1 //===- GVN.cpp - Eliminate redundant values and loads ---------------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This pass performs global value numbering to eliminate fully redundant
11 // instructions. It also performs simple dead load elimination.
12 //
13 // Note that this pass does the value numbering itself; it does not use the
14 // ValueNumbering analysis passes.
15 //
16 //===----------------------------------------------------------------------===//
17
18 #include "llvm/Transforms/Scalar.h"
19 #include "llvm/ADT/DenseMap.h"
20 #include "llvm/ADT/DepthFirstIterator.h"
21 #include "llvm/ADT/Hashing.h"
22 #include "llvm/ADT/MapVector.h"
23 #include "llvm/ADT/PostOrderIterator.h"
24 #include "llvm/ADT/SetVector.h"
25 #include "llvm/ADT/SmallPtrSet.h"
26 #include "llvm/ADT/Statistic.h"
27 #include "llvm/Analysis/AliasAnalysis.h"
28 #include "llvm/Analysis/AssumptionCache.h"
29 #include "llvm/Analysis/CFG.h"
30 #include "llvm/Analysis/ConstantFolding.h"
31 #include "llvm/Analysis/InstructionSimplify.h"
32 #include "llvm/Analysis/Loads.h"
33 #include "llvm/Analysis/MemoryBuiltins.h"
34 #include "llvm/Analysis/MemoryDependenceAnalysis.h"
35 #include "llvm/Analysis/PHITransAddr.h"
36 #include "llvm/Analysis/TargetLibraryInfo.h"
37 #include "llvm/Analysis/ValueTracking.h"
38 #include "llvm/IR/DataLayout.h"
39 #include "llvm/IR/Dominators.h"
40 #include "llvm/IR/GlobalVariable.h"
41 #include "llvm/IR/IRBuilder.h"
42 #include "llvm/IR/IntrinsicInst.h"
43 #include "llvm/IR/LLVMContext.h"
44 #include "llvm/IR/Metadata.h"
45 #include "llvm/IR/PatternMatch.h"
46 #include "llvm/Support/Allocator.h"
47 #include "llvm/Support/CommandLine.h"
48 #include "llvm/Support/Debug.h"
49 #include "llvm/Support/raw_ostream.h"
50 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
51 #include "llvm/Transforms/Utils/Local.h"
52 #include "llvm/Transforms/Utils/SSAUpdater.h"
53 #include <vector>
54 using namespace llvm;
55 using namespace PatternMatch;
56
57 #define DEBUG_TYPE "gvn"
58
59 STATISTIC(NumGVNInstr, "Number of instructions deleted");
60 STATISTIC(NumGVNLoad, "Number of loads deleted");
61 STATISTIC(NumGVNPRE, "Number of instructions PRE'd");
62 STATISTIC(NumGVNBlocks, "Number of blocks merged");
63 STATISTIC(NumGVNSimpl, "Number of instructions simplified");
64 STATISTIC(NumGVNEqProp, "Number of equalities propagated");
65 STATISTIC(NumPRELoad, "Number of loads PRE'd");
66
67 static cl::opt<bool> EnablePRE("enable-pre",
68 cl::init(true), cl::Hidden);
69 static cl::opt<bool> EnableLoadPRE("enable-load-pre", cl::init(true));
70
71 // Maximum allowed recursion depth.
72 static cl::opt<uint32_t>
73 MaxRecurseDepth("max-recurse-depth", cl::Hidden, cl::init(1000), cl::ZeroOrMore,
74 cl::desc("Max recurse depth (default = 1000)"));
75
76 //===----------------------------------------------------------------------===//
77 // ValueTable Class
78 //===----------------------------------------------------------------------===//
79
80 /// This class holds the mapping between values and value numbers. It is used
81 /// as an efficient mechanism to determine the expression-wise equivalence of
82 /// two values.
83 namespace {
84 struct Expression {
85 uint32_t opcode;
86 Type *type;
87 SmallVector<uint32_t, 4> varargs;
88
Expression__anon4436d4200111::Expression89 Expression(uint32_t o = ~2U) : opcode(o) { }
90
operator ==__anon4436d4200111::Expression91 bool operator==(const Expression &other) const {
92 if (opcode != other.opcode)
93 return false;
94 if (opcode == ~0U || opcode == ~1U)
95 return true;
96 if (type != other.type)
97 return false;
98 if (varargs != other.varargs)
99 return false;
100 return true;
101 }
102
hash_value(const Expression & Value)103 friend hash_code hash_value(const Expression &Value) {
104 return hash_combine(Value.opcode, Value.type,
105 hash_combine_range(Value.varargs.begin(),
106 Value.varargs.end()));
107 }
108 };
109
110 class ValueTable {
111 DenseMap<Value*, uint32_t> valueNumbering;
112 DenseMap<Expression, uint32_t> expressionNumbering;
113 AliasAnalysis *AA;
114 MemoryDependenceAnalysis *MD;
115 DominatorTree *DT;
116
117 uint32_t nextValueNumber;
118
119 Expression create_expression(Instruction* I);
120 Expression create_cmp_expression(unsigned Opcode,
121 CmpInst::Predicate Predicate,
122 Value *LHS, Value *RHS);
123 Expression create_extractvalue_expression(ExtractValueInst* EI);
124 uint32_t lookup_or_add_call(CallInst* C);
125 public:
ValueTable()126 ValueTable() : nextValueNumber(1) { }
127 uint32_t lookup_or_add(Value *V);
128 uint32_t lookup(Value *V) const;
129 uint32_t lookup_or_add_cmp(unsigned Opcode, CmpInst::Predicate Pred,
130 Value *LHS, Value *RHS);
131 void add(Value *V, uint32_t num);
132 void clear();
133 void erase(Value *v);
setAliasAnalysis(AliasAnalysis * A)134 void setAliasAnalysis(AliasAnalysis* A) { AA = A; }
getAliasAnalysis() const135 AliasAnalysis *getAliasAnalysis() const { return AA; }
setMemDep(MemoryDependenceAnalysis * M)136 void setMemDep(MemoryDependenceAnalysis* M) { MD = M; }
setDomTree(DominatorTree * D)137 void setDomTree(DominatorTree* D) { DT = D; }
getNextUnusedValueNumber()138 uint32_t getNextUnusedValueNumber() { return nextValueNumber; }
139 void verifyRemoved(const Value *) const;
140 };
141 }
142
143 namespace llvm {
144 template <> struct DenseMapInfo<Expression> {
getEmptyKeyllvm::DenseMapInfo145 static inline Expression getEmptyKey() {
146 return ~0U;
147 }
148
getTombstoneKeyllvm::DenseMapInfo149 static inline Expression getTombstoneKey() {
150 return ~1U;
151 }
152
getHashValuellvm::DenseMapInfo153 static unsigned getHashValue(const Expression e) {
154 using llvm::hash_value;
155 return static_cast<unsigned>(hash_value(e));
156 }
isEqualllvm::DenseMapInfo157 static bool isEqual(const Expression &LHS, const Expression &RHS) {
158 return LHS == RHS;
159 }
160 };
161
162 }
163
164 //===----------------------------------------------------------------------===//
165 // ValueTable Internal Functions
166 //===----------------------------------------------------------------------===//
167
create_expression(Instruction * I)168 Expression ValueTable::create_expression(Instruction *I) {
169 Expression e;
170 e.type = I->getType();
171 e.opcode = I->getOpcode();
172 for (Instruction::op_iterator OI = I->op_begin(), OE = I->op_end();
173 OI != OE; ++OI)
174 e.varargs.push_back(lookup_or_add(*OI));
175 if (I->isCommutative()) {
176 // Ensure that commutative instructions that only differ by a permutation
177 // of their operands get the same value number by sorting the operand value
178 // numbers. Since all commutative instructions have two operands it is more
179 // efficient to sort by hand rather than using, say, std::sort.
180 assert(I->getNumOperands() == 2 && "Unsupported commutative instruction!");
181 if (e.varargs[0] > e.varargs[1])
182 std::swap(e.varargs[0], e.varargs[1]);
183 }
184
185 if (CmpInst *C = dyn_cast<CmpInst>(I)) {
186 // Sort the operand value numbers so x<y and y>x get the same value number.
187 CmpInst::Predicate Predicate = C->getPredicate();
188 if (e.varargs[0] > e.varargs[1]) {
189 std::swap(e.varargs[0], e.varargs[1]);
190 Predicate = CmpInst::getSwappedPredicate(Predicate);
191 }
192 e.opcode = (C->getOpcode() << 8) | Predicate;
193 } else if (InsertValueInst *E = dyn_cast<InsertValueInst>(I)) {
194 for (InsertValueInst::idx_iterator II = E->idx_begin(), IE = E->idx_end();
195 II != IE; ++II)
196 e.varargs.push_back(*II);
197 }
198
199 return e;
200 }
201
create_cmp_expression(unsigned Opcode,CmpInst::Predicate Predicate,Value * LHS,Value * RHS)202 Expression ValueTable::create_cmp_expression(unsigned Opcode,
203 CmpInst::Predicate Predicate,
204 Value *LHS, Value *RHS) {
205 assert((Opcode == Instruction::ICmp || Opcode == Instruction::FCmp) &&
206 "Not a comparison!");
207 Expression e;
208 e.type = CmpInst::makeCmpResultType(LHS->getType());
209 e.varargs.push_back(lookup_or_add(LHS));
210 e.varargs.push_back(lookup_or_add(RHS));
211
212 // Sort the operand value numbers so x<y and y>x get the same value number.
213 if (e.varargs[0] > e.varargs[1]) {
214 std::swap(e.varargs[0], e.varargs[1]);
215 Predicate = CmpInst::getSwappedPredicate(Predicate);
216 }
217 e.opcode = (Opcode << 8) | Predicate;
218 return e;
219 }
220
create_extractvalue_expression(ExtractValueInst * EI)221 Expression ValueTable::create_extractvalue_expression(ExtractValueInst *EI) {
222 assert(EI && "Not an ExtractValueInst?");
223 Expression e;
224 e.type = EI->getType();
225 e.opcode = 0;
226
227 IntrinsicInst *I = dyn_cast<IntrinsicInst>(EI->getAggregateOperand());
228 if (I != nullptr && EI->getNumIndices() == 1 && *EI->idx_begin() == 0 ) {
229 // EI might be an extract from one of our recognised intrinsics. If it
230 // is we'll synthesize a semantically equivalent expression instead on
231 // an extract value expression.
232 switch (I->getIntrinsicID()) {
233 case Intrinsic::sadd_with_overflow:
234 case Intrinsic::uadd_with_overflow:
235 e.opcode = Instruction::Add;
236 break;
237 case Intrinsic::ssub_with_overflow:
238 case Intrinsic::usub_with_overflow:
239 e.opcode = Instruction::Sub;
240 break;
241 case Intrinsic::smul_with_overflow:
242 case Intrinsic::umul_with_overflow:
243 e.opcode = Instruction::Mul;
244 break;
245 default:
246 break;
247 }
248
249 if (e.opcode != 0) {
250 // Intrinsic recognized. Grab its args to finish building the expression.
251 assert(I->getNumArgOperands() == 2 &&
252 "Expect two args for recognised intrinsics.");
253 e.varargs.push_back(lookup_or_add(I->getArgOperand(0)));
254 e.varargs.push_back(lookup_or_add(I->getArgOperand(1)));
255 return e;
256 }
257 }
258
259 // Not a recognised intrinsic. Fall back to producing an extract value
260 // expression.
261 e.opcode = EI->getOpcode();
262 for (Instruction::op_iterator OI = EI->op_begin(), OE = EI->op_end();
263 OI != OE; ++OI)
264 e.varargs.push_back(lookup_or_add(*OI));
265
266 for (ExtractValueInst::idx_iterator II = EI->idx_begin(), IE = EI->idx_end();
267 II != IE; ++II)
268 e.varargs.push_back(*II);
269
270 return e;
271 }
272
273 //===----------------------------------------------------------------------===//
274 // ValueTable External Functions
275 //===----------------------------------------------------------------------===//
276
277 /// add - Insert a value into the table with a specified value number.
add(Value * V,uint32_t num)278 void ValueTable::add(Value *V, uint32_t num) {
279 valueNumbering.insert(std::make_pair(V, num));
280 }
281
lookup_or_add_call(CallInst * C)282 uint32_t ValueTable::lookup_or_add_call(CallInst *C) {
283 if (AA->doesNotAccessMemory(C)) {
284 Expression exp = create_expression(C);
285 uint32_t &e = expressionNumbering[exp];
286 if (!e) e = nextValueNumber++;
287 valueNumbering[C] = e;
288 return e;
289 } else if (AA->onlyReadsMemory(C)) {
290 Expression exp = create_expression(C);
291 uint32_t &e = expressionNumbering[exp];
292 if (!e) {
293 e = nextValueNumber++;
294 valueNumbering[C] = e;
295 return e;
296 }
297 if (!MD) {
298 e = nextValueNumber++;
299 valueNumbering[C] = e;
300 return e;
301 }
302
303 MemDepResult local_dep = MD->getDependency(C);
304
305 if (!local_dep.isDef() && !local_dep.isNonLocal()) {
306 valueNumbering[C] = nextValueNumber;
307 return nextValueNumber++;
308 }
309
310 if (local_dep.isDef()) {
311 CallInst* local_cdep = cast<CallInst>(local_dep.getInst());
312
313 if (local_cdep->getNumArgOperands() != C->getNumArgOperands()) {
314 valueNumbering[C] = nextValueNumber;
315 return nextValueNumber++;
316 }
317
318 for (unsigned i = 0, e = C->getNumArgOperands(); i < e; ++i) {
319 uint32_t c_vn = lookup_or_add(C->getArgOperand(i));
320 uint32_t cd_vn = lookup_or_add(local_cdep->getArgOperand(i));
321 if (c_vn != cd_vn) {
322 valueNumbering[C] = nextValueNumber;
323 return nextValueNumber++;
324 }
325 }
326
327 uint32_t v = lookup_or_add(local_cdep);
328 valueNumbering[C] = v;
329 return v;
330 }
331
332 // Non-local case.
333 const MemoryDependenceAnalysis::NonLocalDepInfo &deps =
334 MD->getNonLocalCallDependency(CallSite(C));
335 // FIXME: Move the checking logic to MemDep!
336 CallInst* cdep = nullptr;
337
338 // Check to see if we have a single dominating call instruction that is
339 // identical to C.
340 for (unsigned i = 0, e = deps.size(); i != e; ++i) {
341 const NonLocalDepEntry *I = &deps[i];
342 if (I->getResult().isNonLocal())
343 continue;
344
345 // We don't handle non-definitions. If we already have a call, reject
346 // instruction dependencies.
347 if (!I->getResult().isDef() || cdep != nullptr) {
348 cdep = nullptr;
349 break;
350 }
351
352 CallInst *NonLocalDepCall = dyn_cast<CallInst>(I->getResult().getInst());
353 // FIXME: All duplicated with non-local case.
