1 //===-- ValueEnumerator.cpp - Number values and types for bitcode writer --===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the ValueEnumerator class.
12 //===----------------------------------------------------------------------===//
14 #include "ValueEnumerator.h"
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/ADT/SmallPtrSet.h"
17 #include "llvm/IR/Constants.h"
18 #include "llvm/IR/DerivedTypes.h"
19 #include "llvm/IR/Instructions.h"
20 #include "llvm/IR/Module.h"
21 #include "llvm/IR/ValueSymbolTable.h"
22 #include "llvm/Support/Debug.h"
23 #include "llvm/Support/raw_ostream.h"
27 static bool isIntOrIntVectorValue(const std::pair<const Value*, unsigned> &V) {
28 return V.first->getType()->isIntOrIntVectorTy();
31 /// ValueEnumerator - Enumerate module-level information.
32 ValueEnumerator::ValueEnumerator(const Module *M) {
33 // Enumerate the global variables.
34 for (Module::const_global_iterator I = M->global_begin(),
35 E = M->global_end(); I != E; ++I)
38 // Enumerate the functions.
39 for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
41 EnumerateAttributes(cast<Function>(I)->getAttributes());
44 // Enumerate the aliases.
45 for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
49 // Remember what is the cutoff between globalvalue's and other constants.
50 unsigned FirstConstant = Values.size();
52 // Enumerate the global variable initializers.
53 for (Module::const_global_iterator I = M->global_begin(),
54 E = M->global_end(); I != E; ++I)
55 if (I->hasInitializer())
56 EnumerateValue(I->getInitializer());
58 // Enumerate the aliasees.
59 for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
61 EnumerateValue(I->getAliasee());
63 // Enumerate the prefix data constants.
64 for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I)
65 if (I->hasPrefixData())
66 EnumerateValue(I->getPrefixData());
68 // Insert constants and metadata that are named at module level into the slot
69 // pool so that the module symbol table can refer to them...
70 EnumerateValueSymbolTable(M->getValueSymbolTable());
71 EnumerateNamedMetadata(M);
73 SmallVector<std::pair<unsigned, MDNode*>, 8> MDs;
75 // Enumerate types used by function bodies and argument lists.
76 for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F) {
78 for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
80 EnumerateType(I->getType());
82 for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
83 for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E;++I){
84 for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
86 if (MDNode *MD = dyn_cast<MDNode>(*OI))
87 if (MD->isFunctionLocal() && MD->getFunction())
88 // These will get enumerated during function-incorporation.
90 EnumerateOperandType(*OI);
92 EnumerateType(I->getType());
93 if (const CallInst *CI = dyn_cast<CallInst>(I))
94 EnumerateAttributes(CI->getAttributes());
95 else if (const InvokeInst *II = dyn_cast<InvokeInst>(I))
96 EnumerateAttributes(II->getAttributes());
98 // Enumerate metadata attached with this instruction.
100 I->getAllMetadataOtherThanDebugLoc(MDs);
101 for (unsigned i = 0, e = MDs.size(); i != e; ++i)
102 EnumerateMetadata(MDs[i].second);
104 if (!I->getDebugLoc().isUnknown()) {
106 I->getDebugLoc().getScopeAndInlinedAt(Scope, IA, I->getContext());
107 if (Scope) EnumerateMetadata(Scope);
108 if (IA) EnumerateMetadata(IA);
113 // Optimize constant ordering.
114 OptimizeConstants(FirstConstant, Values.size());
117 unsigned ValueEnumerator::getInstructionID(const Instruction *Inst) const {
118 InstructionMapType::const_iterator I = InstructionMap.find(Inst);
119 assert(I != InstructionMap.end() && "Instruction is not mapped!");
123 void ValueEnumerator::setInstructionID(const Instruction *I) {
124 InstructionMap[I] = InstructionCount++;
127 unsigned ValueEnumerator::getValueID(const Value *V) const {
128 if (isa<MDNode>(V) || isa<MDString>(V)) {
129 ValueMapType::const_iterator I = MDValueMap.find(V);
130 assert(I != MDValueMap.end() && "Value not in slotcalculator!");
134 ValueMapType::const_iterator I = ValueMap.find(V);
135 assert(I != ValueMap.end() && "Value not in slotcalculator!");
139 void ValueEnumerator::dump() const {
140 print(dbgs(), ValueMap, "Default");
142 print(dbgs(), MDValueMap, "MetaData");
146 void ValueEnumerator::print(raw_ostream &OS, const ValueMapType &Map,
147 const char *Name) const {
149 OS << "Map Name: " << Name << "\n";
150 OS << "Size: " << Map.size() << "\n";
151 for (ValueMapType::const_iterator I = Map.begin(),
152 E = Map.end(); I != E; ++I) {
154 const Value *V = I->first;
156 OS << "Value: " << V->getName();
158 OS << "Value: [null]\n";
161 OS << " Uses(" << std::distance(V->use_begin(),V->use_end()) << "):";
162 for (const Use &U : V->uses()) {
163 if (&U != &*V->use_begin())
166 OS << " " << U->getName();
175 /// OptimizeConstants - Reorder constant pool for denser encoding.
