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/Bitcode/ReaderWriter.h"
29 #include "llvm/Target/TargetData.h"
30 #include "llvm/Target/TargetMachine.h"
31 #include "llvm/Target/TargetMachineRegistry.h"
32 #include "llvm/Support/CommandLine.h"
33 #include "llvm/Support/FileUtilities.h"
34 #include "llvm/Support/ManagedStatic.h"
35 #include "llvm/Support/MemoryBuffer.h"
36 #include "llvm/Support/Streams.h"
37 #include "llvm/Support/SystemUtils.h"
38 #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 // Options to control the linking, optimization, and code gen processes
63 static cl::opt<bool> LinkAsLibrary("link-as-library",
64 cl::desc("Link the .bc files together as a library, not an executable"));
66 static cl::alias Relink("r", cl::aliasopt(LinkAsLibrary),
67 cl::desc("Alias for -link-as-library"));
69 static cl::opt<const TargetMachineRegistry::Entry*, false, TargetNameParser>
70 MachineArch("march", cl::desc("Architecture to generate assembly for:"));
72 static cl::opt<bool> Native("native",
73 cl::desc("Generate a native binary instead of a shell script"));
75 static cl::opt<bool>NativeCBE("native-cbe",
76 cl::desc("Generate a native binary with the C backend and GCC"));
78 static cl::list<std::string> PostLinkOpts("post-link-opts",
79 cl::value_desc("path"),
80 cl::desc("Run one or more optimization programs after linking"));
82 static cl::list<std::string> XLinker("Xlinker", cl::value_desc("option"),
83 cl::desc("Pass options to the system linker"));
85 // Compatibility options that llvm-ld ignores but are supported for
86 // compatibility with LD
87 static cl::opt<std::string> CO3("soname", cl::Hidden,
88 cl::desc("Compatibility option: ignored"));
90 static cl::opt<std::string> CO4("version-script", cl::Hidden,
91 cl::desc("Compatibility option: ignored"));
93 static cl::opt<bool> CO5("eh-frame-hdr", cl::Hidden,
94 cl::desc("Compatibility option: ignored"));
96 static cl::opt<std::string> CO6("h", cl::Hidden,
97 cl::desc("Compatibility option: ignored"));
99 static cl::opt<bool> CO7("start-group", cl::Hidden,
100 cl::desc("Compatibility option: ignored"));
102 static cl::opt<bool> CO8("end-group", cl::Hidden,
103 cl::desc("Compatibility option: ignored"));
105 /// This is just for convenience so it doesn't have to be passed around
107 static std::string progname;
109 /// PrintAndExit - Prints a message to standard error and exits with error code
112 /// Message - The message to print to standard error.
114 static void PrintAndExit(const std::string &Message, int errcode = 1) {
115 cerr << progname << ": " << Message << "\n";
120 static void PrintCommand(const std::vector<const char*> &args) {
121 std::vector<const char*>::const_iterator I = args.begin(), E = args.end();
124 cout << "'" << *I << "'" << " ";
125 cout << "\n" << std::flush;
128 /// CopyEnv - This function takes an array of environment variables and makes a
129 /// copy of it. This copy can then be manipulated any way the caller likes
130 /// without affecting the process's real environment.
133 /// envp - An array of C strings containing an environment.
136 /// NULL - An error occurred.
138 /// Otherwise, a pointer to a new array of C strings is returned. Every string
139 /// in the array is a duplicate of the one in the original array (i.e. we do
140 /// not copy the char *'s from one array to another).
142 static char ** CopyEnv(char ** const envp) {
143 // Count the number of entries in the old list;
144 unsigned entries; // The number of entries in the old environment list
145 for (entries = 0; envp[entries] != NULL; entries++)
148 // Add one more entry for the NULL pointer that ends the list.
151 // If there are no entries at all, just return NULL.
155 // Allocate a new environment list.
156 char **newenv = new char* [entries];
157 if ((newenv = new char* [entries]) == NULL)
160 // Make a copy of the list. Don't forget the NULL that ends the list.