354 if (NonLocalDepCall && DT->properlyDominates(I->getBB(), C->getParent())){
355 cdep = NonLocalDepCall;
356 continue;
357 }
358
359 cdep = nullptr;
360 break;
361 }
362
363 if (!cdep) {
364 valueNumbering[C] = nextValueNumber;
365 return nextValueNumber++;
366 }
367
368 if (cdep->getNumArgOperands() != C->getNumArgOperands()) {
369 valueNumbering[C] = nextValueNumber;
370 return nextValueNumber++;
371 }
372 for (unsigned i = 0, e = C->getNumArgOperands(); i < e; ++i) {
373 uint32_t c_vn = lookup_or_add(C->getArgOperand(i));
374 uint32_t cd_vn = lookup_or_add(cdep->getArgOperand(i));
375 if (c_vn != cd_vn) {
376 valueNumbering[C] = nextValueNumber;
377 return nextValueNumber++;
378 }
379 }
380
381 uint32_t v = lookup_or_add(cdep);
382 valueNumbering[C] = v;
383 return v;
384
385 } else {
386 valueNumbering[C] = nextValueNumber;
387 return nextValueNumber++;
388 }
389 }
390
391 /// lookup_or_add - Returns the value number for the specified value, assigning
392 /// it a new number if it did not have one before.
lookup_or_add(Value * V)393 uint32_t ValueTable::lookup_or_add(Value *V) {
394 DenseMap<Value*, uint32_t>::iterator VI = valueNumbering.find(V);
395 if (VI != valueNumbering.end())
396 return VI->second;
397
398 if (!isa<Instruction>(V)) {
399 valueNumbering[V] = nextValueNumber;
400 return nextValueNumber++;
401 }
402
403 Instruction* I = cast<Instruction>(V);
404 Expression exp;
405 switch (I->getOpcode()) {
406 case Instruction::Call:
407 return lookup_or_add_call(cast<CallInst>(I));
408 case Instruction::Add:
409 case Instruction::FAdd:
410 case Instruction::Sub:
411 case Instruction::FSub:
412 case Instruction::Mul:
413 case Instruction::FMul:
414 case Instruction::UDiv:
415 case Instruction::SDiv:
416 case Instruction::FDiv:
417 case Instruction::URem:
418 case Instruction::SRem:
419 case Instruction::FRem:
420 case Instruction::Shl:
421 case Instruction::LShr:
422 case Instruction::AShr:
423 case Instruction::And:
424 case Instruction::Or:
425 case Instruction::Xor:
426 case Instruction::ICmp:
427 case Instruction::FCmp:
428 case Instruction::Trunc:
429 case Instruction::ZExt:
430 case Instruction::SExt:
431 case Instruction::FPToUI:
432 case Instruction::FPToSI:
433 case Instruction::UIToFP:
434 case Instruction::SIToFP:
435 case Instruction::FPTrunc:
436 case Instruction::FPExt:
437 case Instruction::PtrToInt:
438 case Instruction::IntToPtr:
439 case Instruction::BitCast:
440 case Instruction::Select:
441 case Instruction::ExtractElement:
442 case Instruction::InsertElement:
443 case Instruction::ShuffleVector:
444 case Instruction::InsertValue:
445 case Instruction::GetElementPtr:
446 exp = create_expression(I);
447 break;
448 case Instruction::ExtractValue:
449 exp = create_extractvalue_expression(cast<ExtractValueInst>(I));
450 break;
451 default:
452 valueNumbering[V] = nextValueNumber;
453 return nextValueNumber++;
454 }
455
456 uint32_t& e = expressionNumbering[exp];
457 if (!e) e = nextValueNumber++;
458 valueNumbering[V] = e;
459 return e;
460 }
461
462 /// Returns the value number of the specified value. Fails if
463 /// the value has not yet been numbered.
lookup(Value * V) const464 uint32_t ValueTable::lookup(Value *V) const {
465 DenseMap<Value*, uint32_t>::const_iterator VI = valueNumbering.find(V);
466 assert(VI != valueNumbering.end() && "Value not numbered?");
467 return VI->second;
468 }
469
470 /// Returns the value number of the given comparison,
471 /// assigning it a new number if it did not have one before. Useful when
472 /// we deduced the result of a comparison, but don't immediately have an
473 /// instruction realizing that comparison to hand.
lookup_or_add_cmp(unsigned Opcode,CmpInst::Predicate Predicate,Value * LHS,Value * RHS)474 uint32_t ValueTable::lookup_or_add_cmp(unsigned Opcode,
475 CmpInst::Predicate Predicate,
476 Value *LHS, Value *RHS) {
477 Expression exp = create_cmp_expression(Opcode, Predicate, LHS, RHS);
478 uint32_t& e = expressionNumbering[exp];
479 if (!e) e = nextValueNumber++;
480 return e;
481 }
482
483 /// Remove all entries from the ValueTable.
clear()484 void ValueTable::clear() {
485 valueNumbering.clear();
486 expressionNumbering.clear();
487 nextValueNumber = 1;
488 }
489
490 /// Remove a value from the value numbering.
erase(Value * V)491 void ValueTable::erase(Value *V) {
492 valueNumbering.erase(V);
493 }
494
495 /// verifyRemoved - Verify that the value is removed from all internal data
496 /// structures.
verifyRemoved(const Value * V) const497 void ValueTable::verifyRemoved(const Value *V) const {
498 for (DenseMap<Value*, uint32_t>::const_iterator
499 I = valueNumbering.begin(), E = valueNumbering.end(); I != E; ++I) {
500 assert(I->first != V && "Inst still occurs in value numbering map!");
501 }
502 }
503
504 //===----------------------------------------------------------------------===//
505 // GVN Pass
506 //===----------------------------------------------------------------------===//
507
508 namespace {
509 class GVN;
510 struct AvailableValueInBlock {
511 /// BB - The basic block in question.
512 BasicBlock *BB;
513 enum ValType {
514 SimpleVal, // A simple offsetted value that is accessed.
515 LoadVal, // A value produced by a load.
516 MemIntrin, // A memory intrinsic which is loaded from.
517 UndefVal // A UndefValue representing a value from dead block (which
518 // is not yet physically removed from the CFG).
519 };
520
521 /// V - The value that is live out of the block.
522 PointerIntPair<Value *, 2, ValType> Val;
523
524 /// Offset - The byte offset in Val that is interesting for the load query.
525 unsigned Offset;
526
get__anon4436d4200211::AvailableValueInBlock527 static AvailableValueInBlock get(BasicBlock *BB, Value *V,
528 unsigned Offset = 0) {
529 AvailableValueInBlock Res;
530 Res.BB = BB;
531 Res.Val.setPointer(V);
532 Res.Val.setInt(SimpleVal);
533 Res.Offset = Offset;
534 return Res;
535 }
536
getMI__anon4436d4200211::AvailableValueInBlock537 static AvailableValueInBlock getMI(BasicBlock *BB, MemIntrinsic *MI,
538 unsigned Offset = 0) {
539 AvailableValueInBlock Res;
540 Res.BB = BB;
541 Res.Val.setPointer(MI);
542 Res.Val.setInt(MemIntrin);
543 Res.Offset = Offset;
544 return Res;
545 }
546
getLoad__anon4436d4200211::AvailableValueInBlock547 static AvailableValueInBlock getLoad(BasicBlock *BB, LoadInst *LI,
548 unsigned Offset = 0) {
549 AvailableValueInBlock Res;
550 Res.BB = BB;
551 Res.Val.setPointer(LI);
552 Res.Val.setInt(LoadVal);
553 Res.Offset = Offset;
554 return Res;
555 }
556
getUndef__anon4436d4200211::AvailableValueInBlock557 static AvailableValueInBlock getUndef(BasicBlock *BB) {
558 AvailableValueInBlock Res;
559 Res.BB = BB;
560 Res.Val.setPointer(nullptr);
561 Res.Val.setInt(UndefVal);
562 Res.Offset = 0;
563 return Res;
564 }
565
isSimpleValue__anon4436d4200211::AvailableValueInBlock566 bool isSimpleValue() const { return Val.getInt() == SimpleVal; }
isCoercedLoadValue__anon4436d4200211::AvailableValueInBlock567 bool isCoercedLoadValue() const { return Val.getInt() == LoadVal; }
isMemIntrinValue__anon4436d4200211::AvailableValueInBlock568 bool isMemIntrinValue() const { return Val.getInt() == MemIntrin; }
isUndefValue__anon4436d4200211::AvailableValueInBlock569 bool isUndefValue() const { return Val.getInt() == UndefVal; }
570
getSimpleValue__anon4436d4200211::AvailableValueInBlock571 Value *getSimpleValue() const {
572 assert(isSimpleValue() && "Wrong accessor");
573 return Val.getPointer();
574 }
575
getCoercedLoadValue__anon4436d4200211::AvailableValueInBlock576 LoadInst *getCoercedLoadValue() const {
577 assert(isCoercedLoadValue() && "Wrong accessor");
578 return cast<LoadInst>(Val.getPointer());
579 }
580
getMemIntrinValue__anon4436d4200211::AvailableValueInBlock581 MemIntrinsic *getMemIntrinValue() const {
582 assert(isMemIntrinValue() && "Wrong accessor");
583 return cast<MemIntrinsic>(Val.getPointer());
584 }
585
586 /// Emit code into this block to adjust the value defined here to the
587 /// specified type. This handles various coercion cases.
588 Value *MaterializeAdjustedValue(LoadInst *LI, GVN &gvn) const;
589 };
590
591 class GVN : public FunctionPass {
592 bool NoLoads;
593 MemoryDependenceAnalysis *MD;
594 DominatorTree *DT;
595 const TargetLibraryInfo *TLI;
596 AssumptionCache *AC;
597 SetVector<BasicBlock *> DeadBlocks;
598
599 ValueTable VN;
600
601 /// A mapping from value numbers to lists of Value*'s that
602 /// have that value number. Use findLeader to query it.
603 struct LeaderTableEntry {
604 Value *Val;
605 const BasicBlock *BB;
606 LeaderTableEntry *Next;
607 };
608 DenseMap<uint32_t, LeaderTableEntry> LeaderTable;
609 BumpPtrAllocator TableAllocator;
610
611 SmallVector<Instruction*, 8> InstrsToErase;
612
613 typedef SmallVector<NonLocalDepResult, 64> LoadDepVect;
614 typedef SmallVector<AvailableValueInBlock, 64> AvailValInBlkVect;
615 typedef SmallVector<BasicBlock*, 64> UnavailBlkVect;
616
617 public:
618 static char ID; // Pass identification, replacement for typeid
GVN(bool noloads=false)619 explicit GVN(bool noloads = false)
620 : FunctionPass(ID), NoLoads(noloads), MD(nullptr) {
621 initializeGVNPass(*PassRegistry::getPassRegistry());
622 }
623
624 bool runOnFunction(Function &F) override;
625
626 /// This removes the specified instruction from
627 /// our various maps and marks it for deletion.
markInstructionForDeletion(Instruction * I)628 void markInstructionForDeletion(Instruction *I) {
629 VN.erase(I);
630 InstrsToErase.push_back(I);
631 }
632
getDominatorTree() const633 DominatorTree &getDominatorTree() const { return *DT; }
getAliasAnalysis() const634 AliasAnalysis *getAliasAnalysis() const { return VN.getAliasAnalysis(); }
getMemDep() const635 MemoryDependenceAnalysis &getMemDep() const { return *MD; }
636 private:
637 /// Push a new Value to the LeaderTable onto the list for its value number.
addToLeaderTable(uint32_t N,Value * V,const BasicBlock * BB)638 void addToLeaderTable(uint32_t N, Value *V, const BasicBlock *BB) {
639 LeaderTableEntry &Curr = LeaderTable[N];
640 if (!Curr.Val) {
641 Curr.Val = V;
642 Curr.BB = BB;
643 return;
644 }
645
646 LeaderTableEntry *Node = TableAllocator.Allocate<LeaderTableEntry>();
647 Node->Val = V;
648 Node->BB = BB;
649 Node->Next = Curr.Next;
650 Curr.Next = Node;
651 }
652
653 /// Scan the list of values corresponding to a given
654 /// value number, and remove the given instruction if encountered.
removeFromLeaderTable(uint32_t N,Instruction * I,BasicBlock * BB)655 void removeFromLeaderTable(uint32_t N, Instruction *I, BasicBlock *BB) {
656 LeaderTableEntry* Prev = nullptr;
657 LeaderTableEntry* Curr = &LeaderTable[N];
658
659 while (Curr->Val != I || Curr->BB != BB) {
660 Prev = Curr;
661 Curr = Curr->Next;
662 }
663
664 if (Prev) {
665 Prev->Next = Curr->Next;
666 } else {
667 if (!Curr->Next) {
668 Curr->Val = nullptr;
669 Curr->BB = nullptr;
670 } else {
671 LeaderTableEntry* Next = Curr->Next;
672 Curr->Val = Next->Val;
673 Curr->BB = Next->BB;
674 Curr->Next = Next->Next;
675 }
676 }
677 }
678
679 // List of critical edges to be split between iterations.
680 SmallVector<std::pair<TerminatorInst*, unsigned>, 4> toSplit;
681
682 // This transformation requires dominator postdominator info
getAnalysisUsage(AnalysisUsage & AU) const683 void getAnalysisUsage(AnalysisUsage &AU) const override {
684 AU.addRequired<AssumptionCacheTracker>();
685 AU.addRequired<DominatorTreeWrapperPass>();
686 AU.addRequired<TargetLibraryInfoWrapperPass>();
687 if (!NoLoads)
688 AU.addRequired<MemoryDependenceAnalysis>();
689 AU.addRequired<AliasAnalysis>();
690
691 AU.addPreserved<DominatorTreeWrapperPass>();
692 AU.addPreserved<AliasAnalysis>();
693 }
694
695
696 // Helper fuctions of redundant load elimination
697 bool processLoad(LoadInst *L);
698 bool processNonLocalLoad(LoadInst *L);
699 void AnalyzeLoadAvailability(LoadInst *LI, LoadDepVect &Deps,
700 AvailValInBlkVect &ValuesPerBlock,
701 UnavailBlkVect &UnavailableBlocks);
702 bool PerformLoadPRE(LoadInst *LI, AvailValInBlkVect &ValuesPerBlock,
703 UnavailBlkVect &UnavailableBlocks);
704
705 // Other helper routines
706 bool processInstruction(Instruction *I);
707 bool processBlock(BasicBlock *BB);
708 void dump(DenseMap<uint32_t, Value*> &d);
709 bool iterateOnFunction(Function &F);
710 bool performPRE(Function &F);
711 bool performScalarPRE(Instruction *I);
712 bool performScalarPREInsertion(Instruction *Instr, BasicBlock *Pred,
713 unsigned int ValNo);
714 Value *findLeader(const BasicBlock *BB, uint32_t num);
715 void cleanupGlobalSets();
716 void verifyRemoved(const Instruction *I) const;
717 bool splitCriticalEdges();
718 BasicBlock *splitCriticalEdges(BasicBlock *Pred, BasicBlock *Succ);
719 unsigned replaceAllDominatedUsesWith(Value *From, Value *To,
720 const BasicBlockEdge &Root);
721 bool propagateEquality(Value *LHS, Value *RHS, const BasicBlockEdge &Root);
722 bool processFoldableCondBr(BranchInst *BI);
723 void addDeadBlock(BasicBlock *BB);
724 void assignValNumForDeadCode();
725 };
726
727 char GVN::ID = 0;
728 }
729
730 // The public interface to this file...
createGVNPass(bool NoLoads)731 FunctionPass *llvm::createGVNPass(bool NoLoads) {
732 return new GVN(NoLoads);
733 }
734
735 INITIALIZE_PASS_BEGIN(GVN, "gvn", "Global Value Numbering", false, false)
INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)736 INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
737 INITIALIZE_PASS_DEPENDENCY(MemoryDependenceAnalysis)
738 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
739 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
740 INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
741 INITIALIZE_PASS_END(GVN, "gvn", "Global Value Numbering", false, false)
742
743 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
744 void GVN::dump(DenseMap<uint32_t, Value*>& d) {
745 errs() << "{\n";
746 for (DenseMap<uint32_t, Value*>::iterator I = d.begin(),
747 E = d.end(); I != E; ++I) {
748 errs() << I->first << "\n";
749 I->second->dump();
750 }
751 errs() << "}\n";
752 }
753 #endif
754
755 /// Return true if we can prove that the value
756 /// we're analyzing is fully available in the specified block. As we go, keep
757 /// track of which blocks we know are fully alive in FullyAvailableBlocks. This
758 /// map is actually a tri-state map with the following values:
759 /// 0) we know the block *is not* fully available.
760 /// 1) we know the block *is* fully available.
761 /// 2) we do not know whether the block is fully available or not, but we are
762 /// currently speculating that it will be.
763 /// 3) we are speculating for this block and have used that to speculate for
764 /// other blocks.
