1 //===-- ProgramInfo.cpp - Compute and cache info about a program ----------===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the LLVM research group and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the ProgramInfo and related classes, by sorting through
13 //===----------------------------------------------------------------------===//
15 #include "llvm/Debugger/ProgramInfo.h"
16 #include "llvm/Constants.h"
17 #include "llvm/DerivedTypes.h"
18 #include "llvm/Intrinsics.h"
19 #include "llvm/Instructions.h"
20 #include "llvm/Module.h"
21 #include "llvm/Debugger/SourceFile.h"
22 #include "llvm/Debugger/SourceLanguage.h"
23 #include "llvm/Support/FileUtilities.h"
24 #include "llvm/Support/SlowOperationInformer.h"
25 #include "llvm/ADT/STLExtras.h"
30 /// getGlobalVariablesUsing - Return all of the global variables which have the
31 /// specified value in their initializer somewhere.
32 static void getGlobalVariablesUsing(Value *V,
33 std::vector<GlobalVariable*> &Found) {
34 for (Value::use_iterator I = V->use_begin(), E = V->use_end(); I != E; ++I) {
35 if (GlobalVariable *GV = dyn_cast<GlobalVariable>(*I))
37 else if (Constant *C = dyn_cast<Constant>(*I))
38 getGlobalVariablesUsing(C, Found);
42 /// getStringValue - Turn an LLVM constant pointer that eventually points to a
43 /// global into a string value. Return an empty string if we can't do it.
45 static std::string getStringValue(Value *V, unsigned Offset = 0) {
46 if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
47 if (GV->hasInitializer() && isa<ConstantArray>(GV->getInitializer())) {
48 ConstantArray *Init = cast<ConstantArray>(GV->getInitializer());
49 if (Init->isString()) {
50 std::string Result = Init->getAsString();
51 if (Offset < Result.size()) {
52 // If we are pointing INTO The string, erase the beginning...
53 Result.erase(Result.begin(), Result.begin()+Offset);
55 // Take off the null terminator, and any string fragments after it.
56 std::string::size_type NullPos = Result.find_first_of((char)0);
57 if (NullPos != std::string::npos)
58 Result.erase(Result.begin()+NullPos, Result.end());
63 } else if (Constant *C = dyn_cast<Constant>(V)) {
64 if (GlobalValue *GV = dyn_cast<GlobalValue>(C))
65 return getStringValue(GV, Offset);
66 else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
67 if (CE->getOpcode() == Instruction::GetElementPtr) {
68 // Turn a gep into the specified offset.
69 if (CE->getNumOperands() == 3 &&
70 cast<Constant>(CE->getOperand(1))->isNullValue() &&
71 isa<ConstantInt>(CE->getOperand(2))) {
72 return getStringValue(CE->getOperand(0),
73 Offset+cast<ConstantInt>(CE->getOperand(2))->getRawValue());
81 /// getNextStopPoint - Follow the def-use chains of the specified LLVM value,
82 /// traversing the use chains until we get to a stoppoint. When we do, return
83 /// the source location of the stoppoint. If we don't find a stoppoint, return
85 static const GlobalVariable *getNextStopPoint(const Value *V, unsigned &LineNo,
87 // The use-def chains can fork. As such, we pick the lowest numbered one we
89 const GlobalVariable *LastDesc = 0;
90 unsigned LastLineNo = ~0;
91 unsigned LastColNo = ~0;
93 for (Value::use_const_iterator UI = V->use_begin(), E = V->use_end();
95 bool ShouldRecurse = true;
96 if (cast<Instruction>(*UI)->getOpcode() == Instruction::PHI) {
97 // Infinite loops == bad, ignore PHI nodes.
98 ShouldRecurse = false;
99 } else if (const CallInst *CI = dyn_cast<CallInst>(*UI)) {
100 // If we found a stop point, check to see if it is earlier than what we
101 // already have. If so, remember it.
