1//===- CloneFunction.cpp - Clone a function into another function ---------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file implements the CloneFunctionInto interface, which is used as the
10// low-level function cloner. This is used by the CloneFunction and function
11// inliner to do the dirty work of copying the body of a function around.
12//
13//===----------------------------------------------------------------------===//
14
15#include "llvm/ADT/SmallVector.h"
16#include "llvm/ADT/SmallVectorExtras.h"
17#include "llvm/ADT/Statistic.h"
18#include "llvm/Analysis/ConstantFolding.h"
19#include "llvm/Analysis/DomTreeUpdater.h"
20#include "llvm/Analysis/InstructionSimplify.h"
21#include "llvm/Analysis/LoopInfo.h"
22#include "llvm/IR/AttributeMask.h"
23#include "llvm/IR/CFG.h"
24#include "llvm/IR/Constants.h"
25#include "llvm/IR/DebugInfo.h"
26#include "llvm/IR/DerivedTypes.h"
27#include "llvm/IR/Function.h"
28#include "llvm/IR/InstIterator.h"
29#include "llvm/IR/Instructions.h"
30#include "llvm/IR/IntrinsicInst.h"
31#include "llvm/IR/LLVMContext.h"
32#include "llvm/IR/MDBuilder.h"
33#include "llvm/IR/Metadata.h"
34#include "llvm/IR/Module.h"
35#include "llvm/Transforms/Utils/BasicBlockUtils.h"
36#include "llvm/Transforms/Utils/Cloning.h"
37#include "llvm/Transforms/Utils/Local.h"
38#include "llvm/Transforms/Utils/ValueMapper.h"
39#include <cstdint>
40#include <map>
41#include <optional>
42using namespace llvm;
43
44#define DEBUG_TYPE "clone-function"
45
46STATISTIC(RemappedAtomMax, "Highest global NextAtomGroup (after mapping)");
47
48void llvm::mapAtomInstance(const DebugLoc &DL, ValueToValueMapTy &VMap) {
49 uint64_t CurGroup = DL->getAtomGroup();
50 if (!CurGroup)
51 return;
52
53 // Try inserting a new entry. If there's already a mapping for this atom
54 // then there's nothing to do.
55 auto [It, Inserted] = VMap.AtomMap.insert(KV: {{DL.getInlinedAt(), CurGroup}, 0});
56 if (!Inserted)
57 return;
58
59 // Map entry to a new atom group.
60 uint64_t NewGroup = DL->getContext().incNextDILocationAtomGroup();
61 assert(NewGroup > CurGroup && "Next should always be greater than current");
62 It->second = NewGroup;
63
64 RemappedAtomMax = std::max<uint64_t>(a: NewGroup, b: RemappedAtomMax);
65}
66
67static void collectDebugInfoFromInstructions(const Function &F,
68 DebugInfoFinder &DIFinder) {
69 const Module *M = F.getParent();
70 if (!M)
71 return;
72 // Inspect instructions to process e.g. DILexicalBlocks of inlined functions
73 for (const Instruction &I : instructions(F))
74 DIFinder.processInstruction(M: *M, I);
75}
76
77// Create a predicate that matches the metadata that should be identity mapped
78// during function cloning.
79static MetadataPredicate
80createIdentityMDPredicate(const Function &F, CloneFunctionChangeType Changes) {
81 if (Changes >= CloneFunctionChangeType::DifferentModule)
82 return [](const Metadata *MD) { return false; };
83
84 DISubprogram *SPClonedWithinModule = F.getSubprogram();
85
86 // Don't clone inlined subprograms.
87 auto ShouldKeep = [SPClonedWithinModule](const DISubprogram *SP) -> bool {
88 return SP != SPClonedWithinModule;
89 };
90
91 return [=](const Metadata *MD) {
92 // Avoid cloning compile units.
93 if (isa<DICompileUnit>(Val: MD))
94 return true;
95
96 if (auto *SP = dyn_cast<DISubprogram>(Val: MD))
97 return ShouldKeep(SP);
98
99 // If a subprogram isn't going to be cloned skip its lexical blocks as well.
100 if (auto *LScope = dyn_cast<DILocalScope>(Val: MD))
101 return ShouldKeep(LScope->getSubprogram());
102
103 // Avoid cloning local variables of subprograms that won't be cloned.
104 if (auto *DV = dyn_cast<DILocalVariable>(Val: MD))
105 if (auto *S = dyn_cast_or_null<DILocalScope>(Val: DV->getScope()))
106 return ShouldKeep(S->getSubprogram());
107
108 // DIGlobalVariableExpression representing static local variable may be
109 // encountered in DISubprogram's retainedNodes list. Do not remap it, and
110 // remove it from retainedNodes after mapping.
111 if (isa<DIGlobalVariableExpression>(Val: MD))
112 return true;
113
114 // Clone types that are local to subprograms being cloned.
115 // Avoid cloning other types.
116 auto *Type = dyn_cast<DIType>(Val: MD);
117 if (!Type)
118 return false;
119
120 // No need to clone types if subprograms are not cloned.
121 if (SPClonedWithinModule == nullptr)
122 return true;
123
124 // Scopeless types may be derived from local types (e.g. pointers to local
125 // types). They may need cloning.
126 if (const DIDerivedType *DTy = dyn_cast_or_null<DIDerivedType>(Val: Type);
127 DTy && !DTy->getScope())
128 return false;
129
130 auto *LScope = dyn_cast_or_null<DILocalScope>(Val: Type->getScope());
131 if (!LScope)
132 return true;
133
134 if (ShouldKeep(LScope->getSubprogram()))
135 return true;
136
137 return false;
138 };
139}
140
141/// See comments in Cloning.h.
142BasicBlock *llvm::CloneBasicBlock(const BasicBlock *BB, ValueToValueMapTy &VMap,
143 const Twine &NameSuffix, Function *F,
144 ClonedCodeInfo *CodeInfo, bool MapAtoms) {
145 BasicBlock *NewBB = BasicBlock::Create(Context&: BB->getContext(), Name: "", Parent: F);
146 if (BB->hasName())
147 NewBB->setName(BB->getName() + NameSuffix);
148
149 bool hasCalls = false, hasDynamicAllocas = false, hasMemProfMetadata = false;
150
151 // Loop over all instructions, and copy them over.
152 for (const Instruction &I : *BB) {
153 Instruction *NewInst = I.clone();
154 if (I.hasName())
155 NewInst->setName(I.getName() + NameSuffix);
156
157 NewInst->insertBefore(BB&: *NewBB, InsertPos: NewBB->end());
158 NewInst->cloneDebugInfoFrom(From: &I);
159
160 VMap[&I] = NewInst; // Add instruction map to value.
161
162 if (MapAtoms) {
163 if (const DebugLoc &DL = NewInst->getDebugLoc())
164 mapAtomInstance(DL: DL.get(), VMap);
165 }
166
167 if (isa<CallInst>(Val: I) && !I.isDebugOrPseudoInst()) {
168 hasCalls = true;
169 hasMemProfMetadata |= I.hasMetadata(KindID: LLVMContext::MD_memprof);
170 hasMemProfMetadata |= I.hasMetadata(KindID: LLVMContext::MD_callsite);
171 }
172 if (const AllocaInst *AI = dyn_cast<AllocaInst>(Val: &I)) {
173 if (!AI->isStaticAlloca()) {
174 hasDynamicAllocas = true;
175 }
176 }
177 }
178
179 if (CodeInfo) {
180 CodeInfo->ContainsCalls |= hasCalls;
181 CodeInfo->ContainsMemProfMetadata |= hasMemProfMetadata;
182 CodeInfo->ContainsDynamicAllocas |= hasDynamicAllocas;
183 }
184 return NewBB;
185}
186
187void llvm::CloneFunctionAttributesInto(Function *NewFunc,
188 const Function *OldFunc,
189 ValueToValueMapTy &VMap,
190 bool ModuleLevelChanges,
191 ValueMapTypeRemapper *TypeMapper,
192 ValueMaterializer *Materializer) {
193 // Copy all attributes other than those stored in Function's AttributeList
194 // which holds e.g. parameters and return value attributes.