IsValueFullyAvailableInBlock(BasicBlock * BB,DenseMap<BasicBlock *,char> & FullyAvailableBlocks,uint32_t RecurseDepth)765 static bool IsValueFullyAvailableInBlock(BasicBlock *BB,
766 DenseMap<BasicBlock*, char> &FullyAvailableBlocks,
767 uint32_t RecurseDepth) {
768 if (RecurseDepth > MaxRecurseDepth)
769 return false;
770
771 // Optimistically assume that the block is fully available and check to see
772 // if we already know about this block in one lookup.
773 std::pair<DenseMap<BasicBlock*, char>::iterator, char> IV =
774 FullyAvailableBlocks.insert(std::make_pair(BB, 2));
775
776 // If the entry already existed for this block, return the precomputed value.
777 if (!IV.second) {
778 // If this is a speculative "available" value, mark it as being used for
779 // speculation of other blocks.
780 if (IV.first->second == 2)
781 IV.first->second = 3;
782 return IV.first->second != 0;
783 }
784
785 // Otherwise, see if it is fully available in all predecessors.
786 pred_iterator PI = pred_begin(BB), PE = pred_end(BB);
787
788 // If this block has no predecessors, it isn't live-in here.
789 if (PI == PE)
790 goto SpeculationFailure;
791
792 for (; PI != PE; ++PI)
793 // If the value isn't fully available in one of our predecessors, then it
794 // isn't fully available in this block either. Undo our previous
795 // optimistic assumption and bail out.
796 if (!IsValueFullyAvailableInBlock(*PI, FullyAvailableBlocks,RecurseDepth+1))
797 goto SpeculationFailure;
798
799 return true;
800
801 // If we get here, we found out that this is not, after
802 // all, a fully-available block. We have a problem if we speculated on this and
803 // used the speculation to mark other blocks as available.
804 SpeculationFailure:
805 char &BBVal = FullyAvailableBlocks[BB];
806
807 // If we didn't speculate on this, just return with it set to false.
808 if (BBVal == 2) {
809 BBVal = 0;
810 return false;
811 }
812
813 // If we did speculate on this value, we could have blocks set to 1 that are
814 // incorrect. Walk the (transitive) successors of this block and mark them as
815 // 0 if set to one.
816 SmallVector<BasicBlock*, 32> BBWorklist;
817 BBWorklist.push_back(BB);
818
819 do {
820 BasicBlock *Entry = BBWorklist.pop_back_val();
821 // Note that this sets blocks to 0 (unavailable) if they happen to not
822 // already be in FullyAvailableBlocks. This is safe.
823 char &EntryVal = FullyAvailableBlocks[Entry];
824 if (EntryVal == 0) continue; // Already unavailable.
825
826 // Mark as unavailable.
827 EntryVal = 0;
828
829 BBWorklist.append(succ_begin(Entry), succ_end(Entry));
830 } while (!BBWorklist.empty());
831
832 return false;
833 }
834
835
836 /// Return true if CoerceAvailableValueToLoadType will succeed.
CanCoerceMustAliasedValueToLoad(Value * StoredVal,Type * LoadTy,const DataLayout & DL)837 static bool CanCoerceMustAliasedValueToLoad(Value *StoredVal,
838 Type *LoadTy,
839 const DataLayout &DL) {
840 // If the loaded or stored value is an first class array or struct, don't try
841 // to transform them. We need to be able to bitcast to integer.
842 if (LoadTy->isStructTy() || LoadTy->isArrayTy() ||
843 StoredVal->getType()->isStructTy() ||
844 StoredVal->getType()->isArrayTy())
845 return false;
846
847 // The store has to be at least as big as the load.
848 if (DL.getTypeSizeInBits(StoredVal->getType()) <
849 DL.getTypeSizeInBits(LoadTy))
850 return false;
851
852 return true;
853 }
854
855 /// If we saw a store of a value to memory, and
856 /// then a load from a must-aliased pointer of a different type, try to coerce
857 /// the stored value. LoadedTy is the type of the load we want to replace and
858 /// InsertPt is the place to insert new instructions.
859 ///
860 /// If we can't do it, return null.
CoerceAvailableValueToLoadType(Value * StoredVal,Type * LoadedTy,Instruction * InsertPt,const DataLayout & DL)861 static Value *CoerceAvailableValueToLoadType(Value *StoredVal,
862 Type *LoadedTy,
863 Instruction *InsertPt,
864 const DataLayout &DL) {
865 if (!CanCoerceMustAliasedValueToLoad(StoredVal, LoadedTy, DL))
866 return nullptr;
867
868 // If this is already the right type, just return it.
869 Type *StoredValTy = StoredVal->getType();
870
871 uint64_t StoreSize = DL.getTypeSizeInBits(StoredValTy);
872 uint64_t LoadSize = DL.getTypeSizeInBits(LoadedTy);
873
874 // If the store and reload are the same size, we can always reuse it.
875 if (StoreSize == LoadSize) {
876 // Pointer to Pointer -> use bitcast.
877 if (StoredValTy->getScalarType()->isPointerTy() &&
878 LoadedTy->getScalarType()->isPointerTy())
879 return new BitCastInst(StoredVal, LoadedTy, "", InsertPt);
880
881 // Convert source pointers to integers, which can be bitcast.
882 if (StoredValTy->getScalarType()->isPointerTy()) {
883 StoredValTy = DL.getIntPtrType(StoredValTy);
884 StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
885 }
886
887 Type *TypeToCastTo = LoadedTy;
888 if (TypeToCastTo->getScalarType()->isPointerTy())
889 TypeToCastTo = DL.getIntPtrType(TypeToCastTo);
890
891 if (StoredValTy != TypeToCastTo)
892 StoredVal = new BitCastInst(StoredVal, TypeToCastTo, "", InsertPt);
893
894 // Cast to pointer if the load needs a pointer type.
895 if (LoadedTy->getScalarType()->isPointerTy())
896 StoredVal = new IntToPtrInst(StoredVal, LoadedTy, "", InsertPt);
897
898 return StoredVal;
899 }
900
901 // If the loaded value is smaller than the available value, then we can
902 // extract out a piece from it. If the available value is too small, then we
903 // can't do anything.
904 assert(StoreSize >= LoadSize && "CanCoerceMustAliasedValueToLoad fail");
905
906 // Convert source pointers to integers, which can be manipulated.
907 if (StoredValTy->getScalarType()->isPointerTy()) {
908 StoredValTy = DL.getIntPtrType(StoredValTy);
909 StoredVal = new PtrToIntInst(StoredVal, StoredValTy, "", InsertPt);
910 }
911
912 // Convert vectors and fp to integer, which can be manipulated.
913 if (!StoredValTy->isIntegerTy()) {
914 StoredValTy = IntegerType::get(StoredValTy->getContext(), StoreSize);
915 StoredVal = new BitCastInst(StoredVal, StoredValTy, "", InsertPt);
916 }
917
918 // If this is a big-endian system, we need to shift the value down to the low
919 // bits so that a truncate will work.
920 if (DL.isBigEndian()) {
921 Constant *Val = ConstantInt::get(StoredVal->getType(), StoreSize-LoadSize);
922 StoredVal = BinaryOperator::CreateLShr(StoredVal, Val, "tmp", InsertPt);
923 }
924
925 // Truncate the integer to the right size now.
926 Type *NewIntTy = IntegerType::get(StoredValTy->getContext(), LoadSize);
927 StoredVal = new TruncInst(StoredVal, NewIntTy, "trunc", InsertPt);
928
929 if (LoadedTy == NewIntTy)
930 return StoredVal;
931
932 // If the result is a pointer, inttoptr.
933 if (LoadedTy->getScalarType()->isPointerTy())
934 return new IntToPtrInst(StoredVal, LoadedTy, "inttoptr", InsertPt);
935
936 // Otherwise, bitcast.
937 return new BitCastInst(StoredVal, LoadedTy, "bitcast", InsertPt);
938 }
939
940 /// This function is called when we have a
941 /// memdep query of a load that ends up being a clobbering memory write (store,
942 /// memset, memcpy, memmove). This means that the write *may* provide bits used
943 /// by the load but we can't be sure because the pointers don't mustalias.
944 ///
945 /// Check this case to see if there is anything more we can do before we give
946 /// up. This returns -1 if we have to give up, or a byte number in the stored
947 /// value of the piece that feeds the load.
AnalyzeLoadFromClobberingWrite(Type * LoadTy,Value * LoadPtr,Value * WritePtr,uint64_t WriteSizeInBits,const DataLayout & DL)948 static int AnalyzeLoadFromClobberingWrite(Type *LoadTy, Value *LoadPtr,
949 Value *WritePtr,
950 uint64_t WriteSizeInBits,
951 const DataLayout &DL) {
952 // If the loaded or stored value is a first class array or struct, don't try
953 // to transform them. We need to be able to bitcast to integer.
954 if (LoadTy->isStructTy() || LoadTy->isArrayTy())
955 return -1;
956
957 int64_t StoreOffset = 0, LoadOffset = 0;
958 Value *StoreBase =
959 GetPointerBaseWithConstantOffset(WritePtr, StoreOffset, DL);
960 Value *LoadBase = GetPointerBaseWithConstantOffset(LoadPtr, LoadOffset, DL);
961 if (StoreBase != LoadBase)
962 return -1;
963
964 // If the load and store are to the exact same address, they should have been
965 // a must alias. AA must have gotten confused.
966 // FIXME: Study to see if/when this happens. One case is forwarding a memset
967 // to a load from the base of the memset.
968 #if 0
969 if (LoadOffset == StoreOffset) {
970 dbgs() << "STORE/LOAD DEP WITH COMMON POINTER MISSED:\n"
971 << "Base = " << *StoreBase << "\n"
972 << "Store Ptr = " << *WritePtr << "\n"
973 << "Store Offs = " << StoreOffset << "\n"
974 << "Load Ptr = " << *LoadPtr << "\n";
975 abort();
976 }
977 #endif
978
979 // If the load and store don't overlap at all, the store doesn't provide
980 // anything to the load. In this case, they really don't alias at all, AA
981 // must have gotten confused.
982 uint64_t LoadSize = DL.getTypeSizeInBits(LoadTy);
983
984 if ((WriteSizeInBits & 7) | (LoadSize & 7))
985 return -1;
986 uint64_t StoreSize = WriteSizeInBits >> 3; // Convert to bytes.
987 LoadSize >>= 3;
988
989
990 bool isAAFailure = false;
991 if (StoreOffset < LoadOffset)
992 isAAFailure = StoreOffset+int64_t(StoreSize) <= LoadOffset;
993 else
994 isAAFailure = LoadOffset+int64_t(LoadSize) <= StoreOffset;
995
996 if (isAAFailure) {
997 #if 0
998 dbgs() << "STORE LOAD DEP WITH COMMON BASE:\n"
999 << "Base = " << *StoreBase << "\n"
1000 << "Store Ptr = " << *WritePtr << "\n"
1001 << "Store Offs = " << StoreOffset << "\n"
1002 << "Load Ptr = " << *LoadPtr << "\n";
1003 abort();
1004 #endif
1005 return -1;
1006 }
1007
1008 // If the Load isn't completely contained within the stored bits, we don't
1009 // have all the bits to feed it. We could do something crazy in the future
1010 // (issue a smaller load then merge the bits in) but this seems unlikely to be
1011 // valuable.
1012 if (StoreOffset > LoadOffset ||
1013 StoreOffset+StoreSize < LoadOffset+LoadSize)
1014 return -1;
1015
1016 // Okay, we can do this transformation. Return the number of bytes into the
1017 // store that the load is.
1018 return LoadOffset-StoreOffset;
1019 }
1020
1021 /// This function is called when we have a
1022 /// memdep query of a load that ends up being a clobbering store.
AnalyzeLoadFromClobberingStore(Type * LoadTy,Value * LoadPtr,StoreInst * DepSI)1023 static int AnalyzeLoadFromClobberingStore(Type *LoadTy, Value *LoadPtr,
1024 StoreInst *DepSI) {
1025 // Cannot handle reading from store of first-class aggregate yet.
1026 if (DepSI->getValueOperand()->getType()->isStructTy() ||
1027 DepSI->getValueOperand()->getType()->isArrayTy())
1028 return -1;
1029
1030 const DataLayout &DL = DepSI->getModule()->getDataLayout();
1031 Value *StorePtr = DepSI->getPointerOperand();
1032 uint64_t StoreSize =DL.getTypeSizeInBits(DepSI->getValueOperand()->getType());
1033 return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr,
1034 StorePtr, StoreSize, DL);
1035 }
1036
1037 /// This function is called when we have a
1038 /// memdep query of a load that ends up being clobbered by another load. See if
1039 /// the other load can feed into the second load.
AnalyzeLoadFromClobberingLoad(Type * LoadTy,Value * LoadPtr,LoadInst * DepLI,const DataLayout & DL)1040 static int AnalyzeLoadFromClobberingLoad(Type *LoadTy, Value *LoadPtr,
1041 LoadInst *DepLI, const DataLayout &DL){
1042 // Cannot handle reading from store of first-class aggregate yet.
1043 if (DepLI->getType()->isStructTy() || DepLI->getType()->isArrayTy())
1044 return -1;
1045
1046 Value *DepPtr = DepLI->getPointerOperand();
1047 uint64_t DepSize = DL.getTypeSizeInBits(DepLI->getType());
1048 int R = AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, DepPtr, DepSize, DL);
1049 if (R != -1) return R;
1050
1051 // If we have a load/load clobber an DepLI can be widened to cover this load,
1052 // then we should widen it!
1053 int64_t LoadOffs = 0;
1054 const Value *LoadBase =
1055 GetPointerBaseWithConstantOffset(LoadPtr, LoadOffs, DL);
1056 unsigned LoadSize = DL.getTypeStoreSize(LoadTy);
1057
1058 unsigned Size = MemoryDependenceAnalysis::getLoadLoadClobberFullWidthSize(
1059 LoadBase, LoadOffs, LoadSize, DepLI);
1060 if (Size == 0) return -1;
1061
1062 return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, DepPtr, Size*8, DL);
1063 }
1064
1065
1066
AnalyzeLoadFromClobberingMemInst(Type * LoadTy,Value * LoadPtr,MemIntrinsic * MI,const DataLayout & DL)1067 static int AnalyzeLoadFromClobberingMemInst(Type *LoadTy, Value *LoadPtr,
1068 MemIntrinsic *MI,
1069 const DataLayout &DL) {
1070 // If the mem operation is a non-constant size, we can't handle it.
1071 ConstantInt *SizeCst = dyn_cast<ConstantInt>(MI->getLength());
1072 if (!SizeCst) return -1;
1073 uint64_t MemSizeInBits = SizeCst->getZExtValue()*8;
1074
1075 // If this is memset, we just need to see if the offset is valid in the size
1076 // of the memset..
1077 if (MI->getIntrinsicID() == Intrinsic::memset)
1078 return AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr, MI->getDest(),
1079 MemSizeInBits, DL);
1080
1081 // If we have a memcpy/memmove, the only case we can handle is if this is a
1082 // copy from constant memory. In that case, we can read directly from the
1083 // constant memory.
1084 MemTransferInst *MTI = cast<MemTransferInst>(MI);
1085
1086 Constant *Src = dyn_cast<Constant>(MTI->getSource());
1087 if (!Src) return -1;
1088
1089 GlobalVariable *GV = dyn_cast<GlobalVariable>(GetUnderlyingObject(Src, DL));
1090 if (!GV || !GV->isConstant()) return -1;
1091
1092 // See if the access is within the bounds of the transfer.
1093 int Offset = AnalyzeLoadFromClobberingWrite(LoadTy, LoadPtr,
1094 MI->getDest(), MemSizeInBits, DL);
1095 if (Offset == -1)
1096 return Offset;
1097
1098 unsigned AS = Src->getType()->getPointerAddressSpace();
1099 // Otherwise, see if we can constant fold a load from the constant with the
1100 // offset applied as appropriate.
