550 auto &LV = ValueState[&I];
1551 mergeInValue(IV&: LV, V: &I, MergeWithV: ValueLatticeElement::get(C));
1552 return;
1553 }
1554 }
1555
1556 // Ignore bitcasts, as they may change the number of vector elements.
1557 if (I.getDestTy()->isIntOrIntVectorTy() &&
1558 I.getSrcTy()->isIntOrIntVectorTy() &&
1559 I.getOpcode() != Instruction::BitCast) {
1560 ConstantRange OpRange =
1561 OpSt.asConstantRange(Ty: I.getSrcTy(), /*UndefAllowed=*/false);
1562 auto &LV = getValueState(V: &I);
1563
1564 Type *DestTy = I.getDestTy();
1565 ConstantRange Res = ConstantRange::getEmpty(BitWidth: DestTy->getScalarSizeInBits());
1566 if (auto *Trunc = dyn_cast<TruncInst>(Val: &I))
1567 Res = OpRange.truncate(BitWidth: DestTy->getScalarSizeInBits(),
1568 NoWrapKind: Trunc->getNoWrapKind());
1569 else
1570 Res = OpRange.castOp(CastOp: I.getOpcode(), BitWidth: DestTy->getScalarSizeInBits());
1571 mergeInValue(IV&: LV, V: &I, MergeWithV: ValueLatticeElement::getRange(CR: Res));
1572 } else
1573 markOverdefined(V: &I);
1574}
1575
1576void SCCPInstVisitor::handleExtractOfWithOverflow(ExtractValueInst &EVI,
1577 const WithOverflowInst *WO,
1578 unsigned Idx) {
1579 Value *LHS = WO->getLHS(), *RHS = WO->getRHS();
1580 Type *Ty = LHS->getType();
1581
1582 addAdditionalUser(V: LHS, U: &EVI);
1583 addAdditionalUser(V: RHS, U: &EVI);
1584
1585 const ValueLatticeElement &L = getValueState(V: LHS);
1586 if (L.isUnknownOrUndef())
1587 return; // Wait to resolve.
1588 ConstantRange LR = L.asConstantRange(Ty, /*UndefAllowed=*/false);
1589
1590 const ValueLatticeElement &R = getValueState(V: RHS);
1591 if (R.isUnknownOrUndef())
1592 return; // Wait to resolve.
1593
1594 ConstantRange RR = R.asConstantRange(Ty, /*UndefAllowed=*/false);
1595 if (Idx == 0) {
1596 ConstantRange Res = LR.binaryOp(BinOp: WO->getBinaryOp(), Other: RR);
1597 mergeInValue(IV&: ValueState[&EVI], V: &EVI, MergeWithV: ValueLatticeElement::getRange(CR: Res));
1598 } else {
1599 assert(Idx == 1 && "Index can only be 0 or 1");
1600 ConstantRange NWRegion = ConstantRange::makeGuaranteedNoWrapRegion(
1601 BinOp: WO->getBinaryOp(), Other: RR, NoWrapKind: WO->getNoWrapKind());
1602 if (NWRegion.contains(CR: LR))
1603 return (void)markConstant(V: &EVI, C: ConstantInt::getFalse(Ty: EVI.getType()));
1604 markOverdefined(V: &EVI);
1605 }
1606}
1607
1608void SCCPInstVisitor::visitExtractValueInst(ExtractValueInst &EVI) {
1609 // If this returns a struct, mark all elements over defined, we don't track
1610 // structs in structs.
1611 if (EVI.getType()->isStructTy())
1612 return (void)markOverdefined(V: &EVI);
1613
1614 // resolvedUndefsIn might mark I as overdefined. Bail out, even if we would
1615 // discover a concrete value later.
1616 if (ValueState[&EVI].isOverdefined())
1617 return (void)markOverdefined(V: &EVI);
1618
1619 // If this is extracting from more than one level of struct, we don't know.
1620 if (EVI.getNumIndices() != 1)
1621 return (void)markOverdefined(V: &EVI);
1622
1623 Value *AggVal = EVI.getAggregateOperand();
1624 if (AggVal->getType()->isStructTy()) {
1625 unsigned i = *EVI.idx_begin();
1626 if (auto *WO = dyn_cast<WithOverflowInst>(Val: AggVal))
1627 return handleExtractOfWithOverflow(EVI, WO, Idx: i);
1628 ValueLatticeElement EltVal = getStructValueState(V: AggVal, i);
1629 mergeInValue(IV&: ValueState[&EVI], V: &EVI, MergeWithV: EltVal);
1630 } else {
1631 // Otherwise, must be extracting from an array.
1632 return (void)markOverdefined(V: &EVI);
1633 }
1634}
1635
1636void SCCPInstVisitor::visitInsertValueInst(InsertValueInst &IVI) {
1637 auto *STy = dyn_cast<StructType>(Val: IVI.getType());
1638 if (!STy)
1639 return (void)markOverdefined(V: &IVI);
1640
1641 // resolvedUndefsIn might mark I as overdefined. Bail out, even if we would
1642 // discover a concrete value later.
1643 if (ValueState[&IVI].isOverdefined())
1644 return (void)markOverdefined(V: &IVI);
1645
1646 // If this has more than one index, we can't handle it, drive all results to
1647 // undef.
1648 if (IVI.getNumIndices() != 1)
1649 return (void)markOverdefined(V: &IVI);
1650
1651 Value *Aggr = IVI.getAggregateOperand();
1652 unsigned Idx = *IVI.idx_begin();
1653
1654 // Compute the result based on what we're inserting.
1655 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
1656 // This passes through all values that aren't the inserted element.
1657 if (i != Idx) {
1658 ValueLatticeElement EltVal = getStructValueState(V: Aggr, i);
1659 mergeInValue(IV&: getStructValueState(V: &IVI, i), V: &IVI, MergeWithV: EltVal);
1660 continue;
1661 }
1662
1663 Value *Val = IVI.getInsertedValueOperand();
1664 if (Val->getType()->isStructTy())
1665 // We don't track structs in structs.