195 AttributeList NewAttrs = NewFunc->getAttributes();
196 NewFunc->copyAttributesFrom(Src: OldFunc);
197 NewFunc->setAttributes(NewAttrs);
198
199 const RemapFlags FuncGlobalRefFlags =
200 ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges;
201
202 // Fix up the personality function that got copied over.
203 if (OldFunc->hasPersonalityFn())
204 NewFunc->setPersonalityFn(MapValue(V: OldFunc->getPersonalityFn(), VM&: VMap,
205 Flags: FuncGlobalRefFlags, TypeMapper,
206 Materializer));
207
208 if (OldFunc->hasPrefixData()) {
209 NewFunc->setPrefixData(MapValue(V: OldFunc->getPrefixData(), VM&: VMap,
210 Flags: FuncGlobalRefFlags, TypeMapper,
211 Materializer));
212 }
213
214 if (OldFunc->hasPrologueData()) {
215 NewFunc->setPrologueData(MapValue(V: OldFunc->getPrologueData(), VM&: VMap,
216 Flags: FuncGlobalRefFlags, TypeMapper,
217 Materializer));
218 }
219
220 SmallVector<AttributeSet, 4> NewArgAttrs(NewFunc->arg_size());
221 AttributeList OldAttrs = OldFunc->getAttributes();
222
223 // Clone any argument attributes that are present in the VMap.
224 for (const Argument &OldArg : OldFunc->args()) {
225 if (Argument *NewArg = dyn_cast<Argument>(Val&: VMap[&OldArg])) {
226 // Remap the parameter indices.
227 NewArgAttrs[NewArg->getArgNo()] =
228 OldAttrs.getParamAttrs(ArgNo: OldArg.getArgNo());
229 }
230 }
231
232 NewFunc->setAttributes(
233 AttributeList::get(C&: NewFunc->getContext(), FnAttrs: OldAttrs.getFnAttrs(),
234 RetAttrs: OldAttrs.getRetAttrs(), ArgAttrs: NewArgAttrs));
235}
236
237void llvm::CloneFunctionMetadataInto(Function &NewFunc, const Function &OldFunc,
238 ValueToValueMapTy &VMap,
239 RemapFlags RemapFlag,
240 ValueMapTypeRemapper *TypeMapper,
241 ValueMaterializer *Materializer,
242 const MetadataPredicate *IdentityMD) {
243 SmallVector<std::pair<unsigned, MDNode *>, 1> MDs;
244 OldFunc.getAllMetadata(MDs);
245 for (const auto &[Kind, MD] : MDs) {
246 NewFunc.addMetadata(KindID: Kind, MD&: *MapMetadata(MD, VM&: VMap, Flags: RemapFlag, TypeMapper,
247 Materializer, IdentityMD));
248 }
249}
250
251void llvm::CloneFunctionBodyInto(Function &NewFunc, const Function &OldFunc,
252 ValueToValueMapTy &VMap, RemapFlags RemapFlag,
253 SmallVectorImpl<ReturnInst *> &Returns,
254 const char *NameSuffix,
255 ClonedCodeInfo *CodeInfo,
256 ValueMapTypeRemapper *TypeMapper,
257 ValueMaterializer *Materializer,
258 const MetadataPredicate *IdentityMD) {
259 if (OldFunc.isDeclaration())
260 return;
261
262 // Loop over all of the basic blocks in the function, cloning them as
263 // appropriate. Note that we save BE this way in order to handle cloning of
264 // recursive functions into themselves.
265 for (const BasicBlock &BB : OldFunc) {
266 // Create a new basic block and copy instructions into it!
267 BasicBlock *CBB =
268 CloneBasicBlock(BB: &BB, VMap, NameSuffix, F: &NewFunc, CodeInfo);
269
270 // Add basic block mapping.
271 VMap[&BB] = CBB;
272
273 // It is only legal to clone a function if a block address within that
274 // function is never referenced outside of the function. Given that, we
275 // want to map block addresses from the old function to block addresses in
276 // the clone. (This is different from the generic ValueMapper
277 // implementation, which generates an invalid blockaddress when
278 // cloning a function.)
279 if (BB.hasAddressTaken()) {
280 Constant *OldBBAddr = BlockAddress::get(F: const_cast<Function *>(&OldFunc),
281 BB: const_cast<BasicBlock *>(&BB));
282 VMap[OldBBAddr] = BlockAddress::get(F: &NewFunc, BB: CBB);
283 }
284
285 // Note return instructions for the caller.
286 if (ReturnInst *RI = dyn_cast<ReturnInst>(Val: CBB->getTerminator()))
287 Returns.push_back(Elt: RI);
288 }
289
290 // Loop over all of the instructions in the new function, fixing up operand
291 // references as we go. This uses VMap to do all the hard work.
292 for (Function::iterator
293 BB = cast<BasicBlock>(Val&: VMap[&OldFunc.front()])->getIterator(),
294 BE = NewFunc.end();
295 BB != BE; ++BB)
296 // Loop over all instructions, fixing each one as we find it, and any
297 // attached debug-info records.
298 for (Instruction &II : *BB) {
299 RemapInstruction(I: &II, VM&: VMap, Flags: RemapFlag, TypeMapper, Materializer,
300 IdentityMD);
301 RemapDbgRecordRange(M: II.getModule(), Range: II.getDbgRecordRange(), VM&: VMap,
302 Flags: RemapFlag, TypeMapper, Materializer, IdentityMD);
303 }
304}
305
306// Clone OldFunc into NewFunc, transforming the old arguments into references to
307// VMap values.
308void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc,
309 ValueToValueMapTy &VMap,
310 CloneFunctionChangeType Changes,
311 SmallVectorImpl<ReturnInst *> &Returns,
312 const char *NameSuffix, ClonedCodeInfo *CodeInfo,
313 ValueMapTypeRemapper *TypeMapper,
314 ValueMaterializer *Materializer) {
315 assert(NameSuffix && "NameSuffix cannot be null!");
316
317#ifndef NDEBUG
318 for (const Argument &I : OldFunc->args())
319 assert(VMap.count(&I) && "No mapping from source argument specified!");
320#endif
321
322 bool ModuleLevelChanges = Changes > CloneFunctionChangeType::LocalChangesOnly;
323
324 CloneFunctionAttributesInto(NewFunc, OldFunc, VMap, ModuleLevelChanges,
325 TypeMapper, Materializer);
326
327 // Everything else beyond this point deals with function instructions,
328 // so if we are dealing with a function declaration, we're done.
329 if (OldFunc->isDeclaration())
330 return;
331
332 if (Changes < CloneFunctionChangeType::DifferentModule) {
333 assert((NewFunc->getParent() == nullptr ||
334 NewFunc->getParent() == OldFunc->getParent()) &&
335 "Expected NewFunc to have the same parent, or no parent");
336 } else {
337 assert((NewFunc->getParent() == nullptr ||
338 NewFunc->getParent() != OldFunc->getParent()) &&
339 "Expected NewFunc to have different parents, or no parent");
340
341 if (Changes == CloneFunctionChangeType::DifferentModule) {
342 assert(NewFunc->getParent() &&
343 "Need parent of new function to maintain debug info invariants");
344 }
345 }
346
347 MetadataPredicate IdentityMD = createIdentityMDPredicate(F: *OldFunc, Changes);
348
349 // Cloning is always a Module level operation, since Metadata needs to be
350 // cloned.
