1 //===- llvm-ld.cpp - LLVM 'ld' compatible linker --------------------------===//
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 utility is intended to be compatible with GCC, and follows standard
11 // system 'ld' conventions. As such, the default output file is ./a.out.
12 // Additionally, this program outputs a shell script that is used to invoke LLI
13 // to execute the program. In this manner, the generated executable (a.out for
14 // example), is directly executable, whereas the bytecode file actually lives in
15 // the a.out.bc file generated by this program. Also, Force is on by default.
17 // Note that if someone (or a script) deletes the executable program generated,
18 // the .bc file will be left around. Considering that this is a temporary hack,
19 // I'm not too worried about this.
21 //===----------------------------------------------------------------------===//
23 #include "llvm/LinkAllVMCore.h"
24 #include "llvm/Linker.h"
25 #include "llvm/System/Program.h"
26 #include "llvm/Module.h"
27 #include "llvm/PassManager.h"
28 #include "llvm/Bytecode/Reader.h"
29 #include "llvm/Bytecode/Writer.h"
30 #include "llvm/Target/TargetData.h"
31 #include "llvm/Target/TargetMachine.h"
32 #include "llvm/Target/TargetMachineRegistry.h"
33 #include "llvm/Support/CommandLine.h"
34 #include "llvm/Support/FileUtilities.h"
35 #include "llvm/Support/SystemUtils.h"
36 #include "llvm/System/Signals.h"
43 // Input/Output Options
44 static cl::list<std::string> InputFilenames(cl::Positional, cl::OneOrMore,
45 cl::desc("<input bytecode files>"));
47 static cl::opt<std::string> OutputFilename("o", cl::init("a.out"),
48 cl::desc("Override output filename"),
49 cl::value_desc("filename"));
51 static cl::opt<bool> Verbose("v",
52 cl::desc("Print information about actions taken"));
54 static cl::list<std::string> LibPaths("L", cl::Prefix,
55 cl::desc("Specify a library search path"),
56 cl::value_desc("directory"));
58 static cl::list<std::string> Libraries("l", cl::Prefix,
59 cl::desc("Specify libraries to link to"),
60 cl::value_desc("library prefix"));
62 static cl::opt<bool> LinkAsLibrary("link-as-library",
63 cl::desc("Link the .bc files together as a library, not an executable"));
65 static cl::alias Relink("r", cl::aliasopt(LinkAsLibrary),
66 cl::desc("Alias for -link-as-library"));
68 static cl::opt<const TargetMachineRegistry::Entry*, false, TargetNameParser>
69 MachineArch("march", cl::desc("Architecture to generate assembly for:"));
71 static cl::opt<bool> Native("native",
72 cl::desc("Generate a native binary instead of a shell script"));
74 static cl::opt<bool>NativeCBE("native-cbe",
75 cl::desc("Generate a native binary with the C backend and GCC"));
77 static cl::opt<bool>DisableCompression("disable-compression",cl::init(false),
78 cl::desc("Disable writing of compressed bytecode files"));
80 static cl::list<std::string> PostLinkOpts("post-link-opts",
81 cl::value_desc("path"),
82 cl::desc("Run one or more optimization programs after linking"));
84 static cl::list<std::string> XLinker("Xlinker", cl::value_desc("option"),
85 cl::desc("Pass options to the system linker"));
87 // Compatibility options that are ignored but supported by LD
88 static cl::opt<std::string> CO3("soname", cl::Hidden,
89 cl::desc("Compatibility option: ignored"));
91 static cl::opt<std::string> CO4("version-script", cl::Hidden,
92 cl::desc("Compatibility option: ignored"));
94 static cl::opt<bool> CO5("eh-frame-hdr", cl::Hidden,
95 cl::desc("Compatibility option: ignored"));
97 static cl::opt<std::string> CO6("h", cl::Hidden,
98 cl::desc("Compatibility option: ignored"));
101 /// This is just for convenience so it doesn't have to be passed around
103 static std::string progname;
105 /// PrintAndReturn - Prints a message to standard error and returns true.
108 /// progname - The name of the program (i.e. argv[0]).
109 /// Message - The message to print to standard error.