1666 markOverdefined(IV&: getStructValueState(V: &IVI, i), V: &IVI);
1667 else {
1668 ValueLatticeElement InVal = getValueState(V: Val);
1669 mergeInValue(IV&: getStructValueState(V: &IVI, i), V: &IVI, MergeWithV: InVal);
1670 }
1671 }
1672}
1673
1674void SCCPInstVisitor::visitSelectInst(SelectInst &I) {
1675 // If this select returns a struct, just mark the result overdefined.
1676 // TODO: We could do a lot better than this if code actually uses this.
1677 if (I.getType()->isStructTy())
1678 return (void)markOverdefined(V: &I);
1679
1680 // resolvedUndefsIn might mark I as overdefined. Bail out, even if we would
1681 // discover a concrete value later.
1682 if (ValueState[&I].isOverdefined())
1683 return (void)markOverdefined(V: &I);
1684
1685 const ValueLatticeElement &CondValue = getValueState(V: I.getCondition());
1686 if (CondValue.isUnknownOrUndef())
1687 return;
1688
1689 if (ConstantInt *CondCB =
1690 getConstantInt(IV: CondValue, Ty: I.getCondition()->getType())) {
1691 Value *OpVal = CondCB->isZero() ? I.getFalseValue() : I.getTrueValue();
1692 const ValueLatticeElement &OpValState = getValueState(V: OpVal);
1693 // Safety: ValueState[&I] doesn't invalidate OpValState since it is already
1694 // in the map.
1695 assert(ValueState.contains(&I) && "&I is not in ValueState map.");
1696 mergeInValue(IV&: ValueState[&I], V: &I, MergeWithV: OpValState);
1697 return;
1698 }
1699
1700 // Otherwise, the condition is overdefined or a constant we can't evaluate.
1701 // See if we can produce something better than overdefined based on the T/F
1702 // value.
1703 ValueLatticeElement TVal = getValueState(V: I.getTrueValue());
1704 ValueLatticeElement FVal = getValueState(V: I.getFalseValue());
1705
1706 ValueLatticeElement &State = ValueState[&I];
1707 bool Changed = State.mergeIn(RHS: TVal);
1708 Changed |= State.mergeIn(RHS: FVal);
1709 if (Changed)
1710 pushUsersToWorkListMsg(IV&: State, V: &I);
1711}
1712
1713// Handle Unary Operators.
1714void SCCPInstVisitor::visitUnaryOperator(Instruction &I) {
1715 ValueLatticeElement V0State = getValueState(V: I.getOperand(i: 0));
1716
1717 ValueLatticeElement &IV = ValueState[&I];
1718 // resolvedUndefsIn might mark I as overdefined. Bail out, even if we would
1719 // discover a concrete value later.
1720 if (IV.isOverdefined())
1721 return (void)markOverdefined(V: &I);
1722
1723 // If something is unknown/undef, wait for it to resolve.
1724 if (V0State.isUnknownOrUndef())
1725 return;
1726
1727 if (SCCPSolver::isConstant(LV: V0State))
1728 if (Constant *C = ConstantFoldUnaryOpOperand(
1729 Opcode: I.getOpcode(), Op: getConstant(LV: V0State, Ty: I.getType()), DL))
1730 return (void)markConstant(IV, V: &I, C);
1731
1732 markOverdefined(V: &I);
1733}
1734
1735void SCCPInstVisitor::visitFreezeInst(FreezeInst &I) {
1736 // If this freeze returns a struct, just mark the result overdefined.
1737 // TODO: We could do a lot better than this.
1738 if (I.getType()->isStructTy())
1739 return (void)markOverdefined(V: &I);
1740
1741 ValueLatticeElement V0State = getValueState(V: I.getOperand(i_nocapture: 0));
1742 ValueLatticeElement &IV = ValueState[&I];
1743 // resolvedUndefsIn might mark I as overdefined. Bail out, even if we would
1744 // discover a concrete value later.
1745 if (IV.isOverdefined())
1746 return (void)markOverdefined(V: &I);
1747
1748 // If something is unknown/undef, wait for it to resolve.
1749 if (V0State.isUnknownOrUndef())
1750 return;
1751
1752 if (SCCPSolver::isConstant(LV: V0State) &&
1753 isGuaranteedNotToBeUndefOrPoison(V: getConstant(LV: V0State, Ty: I.getType())))
1754 return (void)markConstant(IV, V: &I, C: getConstant(LV: V0State, Ty: I.getType()));
1755
1756 markOverdefined(V: &I);
1757}
1758
1759// Handle Binary Operators.
1760void SCCPInstVisitor::visitBinaryOperator(Instruction &I) {
1761 ValueLatticeElement V1State = getValueState(V: I.getOperand(i: 0));
1762 ValueLatticeElement V2State = getValueState(V: I.getOperand(i: 1));
1763
1764 ValueLatticeElement &IV = ValueState[&I];
1765 if (IV.isOverdefined())
1766 return;
1767
1768 // If something is undef, wait for it to resolve.
1769 if (V1State.isUnknownOrUndef() || V2State.isUnknownOrUndef())
1770 return;
1771
1772 if (V1State.isOverdefined() && V2State.isOverdefined())
1773 return (void)markOverdefined(V: &I);
1774
1775 // If either of the operands is a constant, try to fold it to a constant.
1776 // TODO: Use information from notconstant better.
