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/Constants.h"
16 #include "llvm/DerivedTypes.h"
17 #include "llvm/Module.h"
18 #include "llvm/TypeSymbolTable.h"
19 #include "llvm/ValueSymbolTable.h"
20 #include "llvm/Instructions.h"
24 static bool isSingleValueType(const std::pair<const llvm::Type*,
26 return P.first->isSingleValueType();
29 static bool isIntegerValue(const std::pair<const Value*, unsigned> &V) {
30 return isa<IntegerType>(V.first->getType());
33 static bool CompareByFrequency(const std::pair<const llvm::Type*,
35 const std::pair<const llvm::Type*,
37 return P1.second > P2.second;
40 /// ValueEnumerator - Enumerate module-level information.
41 ValueEnumerator::ValueEnumerator(const Module *M) {
44 // Enumerate the global variables.
45 for (Module::const_global_iterator I = M->global_begin(),
46 E = M->global_end(); I != E; ++I)
49 // Enumerate the functions.
50 for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I) {
52 EnumerateAttributes(cast<Function>(I)->getAttributes());
55 // Enumerate the aliases.
56 for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
60 // Remember what is the cutoff between globalvalue's and other constants.
61 unsigned FirstConstant = Values.size();
63 // Enumerate the global variable initializers.
64 for (Module::const_global_iterator I = M->global_begin(),
65 E = M->global_end(); I != E; ++I)
66 if (I->hasInitializer())
67 EnumerateValue(I->getInitializer());
69 // Enumerate the aliasees.
70 for (Module::const_alias_iterator I = M->alias_begin(), E = M->alias_end();
72 EnumerateValue(I->getAliasee());
74 // Enumerate types used by the type symbol table.
75 EnumerateTypeSymbolTable(M->getTypeSymbolTable());
77 // Insert constants and metadata that are named at module level into the slot
78 // pool so that the module symbol table can refer to them...
79 EnumerateValueSymbolTable(M->getValueSymbolTable());
80 EnumerateMDSymbolTable(M->getMDSymbolTable());
82 SmallVector<std::pair<unsigned, MDNode*>, 8> MDs;
84 // Enumerate types used by function bodies and argument lists.
85 for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F) {
87 for (Function::const_arg_iterator I = F->arg_begin(), E = F->arg_end();
89 EnumerateType(I->getType());
91 for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
92 for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E;++I){
93 for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
95 EnumerateOperandType(*OI);
96 EnumerateType(I->getType());
97 if (const CallInst *CI = dyn_cast<CallInst>(I))
98 EnumerateAttributes(CI->getAttributes());
99 else if (const InvokeInst *II = dyn_cast<InvokeInst>(I))
100 EnumerateAttributes(II->getAttributes());
102 // Enumerate metadata attached with this instruction.
104 I->getAllMetadata(MDs);
105 for (unsigned i = 0, e = MDs.size(); i != e; ++i)
106 EnumerateMetadata(MDs[i].second);
110 // Optimize constant ordering.
111 OptimizeConstants(FirstConstant, Values.size());
113 // Sort the type table by frequency so that most commonly used types are early
114 // in the table (have low bit-width).
115 std::stable_sort(Types.begin(), Types.end(), CompareByFrequency);
117 // Partition the Type ID's so that the single-value types occur before the
118 // aggregate types. This allows the aggregate types to be dropped from the
119 // type table after parsing the global variable initializers.
120 std::partition(Types.begin(), Types.end(), isSingleValueType);
122 // Now that we rearranged the type table, rebuild TypeMap.
123 for (unsigned i = 0, e = Types.size(); i != e; ++i)
124 TypeMap[Types[i].first] = i+1;
127 unsigned ValueEnumerator::getInstructionID(const Instruction *Inst) const {
128 InstructionMapType::const_iterator I = InstructionMap.find(Inst);
129 assert (I != InstructionMap.end() && "Instruction is not mapped!");
133 void ValueEnumerator::setInstructionID(const Instruction *I) {
134 InstructionMap[I] = InstructionCount++;
137 unsigned ValueEnumerator::getValueID(const Value *V) const {
138 if (isa<MetadataBase>(V)) {
139 ValueMapType::const_iterator I = MDValueMap.find(V);
140 assert(I != MDValueMap.end() && "Value not in slotcalculator!");
144 ValueMapType::const_iterator I = ValueMap.find(V);
145 assert(I != ValueMap.end() && "Value not in slotcalculator!");
149 // Optimize constant ordering.
151 struct CstSortPredicate {
153 explicit CstSortPredicate(ValueEnumerator &ve) : VE(ve) {}
154 bool operator()(const std::pair<const Value*, unsigned> &LHS,
155 const std::pair<const Value*, unsigned> &RHS) {
157 if (LHS.first->getType() != RHS.first->getType())
158 return VE.getTypeID(LHS.first->getType()) <
159 VE.getTypeID(RHS.first->getType());
160 // Then by frequency.