351 const RemapFlags RemapFlag = RF_None;
352
353 CloneFunctionMetadataInto(NewFunc&: *NewFunc, OldFunc: *OldFunc, VMap, RemapFlag, TypeMapper,
354 Materializer, IdentityMD: &IdentityMD);
355
356 CloneFunctionBodyInto(NewFunc&: *NewFunc, OldFunc: *OldFunc, VMap, RemapFlag, Returns,
357 NameSuffix, CodeInfo, TypeMapper, Materializer,
358 IdentityMD: &IdentityMD);
359
360 // DIGlobalVariableExpressions representing static locals stay in the scope of
361 // OldSP after function cloning. Remove them from retainedNodes of NewSP.
362 if (DISubprogram *NewSP = NewFunc->getSubprogram())
363 NewSP->cleanupRetainedNodesIf(Pred: [NewSP](Metadata *N) {
364 auto *GVE = dyn_cast_or_null<DIGlobalVariableExpression>(Val: N);
365 return !GVE ||
366 DISubprogram::getRetainedNodeScope(N: GVE)->getSubprogram() == NewSP;
367 });
368
369 // Only update !llvm.dbg.cu for DifferentModule (not CloneModule). In the
370 // same module, the compile unit will already be listed (or not). When
371 // cloning a module, CloneModule() will handle creating the named metadata.
372 if (Changes != CloneFunctionChangeType::DifferentModule)
373 return;
374
375 // Update !llvm.dbg.cu with compile units added to the new module if this
376 // function is being cloned in isolation.
377 //
378 // FIXME: This is making global / module-level changes, which doesn't seem
379 // like the right encapsulation Consider dropping the requirement to update
380 // !llvm.dbg.cu (either obsoleting the node, or restricting it to
381 // non-discardable compile units) instead of discovering compile units by
382 // visiting the metadata attached to global values, which would allow this
383 // code to be deleted. Alternatively, perhaps give responsibility for this
384 // update to CloneFunctionInto's callers.
385 Module *NewModule = NewFunc->getParent();
386 NamedMDNode *NMD = NewModule->getOrInsertNamedMetadata(Name: "llvm.dbg.cu");
387 // Avoid multiple insertions of the same DICompileUnit to NMD.
388 SmallPtrSet<const void *, 8> Visited(llvm::from_range, NMD->operands());
389
390 // Collect and clone all the compile units referenced from the instructions in
391 // the function (e.g. as instructions' scope).
392 DebugInfoFinder DIFinder;
393 collectDebugInfoFromInstructions(F: *OldFunc, DIFinder);
394 for (DICompileUnit *Unit : DIFinder.compile_units()) {
395 MDNode *MappedUnit =
396 MapMetadata(MD: Unit, VM&: VMap, Flags: RF_None, TypeMapper, Materializer);
397 if (Visited.insert(Ptr: MappedUnit).second)
398 NMD->addOperand(M: MappedUnit);
399 }
400}
401
402/// Return a copy of the specified function and add it to that function's
403/// module. Also, any references specified in the VMap are changed to refer to
404/// their mapped value instead of the original one. If any of the arguments to
405/// the function are in the VMap, the arguments are deleted from the resultant
406/// function. The VMap is updated to include mappings from all of the
407/// instructions and basicblocks in the function from their old to new values.
408///
409Function *llvm::CloneFunction(Function *F, ValueToValueMapTy &VMap,
410 ClonedCodeInfo *CodeInfo) {
411 std::vector<Type *> ArgTypes;
412
413 // The user might be deleting arguments to the function by specifying them in
414 // the VMap. If so, we need to not add the arguments to the arg ty vector
415 //
416 for (const Argument &I : F->args())
417 if (VMap.count(Val: &I) == 0) // Haven't mapped the argument to anything yet?
418 ArgTypes.push_back(x: I.getType());
419
420 // Create a new function type...
421 FunctionType *FTy =
422 FunctionType::get(Result: F->getFunctionType()->getReturnType(), Params: ArgTypes,
423 isVarArg: F->getFunctionType()->isVarArg());
424
425 // Create the new function...
426 Function *NewF = Function::Create(Ty: FTy, Linkage: F->getLinkage(), AddrSpace: F->getAddressSpace(),
427 N: F->getName(), M: F->getParent());
428
429 // Loop over the arguments, copying the names of the mapped arguments over...
430 Function::arg_iterator DestI = NewF->arg_begin();
431 for (const Argument &I : F->args())
432 if (VMap.count(Val: &I) == 0) { // Is this argument preserved?
433 DestI->setName(I.getName()); // Copy the name over...
434 VMap[&I] = &*DestI++; // Add mapping to VMap
435 }
436
437 SmallVector<ReturnInst *, 8> Returns; // Ignore returns cloned.
438 CloneFunctionInto(NewFunc: NewF, OldFunc: F, VMap, Changes: CloneFunctionChangeType::LocalChangesOnly,
439 Returns, NameSuffix: "", CodeInfo);
440
441 return NewF;
442}
443
444namespace {
445/// This is a private class used to implement CloneAndPruneFunctionInto.
446struct PruningFunctionCloner {
447 Function *NewFunc;
448 const Function *OldFunc;
449 ValueToValueMapTy &VMap;
450 bool ModuleLevelChanges;
451 const char *NameSuffix;
452 ClonedCodeInfo &CodeInfo;
453 bool HostFuncIsStrictFP;
454
455 Instruction *cloneInstruction(BasicBlock::const_iterator II);
456
457public:
458 PruningFunctionCloner(Function *newFunc, const Function *oldFunc,
459 ValueToValueMapTy &valueMap, bool moduleLevelChanges,
460 const char *nameSuffix, ClonedCodeInfo &codeInfo)
461 : NewFunc(newFunc), OldFunc(oldFunc), VMap(valueMap),
462 ModuleLevelChanges(moduleLevelChanges), NameSuffix(nameSuffix),
463 CodeInfo(codeInfo) {
464 HostFuncIsStrictFP =
465 newFunc->getAttributes().hasFnAttr(Kind: Attribute::StrictFP);
466 }
467
468 /// The specified block is found to be reachable, clone it and
469 /// anything that it can reach.
470 void CloneBlock(const BasicBlock *BB, BasicBlock::const_iterator StartingInst,
471 std::vector<const BasicBlock *> &ToClone);
472};
473} // namespace
474
475Instruction *
476PruningFunctionCloner::cloneInstruction(BasicBlock::const_iterator II) {
477 if (!HostFuncIsStrictFP)
478 return II->clone();
479
480 const Instruction &OldInst = *II;
481 Intrinsic::ID CIID = getConstrainedIntrinsicID(Instr: OldInst);
482 if (CIID == Intrinsic::not_intrinsic)
483 return II->clone();
484
485 // Instead of cloning the instruction, a call to constrained intrinsic should
486 // be created. Assume the first arguments of constrained intrinsics are the
487 // same as the operands of original instruction.
488
489 // Create intrinsic call.