1101 Src = ConstantExpr::getBitCast(Src,
1102 Type::getInt8PtrTy(Src->getContext(), AS));
1103 Constant *OffsetCst =
1104 ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset);
1105 Src = ConstantExpr::getGetElementPtr(Type::getInt8Ty(Src->getContext()), Src,
1106 OffsetCst);
1107 Src = ConstantExpr::getBitCast(Src, PointerType::get(LoadTy, AS));
1108 if (ConstantFoldLoadFromConstPtr(Src, DL))
1109 return Offset;
1110 return -1;
1111 }
1112
1113
1114 /// This function is called when we have a
1115 /// memdep query of a load that ends up being a clobbering store. This means
1116 /// that the store provides bits used by the load but we the pointers don't
1117 /// mustalias. Check this case to see if there is anything more we can do
1118 /// before we give up.
GetStoreValueForLoad(Value * SrcVal,unsigned Offset,Type * LoadTy,Instruction * InsertPt,const DataLayout & DL)1119 static Value *GetStoreValueForLoad(Value *SrcVal, unsigned Offset,
1120 Type *LoadTy,
1121 Instruction *InsertPt, const DataLayout &DL){
1122 LLVMContext &Ctx = SrcVal->getType()->getContext();
1123
1124 uint64_t StoreSize = (DL.getTypeSizeInBits(SrcVal->getType()) + 7) / 8;
1125 uint64_t LoadSize = (DL.getTypeSizeInBits(LoadTy) + 7) / 8;
1126
1127 IRBuilder<> Builder(InsertPt->getParent(), InsertPt);
1128
1129 // Compute which bits of the stored value are being used by the load. Convert
1130 // to an integer type to start with.
1131 if (SrcVal->getType()->getScalarType()->isPointerTy())
1132 SrcVal = Builder.CreatePtrToInt(SrcVal,
1133 DL.getIntPtrType(SrcVal->getType()));
1134 if (!SrcVal->getType()->isIntegerTy())
1135 SrcVal = Builder.CreateBitCast(SrcVal, IntegerType::get(Ctx, StoreSize*8));
1136
1137 // Shift the bits to the least significant depending on endianness.
1138 unsigned ShiftAmt;
1139 if (DL.isLittleEndian())
1140 ShiftAmt = Offset*8;
1141 else
1142 ShiftAmt = (StoreSize-LoadSize-Offset)*8;
1143
1144 if (ShiftAmt)
1145 SrcVal = Builder.CreateLShr(SrcVal, ShiftAmt);
1146
1147 if (LoadSize != StoreSize)
1148 SrcVal = Builder.CreateTrunc(SrcVal, IntegerType::get(Ctx, LoadSize*8));
1149
1150 return CoerceAvailableValueToLoadType(SrcVal, LoadTy, InsertPt, DL);
1151 }
1152
1153 /// This function is called when we have a
1154 /// memdep query of a load that ends up being a clobbering load. This means
1155 /// that the load *may* provide bits used by the load but we can't be sure
1156 /// because the pointers don't mustalias. Check this case to see if there is
1157 /// anything more we can do before we give up.
GetLoadValueForLoad(LoadInst * SrcVal,unsigned Offset,Type * LoadTy,Instruction * InsertPt,GVN & gvn)1158 static Value *GetLoadValueForLoad(LoadInst *SrcVal, unsigned Offset,
1159 Type *LoadTy, Instruction *InsertPt,
1160 GVN &gvn) {
1161 const DataLayout &DL = SrcVal->getModule()->getDataLayout();
1162 // If Offset+LoadTy exceeds the size of SrcVal, then we must be wanting to
1163 // widen SrcVal out to a larger load.
1164 unsigned SrcValSize = DL.getTypeStoreSize(SrcVal->getType());
1165 unsigned LoadSize = DL.getTypeStoreSize(LoadTy);
1166 if (Offset+LoadSize > SrcValSize) {
1167 assert(SrcVal->isSimple() && "Cannot widen volatile/atomic load!");
1168 assert(SrcVal->getType()->isIntegerTy() && "Can't widen non-integer load");
1169 // If we have a load/load clobber an DepLI can be widened to cover this
1170 // load, then we should widen it to the next power of 2 size big enough!
1171 unsigned NewLoadSize = Offset+LoadSize;
1172 if (!isPowerOf2_32(NewLoadSize))
1173 NewLoadSize = NextPowerOf2(NewLoadSize);
1174
1175 Value *PtrVal = SrcVal->getPointerOperand();
1176
1177 // Insert the new load after the old load. This ensures that subsequent
1178 // memdep queries will find the new load. We can't easily remove the old
1179 // load completely because it is already in the value numbering table.
1180 IRBuilder<> Builder(SrcVal->getParent(), ++BasicBlock::iterator(SrcVal));
1181 Type *DestPTy =
1182 IntegerType::get(LoadTy->getContext(), NewLoadSize*8);
1183 DestPTy = PointerType::get(DestPTy,
1184 PtrVal->getType()->getPointerAddressSpace());
1185 Builder.SetCurrentDebugLocation(SrcVal->getDebugLoc());
1186 PtrVal = Builder.CreateBitCast(PtrVal, DestPTy);
1187 LoadInst *NewLoad = Builder.CreateLoad(PtrVal);
1188 NewLoad->takeName(SrcVal);
1189 NewLoad->setAlignment(SrcVal->getAlignment());
1190
1191 DEBUG(dbgs() << "GVN WIDENED LOAD: " << *SrcVal << "\n");
1192 DEBUG(dbgs() << "TO: " << *NewLoad << "\n");
1193
1194 // Replace uses of the original load with the wider load. On a big endian
1195 // system, we need to shift down to get the relevant bits.
1196 Value *RV = NewLoad;
1197 if (DL.isBigEndian())
1198 RV = Builder.CreateLShr(RV,
1199 NewLoadSize*8-SrcVal->getType()->getPrimitiveSizeInBits());
1200 RV = Builder.CreateTrunc(RV, SrcVal->getType());
1201 SrcVal->replaceAllUsesWith(RV);
1202
1203 // We would like to use gvn.markInstructionForDeletion here, but we can't
1204 // because the load is already memoized into the leader map table that GVN
1205 // tracks. It is potentially possible to remove the load from the table,
1206 // but then there all of the operations based on it would need to be
1207 // rehashed. Just leave the dead load around.
1208 gvn.getMemDep().removeInstruction(SrcVal);
1209 SrcVal = NewLoad;
1210 }
1211
1212 return GetStoreValueForLoad(SrcVal, Offset, LoadTy, InsertPt, DL);
1213 }
1214
1215
1216 /// This function is called when we have a
1217 /// memdep query of a load that ends up being a clobbering mem intrinsic.
GetMemInstValueForLoad(MemIntrinsic * SrcInst,unsigned Offset,Type * LoadTy,Instruction * InsertPt,const DataLayout & DL)1218 static Value *GetMemInstValueForLoad(MemIntrinsic *SrcInst, unsigned Offset,
1219 Type *LoadTy, Instruction *InsertPt,
1220 const DataLayout &DL){
1221 LLVMContext &Ctx = LoadTy->getContext();
1222 uint64_t LoadSize = DL.getTypeSizeInBits(LoadTy)/8;
1223
1224 IRBuilder<> Builder(InsertPt->getParent(), InsertPt);
1225
1226 // We know that this method is only called when the mem transfer fully
1227 // provides the bits for the load.
1228 if (MemSetInst *MSI = dyn_cast<MemSetInst>(SrcInst)) {
1229 // memset(P, 'x', 1234) -> splat('x'), even if x is a variable, and
1230 // independently of what the offset is.
1231 Value *Val = MSI->getValue();
1232 if (LoadSize != 1)
1233 Val = Builder.CreateZExt(Val, IntegerType::get(Ctx, LoadSize*8));
1234
1235 Value *OneElt = Val;
1236
1237 // Splat the value out to the right number of bits.
1238 for (unsigned NumBytesSet = 1; NumBytesSet != LoadSize; ) {
1239 // If we can double the number of bytes set, do it.
1240 if (NumBytesSet*2 <= LoadSize) {
1241 Value *ShVal = Builder.CreateShl(Val, NumBytesSet*8);
1242 Val = Builder.CreateOr(Val, ShVal);
1243 NumBytesSet <<= 1;
1244 continue;
1245 }
1246
1247 // Otherwise insert one byte at a time.
1248 Value *ShVal = Builder.CreateShl(Val, 1*8);
1249 Val = Builder.CreateOr(OneElt, ShVal);
1250 ++NumBytesSet;
1251 }
1252
1253 return CoerceAvailableValueToLoadType(Val, LoadTy, InsertPt, DL);
1254 }
1255
1256 // Otherwise, this is a memcpy/memmove from a constant global.
1257 MemTransferInst *MTI = cast<MemTransferInst>(SrcInst);
1258 Constant *Src = cast<Constant>(MTI->getSource());
1259 unsigned AS = Src->getType()->getPointerAddressSpace();
1260
1261 // Otherwise, see if we can constant fold a load from the constant with the
1262 // offset applied as appropriate.
1263 Src = ConstantExpr::getBitCast(Src,
1264 Type::getInt8PtrTy(Src->getContext(), AS));
1265 Constant *OffsetCst =
1266 ConstantInt::get(Type::getInt64Ty(Src->getContext()), (unsigned)Offset);
1267 Src = ConstantExpr::getGetElementPtr(Type::getInt8Ty(Src->getContext()), Src,
1268 OffsetCst);
1269 Src = ConstantExpr::getBitCast(Src, PointerType::get(LoadTy, AS));
1270 return ConstantFoldLoadFromConstPtr(Src, DL);
1271 }
1272
1273
1274 /// Given a set of loads specified by ValuesPerBlock,
1275 /// construct SSA form, allowing us to eliminate LI. This returns the value
1276 /// that should be used at LI's definition site.
ConstructSSAForLoadSet(LoadInst * LI,SmallVectorImpl<AvailableValueInBlock> & ValuesPerBlock,GVN & gvn)1277 static Value *ConstructSSAForLoadSet(LoadInst *LI,
1278 SmallVectorImpl<AvailableValueInBlock> &ValuesPerBlock,
1279 GVN &gvn) {
1280 // Check for the fully redundant, dominating load case. In this case, we can
1281 // just use the dominating value directly.
1282 if (ValuesPerBlock.size() == 1 &&
1283 gvn.getDominatorTree().properlyDominates(ValuesPerBlock[0].BB,
1284 LI->getParent())) {
1285 assert(!ValuesPerBlock[0].isUndefValue() && "Dead BB dominate this block");
1286 return ValuesPerBlock[0].MaterializeAdjustedValue(LI, gvn);
1287 }
1288
1289 // Otherwise, we have to construct SSA form.
1290 SmallVector<PHINode*, 8> NewPHIs;
1291 SSAUpdater SSAUpdate(&NewPHIs);
1292 SSAUpdate.Initialize(LI->getType(), LI->getName());
1293
1294 for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i) {
1295 const AvailableValueInBlock &AV = ValuesPerBlock[i];
1296 BasicBlock *BB = AV.BB;
1297
1298 if (SSAUpdate.HasValueForBlock(BB))
1299 continue;
1300
1301 SSAUpdate.AddAvailableValue(BB, AV.MaterializeAdjustedValue(LI, gvn));
1302 }
1303
1304 // Perform PHI construction.
1305 Value *V = SSAUpdate.GetValueInMiddleOfBlock(LI->getParent());
1306
1307 // If new PHI nodes were created, notify alias analysis.
1308 if (V->getType()->getScalarType()->isPointerTy()) {
1309 AliasAnalysis *AA = gvn.getAliasAnalysis();
1310
1311 for (unsigned i = 0, e = NewPHIs.size(); i != e; ++i)
1312 AA->copyValue(LI, NewPHIs[i]);
1313
1314 // Now that we've copied information to the new PHIs, scan through
1315 // them again and inform alias analysis that we've added potentially
1316 // escaping uses to any values that are operands to these PHIs.
1317 for (unsigned i = 0, e = NewPHIs.size(); i != e; ++i) {
1318 PHINode *P = NewPHIs[i];
1319 for (unsigned ii = 0, ee = P->getNumIncomingValues(); ii != ee; ++ii) {
1320 unsigned jj = PHINode::getOperandNumForIncomingValue(ii);
1321 AA->addEscapingUse(P->getOperandUse(jj));
1322 }
1323 }
1324 }
1325
1326 return V;
1327 }
1328
MaterializeAdjustedValue(LoadInst * LI,GVN & gvn) const1329 Value *AvailableValueInBlock::MaterializeAdjustedValue(LoadInst *LI,
1330 GVN &gvn) const {
1331 Value *Res;
1332 Type *LoadTy = LI->getType();
1333 const DataLayout &DL = LI->getModule()->getDataLayout();
1334 if (isSimpleValue()) {
1335 Res = getSimpleValue();
1336 if (Res->getType() != LoadTy) {
1337 Res = GetStoreValueForLoad(Res, Offset, LoadTy, BB->getTerminator(), DL);
1338
1339 DEBUG(dbgs() << "GVN COERCED NONLOCAL VAL:\nOffset: " << Offset << " "
1340 << *getSimpleValue() << '\n'
1341 << *Res << '\n' << "\n\n\n");
1342 }
1343 } else if (isCoercedLoadValue()) {
1344 LoadInst *Load = getCoercedLoadValue();
1345 if (Load->getType() == LoadTy && Offset == 0) {
1346 Res = Load;
1347 } else {
1348 Res = GetLoadValueForLoad(Load, Offset, LoadTy, BB->getTerminator(),
1349 gvn);
1350
1351 DEBUG(dbgs() << "GVN COERCED NONLOCAL LOAD:\nOffset: " << Offset << " "
1352 << *getCoercedLoadValue() << '\n'
1353 << *Res << '\n' << "\n\n\n");
1354 }
1355 } else if (isMemIntrinValue()) {
1356 Res = GetMemInstValueForLoad(getMemIntrinValue(), Offset, LoadTy,
1357 BB->getTerminator(), DL);
1358 DEBUG(dbgs() << "GVN COERCED NONLOCAL MEM INTRIN:\nOffset: " << Offset
1359 << " " << *getMemIntrinValue() << '\n'
1360 << *Res << '\n' << "\n\n\n");
1361 } else {
1362 assert(isUndefValue() && "Should be UndefVal");
1363 DEBUG(dbgs() << "GVN COERCED NONLOCAL Undef:\n";);
1364 return UndefValue::get(LoadTy);
1365 }
1366 return Res;
1367 }
1368
isLifetimeStart(const Instruction * Inst)1369 static bool isLifetimeStart(const Instruction *Inst) {
1370 if (const IntrinsicInst* II = dyn_cast<IntrinsicInst>(Inst))
1371 return II->getIntrinsicID() == Intrinsic::lifetime_start;
1372 return false;
1373 }
1374
AnalyzeLoadAvailability(LoadInst * LI,LoadDepVect & Deps,AvailValInBlkVect & ValuesPerBlock,UnavailBlkVect & UnavailableBlocks)1375 void GVN::AnalyzeLoadAvailability(LoadInst *LI, LoadDepVect &Deps,
1376 AvailValInBlkVect &ValuesPerBlock,
1377 UnavailBlkVect &UnavailableBlocks) {
1378
1379 // Filter out useless results (non-locals, etc). Keep track of the blocks
1380 // where we have a value available in repl, also keep track of whether we see
1381 // dependencies that produce an unknown value for the load (such as a call
1382 // that could potentially clobber the load).
1383 unsigned NumDeps = Deps.size();
1384 const DataLayout &DL = LI->getModule()->getDataLayout();
1385 for (unsigned i = 0, e = NumDeps; i != e; ++i) {
1386 BasicBlock *DepBB = Deps[i].getBB();
1387 MemDepResult DepInfo = Deps[i].getResult();
1388
1389 if (DeadBlocks.count(DepBB)) {
1390 // Dead dependent mem-op disguise as a load evaluating the same value
1391 // as the load in question.
1392 ValuesPerBlock.push_back(AvailableValueInBlock::getUndef(DepBB));
1393 continue;
1394 }
1395
1396 if (!DepInfo.isDef() && !DepInfo.isClobber()) {
1397 UnavailableBlocks.push_back(DepBB);
1398 continue;
1399 }
1400
1401 if (DepInfo.isClobber()) {
1402 // The address being loaded in this non-local block may not be the same as
1403 // the pointer operand of the load if PHI translation occurs. Make sure
1404 // to consider the right address.