161 return LHS.second > RHS.second;
166 /// OptimizeConstants - Reorder constant pool for denser encoding.
167 void ValueEnumerator::OptimizeConstants(unsigned CstStart, unsigned CstEnd) {
168 if (CstStart == CstEnd || CstStart+1 == CstEnd) return;
170 CstSortPredicate P(*this);
171 std::stable_sort(Values.begin()+CstStart, Values.begin()+CstEnd, P);
173 // Ensure that integer constants are at the start of the constant pool. This
174 // is important so that GEP structure indices come before gep constant exprs.
175 std::partition(Values.begin()+CstStart, Values.begin()+CstEnd,
178 // Rebuild the modified portion of ValueMap.
179 for (; CstStart != CstEnd; ++CstStart)
180 ValueMap[Values[CstStart].first] = CstStart+1;
184 /// EnumerateTypeSymbolTable - Insert all of the types in the specified symbol
186 void ValueEnumerator::EnumerateTypeSymbolTable(const TypeSymbolTable &TST) {
187 for (TypeSymbolTable::const_iterator TI = TST.begin(), TE = TST.end();
189 EnumerateType(TI->second);
192 /// EnumerateValueSymbolTable - Insert all of the values in the specified symbol
193 /// table into the values table.
194 void ValueEnumerator::EnumerateValueSymbolTable(const ValueSymbolTable &VST) {
195 for (ValueSymbolTable::const_iterator VI = VST.begin(), VE = VST.end();
197 EnumerateValue(VI->getValue());
200 /// EnumerateMDSymbolTable - Insert all of the values in the specified metadata
202 void ValueEnumerator::EnumerateMDSymbolTable(const MDSymbolTable &MST) {
203 for (MDSymbolTable::const_iterator MI = MST.begin(), ME = MST.end();
205 EnumerateValue(MI->getValue());
208 void ValueEnumerator::EnumerateNamedMDNode(const NamedMDNode *MD) {
209 // Check to see if it's already in!
210 unsigned &MDValueID = MDValueMap[MD];
212 // Increment use count.
213 MDValues[MDValueID-1].second++;
217 // Enumerate the type of this value.
218 EnumerateType(MD->getType());
220 for (unsigned i = 0, e = MD->getNumOperands(); i != e; ++i)
221 if (MDNode *E = MD->getOperand(i))
223 MDValues.push_back(std::make_pair(MD, 1U));
224 MDValueMap[MD] = Values.size();
227 void ValueEnumerator::EnumerateMetadata(const MetadataBase *MD) {
228 // Check to see if it's already in!
229 unsigned &MDValueID = MDValueMap[MD];
231 // Increment use count.
232 MDValues[MDValueID-1].second++;
236 // Enumerate the type of this value.
237 EnumerateType(MD->getType());
239 if (const MDNode *N = dyn_cast<MDNode>(MD)) {
240 MDValues.push_back(std::make_pair(MD, 1U));
241 MDValueMap[MD] = MDValues.size();
242 MDValueID = MDValues.size();
243 for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
244 if (Value *V = N->getOperand(i))
247 EnumerateType(Type::getVoidTy(MD->getContext()));
253 assert(isa<MDString>(MD) && "Unknown metadata kind");
254 MDValues.push_back(std::make_pair(MD, 1U));
255 MDValueID = MDValues.size();
258 void ValueEnumerator::EnumerateValue(const Value *V) {
259 assert(!V->getType()->isVoidTy() && "Can't insert void values!");
260 if (const MetadataBase *MB = dyn_cast<MetadataBase>(V))
261 return EnumerateMetadata(MB);
262 else if (const NamedMDNode *NMD = dyn_cast<NamedMDNode>(V))
263 return EnumerateNamedMDNode(NMD);
265 // Check to see if it's already in!
266 unsigned &ValueID = ValueMap[V];
268 // Increment use count.
269 Values[ValueID-1].second++;
273 // Enumerate the type of this value.
274 EnumerateType(V->getType());
276 if (const Constant *C = dyn_cast<Constant>(V)) {
277 if (isa<GlobalValue>(C)) {
278 // Initializers for globals are handled explicitly elsewhere.
279 } else if (isa<ConstantArray>(C) && cast<ConstantArray>(C)->isString()) {
280 // Do not enumerate the initializers for an array of simple characters.
281 // The initializers just polute the value table, and we emit the strings
283 } else if (C->getNumOperands()) {
284 // If a constant has operands, enumerate them. This makes sure that if a
285 // constant has uses (for example an array of const ints), that they are
288 // We prefer to enumerate them with values before we enumerate the user
289 // itself. This makes it more likely that we can avoid forward references
290 // in the reader. We know that there can be no cycles in the constants
291 // graph that don't go through a global variable.