490 LLVMContext &Ctx = NewFunc->getContext();
491 SmallVector<Value *, 4> Args;
492 unsigned NumOperands = OldInst.getNumOperands();
493 if (isa<CallInst>(Val: OldInst))
494 --NumOperands;
495 for (unsigned I = 0; I < NumOperands; ++I)
496 Args.push_back(Elt: OldInst.getOperand(i: I));
497
498 if (const auto *CmpI = dyn_cast<FCmpInst>(Val: &OldInst)) {
499 FCmpInst::Predicate Pred = CmpI->getPredicate();
500 StringRef PredName = FCmpInst::getPredicateName(P: Pred);
501 Args.push_back(Elt: MetadataAsValue::get(Context&: Ctx, MD: MDString::get(Context&: Ctx, Str: PredName)));
502 }
503
504 // The last arguments of a constrained intrinsic are metadata that represent
505 // rounding mode (absent in some intrinsics) and exception behavior. The
506 // inlined function uses default settings.
507 if (Intrinsic::hasConstrainedFPRoundingModeOperand(QID: CIID))
508 Args.push_back(
509 Elt: MetadataAsValue::get(Context&: Ctx, MD: MDString::get(Context&: Ctx, Str: "round.tonearest")));
510 Args.push_back(
511 Elt: MetadataAsValue::get(Context&: Ctx, MD: MDString::get(Context&: Ctx, Str: "fpexcept.ignore")));
512
513 SmallVector<Type *> ArgTys = llvm::map_to_vector(C&: Args, F: &Value::getType);
514 Function *IFn = Intrinsic::getOrInsertDeclaration(M: NewFunc->getParent(), IID: CIID,
515 RetTy: OldInst.getType(), ArgTys);
516 return CallInst::Create(Func: IFn, Args, NameStr: OldInst.getName() + ".strict");
517}
518
519/// The specified block is found to be reachable, clone it and
520/// anything that it can reach.
521void PruningFunctionCloner::CloneBlock(
522 const BasicBlock *BB, BasicBlock::const_iterator StartingInst,
523 std::vector<const BasicBlock *> &ToClone) {
524 WeakTrackingVH &BBEntry = VMap[BB];
525
526 // Have we already cloned this block?
527 if (BBEntry)
528 return;
529
530 // Nope, clone it now.
531 BasicBlock *NewBB;
532 Twine NewName(BB->hasName() ? Twine(BB->getName()) + NameSuffix : "");
533 BBEntry = NewBB = BasicBlock::Create(Context&: BB->getContext(), Name: NewName, Parent: NewFunc);
534
535 // It is only legal to clone a function if a block address within that
536 // function is never referenced outside of the function. Given that, we
537 // want to map block addresses from the old function to block addresses in
538 // the clone. (This is different from the generic ValueMapper
539 // implementation, which generates an invalid blockaddress when
540 // cloning a function.)
541 //
542 // Note that we don't need to fix the mapping for unreachable blocks;
543 // the default mapping there is safe.
544 if (BB->hasAddressTaken()) {
545 Constant *OldBBAddr = BlockAddress::get(F: const_cast<Function *>(OldFunc),
546 BB: const_cast<BasicBlock *>(BB));
547 VMap[OldBBAddr] = BlockAddress::get(F: NewFunc, BB: NewBB);
548 }
549
550 bool hasCalls = false, hasDynamicAllocas = false, hasStaticAllocas = false;
551 bool hasMemProfMetadata = false;
552
553 // Keep a cursor pointing at the last place we cloned debug-info records from.
554 BasicBlock::const_iterator DbgCursor = StartingInst;
555 auto CloneDbgRecordsToHere =
556 [&DbgCursor](Instruction *NewInst, BasicBlock::const_iterator II) {
557 // Clone debug-info records onto this instruction. Iterate through any
558 // source-instructions we've cloned and then subsequently optimised
559 // away, so that their debug-info doesn't go missing.
560 for (; DbgCursor != II; ++DbgCursor)
561 NewInst->cloneDebugInfoFrom(From: &*DbgCursor, FromHere: std::nullopt, InsertAtHead: false);
562 NewInst->cloneDebugInfoFrom(From: &*II);
563 DbgCursor = std::next(x: II);
564 };
565
566 // Loop over all instructions, and copy them over, DCE'ing as we go. This
567 // loop doesn't include the terminator.
568 for (BasicBlock::const_iterator II = StartingInst, IE = --BB->end(); II != IE;
569 ++II) {
570
571 // Don't clone fake_use as it may suppress many optimizations
572 // due to inlining, especially SROA.
573 if (auto *IntrInst = dyn_cast<IntrinsicInst>(Val&: II))
574 if (IntrInst->getIntrinsicID() == Intrinsic::fake_use)
575 continue;
576
577 Instruction *NewInst = cloneInstruction(II);
578 NewInst->insertInto(ParentBB: NewBB, It: NewBB->end());
579
580 if (HostFuncIsStrictFP) {
581 // All function calls in the inlined function must get 'strictfp'
582 // attribute to prevent undesirable optimizations.
583 if (auto *Call = dyn_cast<CallInst>(Val: NewInst))
584 Call->addFnAttr(Kind: Attribute::StrictFP);
585 }
586
587 // Eagerly remap operands to the newly cloned instruction, except for PHI
588 // nodes for which we defer processing until we update the CFG.
589 if (!isa<PHINode>(Val: NewInst)) {
590 RemapInstruction(I: NewInst, VM&: VMap,
591 Flags: ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges);
592
593 // Eagerly constant fold the newly cloned instruction. If successful, add
594 // a mapping to the new value. Non-constant operands may be incomplete at
595 // this stage, thus instruction simplification is performed after
596 // processing phi-nodes.
597 if (Value *V = ConstantFoldInstruction(
598 I: NewInst, DL: BB->getDataLayout())) {
599 if (isInstructionTriviallyDead(I: NewInst)) {
600 VMap[&*II] = V;
601 NewInst->eraseFromParent();
602 continue;
603 }
604 }
605 }
606
607 if (auto *CB = dyn_cast<CallBase>(Val&: II); CB && CB->isIndirectCall())
608 CodeInfo.OriginallyIndirectCalls.insert(X: NewInst);
609
610 if (II->hasName())
611 NewInst->setName(II->getName() + NameSuffix);
612 VMap[&*II] = NewInst; // Add instruction map to value.
613 if (isa<CallInst>(Val: II) && !II->isDebugOrPseudoInst()) {
614 hasCalls = true;
615 hasMemProfMetadata |= II->hasMetadata(KindID: LLVMContext::MD_memprof);
616 hasMemProfMetadata |= II->hasMetadata(KindID: LLVMContext::MD_callsite);
617 }
618
619 CloneDbgRecordsToHere(NewInst, II);
620
621 CodeInfo.OrigVMap[&*II] = NewInst;
622 if (auto *CB = dyn_cast<CallBase>(Val: &*II))
623 if (CB->hasOperandBundles())
624 CodeInfo.OperandBundleCallSites.push_back(x: NewInst);
625
626 if (const AllocaInst *AI = dyn_cast<AllocaInst>(Val&: II)) {
627 if (isa<ConstantInt>(Val: AI->getArraySize()))
628 hasStaticAllocas = true;
629 else
630 hasDynamicAllocas = true;
631 }
632 }
633
634 // Finally, clone over the terminator.
635 const Instruction *OldTI = BB->getTerminator();
636 bool TerminatorDone = false;
637 if (const CondBrInst *BI = dyn_cast<CondBrInst>(Val: OldTI)) {
638 // If the condition was a known constant in the callee...
639 ConstantInt *Cond = dyn_cast<ConstantInt>(Val: BI->getCondition());
640 // Or is a known constant in the caller...
641 if (!Cond) {
642 Value *V = VMap.lookup(Val: BI->getCondition());
643 Cond = dyn_cast_or_null<ConstantInt>(Val: V);
644 }
645
646 // Constant fold to uncond branch!