1405 Value *Address = Deps[i].getAddress();
1406
1407 // If the dependence is to a store that writes to a superset of the bits
1408 // read by the load, we can extract the bits we need for the load from the
1409 // stored value.
1410 if (StoreInst *DepSI = dyn_cast<StoreInst>(DepInfo.getInst())) {
1411 if (Address) {
1412 int Offset =
1413 AnalyzeLoadFromClobberingStore(LI->getType(), Address, DepSI);
1414 if (Offset != -1) {
1415 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB,
1416 DepSI->getValueOperand(),
1417 Offset));
1418 continue;
1419 }
1420 }
1421 }
1422
1423 // Check to see if we have something like this:
1424 // load i32* P
1425 // load i8* (P+1)
1426 // if we have this, replace the later with an extraction from the former.
1427 if (LoadInst *DepLI = dyn_cast<LoadInst>(DepInfo.getInst())) {
1428 // If this is a clobber and L is the first instruction in its block, then
1429 // we have the first instruction in the entry block.
1430 if (DepLI != LI && Address) {
1431 int Offset =
1432 AnalyzeLoadFromClobberingLoad(LI->getType(), Address, DepLI, DL);
1433
1434 if (Offset != -1) {
1435 ValuesPerBlock.push_back(AvailableValueInBlock::getLoad(DepBB,DepLI,
1436 Offset));
1437 continue;
1438 }
1439 }
1440 }
1441
1442 // If the clobbering value is a memset/memcpy/memmove, see if we can
1443 // forward a value on from it.
1444 if (MemIntrinsic *DepMI = dyn_cast<MemIntrinsic>(DepInfo.getInst())) {
1445 if (Address) {
1446 int Offset = AnalyzeLoadFromClobberingMemInst(LI->getType(), Address,
1447 DepMI, DL);
1448 if (Offset != -1) {
1449 ValuesPerBlock.push_back(AvailableValueInBlock::getMI(DepBB, DepMI,
1450 Offset));
1451 continue;
1452 }
1453 }
1454 }
1455
1456 UnavailableBlocks.push_back(DepBB);
1457 continue;
1458 }
1459
1460 // DepInfo.isDef() here
1461
1462 Instruction *DepInst = DepInfo.getInst();
1463
1464 // Loading the allocation -> undef.
1465 if (isa<AllocaInst>(DepInst) || isMallocLikeFn(DepInst, TLI) ||
1466 // Loading immediately after lifetime begin -> undef.
1467 isLifetimeStart(DepInst)) {
1468 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB,
1469 UndefValue::get(LI->getType())));
1470 continue;
1471 }
1472
1473 // Loading from calloc (which zero initializes memory) -> zero
1474 if (isCallocLikeFn(DepInst, TLI)) {
1475 ValuesPerBlock.push_back(AvailableValueInBlock::get(
1476 DepBB, Constant::getNullValue(LI->getType())));
1477 continue;
1478 }
1479
1480 if (StoreInst *S = dyn_cast<StoreInst>(DepInst)) {
1481 // Reject loads and stores that are to the same address but are of
1482 // different types if we have to.
1483 if (S->getValueOperand()->getType() != LI->getType()) {
1484 // If the stored value is larger or equal to the loaded value, we can
1485 // reuse it.
1486 if (!CanCoerceMustAliasedValueToLoad(S->getValueOperand(),
1487 LI->getType(), DL)) {
1488 UnavailableBlocks.push_back(DepBB);
1489 continue;
1490 }
1491 }
1492
1493 ValuesPerBlock.push_back(AvailableValueInBlock::get(DepBB,
1494 S->getValueOperand()));
1495 continue;
1496 }
1497
1498 if (LoadInst *LD = dyn_cast<LoadInst>(DepInst)) {
1499 // If the types mismatch and we can't handle it, reject reuse of the load.
1500 if (LD->getType() != LI->getType()) {
1501 // If the stored value is larger or equal to the loaded value, we can
1502 // reuse it.
1503 if (!CanCoerceMustAliasedValueToLoad(LD, LI->getType(), DL)) {
1504 UnavailableBlocks.push_back(DepBB);
1505 continue;
1506 }
1507 }
1508 ValuesPerBlock.push_back(AvailableValueInBlock::getLoad(DepBB, LD));
1509 continue;
1510 }
1511
1512 UnavailableBlocks.push_back(DepBB);
1513 }
1514 }
1515
PerformLoadPRE(LoadInst * LI,AvailValInBlkVect & ValuesPerBlock,UnavailBlkVect & UnavailableBlocks)1516 bool GVN::PerformLoadPRE(LoadInst *LI, AvailValInBlkVect &ValuesPerBlock,
1517 UnavailBlkVect &UnavailableBlocks) {
1518 // Okay, we have *some* definitions of the value. This means that the value
1519 // is available in some of our (transitive) predecessors. Lets think about
1520 // doing PRE of this load. This will involve inserting a new load into the
1521 // predecessor when it's not available. We could do this in general, but
1522 // prefer to not increase code size. As such, we only do this when we know
1523 // that we only have to insert *one* load (which means we're basically moving
1524 // the load, not inserting a new one).
1525
1526 SmallPtrSet<BasicBlock *, 4> Blockers;
1527 for (unsigned i = 0, e = UnavailableBlocks.size(); i != e; ++i)
1528 Blockers.insert(UnavailableBlocks[i]);
1529
1530 // Let's find the first basic block with more than one predecessor. Walk
1531 // backwards through predecessors if needed.
1532 BasicBlock *LoadBB = LI->getParent();
1533 BasicBlock *TmpBB = LoadBB;
1534
1535 while (TmpBB->getSinglePredecessor()) {
1536 TmpBB = TmpBB->getSinglePredecessor();
1537 if (TmpBB == LoadBB) // Infinite (unreachable) loop.
1538 return false;
1539 if (Blockers.count(TmpBB))
1540 return false;
1541
1542 // If any of these blocks has more than one successor (i.e. if the edge we
1543 // just traversed was critical), then there are other paths through this
1544 // block along which the load may not be anticipated. Hoisting the load
1545 // above this block would be adding the load to execution paths along
1546 // which it was not previously executed.
1547 if (TmpBB->getTerminator()->getNumSuccessors() != 1)
1548 return false;
1549 }
1550
1551 assert(TmpBB);
1552 LoadBB = TmpBB;
1553
1554 // Check to see how many predecessors have the loaded value fully
1555 // available.
1556 MapVector<BasicBlock *, Value *> PredLoads;
1557 DenseMap<BasicBlock*, char> FullyAvailableBlocks;
1558 for (unsigned i = 0, e = ValuesPerBlock.size(); i != e; ++i)
1559 FullyAvailableBlocks[ValuesPerBlock[i].BB] = true;
1560 for (unsigned i = 0, e = UnavailableBlocks.size(); i != e; ++i)
1561 FullyAvailableBlocks[UnavailableBlocks[i]] = false;
1562
1563 SmallVector<BasicBlock *, 4> CriticalEdgePred;
1564 for (pred_iterator PI = pred_begin(LoadBB), E = pred_end(LoadBB);
1565 PI != E; ++PI) {
1566 BasicBlock *Pred = *PI;
1567 if (IsValueFullyAvailableInBlock(Pred, FullyAvailableBlocks, 0)) {
1568 continue;
1569 }
1570
1571 if (Pred->getTerminator()->getNumSuccessors() != 1) {
1572 if (isa<IndirectBrInst>(Pred->getTerminator())) {
1573 DEBUG(dbgs() << "COULD NOT PRE LOAD BECAUSE OF INDBR CRITICAL EDGE '"
1574 << Pred->getName() << "': " << *LI << '\n');
1575 return false;
1576 }
1577
1578 if (LoadBB->isLandingPad()) {
1579 DEBUG(dbgs()
1580 << "COULD NOT PRE LOAD BECAUSE OF LANDING PAD CRITICAL EDGE '"
1581 << Pred->getName() << "': " << *LI << '\n');
1582 return false;
1583 }
1584
1585 CriticalEdgePred.push_back(Pred);
1586 } else {
1587 // Only add the predecessors that will not be split for now.
1588 PredLoads[Pred] = nullptr;
1589 }
1590 }
1591
1592 // Decide whether PRE is profitable for this load.
1593 unsigned NumUnavailablePreds = PredLoads.size() + CriticalEdgePred.size();
1594 assert(NumUnavailablePreds != 0 &&
1595 "Fully available value should already be eliminated!");
1596
1597 // If this load is unavailable in multiple predecessors, reject it.
1598 // FIXME: If we could restructure the CFG, we could make a common pred with
1599 // all the preds that don't have an available LI and insert a new load into
1600 // that one block.
1601 if (NumUnavailablePreds != 1)
1602 return false;
1603
1604 // Split critical edges, and update the unavailable predecessors accordingly.
1605 for (BasicBlock *OrigPred : CriticalEdgePred) {
1606 BasicBlock *NewPred = splitCriticalEdges(OrigPred, LoadBB);
1607 assert(!PredLoads.count(OrigPred) && "Split edges shouldn't be in map!");
1608 PredLoads[NewPred] = nullptr;
1609 DEBUG(dbgs() << "Split critical edge " << OrigPred->getName() << "->"
1610 << LoadBB->getName() << '\n');
1611 }
1612
1613 // Check if the load can safely be moved to all the unavailable predecessors.
1614 bool CanDoPRE = true;
1615 const DataLayout &DL = LI->getModule()->getDataLayout();
1616 SmallVector<Instruction*, 8> NewInsts;
1617 for (auto &PredLoad : PredLoads) {
1618 BasicBlock *UnavailablePred = PredLoad.first;
1619
1620 // Do PHI translation to get its value in the predecessor if necessary. The
1621 // returned pointer (if non-null) is guaranteed to dominate UnavailablePred.
1622
1623 // If all preds have a single successor, then we know it is safe to insert
1624 // the load on the pred (?!?), so we can insert code to materialize the
1625 // pointer if it is not available.
1626 PHITransAddr Address(LI->getPointerOperand(), DL, AC);
1627 Value *LoadPtr = nullptr;
1628 LoadPtr = Address.PHITranslateWithInsertion(LoadBB, UnavailablePred,
1629 *DT, NewInsts);
1630
1631 // If we couldn't find or insert a computation of this phi translated value,
1632 // we fail PRE.
1633 if (!LoadPtr) {
1634 DEBUG(dbgs() << "COULDN'T INSERT PHI TRANSLATED VALUE OF: "
1635 << *LI->getPointerOperand() << "\n");
1636 CanDoPRE = false;
1637 break;
1638 }
1639
1640 PredLoad.second = LoadPtr;
1641 }
1642
1643 if (!CanDoPRE) {
1644 while (!NewInsts.empty()) {
1645 Instruction *I = NewInsts.pop_back_val();
1646 if (MD) MD->removeInstruction(I);
1647 I->eraseFromParent();
1648 }
1649 // HINT: Don't revert the edge-splitting as following transformation may
1650 // also need to split these critical edges.
1651 return !CriticalEdgePred.empty();
1652 }
1653
1654 // Okay, we can eliminate this load by inserting a reload in the predecessor
1655 // and using PHI construction to get the value in the other predecessors, do
1656 // it.
1657 DEBUG(dbgs() << "GVN REMOVING PRE LOAD: " << *LI << '\n');
1658 DEBUG(if (!NewInsts.empty())
1659 dbgs() << "INSERTED " << NewInsts.size() << " INSTS: "
1660 << *NewInsts.back() << '\n');
1661
1662 // Assign value numbers to the new instructions.
1663 for (unsigned i = 0, e = NewInsts.size(); i != e; ++i) {
1664 // FIXME: We really _ought_ to insert these value numbers into their
1665 // parent's availability map. However, in doing so, we risk getting into
1666 // ordering issues. If a block hasn't been processed yet, we would be
1667 // marking a value as AVAIL-IN, which isn't what we intend.
1668 VN.lookup_or_add(NewInsts[i]);
1669 }
1670
1671 for (const auto &PredLoad : PredLoads) {
1672 BasicBlock *UnavailablePred = PredLoad.first;
1673 Value *LoadPtr = PredLoad.second;
1674
1675 Instruction *NewLoad = new LoadInst(LoadPtr, LI->getName()+".pre", false,
1676 LI->getAlignment(),
1677 UnavailablePred->getTerminator());
1678
1679 // Transfer the old load's AA tags to the new load.
1680 AAMDNodes Tags;
1681 LI->getAAMetadata(Tags);
1682 if (Tags)
1683 NewLoad->setAAMetadata(Tags);
1684
1685 // Transfer DebugLoc.
1686 NewLoad->setDebugLoc(LI->getDebugLoc());
1687
1688 // Add the newly created load.
1689 ValuesPerBlock.push_back(AvailableValueInBlock::get(UnavailablePred,
1690 NewLoad));
1691 MD->invalidateCachedPointerInfo(LoadPtr);
1692 DEBUG(dbgs() << "GVN INSERTED " << *NewLoad << '\n');
1693 }
1694
1695 // Perform PHI construction.
1696 Value *V = ConstructSSAForLoadSet(LI, ValuesPerBlock, *this);
1697 LI->replaceAllUsesWith(V);
1698 if (isa<PHINode>(V))
1699 V->takeName(LI);
1700 if (V->getType()->getScalarType()->isPointerTy())
1701 MD->invalidateCachedPointerInfo(V);
1702 markInstructionForDeletion(LI);
1703 ++NumPRELoad;
1704 return true;
1705 }
1706
1707 /// Attempt to eliminate a load whose dependencies are
1708 /// non-local by performing PHI construction.
processNonLocalLoad(LoadInst * LI)1709 bool GVN::processNonLocalLoad(LoadInst *LI) {
1710 // Step 1: Find the non-local dependencies of the load.
1711 LoadDepVect Deps;
1712 MD->getNonLocalPointerDependency(LI, Deps);
1713
1714 // If we had to process more than one hundred blocks to find the
1715 // dependencies, this load isn't worth worrying about. Optimizing
1716 // it will be too expensive.
1717 unsigned NumDeps = Deps.size();
1718 if (NumDeps > 100)
1719 return false;
1720
1721 // If we had a phi translation failure, we'll have a single entry which is a
1722 // clobber in the current block. Reject this early.
1723 if (NumDeps == 1 &&
1724 !Deps[0].getResult().isDef() && !Deps[0].getResult().isClobber()) {
1725 DEBUG(
1726 dbgs() << "GVN: non-local load ";
1727 LI->printAsOperand(dbgs());
1728 dbgs() << " has unknown dependencies\n";
1729 );
1730 return false;
1731 }
1732
1733 // If this load follows a GEP, see if we can PRE the indices before analyzing.
1734 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(LI->getOperand(0))) {
1735 for (GetElementPtrInst::op_iterator OI = GEP->idx_begin(),
1736 OE = GEP->idx_end();
1737 OI != OE; ++OI)
1738 if (Instruction *I = dyn_cast<Instruction>(OI->get()))
1739 performScalarPRE(I);
1740 }
1741
1742 // Step 2: Analyze the availability of the load
1743 AvailValInBlkVect ValuesPerBlock;
1744 UnavailBlkVect UnavailableBlocks;
1745 AnalyzeLoadAvailability(LI, Deps, ValuesPerBlock, UnavailableBlocks);
1746
1747 // If we have no predecessors that produce a known value for this load, exit
1748 // early.
1749 if (ValuesPerBlock.empty())
1750 return false;
1751
1752 // Step 3: Eliminate fully redundancy.
1753 //
1754 // If all of the instructions we depend on produce a known value for this
1755 // load, then it is fully redundant and we can use PHI insertion to compute
1756 // its value. Insert PHIs and remove the fully redundant value now.
1757 if (UnavailableBlocks.empty()) {
1758 DEBUG(dbgs() << "GVN REMOVING NONLOCAL LOAD: " << *LI << '\n');
1759
1760 // Perform PHI construction.