162 while (envp[entries] != NULL) {
163 newenv[entries] = new char[strlen (envp[entries]) + 1];
164 strcpy (newenv[entries], envp[entries]);
167 newenv[entries] = NULL;
173 /// RemoveEnv - Remove the specified environment variable from the environment
177 /// name - The name of the variable to remove. It cannot be NULL.
178 /// envp - The array of environment variables. It cannot be NULL.
181 /// This is mainly done because functions to remove items from the environment
182 /// are not available across all platforms. In particular, Solaris does not
183 /// seem to have an unsetenv() function or a setenv() function (or they are
184 /// undocumented if they do exist).
186 static void RemoveEnv(const char * name, char ** const envp) {
187 for (unsigned index=0; envp[index] != NULL; index++) {
188 // Find the first equals sign in the array and make it an EOS character.
189 char *p = strchr (envp[index], '=');
195 // Compare the two strings. If they are equal, zap this string.
196 // Otherwise, restore it.
197 if (!strcmp(name, envp[index]))
206 /// GenerateBytecode - generates a bytecode file from the module provided
207 void GenerateBytecode(Module* M, const std::string& FileName) {
210 cout << "Generating Bytecode To " << FileName << '\n';
212 // Create the output file.
213 std::ios::openmode io_mode = std::ios::out | std::ios::trunc |
215 std::ofstream Out(FileName.c_str(), io_mode);
217 PrintAndExit("error opening '" + FileName + "' for writing!");
219 // Ensure that the bytecode file gets removed from the disk if we get a
220 // terminating signal.
221 sys::RemoveFileOnSignal(sys::Path(FileName));
224 WriteBitcodeToFile(M, Out);
226 // Close the bytecode file.
230 /// GenerateAssembly - generates a native assembly language source file from the
231 /// specified bytecode file.
234 /// InputFilename - The name of the input bytecode file.
235 /// OutputFilename - The name of the file to generate.
236 /// llc - The pathname to use for LLC.
237 /// envp - The environment to use when running LLC.
239 /// Return non-zero value on error.
241 static int GenerateAssembly(const std::string &OutputFilename,
242 const std::string &InputFilename,
243 const sys::Path &llc,
244 std::string &ErrMsg ) {
245 // Run LLC to convert the bytecode file into assembly code.
246 std::vector<const char*> args;
247 args.push_back(llc.c_str());
248 args.push_back("-f");
249 args.push_back("-o");
250 args.push_back(OutputFilename.c_str());
251 args.push_back(InputFilename.c_str());
255 cout << "Generating Assembly With: \n";
259 return sys::Program::ExecuteAndWait(llc, &args[0], 0, 0, 0, 0, &ErrMsg);
262 /// GenerateCFile - generates a C source file from the specified bytecode file.
263 static int GenerateCFile(const std::string &OutputFile,
264 const std::string &InputFile,
265 const sys::Path &llc,
266 std::string& ErrMsg) {
267 // Run LLC to convert the bytecode file into C.
268 std::vector<const char*> args;
269 args.push_back(llc.c_str());
270 args.push_back("-march=c");
271 args.push_back("-f");
272 args.push_back("-o");
273 args.push_back(OutputFile.c_str());
274 args.push_back(InputFile.c_str());
278 cout << "Generating C Source With: \n";
282 return sys::Program::ExecuteAndWait(llc, &args[0], 0, 0, 0, 0, &ErrMsg);
285 /// GenerateNative - generates a native object file from the
286 /// specified bytecode file.
289 /// InputFilename - The name of the input bytecode file.
290 /// OutputFilename - The name of the file to generate.
291 /// NativeLinkItems - The native libraries, files, code with which to link
292 /// LibPaths - The list of directories in which to find libraries.
293 /// gcc - The pathname to use for GGC.
294 /// envp - A copy of the process's current environment.
299 /// Returns non-zero value on error.
301 static int GenerateNative(const std::string &OutputFilename,
302 const std::string &InputFilename,
303 const Linker::ItemList &LinkItems,
304 const sys::Path &gcc, char ** const envp,
305 std::string& ErrMsg) {
306 // Remove these environment variables from the environment of the
307 // programs that we will execute. It appears that GCC sets these
308 // environment variables so that the programs it uses can configure
309 // themselves identically.
311 // However, when we invoke GCC below, we want it to use its normal
312 // configuration. Hence, we must sanitize its environment.