176 void ValueEnumerator::OptimizeConstants(unsigned CstStart, unsigned CstEnd) {
177 if (CstStart == CstEnd || CstStart+1 == CstEnd) return;
179 std::stable_sort(Values.begin() + CstStart, Values.begin() + CstEnd,
180 [this](const std::pair<const Value *, unsigned> &LHS,
181 const std::pair<const Value *, unsigned> &RHS) {
183 if (LHS.first->getType() != RHS.first->getType())
184 return getTypeID(LHS.first->getType()) < getTypeID(RHS.first->getType());
185 // Then by frequency.
186 return LHS.second > RHS.second;
189 // Ensure that integer and vector of integer constants are at the start of the
190 // constant pool. This is important so that GEP structure indices come before
191 // gep constant exprs.
192 std::partition(Values.begin()+CstStart, Values.begin()+CstEnd,
193 isIntOrIntVectorValue);
195 // Rebuild the modified portion of ValueMap.
196 for (; CstStart != CstEnd; ++CstStart)
197 ValueMap[Values[CstStart].first] = CstStart+1;
201 /// EnumerateValueSymbolTable - Insert all of the values in the specified symbol
202 /// table into the values table.
203 void ValueEnumerator::EnumerateValueSymbolTable(const ValueSymbolTable &VST) {
204 for (ValueSymbolTable::const_iterator VI = VST.begin(), VE = VST.end();
206 EnumerateValue(VI->getValue());
209 /// EnumerateNamedMetadata - Insert all of the values referenced by
210 /// named metadata in the specified module.
211 void ValueEnumerator::EnumerateNamedMetadata(const Module *M) {
212 for (Module::const_named_metadata_iterator I = M->named_metadata_begin(),
213 E = M->named_metadata_end(); I != E; ++I)
214 EnumerateNamedMDNode(I);
217 void ValueEnumerator::EnumerateNamedMDNode(const NamedMDNode *MD) {
218 for (unsigned i = 0, e = MD->getNumOperands(); i != e; ++i)
219 EnumerateMetadata(MD->getOperand(i));
222 /// EnumerateMDNodeOperands - Enumerate all non-function-local values
223 /// and types referenced by the given MDNode.
224 void ValueEnumerator::EnumerateMDNodeOperands(const MDNode *N) {
225 for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
226 if (Value *V = N->getOperand(i)) {
227 if (isa<MDNode>(V) || isa<MDString>(V))
228 EnumerateMetadata(V);
229 else if (!isa<Instruction>(V) && !isa<Argument>(V))
232 EnumerateType(Type::getVoidTy(N->getContext()));
236 void ValueEnumerator::EnumerateMetadata(const Value *MD) {
237 assert((isa<MDNode>(MD) || isa<MDString>(MD)) && "Invalid metadata kind");
239 // Enumerate the type of this value.
240 EnumerateType(MD->getType());
242 const MDNode *N = dyn_cast<MDNode>(MD);
244 // In the module-level pass, skip function-local nodes themselves, but
245 // do walk their operands.
246 if (N && N->isFunctionLocal() && N->getFunction()) {
247 EnumerateMDNodeOperands(N);
251 // Check to see if it's already in!
252 unsigned &MDValueID = MDValueMap[MD];
254 // Increment use count.
255 MDValues[MDValueID-1].second++;
258 MDValues.push_back(std::make_pair(MD, 1U));
259 MDValueID = MDValues.size();
261 // Enumerate all non-function-local operands.
263 EnumerateMDNodeOperands(N);
266 /// EnumerateFunctionLocalMetadataa - Incorporate function-local metadata
267 /// information reachable from the given MDNode.