1777 if ((V1State.isConstant() || V2State.isConstant())) {
1778 Value *V1 = SCCPSolver::isConstant(LV: V1State)
1779 ? getConstant(LV: V1State, Ty: I.getOperand(i: 0)->getType())
1780 : I.getOperand(i: 0);
1781 Value *V2 = SCCPSolver::isConstant(LV: V2State)
1782 ? getConstant(LV: V2State, Ty: I.getOperand(i: 1)->getType())
1783 : I.getOperand(i: 1);
1784 Value *R = simplifyBinOp(Opcode: I.getOpcode(), LHS: V1, RHS: V2, Q: SimplifyQuery(DL, &I));
1785 auto *C = dyn_cast_or_null<Constant>(Val: R);
1786 if (C) {
1787 // Conservatively assume that the result may be based on operands that may
1788 // be undef. Note that we use mergeInValue to combine the constant with
1789 // the existing lattice value for I, as different constants might be found
1790 // after one of the operands go to overdefined, e.g. due to one operand
1791 // being a special floating value.
1792 ValueLatticeElement NewV;
1793 NewV.markConstant(V: C, /*MayIncludeUndef=*/true);
1794 return (void)mergeInValue(IV&: ValueState[&I], V: &I, MergeWithV: NewV);
1795 }
1796 }
1797
1798 // Only use ranges for binary operators on integers.
1799 if (!I.getType()->isIntOrIntVectorTy())
1800 return markOverdefined(V: &I);
1801
1802 // Try to simplify to a constant range.
1803 ConstantRange A =
1804 V1State.asConstantRange(Ty: I.getType(), /*UndefAllowed=*/false);
1805 ConstantRange B =
1806 V2State.asConstantRange(Ty: I.getType(), /*UndefAllowed=*/false);
1807
1808 auto *BO = cast<BinaryOperator>(Val: &I);
1809 ConstantRange R = ConstantRange::getEmpty(BitWidth: I.getType()->getScalarSizeInBits());
1810 if (auto *OBO = dyn_cast<OverflowingBinaryOperator>(Val: BO))
1811 R = A.overflowingBinaryOp(BinOp: BO->getOpcode(), Other: B, NoWrapKind: OBO->getNoWrapKind());
1812 else
1813 R = A.binaryOp(BinOp: BO->getOpcode(), Other: B);
1814 mergeInValue(IV&: ValueState[&I], V: &I, MergeWithV: ValueLatticeElement::getRange(CR: R));
1815
1816 // TODO: Currently we do not exploit special values that produce something
1817 // better than overdefined with an overdefined operand for vector or floating
1818 // point types, like and <4 x i32> overdefined, zeroinitializer.
1819}
1820
1821// Handle ICmpInst instruction.
1822void SCCPInstVisitor::visitCmpInst(CmpInst &I) {
1823 // Do not cache this lookup, getValueState calls later in the function might
1824 // invalidate the reference.
1825 if (ValueState[&I].isOverdefined())
1826 return (void)markOverdefined(V: &I);
1827
1828 Value *Op1 = I.getOperand(i_nocapture: 0);
1829 Value *Op2 = I.getOperand(i_nocapture: 1);
1830
1831 // For parameters, use ParamState which includes constant range info if
1832 // available.
1833 auto V1State = getValueState(V: Op1);
1834 auto V2State = getValueState(V: Op2);
1835
1836 Constant *C = V1State.getCompare(Pred: I.getPredicate(), Ty: I.getType(), Other: V2State, DL);
1837 if (C) {
1838 ValueLatticeElement CV;
1839 CV.markConstant(V: C);
1840 mergeInValue(IV&: ValueState[&I], V: &I, MergeWithV: CV);
1841 return;
1842 }
1843
1844 // If operands are still unknown, wait for it to resolve.
1845 if ((V1State.isUnknownOrUndef() || V2State.isUnknownOrUndef()) &&
1846 !SCCPSolver::isConstant(LV: ValueState[&I]))
1847 return;
1848
1849 markOverdefined(V: &I);
1850}
1851
1852// Handle getelementptr instructions. If all operands are constants then we
1853// can turn this into a getelementptr ConstantExpr.
1854void SCCPInstVisitor::visitGetElementPtrInst(GetElementPtrInst &I) {
1855 if (ValueState[&I].isOverdefined())
1856 return (void)markOverdefined(V: &I);
1857
1858 const ValueLatticeElement &PtrState = getValueState(V: I.getPointerOperand());
1859 if (PtrState.isUnknownOrUndef())
1860 return;
1861
1862 // gep inbounds/nuw of non-null is non-null.
1863 if (PtrState.isNotConstant() && PtrState.getNotConstant()->isNullValue()) {
1864 if (I.hasNoUnsignedWrap() ||
1865 (I.isInBounds() &&
1866 !NullPointerIsDefined(F: I.getFunction(), AS: I.getAddressSpace())))
1867 return (void)markNotNull(IV&: ValueState[&I], V: &I);
1868 return (void)markOverdefined(V: &I);
1869 }
1870
1871 SmallVector<Constant *, 8> Operands;
1872 Operands.reserve(N: I.getNumOperands());
1873 bool PtrMayHaveDifferentProvenance = PtrState.mayHaveDifferentProvenance();
1874
1875 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) {
1876 const ValueLatticeElement &State = getValueState(V: I.getOperand(i_nocapture: i));
1877 if (State.isUnknownOrUndef())
1878 return; // Operands are not resolved yet.
1879
1880 if (Constant *C = getConstant(LV: State, Ty: I.getOperand(i_nocapture: i)->getType())) {
1881 Operands.push_back(Elt: C);
1882 continue;
1883 }
1884
1885 return (void)markOverdefined(V: &I);
1886 }
1887
1888 if (Constant *C = ConstantFoldInstOperands(I: &I, Ops: Operands, DL)) {
1889 mergeInValue(IV&: ValueState[&I], V: &I, MergeWithV: ValueLatticeElement::get(C));
1890 // The pointer operand's lattice has found to be a constant, however, the
1891 // returned pointer of the GEP may not be freely substituted, as it may have
1892 // been derived from a pointer with potentially different provenance.