313 char ** clean_env = CopyEnv(envp);
314 if (clean_env == NULL)
316 RemoveEnv("LIBRARY_PATH", clean_env);
317 RemoveEnv("COLLECT_GCC_OPTIONS", clean_env);
318 RemoveEnv("GCC_EXEC_PREFIX", clean_env);
319 RemoveEnv("COMPILER_PATH", clean_env);
320 RemoveEnv("COLLECT_GCC", clean_env);
323 // Run GCC to assemble and link the program into native code.
326 // We can't just assemble and link the file with the system assembler
327 // and linker because we don't know where to put the _start symbol.
328 // GCC mysteriously knows how to do it.
329 std::vector<std::string> args;
330 args.push_back(gcc.c_str());
331 args.push_back("-fno-strict-aliasing");
332 args.push_back("-O3");
333 args.push_back("-o");
334 args.push_back(OutputFilename);
335 args.push_back(InputFilename);
337 // Add in the library paths
338 for (unsigned index = 0; index < LibPaths.size(); index++) {
339 args.push_back("-L");
340 args.push_back(LibPaths[index]);
343 // Add the requested options
344 for (unsigned index = 0; index < XLinker.size(); index++) {
345 args.push_back(XLinker[index]);
346 args.push_back(Libraries[index]);
349 // Add in the libraries to link.
350 for (unsigned index = 0; index < LinkItems.size(); index++)
351 if (LinkItems[index].first != "crtend") {
352 if (LinkItems[index].second)
353 args.push_back("-l" + LinkItems[index].first);
355 args.push_back(LinkItems[index].first);
359 // Now that "args" owns all the std::strings for the arguments, call the c_str
360 // method to get the underlying string array. We do this game so that the
361 // std::string array is guaranteed to outlive the const char* array.
362 std::vector<const char *> Args;
363 for (unsigned i = 0, e = args.size(); i != e; ++i)
364 Args.push_back(args[i].c_str());
368 cout << "Generating Native Executable With:\n";
372 // Run the compiler to assembly and link together the program.
373 int R = sys::Program::ExecuteAndWait(
374 gcc, &Args[0], (const char**)clean_env, 0, 0, 0, &ErrMsg);
379 /// EmitShellScript - Output the wrapper file that invokes the JIT on the LLVM
380 /// bytecode file for the program.
381 static void EmitShellScript(char **argv) {
383 cout << "Emitting Shell Script\n";
384 #if defined(_WIN32) || defined(__CYGWIN__)
385 // Windows doesn't support #!/bin/sh style shell scripts in .exe files. To
386 // support windows systems, we copy the llvm-stub.exe executable from the
387 // build tree to the destination file.
389 sys::Path llvmstub = FindExecutable("llvm-stub.exe", argv[0]);
390 if (llvmstub.isEmpty())
391 PrintAndExit("Could not find llvm-stub.exe executable!");
393 if (0 != sys::CopyFile(sys::Path(OutputFilename), llvmstub, &ErrMsg))
394 PrintAndExit(ErrMsg);
399 // Output the script to start the program...
400 std::ofstream Out2(OutputFilename.c_str());
402 PrintAndExit("error opening '" + OutputFilename + "' for writing!");
404 Out2 << "#!/bin/sh\n";
405 // Allow user to setenv LLVMINTERP if lli is not in their PATH.
406 Out2 << "lli=${LLVMINTERP-lli}\n";
407 Out2 << "exec $lli \\\n";
408 // gcc accepts -l<lib> and implicitly searches /lib and /usr/lib.
409 LibPaths.push_back("/lib");
410 LibPaths.push_back("/usr/lib");
411 LibPaths.push_back("/usr/X11R6/lib");
412 // We don't need to link in libc! In fact, /usr/lib/libc.so may not be a
413 // shared object at all! See RH 8: plain text.
414 std::vector<std::string>::iterator libc =
415 std::find(Libraries.begin(), Libraries.end(), "c");
416 if (libc != Libraries.end()) Libraries.erase(libc);
417 // List all the shared object (native) libraries this executable will need
418 // on the command line, so that we don't have to do this manually!