102 if (const Function *F = CI->getCalledFunction())
103 if (F->getIntrinsicID() == Intrinsic::dbg_stoppoint) {
104 unsigned CurLineNo = ~0, CurColNo = ~0;
105 const GlobalVariable *CurDesc = 0;
106 if (const ConstantInt *C = dyn_cast<ConstantInt>(CI->getOperand(2)))
107 CurLineNo = C->getRawValue();
108 if (const ConstantInt *C = dyn_cast<ConstantInt>(CI->getOperand(3)))
109 CurColNo = C->getRawValue();
110 const Value *Op = CI->getOperand(4);
112 if ((CurDesc = dyn_cast<GlobalVariable>(Op)) &&
113 (LineNo < LastLineNo ||
114 (LineNo == LastLineNo && ColNo < LastColNo))) {
116 LastLineNo = CurLineNo;
117 LastColNo = CurColNo;
119 ShouldRecurse = false;
124 // If this is not a phi node or a stopping point, recursively scan the users
125 // of this instruction to skip over region.begin's and the like.
127 unsigned CurLineNo, CurColNo;
128 if (const GlobalVariable *GV = getNextStopPoint(*UI, CurLineNo,CurColNo)){
129 if (LineNo < LastLineNo || (LineNo == LastLineNo && ColNo < LastColNo)){
131 LastLineNo = CurLineNo;
132 LastColNo = CurColNo;
139 LineNo = LastLineNo != ~0U ? LastLineNo : 0;
140 ColNo = LastColNo != ~0U ? LastColNo : 0;
146 //===----------------------------------------------------------------------===//
147 // SourceFileInfo implementation
150 SourceFileInfo::SourceFileInfo(const GlobalVariable *Desc,
151 const SourceLanguage &Lang)
152 : Language(&Lang), Descriptor(Desc) {
156 if (Desc && Desc->hasInitializer())
157 if (ConstantStruct *CS = dyn_cast<ConstantStruct>(Desc->getInitializer()))
158 if (CS->getNumOperands() > 4) {
159 if (ConstantUInt *CUI = dyn_cast<ConstantUInt>(CS->getOperand(1)))
160 Version = CUI->getValue();
162 BaseName = getStringValue(CS->getOperand(3));
163 Directory = getStringValue(CS->getOperand(4));
167 SourceFileInfo::~SourceFileInfo() {
171 SourceFile &SourceFileInfo::getSourceText() const {
172 // FIXME: this should take into account the source search directories!
173 if (SourceText == 0) // Read the file in if we haven't already.
174 if (!Directory.empty() && FileOpenable(Directory+"/"+BaseName))
175 SourceText = new SourceFile(Directory+"/"+BaseName, Descriptor);
177 SourceText = new SourceFile(BaseName, Descriptor);
182 //===----------------------------------------------------------------------===//
183 // SourceFunctionInfo implementation
185 SourceFunctionInfo::SourceFunctionInfo(ProgramInfo &PI,
186 const GlobalVariable *Desc)
189 if (Desc && Desc->hasInitializer())
190 if (ConstantStruct *CS = dyn_cast<ConstantStruct>(Desc->getInitializer()))
191 if (CS->getNumOperands() > 2) {
192 // Entry #1 is the file descriptor.
193 if (const GlobalVariable *GV =
194 dyn_cast<GlobalVariable>(CS->getOperand(1)))
195 SourceFile = &PI.getSourceFile(GV);
197 // Entry #2 is the function name.
198 Name = getStringValue(CS->getOperand(2));
202 /// getSourceLocation - This method returns the location of the first stopping
203 /// point in the function.
204 void SourceFunctionInfo::getSourceLocation(unsigned &RetLineNo,
205 unsigned &RetColNo) const {
206 // If we haven't computed this yet...
208 // Look at all of the users of the function descriptor, looking for calls to
209 // %llvm.dbg.func.start.