268 void ValueEnumerator::EnumerateFunctionLocalMetadata(const MDNode *N) {
269 assert(N->isFunctionLocal() && N->getFunction() &&
270 "EnumerateFunctionLocalMetadata called on non-function-local mdnode!");
272 // Enumerate the type of this value.
273 EnumerateType(N->getType());
275 // Check to see if it's already in!
276 unsigned &MDValueID = MDValueMap[N];
278 // Increment use count.
279 MDValues[MDValueID-1].second++;
282 MDValues.push_back(std::make_pair(N, 1U));
283 MDValueID = MDValues.size();
285 // To incoroporate function-local information visit all function-local
286 // MDNodes and all function-local values they reference.
287 for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
288 if (Value *V = N->getOperand(i)) {
289 if (MDNode *O = dyn_cast<MDNode>(V)) {
290 if (O->isFunctionLocal() && O->getFunction())
291 EnumerateFunctionLocalMetadata(O);
292 } else if (isa<Instruction>(V) || isa<Argument>(V))
296 // Also, collect all function-local MDNodes for easy access.
297 FunctionLocalMDs.push_back(N);
300 void ValueEnumerator::EnumerateValue(const Value *V) {
301 assert(!V->getType()->isVoidTy() && "Can't insert void values!");
302 assert(!isa<MDNode>(V) && !isa<MDString>(V) &&
303 "EnumerateValue doesn't handle Metadata!");
305 // Check to see if it's already in!
306 unsigned &ValueID = ValueMap[V];
308 // Increment use count.
309 Values[ValueID-1].second++;
313 // Enumerate the type of this value.
314 EnumerateType(V->getType());
316 if (const Constant *C = dyn_cast<Constant>(V)) {
317 if (isa<GlobalValue>(C)) {
318 // Initializers for globals are handled explicitly elsewhere.
319 } else if (C->getNumOperands()) {
320 // If a constant has operands, enumerate them. This makes sure that if a
321 // constant has uses (for example an array of const ints), that they are
324 // We prefer to enumerate them with values before we enumerate the user
325 // itself. This makes it more likely that we can avoid forward references
326 // in the reader. We know that there can be no cycles in the constants
327 // graph that don't go through a global variable.
328 for (User::const_op_iterator I = C->op_begin(), E = C->op_end();
330 if (!isa<BasicBlock>(*I)) // Don't enumerate BB operand to BlockAddress.
333 // Finally, add the value. Doing this could make the ValueID reference be
334 // dangling, don't reuse it.
335 Values.push_back(std::make_pair(V, 1U));
336 ValueMap[V] = Values.size();
342 Values.push_back(std::make_pair(V, 1U));
343 ValueID = Values.size();
347 void ValueEnumerator::EnumerateType(Type *Ty) {
348 unsigned *TypeID = &TypeMap[Ty];
350 // We've already seen this type.
354 // If it is a non-anonymous struct, mark the type as being visited so that we
355 // don't recursively visit it. This is safe because we allow forward
356 // references of these in the bitcode reader.
357 if (StructType *STy = dyn_cast<StructType>(Ty))
358 if (!STy->isLiteral())
361 // Enumerate all of the subtypes before we enumerate this type. This ensures
362 // that the type will be enumerated in an order that can be directly built.
363 for (Type::subtype_iterator I = Ty->subtype_begin(), E = Ty->subtype_end();
367 // Refresh the TypeID pointer in case the table rehashed.
368 TypeID = &TypeMap[Ty];
370 // Check to see if we got the pointer another way. This can happen when
371 // enumerating recursive types that hit the base case deeper than they start.
373 // If this is actually a struct that we are treating as forward ref'able,
374 // then emit the definition now that all of its contents are available.
375 if (*TypeID && *TypeID != ~0U)
378 // Add this type now that its contents are all happily enumerated.
381 *TypeID = Types.size();
384 // Enumerate the types for the specified value. If the value is a constant,
385 // walk through it, enumerating the types of the constant.
386 void ValueEnumerator::EnumerateOperandType(const Value *V) {
387 EnumerateType(V->getType());
389 if (const Constant *C = dyn_cast<Constant>(V)) {
390 // If this constant is already enumerated, ignore it, we know its type must
392 if (ValueMap.count(V)) return;
394 // This constant may have operands, make sure to enumerate the types in
396 for (unsigned i = 0, e = C->getNumOperands(); i != e; ++i) {
397 const Value *Op = C->getOperand(i);
399 // Don't enumerate basic blocks here, this happens as operands to
401 if (isa<BasicBlock>(Op)) continue;
403 EnumerateOperandType(Op);
406 if (const MDNode *N = dyn_cast<MDNode>(V)) {
407 for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
408 if (Value *Elem = N->getOperand(i))
409 EnumerateOperandType(Elem);
411 } else if (isa<MDString>(V) || isa<MDNode>(V))
412 EnumerateMetadata(V);
415 void ValueEnumerator::EnumerateAttributes(AttributeSet PAL) {
416 if (PAL.isEmpty()) return; // null is always 0.