1761 Value *V = ConstructSSAForLoadSet(LI, ValuesPerBlock, *this);
1762 LI->replaceAllUsesWith(V);
1763
1764 if (isa<PHINode>(V))
1765 V->takeName(LI);
1766 if (V->getType()->getScalarType()->isPointerTy())
1767 MD->invalidateCachedPointerInfo(V);
1768 markInstructionForDeletion(LI);
1769 ++NumGVNLoad;
1770 return true;
1771 }
1772
1773 // Step 4: Eliminate partial redundancy.
1774 if (!EnablePRE || !EnableLoadPRE)
1775 return false;
1776
1777 return PerformLoadPRE(LI, ValuesPerBlock, UnavailableBlocks);
1778 }
1779
1780
patchReplacementInstruction(Instruction * I,Value * Repl)1781 static void patchReplacementInstruction(Instruction *I, Value *Repl) {
1782 // Patch the replacement so that it is not more restrictive than the value
1783 // being replaced.
1784 BinaryOperator *Op = dyn_cast<BinaryOperator>(I);
1785 BinaryOperator *ReplOp = dyn_cast<BinaryOperator>(Repl);
1786 if (Op && ReplOp && isa<OverflowingBinaryOperator>(Op) &&
1787 isa<OverflowingBinaryOperator>(ReplOp)) {
1788 if (ReplOp->hasNoSignedWrap() && !Op->hasNoSignedWrap())
1789 ReplOp->setHasNoSignedWrap(false);
1790 if (ReplOp->hasNoUnsignedWrap() && !Op->hasNoUnsignedWrap())
1791 ReplOp->setHasNoUnsignedWrap(false);
1792 }
1793 if (Instruction *ReplInst = dyn_cast<Instruction>(Repl)) {
1794 // FIXME: If both the original and replacement value are part of the
1795 // same control-flow region (meaning that the execution of one
1796 // guarentees the executation of the other), then we can combine the
1797 // noalias scopes here and do better than the general conservative
1798 // answer used in combineMetadata().
1799
1800 // In general, GVN unifies expressions over different control-flow
1801 // regions, and so we need a conservative combination of the noalias
1802 // scopes.
1803 unsigned KnownIDs[] = {
1804 LLVMContext::MD_tbaa,
1805 LLVMContext::MD_alias_scope,
1806 LLVMContext::MD_noalias,
1807 LLVMContext::MD_range,
1808 LLVMContext::MD_fpmath,
1809 LLVMContext::MD_invariant_load,
1810 };
1811 combineMetadata(ReplInst, I, KnownIDs);
1812 }
1813 }
1814
patchAndReplaceAllUsesWith(Instruction * I,Value * Repl)1815 static void patchAndReplaceAllUsesWith(Instruction *I, Value *Repl) {
1816 patchReplacementInstruction(I, Repl);
1817 I->replaceAllUsesWith(Repl);
1818 }
1819
1820 /// Attempt to eliminate a load, first by eliminating it
1821 /// locally, and then attempting non-local elimination if that fails.
processLoad(LoadInst * L)1822 bool GVN::processLoad(LoadInst *L) {
1823 if (!MD)
1824 return false;
1825
1826 if (!L->isSimple())
1827 return false;
1828
1829 if (L->use_empty()) {
1830 markInstructionForDeletion(L);
1831 return true;
1832 }
1833
1834 // ... to a pointer that has been loaded from before...
1835 MemDepResult Dep = MD->getDependency(L);
1836 const DataLayout &DL = L->getModule()->getDataLayout();
1837
1838 // If we have a clobber and target data is around, see if this is a clobber
1839 // that we can fix up through code synthesis.
1840 if (Dep.isClobber()) {
1841 // Check to see if we have something like this:
1842 // store i32 123, i32* %P
1843 // %A = bitcast i32* %P to i8*
1844 // %B = gep i8* %A, i32 1
1845 // %C = load i8* %B
1846 //
1847 // We could do that by recognizing if the clobber instructions are obviously
1848 // a common base + constant offset, and if the previous store (or memset)
1849 // completely covers this load. This sort of thing can happen in bitfield
1850 // access code.
1851 Value *AvailVal = nullptr;
1852 if (StoreInst *DepSI = dyn_cast<StoreInst>(Dep.getInst())) {
1853 int Offset = AnalyzeLoadFromClobberingStore(
1854 L->getType(), L->getPointerOperand(), DepSI);
1855 if (Offset != -1)
1856 AvailVal = GetStoreValueForLoad(DepSI->getValueOperand(), Offset,
1857 L->getType(), L, DL);
1858 }
1859
1860 // Check to see if we have something like this:
1861 // load i32* P
1862 // load i8* (P+1)
1863 // if we have this, replace the later with an extraction from the former.
1864 if (LoadInst *DepLI = dyn_cast<LoadInst>(Dep.getInst())) {
1865 // If this is a clobber and L is the first instruction in its block, then
1866 // we have the first instruction in the entry block.
1867 if (DepLI == L)
1868 return false;
1869
1870 int Offset = AnalyzeLoadFromClobberingLoad(
1871 L->getType(), L->getPointerOperand(), DepLI, DL);
1872 if (Offset != -1)
1873 AvailVal = GetLoadValueForLoad(DepLI, Offset, L->getType(), L, *this);
1874 }
1875
1876 // If the clobbering value is a memset/memcpy/memmove, see if we can forward
1877 // a value on from it.
1878 if (MemIntrinsic *DepMI = dyn_cast<MemIntrinsic>(Dep.getInst())) {
1879 int Offset = AnalyzeLoadFromClobberingMemInst(
1880 L->getType(), L->getPointerOperand(), DepMI, DL);
1881 if (Offset != -1)
1882 AvailVal = GetMemInstValueForLoad(DepMI, Offset, L->getType(), L, DL);
1883 }
1884
1885 if (AvailVal) {
1886 DEBUG(dbgs() << "GVN COERCED INST:\n" << *Dep.getInst() << '\n'
1887 << *AvailVal << '\n' << *L << "\n\n\n");
1888
1889 // Replace the load!
1890 L->replaceAllUsesWith(AvailVal);
1891 if (AvailVal->getType()->getScalarType()->isPointerTy())
1892 MD->invalidateCachedPointerInfo(AvailVal);
1893 markInstructionForDeletion(L);
1894 ++NumGVNLoad;
1895 return true;
1896 }
1897 }
1898
1899 // If the value isn't available, don't do anything!
1900 if (Dep.isClobber()) {
1901 DEBUG(
1902 // fast print dep, using operator<< on instruction is too slow.
1903 dbgs() << "GVN: load ";
1904 L->printAsOperand(dbgs());
1905 Instruction *I = Dep.getInst();
1906 dbgs() << " is clobbered by " << *I << '\n';
1907 );
1908 return false;
1909 }
1910
1911 // If it is defined in another block, try harder.
1912 if (Dep.isNonLocal())
1913 return processNonLocalLoad(L);
1914
1915 if (!Dep.isDef()) {
1916 DEBUG(
1917 // fast print dep, using operator<< on instruction is too slow.
1918 dbgs() << "GVN: load ";
1919 L->printAsOperand(dbgs());
1920 dbgs() << " has unknown dependence\n";
1921 );
1922 return false;
1923 }
1924
1925 Instruction *DepInst = Dep.getInst();
1926 if (StoreInst *DepSI = dyn_cast<StoreInst>(DepInst)) {
1927 Value *StoredVal = DepSI->getValueOperand();
1928
1929 // The store and load are to a must-aliased pointer, but they may not
1930 // actually have the same type. See if we know how to reuse the stored
1931 // value (depending on its type).
1932 if (StoredVal->getType() != L->getType()) {
1933 StoredVal =
1934 CoerceAvailableValueToLoadType(StoredVal, L->getType(), L, DL);
1935 if (!StoredVal)
1936 return false;
1937
1938 DEBUG(dbgs() << "GVN COERCED STORE:\n" << *DepSI << '\n' << *StoredVal
1939 << '\n' << *L << "\n\n\n");
1940 }
1941
1942 // Remove it!
1943 L->replaceAllUsesWith(StoredVal);
1944 if (StoredVal->getType()->getScalarType()->isPointerTy())
1945 MD->invalidateCachedPointerInfo(StoredVal);
1946 markInstructionForDeletion(L);
1947 ++NumGVNLoad;
1948 return true;
1949 }
1950
1951 if (LoadInst *DepLI = dyn_cast<LoadInst>(DepInst)) {
1952 Value *AvailableVal = DepLI;
1953
1954 // The loads are of a must-aliased pointer, but they may not actually have
1955 // the same type. See if we know how to reuse the previously loaded value
1956 // (depending on its type).
1957 if (DepLI->getType() != L->getType()) {
1958 AvailableVal = CoerceAvailableValueToLoadType(DepLI, L->getType(), L, DL);
1959 if (!AvailableVal)
1960 return false;
1961
1962 DEBUG(dbgs() << "GVN COERCED LOAD:\n" << *DepLI << "\n" << *AvailableVal
1963 << "\n" << *L << "\n\n\n");
1964 }
1965
1966 // Remove it!
1967 patchAndReplaceAllUsesWith(L, AvailableVal);
1968 if (DepLI->getType()->getScalarType()->isPointerTy())
1969 MD->invalidateCachedPointerInfo(DepLI);
1970 markInstructionForDeletion(L);
1971 ++NumGVNLoad;
1972 return true;
1973 }
1974
1975 // If this load really doesn't depend on anything, then we must be loading an
1976 // undef value. This can happen when loading for a fresh allocation with no
1977 // intervening stores, for example.
1978 if (isa<AllocaInst>(DepInst) || isMallocLikeFn(DepInst, TLI)) {
1979 L->replaceAllUsesWith(UndefValue::get(L->getType()));
1980 markInstructionForDeletion(L);
1981 ++NumGVNLoad;
1982 return true;
1983 }
1984
1985 // If this load occurs either right after a lifetime begin,
1986 // then the loaded value is undefined.
1987 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(DepInst)) {
1988 if (II->getIntrinsicID() == Intrinsic::lifetime_start) {
1989 L->replaceAllUsesWith(UndefValue::get(L->getType()));
1990 markInstructionForDeletion(L);
1991 ++NumGVNLoad;
1992 return true;
1993 }
1994 }
1995
1996 // If this load follows a calloc (which zero initializes memory),
1997 // then the loaded value is zero
1998 if (isCallocLikeFn(DepInst, TLI)) {
1999 L->replaceAllUsesWith(Constant::getNullValue(L->getType()));
2000 markInstructionForDeletion(L);
2001 ++NumGVNLoad;
2002 return true;
2003 }
2004
2005 return false;
2006 }
2007
2008 // In order to find a leader for a given value number at a
2009 // specific basic block, we first obtain the list of all Values for that number,
2010 // and then scan the list to find one whose block dominates the block in
2011 // question. This is fast because dominator tree queries consist of only
2012 // a few comparisons of DFS numbers.
findLeader(const BasicBlock * BB,uint32_t num)2013 Value *GVN::findLeader(const BasicBlock *BB, uint32_t num) {
2014 LeaderTableEntry Vals = LeaderTable[num];
2015 if (!Vals.Val) return nullptr;
2016
2017 Value *Val = nullptr;
2018 if (DT->dominates(Vals.BB, BB)) {
2019 Val = Vals.Val;
2020 if (isa<Constant>(Val)) return Val;
2021 }
2022
2023 LeaderTableEntry* Next = Vals.Next;
2024 while (Next) {
2025 if (DT->dominates(Next->BB, BB)) {
2026 if (isa<Constant>(Next->Val)) return Next->Val;
2027 if (!Val) Val = Next->Val;
2028 }
2029
2030 Next = Next->Next;
2031 }
2032
2033 return Val;
2034 }
2035
2036 /// Replace all uses of 'From' with 'To' if the use is dominated by the given
2037 /// basic block. Returns the number of uses that were replaced.
replaceAllDominatedUsesWith(Value * From,Value * To,const BasicBlockEdge & Root)2038 unsigned GVN::replaceAllDominatedUsesWith(Value *From, Value *To,
2039 const BasicBlockEdge &Root) {
2040 unsigned Count = 0;
2041 for (Value::use_iterator UI = From->use_begin(), UE = From->use_end();
2042 UI != UE; ) {
2043 Use &U = *UI++;
2044
2045 if (DT->dominates(Root, U)) {
2046 U.set(To);
2047 ++Count;
2048 }
2049 }
2050 return Count;
2051 }
2052
2053 /// There is an edge from 'Src' to 'Dst'. Return
2054 /// true if every path from the entry block to 'Dst' passes via this edge. In
2055 /// particular 'Dst' must not be reachable via another edge from 'Src'.
isOnlyReachableViaThisEdge(const BasicBlockEdge & E,DominatorTree * DT)2056 static bool isOnlyReachableViaThisEdge(const BasicBlockEdge &E,
2057 DominatorTree *DT) {
2058 // While in theory it is interesting to consider the case in which Dst has
2059 // more than one predecessor, because Dst might be part of a loop which is
2060 // only reachable from Src, in practice it is pointless since at the time
2061 // GVN runs all such loops have preheaders, which means that Dst will have
2062 // been changed to have only one predecessor, namely Src.
2063 const BasicBlock *Pred = E.getEnd()->getSinglePredecessor();
2064 const BasicBlock *Src = E.getStart();
2065 assert((!Pred || Pred == Src) && "No edge between these basic blocks!");
2066 (void)Src;
2067 return Pred != nullptr;
2068 }
2069
2070 /// The given values are known to be equal in every block
2071 /// dominated by 'Root'. Exploit this, for example by replacing 'LHS' with
2072 /// 'RHS' everywhere in the scope. Returns whether a change was made.
propagateEquality(Value * LHS,Value * RHS,const BasicBlockEdge & Root)2073 bool GVN::propagateEquality(Value *LHS, Value *RHS,
2074 const BasicBlockEdge &Root) {
2075 SmallVector<std::pair<Value*, Value*>, 4> Worklist;
2076 Worklist.push_back(std::make_pair(LHS, RHS));
2077 bool Changed = false;
2078 // For speed, compute a conservative fast approximation to
2079 // DT->dominates(Root, Root.getEnd());
2080 bool RootDominatesEnd = isOnlyReachableViaThisEdge(Root, DT);
2081
2082 while (!Worklist.empty()) {
2083 std::pair<Value*, Value*> Item = Worklist.pop_back_val();
2084 LHS = Item.first; RHS = Item.second;
2085
2086 if (LHS == RHS) continue;
2087 assert(LHS->getType() == RHS->getType() && "Equality but unequal types!");
2088
2089 // Don't try to propagate equalities between constants.
2090 if (isa<Constant>(LHS) && isa<Constant>(RHS)) continue;
2091
2092 // Prefer a constant on the right-hand side, or an Argument if no constants.
2093 if (isa<Constant>(LHS) || (isa<Argument>(LHS) && !isa<Constant>(RHS)))
2094 std::swap(LHS, RHS);
2095 assert((isa<Argument>(LHS) || isa<Instruction>(LHS)) && "Unexpected value!");
2096
2097 // If there is no obvious reason to prefer the left-hand side over the
2098 // right-hand side, ensure the longest lived term is on the right-hand side,
2099 // so the shortest lived term will be replaced by the longest lived.
2100 // This tends to expose more simplifications.
2101 uint32_t LVN = VN.lookup_or_add(LHS);
2102 if ((isa<Argument>(LHS) && isa<Argument>(RHS)) ||
2103 (isa<Instruction>(LHS) && isa<Instruction>(RHS))) {
2104 // Move the 'oldest' value to the right-hand side, using the value number
2105 // as a proxy for age.
2106 uint32_t RVN = VN.lookup_or_add(RHS);
2107 if (LVN < RVN) {
2108 std::swap(LHS, RHS);
2109 LVN = RVN;
2110 }
2111 }
2112
2113 // If value numbering later sees that an instruction in the scope is equal
2114 // to 'LHS' then ensure it will be turned into 'RHS'. In order to preserve
2115 // the invariant that instructions only occur in the leader table for their
2116 // own value number (this is used by removeFromLeaderTable), do not do this
2117 // if RHS is an instruction (if an instruction in the scope is morphed into
2118 // LHS then it will be turned into RHS by the next GVN iteration anyway, so
2119 // using the leader table is about compiling faster, not optimizing better).