210 for (Value::use_const_iterator UI = Descriptor->use_begin(),
211 E = Descriptor->use_end(); UI != E; ++UI)
212 if (const CallInst *CI = dyn_cast<CallInst>(*UI))
213 if (const Function *F = CI->getCalledFunction())
214 if (F->getIntrinsicID() == Intrinsic::dbg_func_start) {
215 // We found the start of the function. Check to see if there are
216 // any stop points on the use-list of the function start.
217 const GlobalVariable *SD = getNextStopPoint(CI, LineNo, ColNo);
218 if (SD) { // We found the first stop point!
219 // This is just a sanity check.
220 if (getSourceFile().getDescriptor() != SD)
221 std::cout << "WARNING: first line of function is not in the"
222 " file that the function descriptor claims it is in.\n";
227 RetLineNo = LineNo; RetColNo = ColNo;
230 //===----------------------------------------------------------------------===//
231 // ProgramInfo implementation
234 ProgramInfo::ProgramInfo(Module *m) : M(m) {
235 assert(M && "Cannot create program information with a null module!");
236 ProgramTimeStamp = getFileTimestamp(M->getModuleIdentifier());
238 SourceFilesIsComplete = false;
239 SourceFunctionsIsComplete = false;
242 ProgramInfo::~ProgramInfo() {
243 // Delete cached information about source program objects...
244 for (std::map<const GlobalVariable*, SourceFileInfo*>::iterator
245 I = SourceFiles.begin(), E = SourceFiles.end(); I != E; ++I)
247 for (std::map<const GlobalVariable*, SourceFunctionInfo*>::iterator
248 I = SourceFunctions.begin(), E = SourceFunctions.end(); I != E; ++I)
251 // Delete the source language caches.
252 for (unsigned i = 0, e = LanguageCaches.size(); i != e; ++i)
253 delete LanguageCaches[i].second;
257 //===----------------------------------------------------------------------===//
258 // SourceFileInfo tracking...
261 /// getSourceFile - Return source file information for the specified source file
262 /// descriptor object, adding it to the collection as needed. This method
263 /// always succeeds (is unambiguous), and is always efficient.
265 const SourceFileInfo &
266 ProgramInfo::getSourceFile(const GlobalVariable *Desc) {
267 SourceFileInfo *&Result = SourceFiles[Desc];
268 if (Result) return *Result;
270 // Figure out what language this source file comes from...
271 unsigned LangID = 0; // Zero is unknown language
272 if (Desc && Desc->hasInitializer())
273 if (ConstantStruct *CS = dyn_cast<ConstantStruct>(Desc->getInitializer()))
274 if (CS->getNumOperands() > 2)
275 if (ConstantUInt *CUI = dyn_cast<ConstantUInt>(CS->getOperand(2)))
276 LangID = CUI->getValue();
278 const SourceLanguage &Lang = SourceLanguage::get(LangID);
279 SourceFileInfo *New = Lang.createSourceFileInfo(Desc, *this);
281 // FIXME: this should check to see if there is already a Filename/WorkingDir
282 // pair that matches this one. If so, we shouldn't create the duplicate!
284 SourceFileIndex.insert(std::make_pair(New->getBaseName(), New));
285 return *(Result = New);
289 /// getSourceFiles - Index all of the source files in the program and return
290 /// a mapping of it. This information is lazily computed the first time
291 /// that it is requested. Since this information can take a long time to
292 /// compute, the user is given a chance to cancel it. If this occurs, an
293 /// exception is thrown.
294 const std::map<const GlobalVariable*, SourceFileInfo*> &
295 ProgramInfo::getSourceFiles(bool RequiresCompleteMap) {
296 // If we have a fully populated map, or if the client doesn't need one, just
297 // return what we have.
298 if (SourceFilesIsComplete || !RequiresCompleteMap)
301 // Ok, all of the source file descriptors (compile_unit in dwarf terms),
302 // should be on the use list of the llvm.dbg.translation_units global.