419 unsigned &Entry = AttributeMap[PAL];
421 // Never saw this before, add it.
422 Attribute.push_back(PAL);
423 Entry = Attribute.size();
426 // Do lookups for all attribute groups.
427 for (unsigned i = 0, e = PAL.getNumSlots(); i != e; ++i) {
428 AttributeSet AS = PAL.getSlotAttributes(i);
429 unsigned &Entry = AttributeGroupMap[AS];
431 AttributeGroups.push_back(AS);
432 Entry = AttributeGroups.size();
437 void ValueEnumerator::incorporateFunction(const Function &F) {
438 InstructionCount = 0;
439 NumModuleValues = Values.size();
440 NumModuleMDValues = MDValues.size();
442 // Adding function arguments to the value table.
443 for (Function::const_arg_iterator I = F.arg_begin(), E = F.arg_end();
447 FirstFuncConstantID = Values.size();
449 // Add all function-level constants to the value table.
450 for (Function::const_iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
451 for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E; ++I)
452 for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
454 if ((isa<Constant>(*OI) && !isa<GlobalValue>(*OI)) ||
458 BasicBlocks.push_back(BB);
459 ValueMap[BB] = BasicBlocks.size();
462 // Optimize the constant layout.
463 OptimizeConstants(FirstFuncConstantID, Values.size());
465 // Add the function's parameter attributes so they are available for use in
466 // the function's instruction.
467 EnumerateAttributes(F.getAttributes());
469 FirstInstID = Values.size();
471 SmallVector<MDNode *, 8> FnLocalMDVector;
472 // Add all of the instructions.
473 for (Function::const_iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
474 for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E; ++I) {
475 for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
477 if (MDNode *MD = dyn_cast<MDNode>(*OI))
478 if (MD->isFunctionLocal() && MD->getFunction())
479 // Enumerate metadata after the instructions they might refer to.
480 FnLocalMDVector.push_back(MD);
483 SmallVector<std::pair<unsigned, MDNode*>, 8> MDs;
484 I->getAllMetadataOtherThanDebugLoc(MDs);
485 for (unsigned i = 0, e = MDs.size(); i != e; ++i) {
486 MDNode *N = MDs[i].second;
487 if (N->isFunctionLocal() && N->getFunction())
488 FnLocalMDVector.push_back(N);
491 if (!I->getType()->isVoidTy())
496 // Add all of the function-local metadata.
497 for (unsigned i = 0, e = FnLocalMDVector.size(); i != e; ++i)
498 EnumerateFunctionLocalMetadata(FnLocalMDVector[i]);
501 void ValueEnumerator::purgeFunction() {
502 /// Remove purged values from the ValueMap.
503 for (unsigned i = NumModuleValues, e = Values.size(); i != e; ++i)
504 ValueMap.erase(Values[i].first);
505 for (unsigned i = NumModuleMDValues, e = MDValues.size(); i != e; ++i)
506 MDValueMap.erase(MDValues[i].first);
507 for (unsigned i = 0, e = BasicBlocks.size(); i != e; ++i)
508 ValueMap.erase(BasicBlocks[i]);
510 Values.resize(NumModuleValues);
511 MDValues.resize(NumModuleMDValues);
513 FunctionLocalMDs.clear();
516 static void IncorporateFunctionInfoGlobalBBIDs(const Function *F,
517 DenseMap<const BasicBlock*, unsigned> &IDMap) {
518 unsigned Counter = 0;
519 for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
520 IDMap[BB] = ++Counter;
523 /// getGlobalBasicBlockID - This returns the function-specific ID for the
524 /// specified basic block. This is relatively expensive information, so it
525 /// should only be used by rare constructs such as address-of-label.
526 unsigned ValueEnumerator::getGlobalBasicBlockID(const BasicBlock *BB) const {
527 unsigned &Idx = GlobalBasicBlockIDs[BB];
531 IncorporateFunctionInfoGlobalBBIDs(BB->getParent(), GlobalBasicBlockIDs);
532 return getGlobalBasicBlockID(BB);