2120 // The leader table only tracks basic blocks, not edges. Only add to if we
2121 // have the simple case where the edge dominates the end.
2122 if (RootDominatesEnd && !isa<Instruction>(RHS))
2123 addToLeaderTable(LVN, RHS, Root.getEnd());
2124
2125 // Replace all occurrences of 'LHS' with 'RHS' everywhere in the scope. As
2126 // LHS always has at least one use that is not dominated by Root, this will
2127 // never do anything if LHS has only one use.
2128 if (!LHS->hasOneUse()) {
2129 unsigned NumReplacements = replaceAllDominatedUsesWith(LHS, RHS, Root);
2130 Changed |= NumReplacements > 0;
2131 NumGVNEqProp += NumReplacements;
2132 }
2133
2134 // Now try to deduce additional equalities from this one. For example, if
2135 // the known equality was "(A != B)" == "false" then it follows that A and B
2136 // are equal in the scope. Only boolean equalities with an explicit true or
2137 // false RHS are currently supported.
2138 if (!RHS->getType()->isIntegerTy(1))
2139 // Not a boolean equality - bail out.
2140 continue;
2141 ConstantInt *CI = dyn_cast<ConstantInt>(RHS);
2142 if (!CI)
2143 // RHS neither 'true' nor 'false' - bail out.
2144 continue;
2145 // Whether RHS equals 'true'. Otherwise it equals 'false'.
2146 bool isKnownTrue = CI->isAllOnesValue();
2147 bool isKnownFalse = !isKnownTrue;
2148
2149 // If "A && B" is known true then both A and B are known true. If "A || B"
2150 // is known false then both A and B are known false.
2151 Value *A, *B;
2152 if ((isKnownTrue && match(LHS, m_And(m_Value(A), m_Value(B)))) ||
2153 (isKnownFalse && match(LHS, m_Or(m_Value(A), m_Value(B))))) {
2154 Worklist.push_back(std::make_pair(A, RHS));
2155 Worklist.push_back(std::make_pair(B, RHS));
2156 continue;
2157 }
2158
2159 // If we are propagating an equality like "(A == B)" == "true" then also
2160 // propagate the equality A == B. When propagating a comparison such as
2161 // "(A >= B)" == "true", replace all instances of "A < B" with "false".
2162 if (CmpInst *Cmp = dyn_cast<CmpInst>(LHS)) {
2163 Value *Op0 = Cmp->getOperand(0), *Op1 = Cmp->getOperand(1);
2164
2165 // If "A == B" is known true, or "A != B" is known false, then replace
2166 // A with B everywhere in the scope.
2167 if ((isKnownTrue && Cmp->getPredicate() == CmpInst::ICMP_EQ) ||
2168 (isKnownFalse && Cmp->getPredicate() == CmpInst::ICMP_NE))
2169 Worklist.push_back(std::make_pair(Op0, Op1));
2170
2171 // Handle the floating point versions of equality comparisons too.
2172 if ((isKnownTrue && Cmp->getPredicate() == CmpInst::FCMP_OEQ) ||
2173 (isKnownFalse && Cmp->getPredicate() == CmpInst::FCMP_UNE)) {
2174
2175 // Floating point -0.0 and 0.0 compare equal, so we can only
2176 // propagate values if we know that we have a constant and that
2177 // its value is non-zero.
2178
2179 // FIXME: We should do this optimization if 'no signed zeros' is
2180 // applicable via an instruction-level fast-math-flag or some other
2181 // indicator that relaxed FP semantics are being used.
2182
2183 if (isa<ConstantFP>(Op1) && !cast<ConstantFP>(Op1)->isZero())
2184 Worklist.push_back(std::make_pair(Op0, Op1));
2185 }
2186
2187 // If "A >= B" is known true, replace "A < B" with false everywhere.
2188 CmpInst::Predicate NotPred = Cmp->getInversePredicate();
2189 Constant *NotVal = ConstantInt::get(Cmp->getType(), isKnownFalse);
2190 // Since we don't have the instruction "A < B" immediately to hand, work
2191 // out the value number that it would have and use that to find an
2192 // appropriate instruction (if any).
2193 uint32_t NextNum = VN.getNextUnusedValueNumber();
2194 uint32_t Num = VN.lookup_or_add_cmp(Cmp->getOpcode(), NotPred, Op0, Op1);
2195 // If the number we were assigned was brand new then there is no point in
2196 // looking for an instruction realizing it: there cannot be one!
2197 if (Num < NextNum) {
2198 Value *NotCmp = findLeader(Root.getEnd(), Num);
2199 if (NotCmp && isa<Instruction>(NotCmp)) {
2200 unsigned NumReplacements =
2201 replaceAllDominatedUsesWith(NotCmp, NotVal, Root);
2202 Changed |= NumReplacements > 0;
2203 NumGVNEqProp += NumReplacements;
2204 }
2205 }
2206 // Ensure that any instruction in scope that gets the "A < B" value number
2207 // is replaced with false.
2208 // The leader table only tracks basic blocks, not edges. Only add to if we
2209 // have the simple case where the edge dominates the end.
2210 if (RootDominatesEnd)
2211 addToLeaderTable(Num, NotVal, Root.getEnd());
2212
2213 continue;
2214 }
2215 }
2216
2217 return Changed;
2218 }
2219
2220 /// When calculating availability, handle an instruction
2221 /// by inserting it into the appropriate sets
processInstruction(Instruction * I)2222 bool GVN::processInstruction(Instruction *I) {
2223 // Ignore dbg info intrinsics.
2224 if (isa<DbgInfoIntrinsic>(I))
2225 return false;
2226
2227 // If the instruction can be easily simplified then do so now in preference
2228 // to value numbering it. Value numbering often exposes redundancies, for
2229 // example if it determines that %y is equal to %x then the instruction
2230 // "%z = and i32 %x, %y" becomes "%z = and i32 %x, %x" which we now simplify.
2231 const DataLayout &DL = I->getModule()->getDataLayout();
2232 if (Value *V = SimplifyInstruction(I, DL, TLI, DT, AC)) {
2233 I->replaceAllUsesWith(V);
2234 if (MD && V->getType()->getScalarType()->isPointerTy())
2235 MD->invalidateCachedPointerInfo(V);
2236 markInstructionForDeletion(I);
2237 ++NumGVNSimpl;
2238 return true;
2239 }
2240
2241 if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
2242 if (processLoad(LI))
2243 return true;
2244
2245 unsigned Num = VN.lookup_or_add(LI);
2246 addToLeaderTable(Num, LI, LI->getParent());
2247 return false;
2248 }
2249
2250 // For conditional branches, we can perform simple conditional propagation on
2251 // the condition value itself.
2252 if (BranchInst *BI = dyn_cast<BranchInst>(I)) {
2253 if (!BI->isConditional())
2254 return false;
2255
2256 if (isa<Constant>(BI->getCondition()))
2257 return processFoldableCondBr(BI);
2258
2259 Value *BranchCond = BI->getCondition();
2260 BasicBlock *TrueSucc = BI->getSuccessor(0);
2261 BasicBlock *FalseSucc = BI->getSuccessor(1);
2262 // Avoid multiple edges early.
2263 if (TrueSucc == FalseSucc)
2264 return false;
2265
2266 BasicBlock *Parent = BI->getParent();
2267 bool Changed = false;
2268
2269 Value *TrueVal = ConstantInt::getTrue(TrueSucc->getContext());
2270 BasicBlockEdge TrueE(Parent, TrueSucc);
2271 Changed |= propagateEquality(BranchCond, TrueVal, TrueE);
2272
2273 Value *FalseVal = ConstantInt::getFalse(FalseSucc->getContext());
2274 BasicBlockEdge FalseE(Parent, FalseSucc);
2275 Changed |= propagateEquality(BranchCond, FalseVal, FalseE);
2276
2277 return Changed;
2278 }
2279
2280 // For switches, propagate the case values into the case destinations.
2281 if (SwitchInst *SI = dyn_cast<SwitchInst>(I)) {
2282 Value *SwitchCond = SI->getCondition();
2283 BasicBlock *Parent = SI->getParent();
2284 bool Changed = false;
2285
2286 // Remember how many outgoing edges there are to every successor.
2287 SmallDenseMap<BasicBlock *, unsigned, 16> SwitchEdges;
2288 for (unsigned i = 0, n = SI->getNumSuccessors(); i != n; ++i)
2289 ++SwitchEdges[SI->getSuccessor(i)];
2290
2291 for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
2292 i != e; ++i) {
2293 BasicBlock *Dst = i.getCaseSuccessor();
2294 // If there is only a single edge, propagate the case value into it.
2295 if (SwitchEdges.lookup(Dst) == 1) {
2296 BasicBlockEdge E(Parent, Dst);
2297 Changed |= propagateEquality(SwitchCond, i.getCaseValue(), E);
2298 }
2299 }
2300 return Changed;
2301 }
2302
2303 // Instructions with void type don't return a value, so there's
2304 // no point in trying to find redundancies in them.
2305 if (I->getType()->isVoidTy()) return false;
2306
2307 uint32_t NextNum = VN.getNextUnusedValueNumber();
2308 unsigned Num = VN.lookup_or_add(I);
2309
2310 // Allocations are always uniquely numbered, so we can save time and memory
2311 // by fast failing them.
2312 if (isa<AllocaInst>(I) || isa<TerminatorInst>(I) || isa<PHINode>(I)) {
2313 addToLeaderTable(Num, I, I->getParent());
2314 return false;
2315 }
2316
2317 // If the number we were assigned was a brand new VN, then we don't
2318 // need to do a lookup to see if the number already exists
2319 // somewhere in the domtree: it can't!
2320 if (Num >= NextNum) {
2321 addToLeaderTable(Num, I, I->getParent());
2322 return false;
2323 }
2324
2325 // Perform fast-path value-number based elimination of values inherited from
2326 // dominators.
2327 Value *repl = findLeader(I->getParent(), Num);
2328 if (!repl) {
2329 // Failure, just remember this instance for future use.
2330 addToLeaderTable(Num, I, I->getParent());
2331 return false;
2332 }
2333
2334 // Remove it!
2335 patchAndReplaceAllUsesWith(I, repl);
2336 if (MD && repl->getType()->getScalarType()->isPointerTy())
2337 MD->invalidateCachedPointerInfo(repl);
2338 markInstructionForDeletion(I);
2339 return true;
2340 }
2341
2342 /// runOnFunction - This is the main transformation entry point for a function.
runOnFunction(Function & F)2343 bool GVN::runOnFunction(Function& F) {
2344 if (skipOptnoneFunction(F))
2345 return false;
2346
2347 if (!NoLoads)
2348 MD = &getAnalysis<MemoryDependenceAnalysis>();
2349 DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
2350 AC = &getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F);
2351 TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
2352 VN.setAliasAnalysis(&getAnalysis<AliasAnalysis>());
2353 VN.setMemDep(MD);
2354 VN.setDomTree(DT);
2355
2356 bool Changed = false;
2357 bool ShouldContinue = true;
2358
2359 // Merge unconditional branches, allowing PRE to catch more
2360 // optimization opportunities.
2361 for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; ) {
2362 BasicBlock *BB = FI++;
2363
2364 bool removedBlock = MergeBlockIntoPredecessor(
2365 BB, DT, /* LoopInfo */ nullptr, VN.getAliasAnalysis(), MD);
2366 if (removedBlock) ++NumGVNBlocks;
2367
2368 Changed |= removedBlock;
2369 }
2370
2371 unsigned Iteration = 0;
2372 while (ShouldContinue) {
2373 DEBUG(dbgs() << "GVN iteration: " << Iteration << "\n");
2374 ShouldContinue = iterateOnFunction(F);
2375 Changed |= ShouldContinue;
2376 ++Iteration;
2377 }
2378
2379 if (EnablePRE) {
2380 // Fabricate val-num for dead-code in order to suppress assertion in
2381 // performPRE().
2382 assignValNumForDeadCode();
2383 bool PREChanged = true;
2384 while (PREChanged) {
2385 PREChanged = performPRE(F);
2386 Changed |= PREChanged;
2387 }
2388 }
2389
2390 // FIXME: Should perform GVN again after PRE does something. PRE can move
2391 // computations into blocks where they become fully redundant. Note that
2392 // we can't do this until PRE's critical edge splitting updates memdep.
2393 // Actually, when this happens, we should just fully integrate PRE into GVN.
2394
2395 cleanupGlobalSets();
2396 // Do not cleanup DeadBlocks in cleanupGlobalSets() as it's called for each
2397 // iteration.
2398 DeadBlocks.clear();
2399
2400 return Changed;
2401 }
2402
2403
processBlock(BasicBlock * BB)2404 bool GVN::processBlock(BasicBlock *BB) {
2405 // FIXME: Kill off InstrsToErase by doing erasing eagerly in a helper function
2406 // (and incrementing BI before processing an instruction).
2407 assert(InstrsToErase.empty() &&
2408 "We expect InstrsToErase to be empty across iterations");
2409 if (DeadBlocks.count(BB))
2410 return false;
2411
2412 bool ChangedFunction = false;
2413
2414 for (BasicBlock::iterator BI = BB->begin(), BE = BB->end();
2415 BI != BE;) {
2416 ChangedFunction |= processInstruction(BI);
2417 if (InstrsToErase.empty()) {
2418 ++BI;
2419 continue;
2420 }
2421
2422 // If we need some instructions deleted, do it now.
2423 NumGVNInstr += InstrsToErase.size();
2424
2425 // Avoid iterator invalidation.
2426 bool AtStart = BI == BB->begin();
2427 if (!AtStart)
2428 --BI;
2429
2430 for (SmallVectorImpl<Instruction *>::iterator I = InstrsToErase.begin(),
2431 E = InstrsToErase.end(); I != E; ++I) {
2432 DEBUG(dbgs() << "GVN removed: " << **I << '\n');
2433 if (MD) MD->removeInstruction(*I);
2434 DEBUG(verifyRemoved(*I));
2435 (*I)->eraseFromParent();
2436 }
2437 InstrsToErase.clear();
2438
2439 if (AtStart)
2440 BI = BB->begin();
2441 else
2442 ++BI;
2443 }
2444
2445 return ChangedFunction;
2446 }
2447
2448 // Instantiate an expression in a predecessor that lacked it.
performScalarPREInsertion(Instruction * Instr,BasicBlock * Pred,unsigned int ValNo)2449 bool GVN::performScalarPREInsertion(Instruction *Instr, BasicBlock *Pred,
2450 unsigned int ValNo) {
2451 // Because we are going top-down through the block, all value numbers
2452 // will be available in the predecessor by the time we need them. Any
2453 // that weren't originally present will have been instantiated earlier
2454 // in this loop.
2455 bool success = true;
2456 for (unsigned i = 0, e = Instr->getNumOperands(); i != e; ++i) {
2457 Value *Op = Instr->getOperand(i);
2458 if (isa<Argument>(Op) || isa<Constant>(Op) || isa<GlobalValue>(Op))
2459 continue;
2460
2461 if (Value *V = findLeader(Pred, VN.lookup(Op))) {
2462 Instr->setOperand(i, V);
2463 } else {
2464 success = false;
2465 break;
2466 }
2467 }
2468
2469 // Fail out if we encounter an operand that is not available in
2470 // the PRE predecessor. This is typically because of loads which
2471 // are not value numbered precisely.
2472 if (!success)
2473 return false;
2474
2475 Instr->insertBefore(Pred->getTerminator());
2476 Instr->setName(Instr->getName() + ".pre");
2477 Instr->setDebugLoc(Instr->getDebugLoc());
2478 VN.add(Instr, ValNo);
2479
2480 // Update the availability map to include the new instruction.