1893 if (PtrMayHaveDifferentProvenance)
1894 ValueState[&I].setMayHaveDifferentProvenance(true);
1895 } else
1896 markOverdefined(V: &I);
1897}
1898
1899void SCCPInstVisitor::visitAllocaInst(AllocaInst &I) {
1900 if (!NullPointerIsDefined(F: I.getFunction(), AS: I.getAddressSpace()))
1901 return (void)markNotNull(IV&: ValueState[&I], V: &I);
1902
1903 markOverdefined(V: &I);
1904}
1905
1906void SCCPInstVisitor::visitStoreInst(StoreInst &SI) {
1907 // If this store is of a struct, ignore it.
1908 if (SI.getOperand(i_nocapture: 0)->getType()->isStructTy())
1909 return;
1910
1911 if (TrackedGlobals.empty() || !isa<GlobalVariable>(Val: SI.getOperand(i_nocapture: 1)))
1912 return;
1913
1914 GlobalVariable *GV = cast<GlobalVariable>(Val: SI.getOperand(i_nocapture: 1));
1915 auto I = TrackedGlobals.find(Val: GV);
1916 if (I == TrackedGlobals.end())
1917 return;
1918
1919 // Get the value we are storing into the global, then merge it.
1920 mergeInValue(IV&: I->second, V: GV, MergeWithV: getValueState(V: SI.getOperand(i_nocapture: 0)),
1921 Opts: ValueLatticeElement::MergeOptions().setCheckWiden(false));
1922 if (I->second.isOverdefined())
1923 TrackedGlobals.erase(I); // No need to keep tracking this!
1924}
1925
1926static ValueLatticeElement getValueFromMetadata(const Instruction *I) {
1927 if (const auto *CB = dyn_cast<CallBase>(Val: I)) {
1928 if (CB->getType()->isIntOrIntVectorTy())
1929 if (std::optional<ConstantRange> Range = CB->getRange())
1930 return ValueLatticeElement::getRange(CR: *Range);
1931 if (CB->getType()->isPointerTy() && CB->isReturnNonNull())
1932 return ValueLatticeElement::getNot(
1933 C: ConstantPointerNull::get(T: cast<PointerType>(Val: I->getType())));
1934 }
1935
1936 if (I->getType()->isIntOrIntVectorTy())
1937 if (MDNode *Ranges = I->getMetadata(KindID: LLVMContext::MD_range))
1938 return ValueLatticeElement::getRange(
1939 CR: getConstantRangeFromMetadata(RangeMD: *Ranges));
1940 if (I->hasMetadata(KindID: LLVMContext::MD_nonnull))
1941 return ValueLatticeElement::getNot(
1942 C: ConstantPointerNull::get(T: cast<PointerType>(Val: I->getType())));
1943
1944 return ValueLatticeElement::getOverdefined();
1945}
1946
1947// Handle load instructions. If the operand is a constant pointer to a constant
1948// global, we can replace the load with the loaded constant value!
1949void SCCPInstVisitor::visitLoadInst(LoadInst &I) {
1950 // If this load is of a struct or the load is volatile, just mark the result
1951 // as overdefined.
1952 if (I.getType()->isStructTy() || I.isVolatile())
1953 return (void)markOverdefined(V: &I);
1954
1955 // resolvedUndefsIn might mark I as overdefined. Bail out, even if we would
1956 // discover a concrete value later.
1957 if (ValueState[&I].isOverdefined())
1958 return (void)markOverdefined(V: &I);
1959
1960 const ValueLatticeElement &PtrVal = getValueState(V: I.getOperand(i_nocapture: 0));
1961 if (PtrVal.isUnknownOrUndef())
1962 return; // The pointer is not resolved yet!
1963
1964 if (SCCPSolver::isConstant(LV: PtrVal)) {
1965 Constant *Ptr = getConstant(LV: PtrVal, Ty: I.getOperand(i_nocapture: 0)->getType());
1966 ValueLatticeElement &IV = ValueState[&I];
1967
1968 // load null is undefined.
1969 if (isa<ConstantPointerNull>(Val: Ptr)) {
1970 if (NullPointerIsDefined(F: I.getFunction(), AS: I.getPointerAddressSpace()))
1971 return (void)markOverdefined(IV, V: &I);
1972 else
1973 return;
1974 }
1975
1976 // Transform load (constant global) into the value loaded.
1977 if (auto *GV = dyn_cast<GlobalVariable>(Val: Ptr)) {
1978 if (!TrackedGlobals.empty()) {
1979 // If we are tracking this global, merge in the known value for it.
1980 auto It = TrackedGlobals.find(Val: GV);
1981 if (It != TrackedGlobals.end()) {
1982 mergeInValue(IV, V: &I, MergeWithV: It->second, Opts: getMaxWidenStepsOpts());
1983 return;
1984 }
1985 }
1986 }
1987
1988 // Transform load from a constant into a constant if possible.
1989 if (Constant *C = ConstantFoldLoadFromConstPtr(C: Ptr, Ty: I.getType(), DL))
1990 return (void)markConstant(IV, V: &I, C);
1991 }
1992
1993 // Fall back to metadata.
1994 mergeInValue(IV&: ValueState[&I], V: &I, MergeWithV: getValueFromMetadata(I: &I));
1995}
1996
1997void SCCPInstVisitor::visitCallBase(CallBase &CB) {
1998 handleCallResult(CB);
1999 handleCallArguments(CB);
2000}
2001
2002void SCCPInstVisitor::handleCallOverdefined(CallBase &CB) {
2003 Function *F = CB.getCalledFunction();
2004
2005 // Void return and not tracking callee, just bail.
2006 if (CB.getType()->isVoidTy())
2007 return;
2008
2009 // Always mark struct return as overdefined.
2010 if (CB.getType()->isStructTy())
2011 return (void)markOverdefined(V: &CB);
2012
2013 // Otherwise, if we have a single return value case, and if the function is
2014 // a declaration, maybe we can constant fold it.
