1 //===- gccld.cpp - LLVM 'ld' compatible linker ----------------------------===//
3 // This utility is intended to be compatible with GCC, and follows standard
4 // system 'ld' conventions. As such, the default output file is ./a.out.
5 // Additionally, this program outputs a shell script that is used to invoke LLI
6 // to execute the program. In this manner, the generated executable (a.out for
7 // example), is directly executable, whereas the bytecode file actually lives in
8 // the a.out.bc file generated by this program. Also, Force is on by default.
10 // Note that if someone (or a script) deletes the executable program generated,
11 // the .bc file will be left around. Considering that this is a temporary hack,
12 // I'm not to worried about this.
14 //===----------------------------------------------------------------------===//
16 #include "llvm/Transforms/Utils/Linker.h"
17 #include "llvm/Module.h"
18 #include "llvm/PassManager.h"
19 #include "llvm/Bytecode/Reader.h"
20 #include "llvm/Bytecode/WriteBytecodePass.h"
21 #include "llvm/Transforms/IPO.h"
22 #include "llvm/Transforms/Scalar.h"
23 #include "Support/CommandLine.h"
24 #include "Support/Signals.h"
29 #include <sys/types.h> // For FileExists
34 InputFilenames(cl::Positional, cl::desc("<input bytecode files>"),
38 OutputFilename("o", cl::desc("Override output filename"), cl::init("a.out"),
39 cl::value_desc("filename"));
42 Verbose("v", cl::desc("Print information about actions taken"));
45 LibPaths("L", cl::desc("Specify a library search path"), cl::Prefix,
46 cl::value_desc("directory"));
49 Libraries("l", cl::desc("Specify libraries to link to"), cl::Prefix,
50 cl::value_desc("library prefix"));
53 Strip("s", cl::desc("Strip symbol info from executable"));
56 NoInternalize("disable-internalize",
57 cl::desc("Do not mark all symbols as internal"));
60 LinkAsLibrary("link-as-library", cl::desc("Link the .bc files together as a"
61 " library, not an executable"));
63 // Compatibility options that are ignored, but support by LD
65 CO3("soname", cl::Hidden, cl::desc("Compatibility option: ignored"));
67 CO4("version-script", cl::Hidden, cl::desc("Compatibility option: ignored"));
69 CO5("eh-frame-hdr", cl::Hidden, cl::desc("Compatibility option: ignored"));
72 // FileExists - Return true if the specified string is an openable file...
73 static inline bool FileExists(const std::string &FN) {
75 return stat(FN.c_str(), &StatBuf) != -1;
79 // LoadObject - Read the specified "object file", which should not search the
80 // library path to find it.
81 static inline std::auto_ptr<Module> LoadObject(const std::string &FN,
82 std::string &OutErrorMessage) {
83 if (Verbose) std::cerr << "Loading '" << FN << "'\n";
84 if (!FileExists(FN)) {
85 OutErrorMessage = "could not find input file '" + FN + "'!";
86 return std::auto_ptr<Module>();
89 std::string ErrorMessage;
90 Module *Result = ParseBytecodeFile(FN, &ErrorMessage);
91 if (Result) return std::auto_ptr<Module>(Result);
93 OutErrorMessage = "Bytecode file '" + FN + "' corrupt!";
94 if (ErrorMessage.size()) OutErrorMessage += ": " + ErrorMessage;
95 return std::auto_ptr<Module>();
99 static Module *LoadSingleLibraryObject(const std::string &Filename) {
100 std::string ErrorMessage;
101 std::auto_ptr<Module> M = LoadObject(Filename, ErrorMessage);
102 if (M.get() == 0 && Verbose) {
103 std::cerr << "Error loading '" + Filename + "'";
104 if (!ErrorMessage.empty()) std::cerr << ": " << ErrorMessage;
112 // LoadLibraryFromDirectory - This looks for a .a, .so, or .bc file in a
113 // particular directory. It returns true if no library is found, otherwise it
114 // puts the loaded modules into the Objects list, and sets isArchive to true if
115 // a .a file was loaded.
117 static inline bool LoadLibraryFromDirectory(const std::string &LibName,
118 const std::string &Directory,
119 std::vector<Module*> &Objects,
121 if (FileExists(Directory + "lib" + LibName + ".a")) {
122 std::string ErrorMessage;
123 if (Verbose) std::cerr << " Loading '" << Directory << "lib"
124 << LibName << ".a'\n";
125 if (!ReadArchiveFile(Directory + "lib" + LibName + ".a", Objects,
126 &ErrorMessage)) { // Read the archive file
128 return false; // Success!