111 static int PrintAndReturn(const std::string &Message) {
112 std::cerr << progname << ": " << Message << "\n";
116 /// CopyEnv - This function takes an array of environment variables and makes a
117 /// copy of it. This copy can then be manipulated any way the caller likes
118 /// without affecting the process's real environment.
121 /// envp - An array of C strings containing an environment.
124 /// NULL - An error occurred.
126 /// Otherwise, a pointer to a new array of C strings is returned. Every string
127 /// in the array is a duplicate of the one in the original array (i.e. we do
128 /// not copy the char *'s from one array to another).
130 static char ** CopyEnv(char ** const envp) {
131 // Count the number of entries in the old list;
132 unsigned entries; // The number of entries in the old environment list
133 for (entries = 0; envp[entries] != NULL; entries++)
136 // Add one more entry for the NULL pointer that ends the list.
139 // If there are no entries at all, just return NULL.
143 // Allocate a new environment list.
144 char **newenv = new char* [entries];
145 if ((newenv = new char* [entries]) == NULL)
148 // Make a copy of the list. Don't forget the NULL that ends the list.
150 while (envp[entries] != NULL) {
151 newenv[entries] = new char[strlen (envp[entries]) + 1];
152 strcpy (newenv[entries], envp[entries]);
155 newenv[entries] = NULL;
161 /// RemoveEnv - Remove the specified environment variable from the environment
165 /// name - The name of the variable to remove. It cannot be NULL.
166 /// envp - The array of environment variables. It cannot be NULL.
169 /// This is mainly done because functions to remove items from the environment
170 /// are not available across all platforms. In particular, Solaris does not
171 /// seem to have an unsetenv() function or a setenv() function (or they are
172 /// undocumented if they do exist).
174 static void RemoveEnv(const char * name, char ** const envp) {
175 for (unsigned index=0; envp[index] != NULL; index++) {
176 // Find the first equals sign in the array and make it an EOS character.
177 char *p = strchr (envp[index], '=');
183 // Compare the two strings. If they are equal, zap this string.
184 // Otherwise, restore it.
185 if (!strcmp(name, envp[index]))
194 /// GenerateBytecode - generates a bytecode file from the module provided
195 void GenerateBytecode(Module* M, const std::string& FileName) {
197 // Create the output file.
198 std::ios::openmode io_mode = std::ios::out | std::ios::trunc |
200 std::ofstream Out(FileName.c_str(), io_mode);
202 PrintAndReturn("error opening '" + FileName + "' for writing!");
206 // Ensure that the bytecode file gets removed from the disk if we get a
207 // terminating signal.
208 sys::RemoveFileOnSignal(sys::Path(FileName));
211 WriteBytecodeToFile(M, Out, !DisableCompression);
213 // Close the bytecode file.
217 /// GenerateAssembly - generates a native assembly language source file from the
218 /// specified bytecode file.
221 /// InputFilename - The name of the input bytecode file.
222 /// OutputFilename - The name of the file to generate.
223 /// llc - The pathname to use for LLC.
224 /// envp - The environment to use when running LLC.
226 /// Return non-zero value on error.
228 static int GenerateAssembly(const std::string &OutputFilename,
229 const std::string &InputFilename,
230 const sys::Path &llc) {
231 // Run LLC to convert the bytecode file into assembly code.
232 std::vector<const char*> args;
233 args.push_back(llc.c_str());
234 args.push_back("-f");
235 args.push_back("-o");
236 args.push_back(OutputFilename.c_str());
237 args.push_back(InputFilename.c_str());
240 return sys::Program::ExecuteAndWait(llc,&args[0]);
243 /// GenerateCFile - generates a C source file from the specified bytecode file.
244 static int GenerateCFile(const std::string &OutputFile,
245 const std::string &InputFile,
246 const sys::Path &llc) {
247 // Run LLC to convert the bytecode file into C.
248 std::vector<const char*> args;
249 args.push_back(llc.c_str());
250 args.push_back("-march=c");
251 args.push_back("-f");
252 args.push_back("-o");
253 args.push_back(OutputFile.c_str());
254 args.push_back(InputFile.c_str());
256 return sys::Program::ExecuteAndWait(llc, &args[0]);
259 /// GenerateNative - generates a native object file from the
260 /// specified bytecode file.