647 if (Cond) {
648 BasicBlock *Dest = BI->getSuccessor(i: !Cond->getZExtValue());
649 auto *NewBI = UncondBrInst::Create(Target: Dest, InsertBefore: NewBB);
650 NewBI->setDebugLoc(BI->getDebugLoc());
651 VMap[OldTI] = NewBI;
652 ToClone.push_back(x: Dest);
653 TerminatorDone = true;
654 }
655 } else if (const SwitchInst *SI = dyn_cast<SwitchInst>(Val: OldTI)) {
656 // If switching on a value known constant in the caller.
657 ConstantInt *Cond = dyn_cast<ConstantInt>(Val: SI->getCondition());
658 if (!Cond) { // Or known constant after constant prop in the callee...
659 Value *V = VMap.lookup(Val: SI->getCondition());
660 Cond = dyn_cast_or_null<ConstantInt>(Val: V);
661 }
662 if (Cond) { // Constant fold to uncond branch!
663 SwitchInst::ConstCaseHandle Case = *SI->findCaseValue(C: Cond);
664 BasicBlock *Dest = const_cast<BasicBlock *>(Case.getCaseSuccessor());
665 auto *NewBI = UncondBrInst::Create(Target: Dest, InsertBefore: NewBB);
666 NewBI->setDebugLoc(SI->getDebugLoc());
667 VMap[OldTI] = NewBI;
668 ToClone.push_back(x: Dest);
669 TerminatorDone = true;
670 }
671 }
672
673 if (!TerminatorDone) {
674 Instruction *NewInst = OldTI->clone();
675 if (OldTI->hasName())
676 NewInst->setName(OldTI->getName() + NameSuffix);
677 NewInst->insertInto(ParentBB: NewBB, It: NewBB->end());
678
679 CloneDbgRecordsToHere(NewInst, OldTI->getIterator());
680
681 VMap[OldTI] = NewInst; // Add instruction map to value.
682
683 CodeInfo.OrigVMap[OldTI] = NewInst;
684 if (auto *CB = dyn_cast<CallBase>(Val: OldTI))
685 if (CB->hasOperandBundles())
686 CodeInfo.OperandBundleCallSites.push_back(x: NewInst);
687
688 // Recursively clone any reachable successor blocks.
689 append_range(C&: ToClone, R: successors(I: BB->getTerminator()));
690 } else {
691 // If we didn't create a new terminator, clone DbgVariableRecords from the
692 // old terminator onto the new terminator.
693 Instruction *NewInst = NewBB->getTerminator();
694 assert(NewInst);
695
696 CloneDbgRecordsToHere(NewInst, OldTI->getIterator());
697 }
698
699 CodeInfo.ContainsCalls |= hasCalls;
700 CodeInfo.ContainsMemProfMetadata |= hasMemProfMetadata;
701 CodeInfo.ContainsDynamicAllocas |= hasDynamicAllocas;
702 CodeInfo.ContainsDynamicAllocas |=
703 hasStaticAllocas && BB != &BB->getParent()->front();
704}
705
706/// This works like CloneAndPruneFunctionInto, except that it does not clone the
707/// entire function. Instead it starts at an instruction provided by the caller
708/// and copies (and prunes) only the code reachable from that instruction.
709void llvm::CloneAndPruneIntoFromInst(Function *NewFunc, const Function *OldFunc,
710 const Instruction *StartingInst,
711 ValueToValueMapTy &VMap,
712 bool ModuleLevelChanges,
713 SmallVectorImpl<ReturnInst *> &Returns,
714 const char *NameSuffix,
715 ClonedCodeInfo &CodeInfo) {
716 assert(NameSuffix && "NameSuffix cannot be null!");
717
718 ValueMapTypeRemapper *TypeMapper = nullptr;
719 ValueMaterializer *Materializer = nullptr;
720
721#ifndef NDEBUG
722 // If the cloning starts at the beginning of the function, verify that
723 // the function arguments are mapped.
724 if (!StartingInst)
725 for (const Argument &II : OldFunc->args())
726 assert(VMap.count(&II) && "No mapping from source argument specified!");
727#endif
728
729 PruningFunctionCloner PFC(NewFunc, OldFunc, VMap, ModuleLevelChanges,
730 NameSuffix, CodeInfo);
731 const BasicBlock *StartingBB;
732 if (StartingInst)
733 StartingBB = StartingInst->getParent();
734 else {
735 StartingBB = &OldFunc->getEntryBlock();
736 StartingInst = &StartingBB->front();
737 }
738
739 // Clone the entry block, and anything recursively reachable from it.
740 std::vector<const BasicBlock *> CloneWorklist;
741 PFC.CloneBlock(BB: StartingBB, StartingInst: StartingInst->getIterator(), ToClone&: CloneWorklist);
742 while (!CloneWorklist.empty()) {
743 const BasicBlock *BB = CloneWorklist.back();
744 CloneWorklist.pop_back();
745 PFC.CloneBlock(BB, StartingInst: BB->begin(), ToClone&: CloneWorklist);
746 }
747
748 // Loop over all of the basic blocks in the old function. If the block was
749 // reachable, we have cloned it and the old block is now in the value map:
750 // insert it into the new function in the right order. If not, ignore it.
751 //
752 // Defer PHI resolution until rest of function is resolved.
753 SmallVector<const PHINode *, 16> PHIToResolve;
754 for (const BasicBlock &BI : *OldFunc) {
755 Value *V = VMap.lookup(Val: &BI);
756 BasicBlock *NewBB = cast_or_null<BasicBlock>(Val: V);
757 if (!NewBB)
758 continue; // Dead block.
759
760 // Move the new block to preserve the order in the original function.
761 NewBB->moveBefore(MovePos: NewFunc->end());
762
763 // Handle PHI nodes specially, as we have to remove references to dead
764 // blocks.
765 for (const PHINode &PN : BI.phis()) {
766 // PHI nodes may have been remapped to non-PHI nodes by the caller or
767 // during the cloning process.
768 if (isa<PHINode>(Val: VMap[&PN]))
769 PHIToResolve.push_back(Elt: &PN);
770 else
771 break;
772 }
773
774 // Finally, remap the terminator instructions, as those can't be remapped
775 // until all BBs are mapped.
776 RemapInstruction(I: NewBB->getTerminator(), VM&: VMap,
777 Flags: ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges,
778 TypeMapper, Materializer);
779 }
780
781 // Defer PHI resolution until rest of function is resolved, PHI resolution
782 // requires the CFG to be up-to-date.
783 for (unsigned phino = 0, e = PHIToResolve.size(); phino != e;) {
784 const PHINode *OPN = PHIToResolve[phino];
785 unsigned NumPreds = OPN->getNumIncomingValues();
786 const BasicBlock *OldBB = OPN->getParent();
787 BasicBlock *NewBB = cast<BasicBlock>(Val&: VMap[OldBB]);
788
789 // Map operands for blocks that are live and remove operands for blocks
790 // that are dead.
791 for (; phino != PHIToResolve.size() &&
792 PHIToResolve[phino]->getParent() == OldBB;
793 ++phino) {
794 OPN = PHIToResolve[phino];
795 PHINode *PN = cast<PHINode>(Val&: VMap[OPN]);
796 for (int64_t pred = NumPreds - 1; pred >= 0; --pred) {
797 Value *V = VMap.lookup(Val: PN->getIncomingBlock(i: pred));
798 if (BasicBlock *MappedBlock = cast_or_null<BasicBlock>(Val: V)) {
799 Value *InVal =
800 MapValue(V: PN->getIncomingValue(i: pred), VM&: VMap,
801 Flags: ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges);
802 assert(InVal && "Unknown input value?");
803 PN->setIncomingValue(i: pred, V: InVal);
804 PN->setIncomingBlock(i: pred, BB: MappedBlock);
805 continue;
806 }
807 PN->removeIncomingValue(Idx: pred, DeletePHIIfEmpty: false);
808 }
809 }
810
811 // The loop above has removed PHI entries for those blocks that are dead
812 // and has updated others. However, if a block is live (i.e. copied over)
813 // but its terminator has been changed to not go to this block, then our
814 // phi nodes will have invalid entries. Update the PHI nodes in this
815 // case.