292 for (User::const_op_iterator I = C->op_begin(), E = C->op_end();
294 if (!isa<BasicBlock>(*I)) // Don't enumerate BB operand to BlockAddress.
297 // Finally, add the value. Doing this could make the ValueID reference be
298 // dangling, don't reuse it.
299 Values.push_back(std::make_pair(V, 1U));
300 ValueMap[V] = Values.size();
306 Values.push_back(std::make_pair(V, 1U));
307 ValueID = Values.size();
311 void ValueEnumerator::EnumerateType(const Type *Ty) {
312 unsigned &TypeID = TypeMap[Ty];
315 // If we've already seen this type, just increase its occurrence count.
316 Types[TypeID-1].second++;
320 // First time we saw this type, add it.
321 Types.push_back(std::make_pair(Ty, 1U));
322 TypeID = Types.size();
324 // Enumerate subtypes.
325 for (Type::subtype_iterator I = Ty->subtype_begin(), E = Ty->subtype_end();
330 // Enumerate the types for the specified value. If the value is a constant,
331 // walk through it, enumerating the types of the constant.
332 void ValueEnumerator::EnumerateOperandType(const Value *V) {
333 EnumerateType(V->getType());
334 if (const Constant *C = dyn_cast<Constant>(V)) {
335 // If this constant is already enumerated, ignore it, we know its type must
337 if (ValueMap.count(V)) return;
339 // This constant may have operands, make sure to enumerate the types in
341 for (unsigned i = 0, e = C->getNumOperands(); i != e; ++i) {
342 const User *Op = C->getOperand(i);
344 // Don't enumerate basic blocks here, this happens as operands to
346 if (isa<BasicBlock>(Op)) continue;
348 EnumerateOperandType(cast<Constant>(Op));
351 if (const MDNode *N = dyn_cast<MDNode>(V)) {
352 for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
353 if (Value *Elem = N->getOperand(i))
354 EnumerateOperandType(Elem);
356 } else if (isa<MDString>(V) || isa<MDNode>(V))
360 void ValueEnumerator::EnumerateAttributes(const AttrListPtr &PAL) {
361 if (PAL.isEmpty()) return; // null is always 0.
363 unsigned &Entry = AttributeMap[PAL.getRawPointer()];
365 // Never saw this before, add it.
366 Attributes.push_back(PAL);
367 Entry = Attributes.size();
372 void ValueEnumerator::incorporateFunction(const Function &F) {
373 NumModuleValues = Values.size();
375 // Adding function arguments to the value table.
376 for(Function::const_arg_iterator I = F.arg_begin(), E = F.arg_end();
380 FirstFuncConstantID = Values.size();
382 // Add all function-level constants to the value table.
383 for (Function::const_iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
384 for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E; ++I)
385 for (User::const_op_iterator OI = I->op_begin(), E = I->op_end();
387 if ((isa<Constant>(*OI) && !isa<GlobalValue>(*OI)) ||
391 BasicBlocks.push_back(BB);
392 ValueMap[BB] = BasicBlocks.size();
395 // Optimize the constant layout.
396 OptimizeConstants(FirstFuncConstantID, Values.size());
398 // Add the function's parameter attributes so they are available for use in
399 // the function's instruction.
400 EnumerateAttributes(F.getAttributes());
402 FirstInstID = Values.size();
404 // Add all of the instructions.
405 for (Function::const_iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
406 for (BasicBlock::const_iterator I = BB->begin(), E = BB->end(); I!=E; ++I) {
407 if (!I->getType()->isVoidTy())
413 void ValueEnumerator::purgeFunction() {
414 /// Remove purged values from the ValueMap.
415 for (unsigned i = NumModuleValues, e = Values.size(); i != e; ++i)
416 ValueMap.erase(Values[i].first);
417 for (unsigned i = 0, e = BasicBlocks.size(); i != e; ++i)
418 ValueMap.erase(BasicBlocks[i]);
420 Values.resize(NumModuleValues);
424 static void IncorporateFunctionInfoGlobalBBIDs(const Function *F,
425 DenseMap<const BasicBlock*, unsigned> &IDMap) {
426 unsigned Counter = 0;
427 for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
428 IDMap[BB] = ++Counter;
431 /// getGlobalBasicBlockID - This returns the function-specific ID for the
432 /// specified basic block. This is relatively expensive information, so it
433 /// should only be used by rare constructs such as address-of-label.
434 unsigned ValueEnumerator::getGlobalBasicBlockID(const BasicBlock *BB) const {
435 unsigned &Idx = GlobalBasicBlockIDs[BB];
439 IncorporateFunctionInfoGlobalBBIDs(BB->getParent(), GlobalBasicBlockIDs);
440 return getGlobalBasicBlockID(BB);