419 for (std::vector<std::string>::iterator i = Libraries.begin(),
420 e = Libraries.end(); i != e; ++i) {
421 sys::Path FullLibraryPath = sys::Path::FindLibrary(*i);
422 if (!FullLibraryPath.isEmpty() && FullLibraryPath.isDynamicLibrary())
423 Out2 << " -load=" << FullLibraryPath.toString() << " \\\n";
425 Out2 << " $0.bc ${1+\"$@\"}\n";
429 // BuildLinkItems -- This function generates a LinkItemList for the LinkItems
430 // linker function by combining the Files and Libraries in the order they were
431 // declared on the command line.
432 static void BuildLinkItems(
433 Linker::ItemList& Items,
434 const cl::list<std::string>& Files,
435 const cl::list<std::string>& Libraries) {
437 // Build the list of linkage items for LinkItems.
439 cl::list<std::string>::const_iterator fileIt = Files.begin();
440 cl::list<std::string>::const_iterator libIt = Libraries.begin();
442 int libPos = -1, filePos = -1;
443 while ( libIt != Libraries.end() || fileIt != Files.end() ) {
444 if (libIt != Libraries.end())
445 libPos = Libraries.getPosition(libIt - Libraries.begin());
448 if (fileIt != Files.end())
449 filePos = Files.getPosition(fileIt - Files.begin());
453 if (filePos != -1 && (libPos == -1 || filePos < libPos)) {
455 Items.push_back(std::make_pair(*fileIt++, false));
456 } else if (libPos != -1 && (filePos == -1 || libPos < filePos)) {
458 Items.push_back(std::make_pair(*libIt++, true));
463 // Rightly this should go in a header file but it just seems such a waste.
465 extern void Optimize(Module*);
468 int main(int argc, char **argv, char **envp) {
469 llvm_shutdown_obj X; // Call llvm_shutdown() on exit.
471 // Initial global variable above for convenience printing of program name.
472 progname = sys::Path(argv[0]).getBasename();
474 // Parse the command line options
475 cl::ParseCommandLineOptions(argc, argv, " llvm linker\n");
476 sys::PrintStackTraceOnErrorSignal();
478 // Construct a Linker (now that Verbose is set)
479 Linker TheLinker(progname, OutputFilename, Verbose);
481 // Keep track of the native link items (versus the bytecode items)
482 Linker::ItemList NativeLinkItems;
484 // Add library paths to the linker
485 TheLinker.addPaths(LibPaths);
486 TheLinker.addSystemPaths();
488 // Remove any consecutive duplicates of the same library...
489 Libraries.erase(std::unique(Libraries.begin(), Libraries.end()),
493 std::vector<sys::Path> Files;
494 for (unsigned i = 0; i < InputFilenames.size(); ++i )
495 Files.push_back(sys::Path(InputFilenames[i]));
496 if (TheLinker.LinkInFiles(Files))
497 return 1; // Error already printed
499 // The libraries aren't linked in but are noted as "dependent" in the
501 for (cl::list<std::string>::const_iterator I = Libraries.begin(),
502 E = Libraries.end(); I != E ; ++I) {
503 TheLinker.getModule()->addLibrary(*I);
506 // Build a list of the items from our command line
507 Linker::ItemList Items;
508 BuildLinkItems(Items, InputFilenames, Libraries);
510 // Link all the items together
511 if (TheLinker.LinkInItems(Items, NativeLinkItems) )
512 return 1; // Error already printed
515 std::auto_ptr<Module> Composite(TheLinker.releaseModule());
517 // Optimize the module
518 Optimize(Composite.get());
520 // Generate the bytecode for the optimized module.
521 std::string RealBytecodeOutput = OutputFilename;
522 if (!LinkAsLibrary) RealBytecodeOutput += ".bc";
523 GenerateBytecode(Composite.get(), RealBytecodeOutput);
525 // If we are not linking a library, generate either a native executable
526 // or a JIT shell script, depending upon what the user wants.
527 if (!LinkAsLibrary) {
528 // If the user wants to run a post-link optimization, run it now.