2015 if (F && F->isDeclaration() && canConstantFoldCallTo(Call: &CB, F)) {
2016 SmallVector<Constant *, 8> Operands;
2017 for (const Use &A : CB.args()) {
2018 if (A.get()->getType()->isStructTy())
2019 return markOverdefined(V: &CB); // Can't handle struct args.
2020 if (A.get()->getType()->isMetadataTy())
2021 continue; // Carried in CB, not allowed in Operands.
2022 const ValueLatticeElement &State = getValueState(V: A);
2023
2024 if (State.isUnknownOrUndef())
2025 return; // Operands are not resolved yet.
2026 if (SCCPSolver::isOverdefined(LV: State))
2027 return (void)markOverdefined(V: &CB);
2028 assert(SCCPSolver::isConstant(State) && "Unknown state!");
2029 Operands.push_back(Elt: getConstant(LV: State, Ty: A->getType()));
2030 }
2031
2032 if (SCCPSolver::isOverdefined(LV: getValueState(V: &CB)))
2033 return (void)markOverdefined(V: &CB);
2034
2035 // If we can constant fold this, mark the result of the call as a
2036 // constant.
2037 if (Constant *C = ConstantFoldCall(Call: &CB, F, Operands, TLI: &GetTLI(*F))) {
2038 mergeInValue(IV&: ValueState[&CB], V: &CB, MergeWithV: ValueLatticeElement::get(C));
2039 return;
2040 }
2041 }
2042
2043 // Fall back to metadata.
2044 mergeInValue(IV&: ValueState[&CB], V: &CB, MergeWithV: getValueFromMetadata(I: &CB));
2045}
2046
2047void SCCPInstVisitor::handleCallArguments(CallBase &CB) {
2048 Function *F = CB.getCalledFunction();
2049 // If this is a local function that doesn't have its address taken, mark its
2050 // entry block executable and merge in the actual arguments to the call into
2051 // the formal arguments of the function.
2052 if (TrackingIncomingArguments.count(Ptr: F)) {
2053 markBlockExecutable(BB: &F->front());
2054
2055 // Propagate information from this call site into the callee.
2056 auto CAI = CB.arg_begin();
2057 for (Function::arg_iterator AI = F->arg_begin(), E = F->arg_end(); AI != E;
2058 ++AI, ++CAI) {
2059 // If this argument is byval, and if the function is not readonly, there
2060 // will be an implicit copy formed of the input aggregate.
2061 if (AI->hasByValAttr() && !F->onlyReadsMemory()) {
2062 markOverdefined(V: &*AI);
2063 continue;
2064 }
2065
2066 if (auto *STy = dyn_cast<StructType>(Val: AI->getType())) {
2067 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
2068 ValueLatticeElement CallArg = getStructValueState(V: *CAI, i);
2069 mergeInValue(IV&: getStructValueState(V: &*AI, i), V: &*AI, MergeWithV: CallArg,
2070 Opts: getMaxWidenStepsOpts());
2071 }
2072 } else {
2073 ValueLatticeElement CallArg =
2074 getValueState(V: *CAI).intersect(Other: getArgAttributeVL(A: &*AI));
2075 mergeInValue(IV&: ValueState[&*AI], V: &*AI, MergeWithV: CallArg, Opts: getMaxWidenStepsOpts());
2076 }
2077 }
2078 }
2079}
2080
2081void SCCPInstVisitor::handlePredicate(Instruction *I, Value *CopyOf,
2082 const PredicateBase *PI) {
2083 ValueLatticeElement CopyOfVal = getValueState(V: CopyOf);
2084 const std::optional<PredicateConstraint> &Constraint = PI->getConstraint();
2085 if (!Constraint) {
2086 mergeInValue(IV&: ValueState[I], V: I, MergeWithV: CopyOfVal);
2087 return;
2088 }
2089
2090 CmpInst::Predicate Pred = Constraint->Predicate;
2091 Value *OtherOp = Constraint->OtherOp;
2092
2093 // Wait until OtherOp is resolved.
2094 if (getValueState(V: OtherOp).isUnknown()) {
2095 addAdditionalUser(V: OtherOp, U: I);
2096 return;
2097 }
2098
2099 ValueLatticeElement CondVal = getValueState(V: OtherOp);
2100 ValueLatticeElement &IV = ValueState[I];
2101 if (CondVal.isConstantRange() || CopyOfVal.isConstantRange()) {
2102 auto ImposedCR =
2103 ConstantRange::getFull(BitWidth: DL.getTypeSizeInBits(Ty: CopyOf->getType()));
2104
2105 // Get the range imposed by the condition.
2106 if (CondVal.isConstantRange())
2107 ImposedCR = ConstantRange::makeAllowedICmpRegion(
2108 Pred, Other: CondVal.getConstantRange());
2109
2110 // Combine range info for the original value with the new range from the
2111 // condition.
2112 auto CopyOfCR = CopyOfVal.asConstantRange(Ty: CopyOf->getType(),
2113 /*UndefAllowed=*/true);
2114 // Treat an unresolved input like a full range.
2115 if (CopyOfCR.isEmptySet())
2116 CopyOfCR = ConstantRange::getFull(BitWidth: CopyOfCR.getBitWidth());
2117 auto NewCR = ImposedCR.intersectWith(CR: CopyOfCR);
2118 // If the existing information is != x, do not use the information from
2119 // a chained predicate, as the != x information is more likely to be
2120 // helpful in practice.
2121 if (!CopyOfCR.contains(CR: NewCR) && CopyOfCR.getSingleMissingElement())
2122 NewCR = std::move(CopyOfCR);
2123
2124 // The new range is based on a branch condition. That guarantees that
2125 // neither of the compare operands can be undef in the branch targets,
2126 // unless we have conditions that are always true/false (e.g. icmp ule
2127 // i32, %a, i32_max). For the latter overdefined/empty range will be
2128 // inferred, but the branch will get folded accordingly anyways.