304 GlobalVariable *Units =
305 M->getGlobalVariable("llvm.dbg.translation_units",
306 StructType::get(std::vector<const Type*>()));
308 throw "Program contains no debugging information!";
310 std::vector<GlobalVariable*> TranslationUnits;
311 getGlobalVariablesUsing(Units, TranslationUnits);
313 SlowOperationInformer SOI("building source files index");
315 // Loop over all of the translation units found, building the SourceFiles
317 for (unsigned i = 0, e = TranslationUnits.size(); i != e; ++i) {
318 getSourceFile(TranslationUnits[i]);
319 SOI.progress(i+1, e);
322 // Ok, if we got this far, then we indexed the whole program.
323 SourceFilesIsComplete = true;
327 /// getSourceFile - Look up the file with the specified name. If there is
328 /// more than one match for the specified filename, prompt the user to pick
329 /// one. If there is no source file that matches the specified name, throw
330 /// an exception indicating that we can't find the file. Otherwise, return
331 /// the file information for that file.
332 const SourceFileInfo &ProgramInfo::getSourceFile(const std::string &Filename) {
333 std::multimap<std::string, SourceFileInfo*>::const_iterator Start, End;
335 tie(Start, End) = SourceFileIndex.equal_range(Filename);
337 if (Start == End) throw "Could not find source file '" + Filename + "'!";
338 const SourceFileInfo &SFI = *Start->second;
340 if (Start == End) return SFI;
342 throw "FIXME: Multiple source files with the same name not implemented!";
346 //===----------------------------------------------------------------------===//
347 // SourceFunctionInfo tracking...
351 /// getFunction - Return function information for the specified function
352 /// descriptor object, adding it to the collection as needed. This method
353 /// always succeeds (is unambiguous), and is always efficient.
355 const SourceFunctionInfo &
356 ProgramInfo::getFunction(const GlobalVariable *Desc) {
357 SourceFunctionInfo *&Result = SourceFunctions[Desc];
358 if (Result) return *Result;
360 // Figure out what language this function comes from...
361 const GlobalVariable *SourceFileDesc = 0;
362 if (Desc && Desc->hasInitializer())
363 if (ConstantStruct *CS = dyn_cast<ConstantStruct>(Desc->getInitializer()))
364 if (CS->getNumOperands() > 0)
365 if (const GlobalVariable *GV =
366 dyn_cast<GlobalVariable>(CS->getOperand(1)))
369 const SourceLanguage &Lang = getSourceFile(SourceFileDesc).getLanguage();
370 return *(Result = Lang.createSourceFunctionInfo(Desc, *this));
374 // getSourceFunctions - Index all of the functions in the program and return
375 // them. This information is lazily computed the first time that it is
376 // requested. Since this information can take a long time to compute, the user
377 // is given a chance to cancel it. If this occurs, an exception is thrown.
378 const std::map<const GlobalVariable*, SourceFunctionInfo*> &
379 ProgramInfo::getSourceFunctions(bool RequiresCompleteMap) {
380 if (SourceFunctionsIsComplete || !RequiresCompleteMap)
381 return SourceFunctions;
383 // Ok, all of the source function descriptors (subprogram in dwarf terms),
384 // should be on the use list of the llvm.dbg.translation_units global.
386 GlobalVariable *Units =
387 M->getGlobalVariable("llvm.dbg.globals",
388 StructType::get(std::vector<const Type*>()));
390 throw "Program contains no debugging information!";
392 std::vector<GlobalVariable*> Functions;
393 getGlobalVariablesUsing(Units, Functions);
395 SlowOperationInformer SOI("building functions index");
397 // Loop over all of the functions found, building the SourceFunctions mapping.
398 for (unsigned i = 0, e = Functions.size(); i != e; ++i) {
399 getFunction(Functions[i]);
400 SOI.progress(i+1, e);
403 // Ok, if we got this far, then we indexed the whole program.
404 SourceFunctionsIsComplete = true;
405 return SourceFunctions;