529 if (!PostLinkOpts.empty()) {
530 std::vector<std::string> opts = PostLinkOpts;
531 for (std::vector<std::string>::iterator I = opts.begin(),
532 E = opts.end(); I != E; ++I) {
534 if (!prog.canExecute()) {
535 prog = sys::Program::FindProgramByName(*I);
537 PrintAndExit(std::string("Optimization program '") + *I +
538 "' is not found or not executable.");
540 // Get the program arguments
541 sys::Path tmp_output("opt_result");
543 if (tmp_output.createTemporaryFileOnDisk(true, &ErrMsg))
544 PrintAndExit(ErrMsg);
547 args[0] = I->c_str();
548 args[1] = RealBytecodeOutput.c_str();
549 args[2] = tmp_output.c_str();
551 if (0 == sys::Program::ExecuteAndWait(prog, args, 0,0,0,0, &ErrMsg)) {
552 if (tmp_output.isBytecodeFile() || tmp_output.isBitcodeFile()) {
553 sys::Path target(RealBytecodeOutput);
554 target.eraseFromDisk();
555 if (tmp_output.renamePathOnDisk(target, &ErrMsg))
556 PrintAndExit(ErrMsg, 2);
558 PrintAndExit("Post-link optimization output is not bytecode");
560 PrintAndExit(ErrMsg);
565 // If the user wants to generate a native executable, compile it from the
568 // Otherwise, create a script that will run the bytecode through the JIT.
570 // Name of the Assembly Language output file
571 sys::Path AssemblyFile ( OutputFilename);
572 AssemblyFile.appendSuffix("s");
574 // Mark the output files for removal if we get an interrupt.
575 sys::RemoveFileOnSignal(AssemblyFile);
576 sys::RemoveFileOnSignal(sys::Path(OutputFilename));
578 // Determine the locations of the llc and gcc programs.
579 sys::Path llc = FindExecutable("llc", argv[0]);
581 PrintAndExit("Failed to find llc");
583 sys::Path gcc = FindExecutable("gcc", argv[0]);
585 PrintAndExit("Failed to find gcc");
587 // Generate an assembly language file for the bytecode.
589 if (0 != GenerateAssembly(AssemblyFile.toString(), RealBytecodeOutput,
591 PrintAndExit(ErrMsg);
593 if (0 != GenerateNative(OutputFilename, AssemblyFile.toString(),
594 NativeLinkItems, gcc, envp, ErrMsg))
595 PrintAndExit(ErrMsg);
597 // Remove the assembly language file.
598 AssemblyFile.eraseFromDisk();
599 } else if (NativeCBE) {
600 sys::Path CFile (OutputFilename);
601 CFile.appendSuffix("cbe.c");
603 // Mark the output files for removal if we get an interrupt.
604 sys::RemoveFileOnSignal(CFile);
605 sys::RemoveFileOnSignal(sys::Path(OutputFilename));
607 // Determine the locations of the llc and gcc programs.
608 sys::Path llc = FindExecutable("llc", argv[0]);
610 PrintAndExit("Failed to find llc");
612 sys::Path gcc = FindExecutable("gcc", argv[0]);
614 PrintAndExit("Failed to find gcc");
616 // Generate an assembly language file for the bytecode.
618 if (0 != GenerateCFile(
619 CFile.toString(), RealBytecodeOutput, llc, ErrMsg))
620 PrintAndExit(ErrMsg);
622 if (0 != GenerateNative(OutputFilename, CFile.toString(),
623 NativeLinkItems, gcc, envp, ErrMsg))
624 PrintAndExit(ErrMsg);
626 // Remove the assembly language file.
627 CFile.eraseFromDisk();
630 EmitShellScript(argv);
633 // Make the script executable...
635 if (sys::Path(OutputFilename).makeExecutableOnDisk(&ErrMsg))
636 PrintAndExit(ErrMsg);
638 // Make the bytecode file readable and directly executable in LLEE as well
639 if (sys::Path(RealBytecodeOutput).makeExecutableOnDisk(&ErrMsg))
640 PrintAndExit(ErrMsg);
642 if (sys::Path(RealBytecodeOutput).makeReadableOnDisk(&ErrMsg))
643 PrintAndExit(ErrMsg);
645 } catch (const std::string& msg) {
648 PrintAndExit("Unexpected unknown exception occurred.", 2);