2129 addAdditionalUser(V: OtherOp, U: I);
2130 mergeInValue(
2131 IV, V: I, MergeWithV: ValueLatticeElement::getRange(CR: NewCR, /*MayIncludeUndef*/ false));
2132 return;
2133 } else if (Pred == CmpInst::ICMP_EQ &&
2134 (CondVal.isConstant() || CondVal.isNotConstant())) {
2135 // For non-integer values or integer constant expressions, only
2136 // propagate equal constants or not-constants.
2137 addAdditionalUser(V: OtherOp, U: I);
2138 if (CopyOf->getType()->isPointerTy())
2139 CondVal.setMayHaveDifferentProvenance(true);
2140 mergeInValue(IV, V: I, MergeWithV: CondVal);
2141 return;
2142 } else if (Pred == CmpInst::ICMP_NE && CondVal.isConstant()) {
2143 // Propagate inequalities.
2144 addAdditionalUser(V: OtherOp, U: I);
2145 mergeInValue(IV, V: I, MergeWithV: ValueLatticeElement::getNot(C: CondVal.getConstant()));
2146 return;
2147 }
2148
2149 return (void)mergeInValue(IV, V: I, MergeWithV: CopyOfVal);
2150}
2151
2152void SCCPInstVisitor::handleCallResult(CallBase &CB) {
2153 Function *F = CB.getCalledFunction();
2154
2155 if (auto *II = dyn_cast<IntrinsicInst>(Val: &CB)) {
2156 if (II->getIntrinsicID() == Intrinsic::vscale) {
2157 unsigned BitWidth = CB.getType()->getScalarSizeInBits();
2158 const ConstantRange Result = getVScaleRange(F: II->getFunction(), BitWidth);
2159 return (void)mergeInValue(IV&: ValueState[II], V: II,
2160 MergeWithV: ValueLatticeElement::getRange(CR: Result));
2161 }
2162 if (II->getIntrinsicID() == Intrinsic::experimental_get_vector_length) {
2163 Value *CountArg = II->getArgOperand(i: 0);
2164 Value *VF = II->getArgOperand(i: 1);
2165 bool Scalable = cast<ConstantInt>(Val: II->getArgOperand(i: 2))->isOne();
2166
2167 // Computation happens in the larger type.
2168 unsigned BitWidth = std::max(a: CountArg->getType()->getScalarSizeInBits(),
2169 b: VF->getType()->getScalarSizeInBits());
2170
2171 ConstantRange Count = getValueState(V: CountArg)
2172 .asConstantRange(Ty: CountArg->getType(), UndefAllowed: false)
2173 .zeroExtend(BitWidth);
2174 ConstantRange MaxLanes = getValueState(V: VF)
2175 .asConstantRange(Ty: VF->getType(), UndefAllowed: false)
2176 .zeroExtend(BitWidth);
2177 if (Scalable)
2178 MaxLanes =
2179 MaxLanes.multiply(Other: getVScaleRange(F: II->getFunction(), BitWidth));
2180
2181 // The result is always less than both Count and MaxLanes.
2182 ConstantRange Result = ConstantRange::getNonEmpty(
2183 Lower: APInt::getZero(numBits: BitWidth),
2184 Upper: APIntOps::umin(A: Count.getUnsignedMax(), B: MaxLanes.getUnsignedMax()) +
2185 1);
2186
2187 // If Count <= MaxLanes, getvectorlength(Count, MaxLanes) = Count
2188 if (Count.icmp(Pred: CmpInst::ICMP_ULE, Other: MaxLanes))
2189 Result = std::move(Count);
2190
2191 Result = Result.truncate(BitWidth: II->getType()->getScalarSizeInBits());
2192 return (void)mergeInValue(IV&: ValueState[II], V: II,
2193 MergeWithV: ValueLatticeElement::getRange(CR: Result));
2194 }
2195
2196 if (ConstantRange::isIntrinsicSupported(IntrinsicID: II->getIntrinsicID())) {
2197 // Compute result range for intrinsics supported by ConstantRange.
2198 // Do this even if we don't know a range for all operands, as we may
2199 // still know something about the result range, e.g. of abs(x).
2200 SmallVector<ConstantRange, 2> OpRanges;
2201 for (Value *Op : II->args()) {
2202 const ValueLatticeElement &State = getValueState(V: Op);
2203 if (State.isUnknownOrUndef())
2204 return;
2205 OpRanges.push_back(
2206 Elt: State.asConstantRange(Ty: Op->getType(), /*UndefAllowed=*/false));
2207 }
2208
2209 ConstantRange Result =
2210 ConstantRange::intrinsic(IntrinsicID: II->getIntrinsicID(), Ops: OpRanges);
2211 return (void)mergeInValue(IV&: ValueState[II], V: II,
2212 MergeWithV: ValueLatticeElement::getRange(CR: Result));
2213 }
2214 }
2215
2216 // The common case is that we aren't tracking the callee, either because we
2217 // are not doing interprocedural analysis or the callee is indirect, or is
2218 // external. Handle these cases first.
2219 if (!F || F->isDeclaration())
2220 return handleCallOverdefined(CB);
2221
2222 // If this is a single/zero retval case, see if we're tracking the function.
2223 if (auto *STy = dyn_cast<StructType>(Val: F->getReturnType())) {
2224 if (!MRVFunctionsTracked.count(Ptr: F))
2225 return handleCallOverdefined(CB); // Not tracking this callee.
2226
2227 // If we are tracking this callee, propagate the result of the function
2228 // into this call site.
2229 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
2230 mergeInValue(IV&: getStructValueState(V: &CB, i), V: &CB,
2231 MergeWithV: TrackedMultipleRetVals[std::make_pair(x&: F, y&: i)],
2232 Opts: getMaxWidenStepsOpts());
2233 } else {
2234 auto TFRVI = TrackedRetVals.find(Key: F);
2235 if (TFRVI == TrackedRetVals.end())
2236 return handleCallOverdefined(CB); // Not tracking this callee.