263 /// InputFilename - The name of the input bytecode file.
264 /// OutputFilename - The name of the file to generate.
265 /// Libraries - The list of libraries with which to link.
266 /// LibPaths - The list of directories in which to find libraries.
267 /// gcc - The pathname to use for GGC.
268 /// envp - A copy of the process's current environment.
273 /// Returns non-zero value on error.
275 static int GenerateNative(const std::string &OutputFilename,
276 const std::string &InputFilename,
277 const std::vector<std::string> &Libraries,
278 const sys::Path &gcc, char ** const envp) {
279 // Remove these environment variables from the environment of the
280 // programs that we will execute. It appears that GCC sets these
281 // environment variables so that the programs it uses can configure
282 // themselves identically.
284 // However, when we invoke GCC below, we want it to use its normal
285 // configuration. Hence, we must sanitize its environment.
286 char ** clean_env = CopyEnv(envp);
287 if (clean_env == NULL)
289 RemoveEnv("LIBRARY_PATH", clean_env);
290 RemoveEnv("COLLECT_GCC_OPTIONS", clean_env);
291 RemoveEnv("GCC_EXEC_PREFIX", clean_env);
292 RemoveEnv("COMPILER_PATH", clean_env);
293 RemoveEnv("COLLECT_GCC", clean_env);
296 // Run GCC to assemble and link the program into native code.
299 // We can't just assemble and link the file with the system assembler
300 // and linker because we don't know where to put the _start symbol.
301 // GCC mysteriously knows how to do it.
302 std::vector<const char*> args;
303 args.push_back(gcc.c_str());
304 args.push_back("-fno-strict-aliasing");
305 args.push_back("-O3");
306 args.push_back("-o");
307 args.push_back(OutputFilename.c_str());
308 args.push_back(InputFilename.c_str());
310 // Add in the library paths
311 for (unsigned index = 0; index < LibPaths.size(); index++) {
312 args.push_back("-L");
313 args.push_back(LibPaths[index].c_str());
316 // Add the requested options
317 for (unsigned index = 0; index < XLinker.size(); index++) {
318 args.push_back(XLinker[index].c_str());
319 args.push_back(Libraries[index].c_str());
322 // Add in the libraries to link.
323 for (unsigned index = 0; index < Libraries.size(); index++)
324 if (Libraries[index] != "crtend") {
325 args.push_back("-l");
326 args.push_back(Libraries[index].c_str());
331 // Run the compiler to assembly and link together the program.
332 int R = sys::Program::ExecuteAndWait(gcc, &args[0], (const char**)clean_env);
337 /// EmitShellScript - Output the wrapper file that invokes the JIT on the LLVM
338 /// bytecode file for the program.
339 static void EmitShellScript(char **argv) {
340 #if defined(_WIN32) || defined(__CYGWIN__)
341 // Windows doesn't support #!/bin/sh style shell scripts in .exe files. To
342 // support windows systems, we copy the llvm-stub.exe executable from the
343 // build tree to the destination file.
344 sys::Path llvmstub = FindExecutable("llvm-stub.exe", argv[0]);
345 if (llvmstub.isEmpty()) {
346 std::cerr << "Could not find llvm-stub.exe executable!\n";
349 sys::CopyFile(sys::Path(OutputFilename), llvmstub);
353 // Output the script to start the program...
354 std::ofstream Out2(OutputFilename.c_str());
356 exit(PrintAndReturn("error opening '" + OutputFilename + "' for writing!"));
358 Out2 << "#!/bin/sh\n";
359 // Allow user to setenv LLVMINTERP if lli is not in their PATH.
360 Out2 << "lli=${LLVMINTERP-lli}\n";
361 Out2 << "exec $lli \\\n";
362 // gcc accepts -l<lib> and implicitly searches /lib and /usr/lib.
363 LibPaths.push_back("/lib");
364 LibPaths.push_back("/usr/lib");
365 LibPaths.push_back("/usr/X11R6/lib");
366 // We don't need to link in libc! In fact, /usr/lib/libc.so may not be a
367 // shared object at all! See RH 8: plain text.