816 PHINode *PN = cast<PHINode>(Val: NewBB->begin());
817 NumPreds = pred_size(BB: NewBB);
818 if (NumPreds != PN->getNumIncomingValues()) {
819 assert(NumPreds < PN->getNumIncomingValues());
820 // Count how many times each predecessor comes to this block.
821 DenseMap<BasicBlock *, unsigned> PredCount;
822 for (BasicBlock *Pred : predecessors(BB: NewBB))
823 ++PredCount[Pred];
824
825 BasicBlock::iterator I = NewBB->begin();
826 DenseMap<BasicBlock *, unsigned> SeenPredCount;
827 SeenPredCount.reserve(NumEntries: PredCount.size());
828 for (; (PN = dyn_cast<PHINode>(Val&: I)); ++I) {
829 SeenPredCount.clear();
830 PN->removeIncomingValueIf(
831 Predicate: [&](unsigned Idx) {
832 BasicBlock *IncomingBlock = PN->getIncomingBlock(i: Idx);
833 auto It = PredCount.find(Val: IncomingBlock);
834 if (It == PredCount.end())
835 return true;
836 unsigned &SeenCount = SeenPredCount[IncomingBlock];
837 if (SeenCount < It->second) {
838 SeenCount++;
839 return false;
840 }
841 return true;
842 },
843 DeletePHIIfEmpty: false);
844 }
845 }
846
847 // If the loops above have made these phi nodes have 0 or 1 operand,
848 // replace them with poison or the input value. We must do this for
849 // correctness, because 0-operand phis are not valid.
850 PN = cast<PHINode>(Val: NewBB->begin());
851 if (PN->getNumIncomingValues() == 0) {
852 BasicBlock::iterator I = NewBB->begin();
853 BasicBlock::const_iterator OldI = OldBB->begin();
854 while ((PN = dyn_cast<PHINode>(Val: I++))) {
855 Value *NV = PoisonValue::get(T: PN->getType());
856 PN->replaceAllUsesWith(V: NV);
857 assert(VMap[&*OldI] == PN && "VMap mismatch");
858 VMap[&*OldI] = NV;
859 PN->eraseFromParent();
860 ++OldI;
861 }
862 }
863 }
864
865 // Drop all incompatible return attributes that cannot be applied to NewFunc
866 // during cloning, so as to allow instruction simplification to reason on the
867 // old state of the function. The original attributes are restored later.
868 AttributeList Attrs = NewFunc->getAttributes();
869 AttributeMask IncompatibleAttrs = AttributeFuncs::typeIncompatible(
870 Ty: OldFunc->getReturnType(), AS: Attrs.getRetAttrs());
871 NewFunc->removeRetAttrs(Attrs: IncompatibleAttrs);
872
873 // As phi-nodes have been now remapped, allow incremental simplification of
874 // newly-cloned instructions.
875 const DataLayout &DL = NewFunc->getDataLayout();
876 for (const BasicBlock &BB : *OldFunc) {
877 for (const Instruction &I : BB) {
878 auto *NewI = dyn_cast_or_null<Instruction>(Val: VMap.lookup(Val: &I));
879 if (!NewI)
880 continue;
881
882 if (Value *V = simplifyInstruction(I: NewI, Q: DL)) {
883 NewI->replaceAllUsesWith(V);
884
885 if (isInstructionTriviallyDead(I: NewI)) {
886 NewI->eraseFromParent();
887 } else {
888 // Did not erase it? Restore the new instruction into VMap previously
889 // dropped by `ValueIsRAUWd`.
890 VMap[&I] = NewI;
891 }
892 }
893 }
894 }
895
896 // Restore attributes.
897 NewFunc->setAttributes(Attrs);
898
899 // Remap debug records operands now that all values have been mapped.
900 // Doing this now (late) preserves use-before-defs in debug records. If
901 // we didn't do this, ValueAsMetadata(use-before-def) operands would be
902 // replaced by empty metadata. This would signal later cleanup passes to
903 // remove the debug records, potentially causing incorrect locations.
904 Function::iterator Begin = cast<BasicBlock>(Val&: VMap[StartingBB])->getIterator();
905 for (BasicBlock &BB : make_range(x: Begin, y: NewFunc->end())) {
906 for (Instruction &I : BB) {
907 RemapDbgRecordRange(M: I.getModule(), Range: I.getDbgRecordRange(), VM&: VMap,
908 Flags: ModuleLevelChanges ? RF_None
909 : RF_NoModuleLevelChanges,
910 TypeMapper, Materializer);
911 }
912 }
913
914 // Simplify conditional branches and switches with a constant operand. We try
915 // to prune these out when cloning, but if the simplification required
916 // looking through PHI nodes, those are only available after forming the full
917 // basic block. That may leave some here, and we still want to prune the dead
918 // code as early as possible.
919 for (BasicBlock &BB : make_range(x: Begin, y: NewFunc->end()))
920 ConstantFoldTerminator(BB: &BB);
921
922 // Some blocks may have become unreachable as a result. Find and delete them.
923 {
924 SmallPtrSet<BasicBlock *, 16> ReachableBlocks;
925 SmallVector<BasicBlock *, 16> Worklist;
926 Worklist.push_back(Elt: &*Begin);
927 while (!Worklist.empty()) {
928 BasicBlock *BB = Worklist.pop_back_val();
929 if (ReachableBlocks.insert(Ptr: BB).second)
930 append_range(C&: Worklist, R: successors(BB));
931 }
932
933 SmallVector<BasicBlock *, 16> UnreachableBlocks;
934 for (BasicBlock &BB : make_range(x: Begin, y: NewFunc->end()))
935 if (!ReachableBlocks.contains(Ptr: &BB))
936 UnreachableBlocks.push_back(Elt: &BB);
937 DeleteDeadBlocks(BBs: UnreachableBlocks);
938 }
939
940 // Now that the inlined function body has been fully constructed, go through
941 // and zap unconditional fall-through branches. This happens all the time when
942 // specializing code: code specialization turns conditional branches into
943 // uncond branches, and this code folds them.
944 Function::iterator I = Begin;
945 while (I != NewFunc->end()) {
946 UncondBrInst *BI = dyn_cast<UncondBrInst>(Val: I->getTerminator());
947 if (!BI) {
948 ++I;
949 continue;
950 }
951
952 BasicBlock *Dest = BI->getSuccessor();
953 if (!Dest->getSinglePredecessor() || Dest->hasAddressTaken()) {
954 ++I;
955 continue;
956 }
957
958 // We shouldn't be able to get single-entry PHI nodes here, as instsimplify
959 // above should have zapped all of them..
960 assert(!isa<PHINode>(Dest->begin()));
961
962 // We know all single-entry PHI nodes in the inlined function have been
963 // removed, so we just need to splice the blocks.
964 BI->eraseFromParent();
965
966 // Make all PHI nodes that referred to Dest now refer to I as their source.
967 Dest->replaceAllUsesWith(V: &*I);
968
969 // Move all the instructions in the succ to the pred.
970 I->splice(ToIt: I->end(), FromBB: Dest);
971
972 // Remove the dest block.
973 Dest->eraseFromParent();
974
975 // Do not increment I, iteratively merge all things this block branches to.