2237
2238 // If so, propagate the return value of the callee into this call result.
2239 mergeInValue(IV&: ValueState[&CB], V: &CB, MergeWithV: TFRVI->second, Opts: getMaxWidenStepsOpts());
2240 }
2241}
2242
2243bool SCCPInstVisitor::isInstFullyOverDefined(Instruction &Inst) {
2244 // For structure Type, we handle each member separately.
2245 // A structure object won't be considered as overdefined when
2246 // there is at least one member that is not overdefined.
2247 if (StructType *STy = dyn_cast<StructType>(Val: Inst.getType())) {
2248 for (unsigned i = 0, e = STy->getNumElements(); i < e; ++i) {
2249 if (!getStructValueState(V: &Inst, i).isOverdefined())
2250 return false;
2251 }
2252 return true;
2253 }
2254
2255 return getValueState(V: &Inst).isOverdefined();
2256}
2257
2258void SCCPInstVisitor::solve() {
2259 // Process the work lists until they are empty!
2260 while (!BBWorkList.empty() || !InstWorkList.empty()) {
2261 // Process the instruction work list.
2262 while (!InstWorkList.empty()) {
2263 Instruction *I = InstWorkList.pop_back_val();
2264 Invalidated.erase(V: I);
2265
2266 LLVM_DEBUG(dbgs() << "\nPopped off I-WL: " << *I << '\n');
2267
2268 visit(I);
2269 }
2270
2271 // Process the basic block work list.
2272 while (!BBWorkList.empty()) {
2273 BasicBlock *BB = BBWorkList.pop_back_val();
2274 BBVisited.insert(Ptr: BB);
2275
2276 LLVM_DEBUG(dbgs() << "\nPopped off BBWL: " << *BB << '\n');
2277 for (Instruction &I : *BB) {
2278 CurI = &I;
2279 visit(I);
2280 }
2281 CurI = nullptr;
2282 }
2283 }
2284}
2285
2286bool SCCPInstVisitor::resolvedUndef(Instruction &I) {
2287 // Look for instructions which produce undef values.
2288 if (I.getType()->isVoidTy())
2289 return false;
2290
2291 if (auto *STy = dyn_cast<StructType>(Val: I.getType())) {
2292 // Only a few things that can be structs matter for undef.
2293
2294 // Tracked calls must never be marked overdefined in resolvedUndefsIn.
2295 if (auto *CB = dyn_cast<CallBase>(Val: &I))
2296 if (Function *F = CB->getCalledFunction())
2297 if (MRVFunctionsTracked.count(Ptr: F))
2298 return false;
2299
2300 // extractvalue and insertvalue don't need to be marked; they are
2301 // tracked as precisely as their operands.
2302 if (isa<ExtractValueInst>(Val: I) || isa<InsertValueInst>(Val: I))
2303 return false;
2304 // Send the results of everything else to overdefined. We could be
2305 // more precise than this but it isn't worth bothering.
2306 for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
2307 ValueLatticeElement &LV = getStructValueState(V: &I, i);
2308 if (LV.isUnknown()) {
2309 markOverdefined(IV&: LV, V: &I);
2310 return true;
2311 }
2312 }
2313 return false;
2314 }
2315
2316 ValueLatticeElement &LV = getValueState(V: &I);
2317 if (!LV.isUnknown())
2318 return false;
2319
2320 // There are two reasons a call can have an undef result
2321 // 1. It could be tracked.
2322 // 2. It could be constant-foldable.
2323 // Because of the way we solve return values, tracked calls must
2324 // never be marked overdefined in resolvedUndefsIn.
2325 if (auto *CB = dyn_cast<CallBase>(Val: &I))
2326 if (Function *F = CB->getCalledFunction())
2327 if (TrackedRetVals.count(Key: F))
2328 return false;
2329
2330 if (isa<LoadInst>(Val: I)) {
2331 // A load here means one of two things: a load of undef from a global,
2332 // a load from an unknown pointer. Either way, having it return undef
2333 // is okay.
2334 return false;
2335 }
2336
2337 markOverdefined(V: &I);
2338 return true;
2339}
2340
2341/// While solving the dataflow for a function, we don't compute a result for
2342/// operations with an undef operand, to allow undef to be lowered to a
2343/// constant later. For example, constant folding of "zext i8 undef to i16"
2344/// would result in "i16 0", and if undef is later lowered to "i8 1", then the
2345/// zext result would become "i16 1" and would result into an overdefined
2346/// lattice value once merged with the previous result. Not computing the
2347/// result of the zext (treating undef the same as unknown) allows us to handle
2348/// a later undef->constant lowering more optimally.
2349///
2350/// However, if the operand remains undef when the solver returns, we do need
2351/// to assign some result to the instruction (otherwise we would treat it as
2352/// unreachable). For simplicity, we mark any instructions that are still
2353/// unknown as overdefined.