368 std::vector<std::string>::iterator libc =
369 std::find(Libraries.begin(), Libraries.end(), "c");
370 if (libc != Libraries.end()) Libraries.erase(libc);
371 // List all the shared object (native) libraries this executable will need
372 // on the command line, so that we don't have to do this manually!
373 for (std::vector<std::string>::iterator i = Libraries.begin(),
374 e = Libraries.end(); i != e; ++i) {
375 sys::Path FullLibraryPath = sys::Path::FindLibrary(*i);
376 if (!FullLibraryPath.isEmpty() && FullLibraryPath.isDynamicLibrary())
377 Out2 << " -load=" << FullLibraryPath.toString() << " \\\n";
379 Out2 << " $0.bc ${1+\"$@\"}\n";
383 // BuildLinkItems -- This function generates a LinkItemList for the LinkItems
384 // linker function by combining the Files and Libraries in the order they were
385 // declared on the command line.
386 static void BuildLinkItems(
387 Linker::ItemList& Items,
388 const cl::list<std::string>& Files,
389 const cl::list<std::string>& Libraries) {
391 // Build the list of linkage items for LinkItems.
393 cl::list<std::string>::const_iterator fileIt = Files.begin();
394 cl::list<std::string>::const_iterator libIt = Libraries.begin();
396 int libPos = -1, filePos = -1;
397 while ( libIt != Libraries.end() || fileIt != Files.end() ) {
398 if (libIt != Libraries.end())
399 libPos = Libraries.getPosition(libIt - Libraries.begin());
402 if (fileIt != Files.end())
403 filePos = Files.getPosition(fileIt - Files.begin());
407 if (filePos != -1 && (libPos == -1 || filePos < libPos)) {
409 Items.push_back(std::make_pair(*fileIt++, false));
410 } else if (libPos != -1 && (filePos == -1 || libPos < filePos)) {
412 Items.push_back(std::make_pair(*libIt++, true));
417 // Rightly this should go in a header file but it just seems such a waste.
419 extern void Optimize(Module*);
422 int main(int argc, char **argv, char **envp) {
424 // Initial global variable above for convenience printing of program name.
425 progname = sys::Path(argv[0]).getBasename();
426 Linker TheLinker(progname, OutputFilename, Verbose);
428 // Parse the command line options
429 cl::ParseCommandLineOptions(argc, argv, " llvm linker\n");
430 sys::PrintStackTraceOnErrorSignal();
432 // Set up the library paths for the Linker
433 TheLinker.addPaths(LibPaths);
434 TheLinker.addSystemPaths();
436 // Remove any consecutive duplicates of the same library...
437 Libraries.erase(std::unique(Libraries.begin(), Libraries.end()),
441 std::vector<sys::Path> Files;
442 for (unsigned i = 0; i < InputFilenames.size(); ++i )
443 Files.push_back(sys::Path(InputFilenames[i]));
444 if (TheLinker.LinkInFiles(Files))
445 return 1; // Error already printed
447 // The libraries aren't linked in but are noted as "dependent" in the
449 for (cl::list<std::string>::const_iterator I = Libraries.begin(),
450 E = Libraries.end(); I != E ; ++I) {
451 TheLinker.getModule()->addLibrary(*I);
454 // Build a list of the items from our command line
455 Linker::ItemList Items;
456 Linker::ItemList NativeItems;
457 BuildLinkItems(Items, InputFilenames, Libraries);
459 // Link all the items together
460 if (TheLinker.LinkInItems(Items,NativeItems) )
464 std::auto_ptr<Module> Composite(TheLinker.releaseModule());
466 // Optimize the module
467 Optimize(Composite.get());
469 // Generate the bytecode for the optimized module.
470 std::string RealBytecodeOutput = OutputFilename;
471 if (!LinkAsLibrary) RealBytecodeOutput += ".bc";
472 GenerateBytecode(Composite.get(), RealBytecodeOutput);
474 // If we are not linking a library, generate either a native executable
475 // or a JIT shell script, depending upon what the user wants.