132 std::cerr << " Error loading archive '" + Directory +"lib"+LibName+".a'";
133 if (!ErrorMessage.empty()) std::cerr << ": " << ErrorMessage;
138 if (FileExists(Directory + "lib" + LibName + ".so"))
139 if (Module *M = LoadSingleLibraryObject(Directory + "lib" + LibName+".so")){
141 Objects.push_back(M);
145 if (FileExists(Directory + "lib" + LibName + ".bc"))
146 if (Module *M = LoadSingleLibraryObject(Directory + "lib" + LibName+".bc")){
148 Objects.push_back(M);
154 // LoadLibrary - This searches for a .a, .so, or .bc file which provides the
155 // LLVM bytecode for the library. It returns true if no library is found,
156 // otherwise it puts the loaded modules into the Objects list, and sets
157 // isArchive to true if a .a file was loaded.
159 static inline bool LoadLibrary(const std::string &LibName,
160 std::vector<Module*> &Objects, bool &isArchive,
161 std::string &ErrorMessage) {
162 std::string Directory;
163 unsigned NextLibPathIdx = 0;
166 // Try loading from the current directory...
167 if (Verbose) std::cerr << " Looking in directory '" << Directory << "'\n";
168 if (!LoadLibraryFromDirectory(LibName, Directory, Objects, isArchive))
171 if (NextLibPathIdx == LibPaths.size()) break;
172 Directory = LibPaths[NextLibPathIdx++]+"/";
175 ErrorMessage = "error linking library '-l" + LibName+ "': library not found!";
179 static void GetAllDefinedSymbols(Module *M,
180 std::set<std::string> &DefinedSymbols) {
181 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
182 if (I->hasName() && !I->isExternal() && !I->hasInternalLinkage())
183 DefinedSymbols.insert(I->getName());
184 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
185 if (I->hasName() && !I->isExternal() && !I->hasInternalLinkage())
186 DefinedSymbols.insert(I->getName());
189 // GetAllUndefinedSymbols - This calculates the set of undefined symbols that
190 // still exist in an LLVM module. This is a bit tricky because there may be two
191 // symbols with the same name, but different LLVM types that will be resolved to
192 // each other, but aren't currently (thus we need to treat it as resolved).
194 static void GetAllUndefinedSymbols(Module *M,
195 std::set<std::string> &UndefinedSymbols) {
196 std::set<std::string> DefinedSymbols;
197 UndefinedSymbols.clear(); // Start out empty
199 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
202 UndefinedSymbols.insert(I->getName());
203 else if (!I->hasInternalLinkage())
204 DefinedSymbols.insert(I->getName());
206 for (Module::giterator I = M->gbegin(), E = M->gend(); I != E; ++I)
209 UndefinedSymbols.insert(I->getName());
210 else if (!I->hasInternalLinkage())
211 DefinedSymbols.insert(I->getName());
214 // Prune out any defined symbols from the undefined symbols set...
215 for (std::set<std::string>::iterator I = UndefinedSymbols.begin();
216 I != UndefinedSymbols.end(); )
217 if (DefinedSymbols.count(*I))
218 UndefinedSymbols.erase(I++); // This symbol really is defined!
220 ++I; // Keep this symbol in the undefined symbols list
224 static bool LinkLibrary(Module *M, const std::string &LibName,
225 std::string &ErrorMessage) {
226 std::vector<Module*> Objects;
228 if (LoadLibrary(LibName, Objects, isArchive, ErrorMessage)) return true;
230 // Figure out which symbols are defined by all of the modules in the .a file
231 std::vector<std::set<std::string> > DefinedSymbols;
232 DefinedSymbols.resize(Objects.size());
233 for (unsigned i = 0; i != Objects.size(); ++i)
234 GetAllDefinedSymbols(Objects[i], DefinedSymbols[i]);
236 std::set<std::string> UndefinedSymbols;
237 GetAllUndefinedSymbols(M, UndefinedSymbols);
240 while (Linked) { // While we are linking in object files, loop.
243 for (unsigned i = 0; i != Objects.size(); ++i) {
244 // Consider whether we need to link in this module... we only need to
245 // link it in if it defines some symbol which is so far undefined.
247 const std::set<std::string> &DefSymbols = DefinedSymbols[i];
249 bool ObjectRequired = false;
250 for (std::set<std::string>::iterator I = UndefinedSymbols.begin(),
251 E = UndefinedSymbols.end(); I != E; ++I)
252 if (DefSymbols.count(*I)) {
254 std::cerr << " Found object providing symbol '" << *I << "'...\n";
255 ObjectRequired = true;
259 // We DO need to link this object into the program...