976 }
977
978 // Make a final pass over the basic blocks from the old function to gather
979 // any return instructions which survived folding. We have to do this here
980 // because we can iteratively remove and merge returns above.
981 for (Function::iterator I = cast<BasicBlock>(Val&: VMap[StartingBB])->getIterator(),
982 E = NewFunc->end();
983 I != E; ++I)
984 if (ReturnInst *RI = dyn_cast<ReturnInst>(Val: I->getTerminator()))
985 Returns.push_back(Elt: RI);
986}
987
988/// This works exactly like CloneFunctionInto,
989/// except that it does some simple constant prop and DCE on the fly. The
990/// effect of this is to copy significantly less code in cases where (for
991/// example) a function call with constant arguments is inlined, and those
992/// constant arguments cause a significant amount of code in the callee to be
993/// dead. Since this doesn't produce an exact copy of the input, it can't be
994/// used for things like CloneFunction or CloneModule.
995void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc,
996 ValueToValueMapTy &VMap,
997 bool ModuleLevelChanges,
998 SmallVectorImpl<ReturnInst *> &Returns,
999 const char *NameSuffix,
1000 ClonedCodeInfo &CodeInfo) {
1001 CloneAndPruneIntoFromInst(NewFunc, OldFunc, StartingInst: &OldFunc->front().front(), VMap,
1002 ModuleLevelChanges, Returns, NameSuffix, CodeInfo);
1003}
1004
1005/// Remaps instructions in \p Blocks using the mapping in \p VMap.
1006void llvm::remapInstructionsInBlocks(ArrayRef<BasicBlock *> Blocks,
1007 ValueToValueMapTy &VMap) {
1008 // Rewrite the code to refer to itself.
1009 for (BasicBlock *BB : Blocks) {
1010 for (Instruction &Inst : *BB) {
1011 RemapDbgRecordRange(M: Inst.getModule(), Range: Inst.getDbgRecordRange(), VM&: VMap,
1012 Flags: RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
1013 RemapInstruction(I: &Inst, VM&: VMap,
1014 Flags: RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
1015 }
1016 }
1017}
1018
1019/// Clones a loop \p OrigLoop. Returns the loop and the blocks in \p
1020/// Blocks.
1021///
1022/// Updates LoopInfo and DominatorTree assuming the loop is dominated by block
1023/// \p LoopDomBB. Insert the new blocks before block specified in \p Before.
1024/// The client needs to further update the CFG and DominatorTree after calling
1025/// this function, to ensure the IR remains valid.
1026Loop *llvm::cloneLoopWithPreheader(BasicBlock *Before, BasicBlock *LoopDomBB,
1027 Loop *OrigLoop, ValueToValueMapTy &VMap,
1028 const Twine &NameSuffix, LoopInfo *LI,
1029 DominatorTree *DT,
1030 SmallVectorImpl<BasicBlock *> &Blocks) {
1031 Function *F = OrigLoop->getHeader()->getParent();
1032 Loop *ParentLoop = OrigLoop->getParentLoop();
1033 DenseMap<Loop *, Loop *> LMap;
1034
1035 Loop *NewLoop = LI->AllocateLoop();
1036 LMap[OrigLoop] = NewLoop;
1037 if (ParentLoop)
1038 ParentLoop->addChildLoop(NewChild: NewLoop);
1039 else
1040 LI->addTopLevelLoop(New: NewLoop);
1041
1042 BasicBlock *OrigPH = OrigLoop->getLoopPreheader();
1043 assert(OrigPH && "No preheader");
1044 BasicBlock *NewPH = CloneBasicBlock(BB: OrigPH, VMap, NameSuffix, F);
1045 // To rename the loop PHIs.
1046 VMap[OrigPH] = NewPH;
1047 Blocks.push_back(Elt: NewPH);
1048
1049 // Update LoopInfo.
1050 if (ParentLoop)
1051 ParentLoop->addBasicBlockToLoop(NewBB: NewPH, LI&: *LI);
1052
1053 // Update DominatorTree.
1054 DT->addNewBlock(BB: NewPH, DomBB: LoopDomBB);
1055
1056 for (Loop *CurLoop : OrigLoop->getLoopsInPreorder()) {
1057 Loop *&NewLoop = LMap[CurLoop];
1058 if (!NewLoop) {
1059 NewLoop = LI->AllocateLoop();
1060
1061 // Establish the parent/child relationship.
1062 Loop *OrigParent = CurLoop->getParentLoop();
1063 assert(OrigParent && "Could not find the original parent loop");
1064 Loop *NewParentLoop = LMap[OrigParent];
1065 assert(NewParentLoop && "Could not find the new parent loop");
1066
1067 NewParentLoop->addChildLoop(NewChild: NewLoop);
1068 }
1069 }
1070
1071 for (BasicBlock *BB : OrigLoop->getBlocks()) {
1072 Loop *CurLoop = LI->getLoopFor(BB);
1073 Loop *&NewLoop = LMap[CurLoop];
1074 assert(NewLoop && "Expecting new loop to be allocated");
1075
1076 BasicBlock *NewBB = CloneBasicBlock(BB, VMap, NameSuffix, F);
1077 VMap[BB] = NewBB;
1078
1079 // Update LoopInfo.
1080 NewLoop->addBasicBlockToLoop(NewBB, LI&: *LI);
1081
1082 // Add DominatorTree node. After seeing all blocks, update to correct
1083 // IDom.
1084 DT->addNewBlock(BB: NewBB, DomBB: NewPH);
1085
1086 Blocks.push_back(Elt: NewBB);
1087 }
1088
1089 for (BasicBlock *BB : OrigLoop->getBlocks()) {
1090 // Update loop headers.
1091 Loop *CurLoop = LI->getLoopFor(BB);
1092 if (BB == CurLoop->getHeader())
1093 LMap[CurLoop]->moveToHeader(BB: cast<BasicBlock>(Val&: VMap[BB]));
1094
1095 // Update DominatorTree.
1096 BasicBlock *IDomBB = DT->getNode(BB)->getIDom()->getBlock();
1097 DT->changeImmediateDominator(BB: cast<BasicBlock>(Val&: VMap[BB]),
1098 NewBB: cast<BasicBlock>(Val&: VMap[IDomBB]));
1099 }
1100
1101 // Move them physically from the end of the block list.
1102 F->splice(ToIt: Before->getIterator(), FromF: F, FromIt: NewPH->getIterator());
1103 F->splice(ToIt: Before->getIterator(), FromF: F, FromBeginIt: NewLoop->getHeader()->getIterator(),
1104 FromEndIt: F->end());
1105
1106 return NewLoop;
1107}
1108
1109/// Duplicate non-Phi instructions from the beginning of block up to
1110/// StopAt instruction into a split block between BB and its predecessor.
1111BasicBlock *llvm::DuplicateInstructionsInSplitBetween(
1112 BasicBlock *BB, BasicBlock *PredBB, Instruction *StopAt,
1113 ValueToValueMapTy &ValueMapping, DomTreeUpdater &DTU) {
1114
1115 assert(count(successors(PredBB), BB) == 1 &&
1116 "There must be a single edge between PredBB and BB!");
1117 // We are going to have to map operands from the original BB block to the new
1118 // copy of the block 'NewBB'. If there are PHI nodes in BB, evaluate them to
1119 // account for entry from PredBB.
1120 BasicBlock::iterator BI = BB->begin();
1121 for (; PHINode *PN = dyn_cast<PHINode>(Val&: BI); ++BI)
1122 ValueMapping[PN] = PN->getIncomingValueForBlock(BB: PredBB);
1123
1124 BasicBlock *NewBB = SplitEdge(From: PredBB, To: BB);
1125 NewBB->setName(PredBB->getName() + ".split");
1126 Instruction *NewTerm = NewBB->getTerminator();
1127
1128 // FIXME: SplitEdge does not yet take a DTU, so we include the split edge
1129 // in the update set here.