476 if (!LinkAsLibrary) {
477 // If the user wants to run a post-link optimization, run it now.
478 if (!PostLinkOpts.empty()) {
479 std::vector<std::string> opts = PostLinkOpts;
480 for (std::vector<std::string>::iterator I = opts.begin(),
481 E = opts.end(); I != E; ++I) {
483 if (!prog.canExecute()) {
484 prog = sys::Program::FindProgramByName(*I);
486 return PrintAndReturn(std::string("Optimization program '") + *I +
487 "' is not found or not executable.");
489 // Get the program arguments
490 sys::Path tmp_output("opt_result");
491 if (!tmp_output.createTemporaryFileOnDisk()) {
492 return PrintAndReturn(
493 "Can't create temporary file for post-link optimization");
496 args[0] = I->c_str();
497 args[1] = RealBytecodeOutput.c_str();
498 args[2] = tmp_output.c_str();
500 if (0 == sys::Program::ExecuteAndWait(prog, args)) {
501 if (tmp_output.isBytecodeFile()) {
502 sys::Path target(RealBytecodeOutput);
503 target.eraseFromDisk();
504 tmp_output.renamePathOnDisk(target);
506 return PrintAndReturn(
507 "Post-link optimization output is not bytecode");
512 // If the user wants to generate a native executable, compile it from the
515 // Otherwise, create a script that will run the bytecode through the JIT.
517 // Name of the Assembly Language output file
518 sys::Path AssemblyFile ( OutputFilename);
519 AssemblyFile.appendSuffix("s");
521 // Mark the output files for removal if we get an interrupt.
522 sys::RemoveFileOnSignal(AssemblyFile);
523 sys::RemoveFileOnSignal(sys::Path(OutputFilename));
525 // Determine the locations of the llc and gcc programs.
526 sys::Path llc = FindExecutable("llc", argv[0]);
528 return PrintAndReturn("Failed to find llc");
530 sys::Path gcc = FindExecutable("gcc", argv[0]);
532 return PrintAndReturn("Failed to find gcc");
534 // Generate an assembly language file for the bytecode.
535 if (Verbose) std::cout << "Generating Assembly Code\n";
536 GenerateAssembly(AssemblyFile.toString(), RealBytecodeOutput, llc);
537 if (Verbose) std::cout << "Generating Native Code\n";
538 GenerateNative(OutputFilename, AssemblyFile.toString(), Libraries,
541 // Remove the assembly language file.
542 AssemblyFile.eraseFromDisk();
543 } else if (NativeCBE) {
544 sys::Path CFile (OutputFilename);
545 CFile.appendSuffix("cbe.c");
547 // Mark the output files for removal if we get an interrupt.
548 sys::RemoveFileOnSignal(CFile);
549 sys::RemoveFileOnSignal(sys::Path(OutputFilename));
551 // Determine the locations of the llc and gcc programs.
552 sys::Path llc = FindExecutable("llc", argv[0]);
554 return PrintAndReturn("Failed to find llc");
556 sys::Path gcc = FindExecutable("gcc", argv[0]);
558 return PrintAndReturn("Failed to find gcc");
560 // Generate an assembly language file for the bytecode.
561 if (Verbose) std::cout << "Generating Assembly Code\n";
562 GenerateCFile(CFile.toString(), RealBytecodeOutput, llc);
563 if (Verbose) std::cout << "Generating Native Code\n";
564 GenerateNative(OutputFilename, CFile.toString(), Libraries, gcc, envp);
566 // Remove the assembly language file.
567 CFile.eraseFromDisk();
570 EmitShellScript(argv);
573 // Make the script executable...
574 sys::Path(OutputFilename).makeExecutableOnDisk();
576 // Make the bytecode file readable and directly executable in LLEE as well
577 sys::Path(RealBytecodeOutput).makeExecutableOnDisk();
578 sys::Path(RealBytecodeOutput).makeReadableOnDisk();
582 } catch (const std::string& msg) {
583 std::cerr << argv[0] << ": " << msg << "\n";
585 std::cerr << argv[0] << ": Unexpected unknown exception occurred.\n";