260 if (ObjectRequired) {
261 if (LinkModules(M, Objects[i], &ErrorMessage))
262 return true; // Couldn't link in the right object file...
264 // Since we have linked in this object, delete it from the list of
265 // objects to consider in this archive file.
266 std::swap(Objects[i], Objects.back());
267 std::swap(DefinedSymbols[i], DefinedSymbols.back());
269 DefinedSymbols.pop_back();
270 --i; // Do not skip an entry
272 // The undefined symbols set should have shrunk.
273 GetAllUndefinedSymbols(M, UndefinedSymbols);
274 Linked = true; // We have linked something in!
282 static int PrintAndReturn(const char *progname, const std::string &Message,
283 const std::string &Extra = "") {
284 std::cerr << progname << Extra << ": " << Message << "\n";
289 int main(int argc, char **argv) {
290 cl::ParseCommandLineOptions(argc, argv, " llvm linker for GCC\n");
292 std::string ErrorMessage;
293 std::auto_ptr<Module> Composite(LoadObject(InputFilenames[0], ErrorMessage));
294 if (Composite.get() == 0)
295 return PrintAndReturn(argv[0], ErrorMessage);
297 for (unsigned i = 1; i < InputFilenames.size(); ++i) {
298 std::auto_ptr<Module> M(LoadObject(InputFilenames[i], ErrorMessage));
300 return PrintAndReturn(argv[0], ErrorMessage);
302 if (Verbose) std::cerr << "Linking in '" << InputFilenames[i] << "'\n";
304 if (LinkModules(Composite.get(), M.get(), &ErrorMessage))
305 return PrintAndReturn(argv[0], ErrorMessage,
306 ": error linking in '" + InputFilenames[i] + "'");
309 // Remove any consecutive duplicates of the same library...
310 Libraries.erase(std::unique(Libraries.begin(), Libraries.end()),
313 // Link in all of the libraries next...
314 for (unsigned i = 0; i != Libraries.size(); ++i) {
315 if (Verbose) std::cerr << "Linking in library: -l" << Libraries[i] << "\n";
316 if (LinkLibrary(Composite.get(), Libraries[i], ErrorMessage))
317 return PrintAndReturn(argv[0], ErrorMessage);
320 // In addition to just linking the input from GCC, we also want to spiff it up
321 // a little bit. Do this now.
325 // Linking modules together can lead to duplicated global constants, only keep
326 // one copy of each constant...
328 Passes.add(createConstantMergePass());
330 // If the -s command line option was specified, strip the symbols out of the
331 // resulting program to make it smaller. -s is a GCC option that we are
335 Passes.add(createSymbolStrippingPass());
337 // Often if the programmer does not specify proper prototypes for the
338 // functions they are calling, they end up calling a vararg version of the
339 // function that does not get a body filled in (the real function has typed
340 // arguments). This pass merges the two functions.
342 Passes.add(createFunctionResolvingPass());
344 if (!NoInternalize) {
345 // Now that composite has been compiled, scan through the module, looking
346 // for a main function. If main is defined, mark all other functions
349 Passes.add(createInternalizePass());
352 // Now that we have optimized the program, discard unreachable functions...
354 Passes.add(createGlobalDCEPass());
356 // Add the pass that writes bytecode to the output file...
357 std::string RealBytecodeOutput = OutputFilename;
358 if (!LinkAsLibrary) RealBytecodeOutput += ".bc";
359 std::ofstream Out(RealBytecodeOutput.c_str());
361 return PrintAndReturn(argv[0], "error opening '" + RealBytecodeOutput +
363 Passes.add(new WriteBytecodePass(&Out)); // Write bytecode to file...
365 // Make sure that the Out file gets unlink'd from the disk if we get a SIGINT
366 RemoveFileOnSignal(RealBytecodeOutput);
368 // Run our queue of passes all at once now, efficiently.
369 Passes.run(*Composite.get());
372 if (!LinkAsLibrary) {
373 // Output the script to start the program...
374 std::ofstream Out2(OutputFilename.c_str());
376 return PrintAndReturn(argv[0], "error opening '" + OutputFilename +
378 Out2 << "#!/bin/sh\nlli -q -abort-on-exception $0.bc $*\n";
381 // Make the script executable...
382 chmod(OutputFilename.c_str(), 0755);