1130 DTU.applyUpdates(Updates: {{DominatorTree::Delete, PredBB, BB},
1131 {DominatorTree::Insert, PredBB, NewBB},
1132 {DominatorTree::Insert, NewBB, BB}});
1133
1134 // Clone the non-phi instructions of BB into NewBB, keeping track of the
1135 // mapping and using it to remap operands in the cloned instructions.
1136 // Stop once we see the terminator too. This covers the case where BB's
1137 // terminator gets replaced and StopAt == BB's terminator.
1138 for (; StopAt != &*BI && BB->getTerminator() != &*BI; ++BI) {
1139 Instruction *New = BI->clone();
1140 New->setName(BI->getName());
1141 New->insertBefore(InsertPos: NewTerm->getIterator());
1142 New->cloneDebugInfoFrom(From: &*BI);
1143 ValueMapping[&*BI] = New;
1144
1145 // Remap operands to patch up intra-block references.
1146 for (unsigned i = 0, e = New->getNumOperands(); i != e; ++i)
1147 if (Instruction *Inst = dyn_cast<Instruction>(Val: New->getOperand(i))) {
1148 auto I = ValueMapping.find(Val: Inst);
1149 if (I != ValueMapping.end())
1150 New->setOperand(i, Val: I->second);
1151 }
1152
1153 // Remap debug variable operands.
1154 remapDebugVariable(Mapping&: ValueMapping, Inst: New);
1155 }
1156
1157 return NewBB;
1158}
1159
1160void llvm::cloneNoAliasScopes(ArrayRef<MDNode *> NoAliasDeclScopes,
1161 DenseMap<MDNode *, MDNode *> &ClonedScopes,
1162 StringRef Ext, LLVMContext &Context) {
1163 MDBuilder MDB(Context);
1164
1165 for (MDNode *ScopeList : NoAliasDeclScopes) {
1166 for (const MDOperand &MDOp : ScopeList->operands()) {
1167 if (MDNode *MD = dyn_cast<MDNode>(Val: MDOp)) {
1168 AliasScopeNode SNANode(MD);
1169
1170 std::string Name;
1171 auto ScopeName = SNANode.getName();
1172 if (!ScopeName.empty())
1173 Name = (Twine(ScopeName) + ":" + Ext).str();
1174 else
1175 Name = std::string(Ext);
1176
1177 MDNode *NewScope = MDB.createAnonymousAliasScope(
1178 Domain: const_cast<MDNode *>(SNANode.getDomain()), Name);
1179 ClonedScopes.insert(KV: std::make_pair(x&: MD, y&: NewScope));
1180 }
1181 }
1182 }
1183}
1184
1185void llvm::adaptNoAliasScopes(Instruction *I,
1186 const DenseMap<MDNode *, MDNode *> &ClonedScopes,
1187 LLVMContext &Context) {
1188 auto CloneScopeList = [&](const MDNode *ScopeList) -> MDNode * {
1189 bool NeedsReplacement = false;
1190 SmallVector<Metadata *, 8> NewScopeList;
1191 for (const MDOperand &MDOp : ScopeList->operands()) {
1192 if (MDNode *MD = dyn_cast<MDNode>(Val: MDOp)) {
1193 if (auto *NewMD = ClonedScopes.lookup(Val: MD)) {
1194 NewScopeList.push_back(Elt: NewMD);
1195 NeedsReplacement = true;
1196 continue;
1197 }
1198 NewScopeList.push_back(Elt: MD);
1199 }
1200 }
1201 if (NeedsReplacement)
1202 return MDNode::get(Context, MDs: NewScopeList);
1203 return nullptr;
1204 };
1205
1206 if (auto *Decl = dyn_cast<NoAliasScopeDeclInst>(Val: I))
1207 if (MDNode *NewScopeList = CloneScopeList(Decl->getScopeList()))
1208 Decl->setScopeList(NewScopeList);
1209
1210 auto replaceWhenNeeded = [&](unsigned MD_ID) {
1211 if (const MDNode *CSNoAlias = I->getMetadata(KindID: MD_ID))
1212 if (MDNode *NewScopeList = CloneScopeList(CSNoAlias))
1213 I->setMetadata(KindID: MD_ID, Node: NewScopeList);
1214 };
1215 replaceWhenNeeded(LLVMContext::MD_noalias);
1216 replaceWhenNeeded(LLVMContext::MD_alias_scope);
1217}
1218
1219void llvm::cloneAndAdaptNoAliasScopes(ArrayRef<MDNode *> NoAliasDeclScopes,
1220 ArrayRef<BasicBlock *> NewBlocks,
1221 LLVMContext &Context, StringRef Ext) {
1222 if (NoAliasDeclScopes.empty())
1223 return;
1224
1225 DenseMap<MDNode *, MDNode *> ClonedScopes;
1226 LLVM_DEBUG(dbgs() << "cloneAndAdaptNoAliasScopes: cloning "
1227 << NoAliasDeclScopes.size() << " node(s)\n");
1228
1229 cloneNoAliasScopes(NoAliasDeclScopes, ClonedScopes, Ext, Context);
1230 // Identify instructions using metadata that needs adaptation
1231 for (BasicBlock *NewBlock : NewBlocks)
1232 for (Instruction &I : *NewBlock)
1233 adaptNoAliasScopes(I: &I, ClonedScopes, Context);
1234}
1235
1236void llvm::cloneAndAdaptNoAliasScopes(ArrayRef<MDNode *> NoAliasDeclScopes,
1237 Instruction *IStart, Instruction *IEnd,
1238 LLVMContext &Context, StringRef Ext) {
1239 if (NoAliasDeclScopes.empty())
1240 return;
1241
1242 DenseMap<MDNode *, MDNode *> ClonedScopes;
1243 LLVM_DEBUG(dbgs() << "cloneAndAdaptNoAliasScopes: cloning "
1244 << NoAliasDeclScopes.size() << " node(s)\n");
1245
1246 cloneNoAliasScopes(NoAliasDeclScopes, ClonedScopes, Ext, Context);
1247 // Identify instructions using metadata that needs adaptation
1248 assert(IStart->getParent() == IEnd->getParent() && "different basic block ?");
1249 auto ItStart = IStart->getIterator();
1250 auto ItEnd = IEnd->getIterator();
1251 ++ItEnd; // IEnd is included, increment ItEnd to get the end of the range
1252 for (auto &I : llvm::make_range(x: ItStart, y: ItEnd))
1253 adaptNoAliasScopes(I: &I, ClonedScopes, Context);
1254}
1255
1256void llvm::identifyNoAliasScopesToClone(
1257 ArrayRef<BasicBlock *> BBs, SmallVectorImpl<MDNode *> &NoAliasDeclScopes) {
1258 for (BasicBlock *BB : BBs)
1259 for (Instruction &I : *BB)
1260 if (auto *Decl = dyn_cast<NoAliasScopeDeclInst>(Val: &I))
1261 NoAliasDeclScopes.push_back(Elt: Decl->getScopeList());
1262}
1263
1264void llvm::identifyNoAliasScopesToClone(
1265 BasicBlock::iterator Start, BasicBlock::iterator End,
1266 SmallVectorImpl<MDNode *> &NoAliasDeclScopes) {
1267 for (Instruction &I : make_range(x: Start, y: End))
1268 if (auto *Decl = dyn_cast<NoAliasScopeDeclInst>(Val: &I))
1269 NoAliasDeclScopes.push_back(Elt: Decl->getScopeList());
1270}
1271