2481 addToLeaderTable(ValNo, Instr, Pred);
2482 return true;
2483 }
2484
performScalarPRE(Instruction * CurInst)2485 bool GVN::performScalarPRE(Instruction *CurInst) {
2486 SmallVector<std::pair<Value*, BasicBlock*>, 8> predMap;
2487
2488 if (isa<AllocaInst>(CurInst) || isa<TerminatorInst>(CurInst) ||
2489 isa<PHINode>(CurInst) || CurInst->getType()->isVoidTy() ||
2490 CurInst->mayReadFromMemory() || CurInst->mayHaveSideEffects() ||
2491 isa<DbgInfoIntrinsic>(CurInst))
2492 return false;
2493
2494 // Don't do PRE on compares. The PHI would prevent CodeGenPrepare from
2495 // sinking the compare again, and it would force the code generator to
2496 // move the i1 from processor flags or predicate registers into a general
2497 // purpose register.
2498 if (isa<CmpInst>(CurInst))
2499 return false;
2500
2501 // We don't currently value number ANY inline asm calls.
2502 if (CallInst *CallI = dyn_cast<CallInst>(CurInst))
2503 if (CallI->isInlineAsm())
2504 return false;
2505
2506 uint32_t ValNo = VN.lookup(CurInst);
2507
2508 // Look for the predecessors for PRE opportunities. We're
2509 // only trying to solve the basic diamond case, where
2510 // a value is computed in the successor and one predecessor,
2511 // but not the other. We also explicitly disallow cases
2512 // where the successor is its own predecessor, because they're
2513 // more complicated to get right.
2514 unsigned NumWith = 0;
2515 unsigned NumWithout = 0;
2516 BasicBlock *PREPred = nullptr;
2517 BasicBlock *CurrentBlock = CurInst->getParent();
2518 predMap.clear();
2519
2520 for (pred_iterator PI = pred_begin(CurrentBlock), PE = pred_end(CurrentBlock);
2521 PI != PE; ++PI) {
2522 BasicBlock *P = *PI;
2523 // We're not interested in PRE where the block is its
2524 // own predecessor, or in blocks with predecessors
2525 // that are not reachable.
2526 if (P == CurrentBlock) {
2527 NumWithout = 2;
2528 break;
2529 } else if (!DT->isReachableFromEntry(P)) {
2530 NumWithout = 2;
2531 break;
2532 }
2533
2534 Value *predV = findLeader(P, ValNo);
2535 if (!predV) {
2536 predMap.push_back(std::make_pair(static_cast<Value *>(nullptr), P));
2537 PREPred = P;
2538 ++NumWithout;
2539 } else if (predV == CurInst) {
2540 /* CurInst dominates this predecessor. */
2541 NumWithout = 2;
2542 break;
2543 } else {
2544 predMap.push_back(std::make_pair(predV, P));
2545 ++NumWith;
2546 }
2547 }
2548
2549 // Don't do PRE when it might increase code size, i.e. when
2550 // we would need to insert instructions in more than one pred.
2551 if (NumWithout > 1 || NumWith == 0)
2552 return false;
2553
2554 // We may have a case where all predecessors have the instruction,
2555 // and we just need to insert a phi node. Otherwise, perform
2556 // insertion.
2557 Instruction *PREInstr = nullptr;
2558
2559 if (NumWithout != 0) {
2560 // Don't do PRE across indirect branch.
2561 if (isa<IndirectBrInst>(PREPred->getTerminator()))
2562 return false;
2563
2564 // We can't do PRE safely on a critical edge, so instead we schedule
2565 // the edge to be split and perform the PRE the next time we iterate
2566 // on the function.
2567 unsigned SuccNum = GetSuccessorNumber(PREPred, CurrentBlock);
2568 if (isCriticalEdge(PREPred->getTerminator(), SuccNum)) {
2569 toSplit.push_back(std::make_pair(PREPred->getTerminator(), SuccNum));
2570 return false;
2571 }
2572 // We need to insert somewhere, so let's give it a shot
2573 PREInstr = CurInst->clone();
2574 if (!performScalarPREInsertion(PREInstr, PREPred, ValNo)) {
2575 // If we failed insertion, make sure we remove the instruction.
2576 DEBUG(verifyRemoved(PREInstr));
2577 delete PREInstr;
2578 return false;
2579 }
2580 }
2581
2582 // Either we should have filled in the PRE instruction, or we should
2583 // not have needed insertions.
2584 assert (PREInstr != nullptr || NumWithout == 0);
2585
2586 ++NumGVNPRE;
2587
2588 // Create a PHI to make the value available in this block.
2589 PHINode *Phi =
2590 PHINode::Create(CurInst->getType(), predMap.size(),
2591 CurInst->getName() + ".pre-phi", CurrentBlock->begin());
2592 for (unsigned i = 0, e = predMap.size(); i != e; ++i) {
2593 if (Value *V = predMap[i].first)
2594 Phi->addIncoming(V, predMap[i].second);
2595 else
2596 Phi->addIncoming(PREInstr, PREPred);
2597 }
2598
2599 VN.add(Phi, ValNo);
2600 addToLeaderTable(ValNo, Phi, CurrentBlock);
2601 Phi->setDebugLoc(CurInst->getDebugLoc());
2602 CurInst->replaceAllUsesWith(Phi);
2603 if (Phi->getType()->getScalarType()->isPointerTy()) {
2604 // Because we have added a PHI-use of the pointer value, it has now
2605 // "escaped" from alias analysis' perspective. We need to inform
2606 // AA of this.
2607 for (unsigned ii = 0, ee = Phi->getNumIncomingValues(); ii != ee; ++ii) {
2608 unsigned jj = PHINode::getOperandNumForIncomingValue(ii);
2609 VN.getAliasAnalysis()->addEscapingUse(Phi->getOperandUse(jj));
2610 }
2611
2612 if (MD)
2613 MD->invalidateCachedPointerInfo(Phi);
2614 }
2615 VN.erase(CurInst);
2616 removeFromLeaderTable(ValNo, CurInst, CurrentBlock);
2617
2618 DEBUG(dbgs() << "GVN PRE removed: " << *CurInst << '\n');
2619 if (MD)
2620 MD->removeInstruction(CurInst);
2621 DEBUG(verifyRemoved(CurInst));
2622 CurInst->eraseFromParent();
2623 ++NumGVNInstr;
2624
2625 return true;
2626 }
2627
2628 /// Perform a purely local form of PRE that looks for diamond
2629 /// control flow patterns and attempts to perform simple PRE at the join point.
performPRE(Function & F)2630 bool GVN::performPRE(Function &F) {
2631 bool Changed = false;
2632 for (BasicBlock *CurrentBlock : depth_first(&F.getEntryBlock())) {
2633 // Nothing to PRE in the entry block.
2634 if (CurrentBlock == &F.getEntryBlock())
2635 continue;
2636
2637 // Don't perform PRE on a landing pad.
2638 if (CurrentBlock->isLandingPad())
2639 continue;
2640
2641 for (BasicBlock::iterator BI = CurrentBlock->begin(),
2642 BE = CurrentBlock->end();
2643 BI != BE;) {
2644 Instruction *CurInst = BI++;
2645 Changed = performScalarPRE(CurInst);
2646 }
2647 }
2648
2649 if (splitCriticalEdges())
2650 Changed = true;
2651
2652 return Changed;
2653 }
2654
2655 /// Split the critical edge connecting the given two blocks, and return
2656 /// the block inserted to the critical edge.
splitCriticalEdges(BasicBlock * Pred,BasicBlock * Succ)2657 BasicBlock *GVN::splitCriticalEdges(BasicBlock *Pred, BasicBlock *Succ) {
2658 BasicBlock *BB = SplitCriticalEdge(
2659 Pred, Succ, CriticalEdgeSplittingOptions(getAliasAnalysis(), DT));
2660 if (MD)
2661 MD->invalidateCachedPredecessors();
2662 return BB;
2663 }
2664
2665 /// Split critical edges found during the previous
2666 /// iteration that may enable further optimization.
splitCriticalEdges()2667 bool GVN::splitCriticalEdges() {
2668 if (toSplit.empty())
2669 return false;
2670 do {
2671 std::pair<TerminatorInst*, unsigned> Edge = toSplit.pop_back_val();
2672 SplitCriticalEdge(Edge.first, Edge.second,
2673 CriticalEdgeSplittingOptions(getAliasAnalysis(), DT));
2674 } while (!toSplit.empty());
2675 if (MD) MD->invalidateCachedPredecessors();
2676 return true;
2677 }
2678
2679 /// Executes one iteration of GVN
iterateOnFunction(Function & F)2680 bool GVN::iterateOnFunction(Function &F) {
2681 cleanupGlobalSets();
2682
2683 // Top-down walk of the dominator tree
2684 bool Changed = false;
2685 // Save the blocks this function have before transformation begins. GVN may
2686 // split critical edge, and hence may invalidate the RPO/DT iterator.
2687 //
2688 std::vector<BasicBlock *> BBVect;
2689 BBVect.reserve(256);
2690 // Needed for value numbering with phi construction to work.
2691 ReversePostOrderTraversal<Function *> RPOT(&F);
2692 for (ReversePostOrderTraversal<Function *>::rpo_iterator RI = RPOT.begin(),
2693 RE = RPOT.end();
2694 RI != RE; ++RI)
2695 BBVect.push_back(*RI);
2696
2697 for (std::vector<BasicBlock *>::iterator I = BBVect.begin(), E = BBVect.end();
2698 I != E; I++)
2699 Changed |= processBlock(*I);
2700
2701 return Changed;
2702 }
2703
cleanupGlobalSets()2704 void GVN::cleanupGlobalSets() {
2705 VN.clear();
2706 LeaderTable.clear();
2707 TableAllocator.Reset();
2708 }
2709
2710 /// Verify that the specified instruction does not occur in our
2711 /// internal data structures.
verifyRemoved(const Instruction * Inst) const2712 void GVN::verifyRemoved(const Instruction *Inst) const {
2713 VN.verifyRemoved(Inst);
2714
2715 // Walk through the value number scope to make sure the instruction isn't
2716 // ferreted away in it.
2717 for (DenseMap<uint32_t, LeaderTableEntry>::const_iterator
2718 I = LeaderTable.begin(), E = LeaderTable.end(); I != E; ++I) {
2719 const LeaderTableEntry *Node = &I->second;
2720 assert(Node->Val != Inst && "Inst still in value numbering scope!");
2721
2722 while (Node->Next) {
2723 Node = Node->Next;
2724 assert(Node->Val != Inst && "Inst still in value numbering scope!");
2725 }
2726 }
2727 }
2728
2729 /// BB is declared dead, which implied other blocks become dead as well. This
2730 /// function is to add all these blocks to "DeadBlocks". For the dead blocks'
2731 /// live successors, update their phi nodes by replacing the operands
2732 /// corresponding to dead blocks with UndefVal.
addDeadBlock(BasicBlock * BB)2733 void GVN::addDeadBlock(BasicBlock *BB) {
2734 SmallVector<BasicBlock *, 4> NewDead;
2735 SmallSetVector<BasicBlock *, 4> DF;
2736
2737 NewDead.push_back(BB);
2738 while (!NewDead.empty()) {
2739 BasicBlock *D = NewDead.pop_back_val();
2740 if (DeadBlocks.count(D))
2741 continue;
2742
2743 // All blocks dominated by D are dead.
2744 SmallVector<BasicBlock *, 8> Dom;
2745 DT->getDescendants(D, Dom);
2746 DeadBlocks.insert(Dom.begin(), Dom.end());
2747
2748 // Figure out the dominance-frontier(D).
2749 for (SmallVectorImpl<BasicBlock *>::iterator I = Dom.begin(),
2750 E = Dom.end(); I != E; I++) {
2751 BasicBlock *B = *I;
2752 for (succ_iterator SI = succ_begin(B), SE = succ_end(B); SI != SE; SI++) {
2753 BasicBlock *S = *SI;
2754 if (DeadBlocks.count(S))
2755 continue;
2756
2757 bool AllPredDead = true;
2758 for (pred_iterator PI = pred_begin(S), PE = pred_end(S); PI != PE; PI++)
2759 if (!DeadBlocks.count(*PI)) {
2760 AllPredDead = false;
2761 break;
2762 }
2763
2764 if (!AllPredDead) {
2765 // S could be proved dead later on. That is why we don't update phi
2766 // operands at this moment.
2767 DF.insert(S);
2768 } else {
2769 // While S is not dominated by D, it is dead by now. This could take
2770 // place if S already have a dead predecessor before D is declared
2771 // dead.
2772 NewDead.push_back(S);
2773 }
2774 }
2775 }
2776 }
2777
2778 // For the dead blocks' live successors, update their phi nodes by replacing
2779 // the operands corresponding to dead blocks with UndefVal.
2780 for(SmallSetVector<BasicBlock *, 4>::iterator I = DF.begin(), E = DF.end();
2781 I != E; I++) {
2782 BasicBlock *B = *I;
2783 if (DeadBlocks.count(B))
2784 continue;
2785
2786 SmallVector<BasicBlock *, 4> Preds(pred_begin(B), pred_end(B));
2787 for (SmallVectorImpl<BasicBlock *>::iterator PI = Preds.begin(),
2788 PE = Preds.end(); PI != PE; PI++) {
2789 BasicBlock *P = *PI;
2790
2791 if (!DeadBlocks.count(P))
2792 continue;
2793
2794 if (isCriticalEdge(P->getTerminator(), GetSuccessorNumber(P, B))) {
2795 if (BasicBlock *S = splitCriticalEdges(P, B))
2796 DeadBlocks.insert(P = S);
2797 }
2798
2799 for (BasicBlock::iterator II = B->begin(); isa<PHINode>(II); ++II) {
2800 PHINode &Phi = cast<PHINode>(*II);
2801 Phi.setIncomingValue(Phi.getBasicBlockIndex(P),
2802 UndefValue::get(Phi.getType()));
2803 }
2804 }
2805 }
2806 }
2807
2808 // If the given branch is recognized as a foldable branch (i.e. conditional
2809 // branch with constant condition), it will perform following analyses and
2810 // transformation.
2811 // 1) If the dead out-coming edge is a critical-edge, split it. Let
2812 // R be the target of the dead out-coming edge.
2813 // 1) Identify the set of dead blocks implied by the branch's dead outcoming
2814 // edge. The result of this step will be {X| X is dominated by R}
2815 // 2) Identify those blocks which haves at least one dead prodecessor. The
2816 // result of this step will be dominance-frontier(R).
2817 // 3) Update the PHIs in DF(R) by replacing the operands corresponding to
2818 // dead blocks with "UndefVal" in an hope these PHIs will optimized away.
2819 //
2820 // Return true iff *NEW* dead code are found.
processFoldableCondBr(BranchInst * BI)2821 bool GVN::processFoldableCondBr(BranchInst *BI) {
2822 if (!BI || BI->isUnconditional())
2823 return false;
2824
2825 ConstantInt *Cond = dyn_cast<ConstantInt>(BI->getCondition());
2826 if (!Cond)
2827 return false;
2828
2829 BasicBlock *DeadRoot = Cond->getZExtValue() ?
2830 BI->getSuccessor(1) : BI->getSuccessor(0);
2831 if (DeadBlocks.count(DeadRoot))
2832 return false;
2833
2834 if (!DeadRoot->getSinglePredecessor())
2835 DeadRoot = splitCriticalEdges(BI->getParent(), DeadRoot);
2836
2837 addDeadBlock(DeadRoot);
2838 return true;
2839 }
2840
2841 // performPRE() will trigger assert if it comes across an instruction without
2842 // associated val-num. As it normally has far more live instructions than dead
2843 // instructions, it makes more sense just to "fabricate" a val-number for the
2844 // dead code than checking if instruction involved is dead or not.
assignValNumForDeadCode()2845 void GVN::assignValNumForDeadCode() {
2846 for (SetVector<BasicBlock *>::iterator I = DeadBlocks.begin(),
2847 E = DeadBlocks.end(); I != E; I++) {
2848 BasicBlock *BB = *I;
2849 for (BasicBlock::iterator II = BB->begin(), EE = BB->end();
2850 II != EE; II++) {
2851 Instruction *Inst = &*II;
2852 unsigned ValNum = VN.lookup_or_add(Inst);
2853 addToLeaderTable(ValNum, Inst, BB);
2854 }
2855 }
2856 }
2857