2354bool SCCPInstVisitor::resolvedUndefsIn(Function &F) {
2355 bool MadeChange = false;
2356 for (BasicBlock &BB : F) {
2357 if (!BBExecutable.count(Ptr: &BB))
2358 continue;
2359
2360 for (Instruction &I : BB)
2361 MadeChange |= resolvedUndef(I);
2362 }
2363
2364 LLVM_DEBUG(if (MadeChange) dbgs()
2365 << "\nResolved undefs in " << F.getName() << '\n');
2366
2367 return MadeChange;
2368}
2369
2370//===----------------------------------------------------------------------===//
2371//
2372// SCCPSolver implementations
2373//
2374SCCPSolver::SCCPSolver(
2375 const DataLayout &DL,
2376 std::function<const TargetLibraryInfo &(Function &)> GetTLI,
2377 LLVMContext &Ctx)
2378 : Visitor(new SCCPInstVisitor(DL, std::move(GetTLI), Ctx)) {}
2379
2380SCCPSolver::~SCCPSolver() = default;
2381
2382const DataLayout &SCCPSolver::getDataLayout() const {
2383 return Visitor->getDataLayout();
2384}
2385
2386void SCCPSolver::addPredicateInfo(Function &F, DominatorTree &DT,
2387 AssumptionCache &AC) {
2388 Visitor->addPredicateInfo(F, DT, AC);
2389}
2390
2391void SCCPSolver::removeSSACopies(Function &F) {
2392 Visitor->removeSSACopies(F);
2393}
2394
2395bool SCCPSolver::markBlockExecutable(BasicBlock *BB) {
2396 return Visitor->markBlockExecutable(BB);
2397}
2398
2399const PredicateBase *SCCPSolver::getPredicateInfoFor(Instruction *I) {
2400 return Visitor->getPredicateInfoFor(I);
2401}
2402
2403void SCCPSolver::trackValueOfGlobalVariable(GlobalVariable *GV) {
2404 Visitor->trackValueOfGlobalVariable(GV);
2405}
2406
2407void SCCPSolver::addTrackedFunction(Function *F) {
2408 Visitor->addTrackedFunction(F);
2409}
2410
2411void SCCPSolver::addToMustPreserveReturnsInFunctions(Function *F) {
2412 Visitor->addToMustPreserveReturnsInFunctions(F);
2413}
2414
2415bool SCCPSolver::mustPreserveReturn(Function *F) {
2416 return Visitor->mustPreserveReturn(F);
2417}
2418
2419void SCCPSolver::addArgumentTrackedFunction(Function *F) {
2420 Visitor->addArgumentTrackedFunction(F);
2421}
2422
2423bool SCCPSolver::isArgumentTrackedFunction(Function *F) {
2424 return Visitor->isArgumentTrackedFunction(F);
2425}
2426
2427const SmallPtrSetImpl<Function *> &
2428SCCPSolver::getArgumentTrackedFunctions() const {
2429 return Visitor->getArgumentTrackedFunctions();
2430}
2431
2432void SCCPSolver::solve() { Visitor->solve(); }
2433
2434bool SCCPSolver::resolvedUndefsIn(Function &F) {
2435 return Visitor->resolvedUndefsIn(F);
2436}
2437
2438void SCCPSolver::solveWhileResolvedUndefsIn(Module &M) {
2439 Visitor->solveWhileResolvedUndefsIn(M);
2440}
2441
2442void
2443SCCPSolver::solveWhileResolvedUndefsIn(SmallVectorImpl<Function *> &WorkList) {
2444 Visitor->solveWhileResolvedUndefsIn(WorkList);
2445}
2446
2447void SCCPSolver::solveWhileResolvedUndefs() {
2448 Visitor->solveWhileResolvedUndefs();
2449}
2450
2451bool SCCPSolver::isBlockExecutable(BasicBlock *BB) const {
2452 return Visitor->isBlockExecutable(BB);
2453}
2454
2455bool SCCPSolver::isEdgeFeasible(BasicBlock *From, BasicBlock *To) const {
2456 return Visitor->isEdgeFeasible(From, To);
2457}
2458
2459std::vector<ValueLatticeElement>
2460SCCPSolver::getStructLatticeValueFor(Value *V) const {
2461 return Visitor->getStructLatticeValueFor(V);
2462}
2463
2464void SCCPSolver::removeLatticeValueFor(Value *V) {
2465 return Visitor->removeLatticeValueFor(V);
2466}
2467
2468void SCCPSolver::resetLatticeValueFor(CallBase *Call) {
2469 Visitor->resetLatticeValueFor(Call);
2470}
2471
2472const ValueLatticeElement &SCCPSolver::getLatticeValueFor(Value *V) const {
2473 return Visitor->getLatticeValueFor(V);
2474}
2475
2476const MapVector<Function *, ValueLatticeElement> &
2477SCCPSolver::getTrackedRetVals() const {
2478 return Visitor->getTrackedRetVals();
2479}
2480
2481const DenseMap<GlobalVariable *, ValueLatticeElement> &
2482SCCPSolver::getTrackedGlobals() const {
2483 return Visitor->getTrackedGlobals();
2484}
2485
2486const SmallPtrSet<Function *, 16> &SCCPSolver::getMRVFunctionsTracked() const {
2487 return Visitor->getMRVFunctionsTracked();
2488}
2489
2490void SCCPSolver::markOverdefined(Value *V) { Visitor->markOverdefined(V); }
2491
2492void SCCPSolver::trackValueOfArgument(Argument *V) {
2493 Visitor->trackValueOfArgument(A: V);
2494}
2495
2496bool SCCPSolver::isStructLatticeConstant(Function *F, StructType *STy) {
2497 return Visitor->isStructLatticeConstant(F, STy);
2498}
2499
2500Constant *SCCPSolver::getConstant(const ValueLatticeElement &LV,
2501 Type *Ty) const {
2502 return Visitor->getConstant(LV, Ty);
2503}
2504
2505Constant *SCCPSolver::getConstantOrNull(Value *V) const {
2506 return Visitor->getConstantOrNull(V);
2507}
2508
2509void SCCPSolver::setLatticeValueForSpecializationArguments(Function *F,
2510 const SmallVectorImpl<ArgInfo> &Args) {
2511 Visitor->setLatticeValueForSpecializationArguments(F, Args);
2512}
2513
2514void SCCPSolver::markFunctionUnreachable(Function *F) {
2515 Visitor->markFunctionUnreachable(F);
2516}
2517
2518void SCCPSolver::visit(Instruction *I) { Visitor->visit(I); }
2519
2520void SCCPSolver::visitCall(CallInst &I) { Visitor->visitCall(I); }
2521