1 //===- llvm-ld.cpp - LLVM 'ld' compatible linker --------------------------===//
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 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 bitcode 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"
44 // Input/Output Options
45 static cl::list<std::string> InputFilenames(cl::Positional, cl::OneOrMore,
46 cl::desc("<input bitcode files>"));
48 static cl::opt<std::string> OutputFilename("o", cl::init("a.out"),
49 cl::desc("Override output filename"),
50 cl::value_desc("filename"));
52 static cl::opt<bool> Verbose("v",
53 cl::desc("Print information about actions taken"));
55 static cl::list<std::string> LibPaths("L", cl::Prefix,
56 cl::desc("Specify a library search path"),
57 cl::value_desc("directory"));
59 static cl::list<std::string> FrameworkPaths("F", cl::Prefix,
60 cl::desc("Specify a framework search path"),
61 cl::value_desc("directory"));
63 static cl::list<std::string> Libraries("l", cl::Prefix,
64 cl::desc("Specify libraries to link to"),
65 cl::value_desc("library prefix"));
67 static cl::list<std::string> Frameworks("framework",
68 cl::desc("Specify frameworks to link to"),
69 cl::value_desc("framework"));
71 // Options to control the linking, optimization, and code gen processes
72 static cl::opt<bool> LinkAsLibrary("link-as-library",
73 cl::desc("Link the .bc files together as a library, not an executable"));
75 static cl::alias Relink("r", cl::aliasopt(LinkAsLibrary),
76 cl::desc("Alias for -link-as-library"));
78 static cl::opt<bool> Native("native",
79 cl::desc("Generate a native binary instead of a shell script"));
81 static cl::opt<bool>NativeCBE("native-cbe",
82 cl::desc("Generate a native binary with the C backend and GCC"));
84 static cl::list<std::string> PostLinkOpts("post-link-opts",
85 cl::value_desc("path"),
86 cl::desc("Run one or more optimization programs after linking"));
88 static cl::list<std::string> XLinker("Xlinker", cl::value_desc("option"),
89 cl::desc("Pass options to the system linker"));
91 // Compatibility options that llvm-ld ignores but are supported for
92 // compatibility with LD
93 static cl::opt<std::string> CO3("soname", cl::Hidden,
94 cl::desc("Compatibility option: ignored"));
96 static cl::opt<std::string> CO4("version-script", cl::Hidden,
97 cl::desc("Compatibility option: ignored"));
99 static cl::opt<bool> CO5("eh-frame-hdr", cl::Hidden,
100 cl::desc("Compatibility option: ignored"));
102 static cl::opt<std::string> CO6("h", cl::Hidden,
103 cl::desc("Compatibility option: ignored"));
105 static cl::opt<bool> CO7("start-group", cl::Hidden,
106 cl::desc("Compatibility option: ignored"));
108 static cl::opt<bool> CO8("end-group", cl::Hidden,
109 cl::desc("Compatibility option: ignored"));
111 static cl::opt<std::string> CO9("m", cl::Hidden,
112 cl::desc("Compatibility option: ignored"));
114 /// This is just for convenience so it doesn't have to be passed around
116 static std::string progname;
118 /// PrintAndExit - Prints a message to standard error and exits with error code
121 /// Message - The message to print to standard error.
123 static void PrintAndExit(const std::string &Message, int errcode = 1) {
124 cerr << progname << ": " << Message << "\n";
129 static void PrintCommand(const std::vector<const char*> &args) {
130 std::vector<const char*>::const_iterator I = args.begin(), E = args.end();
133 cout << "'" << *I << "'" << " ";
134 cout << "\n" << std::flush;
137 /// CopyEnv - This function takes an array of environment variables and makes a
138 /// copy of it. This copy can then be manipulated any way the caller likes
139 /// without affecting the process's real environment.
142 /// envp - An array of C strings containing an environment.
145 /// NULL - An error occurred.
147 /// Otherwise, a pointer to a new array of C strings is returned. Every string
148 /// in the array is a duplicate of the one in the original array (i.e. we do
149 /// not copy the char *'s from one array to another).
151 static char ** CopyEnv(char ** const envp) {
152 // Count the number of entries in the old list;
153 unsigned entries; // The number of entries in the old environment list
154 for (entries = 0; envp[entries] != NULL; entries++)
157 // Add one more entry for the NULL pointer that ends the list.
160 // If there are no entries at all, just return NULL.
164 // Allocate a new environment list.
165 char **newenv = new char* [entries];
166 if ((newenv = new char* [entries]) == NULL)
169 // Make a copy of the list. Don't forget the NULL that ends the list.
171 while (envp[entries] != NULL) {
172 newenv[entries] = new char[strlen (envp[entries]) + 1];
173 strcpy (newenv[entries], envp[entries]);
176 newenv[entries] = NULL;
182 /// RemoveEnv - Remove the specified environment variable from the environment
186 /// name - The name of the variable to remove. It cannot be NULL.
187 /// envp - The array of environment variables. It cannot be NULL.
190 /// This is mainly done because functions to remove items from the environment
191 /// are not available across all platforms. In particular, Solaris does not
192 /// seem to have an unsetenv() function or a setenv() function (or they are
193 /// undocumented if they do exist).
195 static void RemoveEnv(const char * name, char ** const envp) {
196 for (unsigned index=0; envp[index] != NULL; index++) {
197 // Find the first equals sign in the array and make it an EOS character.
198 char *p = strchr (envp[index], '=');
204 // Compare the two strings. If they are equal, zap this string.
205 // Otherwise, restore it.
206 if (!strcmp(name, envp[index]))
215 /// GenerateBitcode - generates a bitcode file from the module provided
216 void GenerateBitcode(Module* M, const std::string& FileName) {
219 cout << "Generating Bitcode To " << FileName << '\n';
221 // Create the output file.
222 std::ios::openmode io_mode = std::ios::out | std::ios::trunc |
224 std::ofstream Out(FileName.c_str(), io_mode);
226 PrintAndExit("error opening '" + FileName + "' for writing!");
228 // Ensure that the bitcode file gets removed from the disk if we get a
229 // terminating signal.
230 sys::RemoveFileOnSignal(sys::Path(FileName));
233 WriteBitcodeToFile(M, Out);
235 // Close the bitcode file.
239 /// GenerateAssembly - generates a native assembly language source file from the
240 /// specified bitcode file.
243 /// InputFilename - The name of the input bitcode file.
244 /// OutputFilename - The name of the file to generate.
245 /// llc - The pathname to use for LLC.
246 /// envp - The environment to use when running LLC.
248 /// Return non-zero value on error.
250 static int GenerateAssembly(const std::string &OutputFilename,
251 const std::string &InputFilename,
252 const sys::Path &llc,
253 std::string &ErrMsg ) {
254 // Run LLC to convert the bitcode file into assembly code.
255 std::vector<const char*> args;
256 args.push_back(llc.c_str());
257 // We will use GCC to assemble the program so set the assembly syntax to AT&T,
258 // regardless of what the target in the bitcode file is.
259 args.push_back("-x86-asm-syntax=att");
260 args.push_back("-f");
261 args.push_back("-o");
262 args.push_back(OutputFilename.c_str());
263 args.push_back(InputFilename.c_str());
267 cout << "Generating Assembly With: \n";
271 return sys::Program::ExecuteAndWait(llc, &args[0], 0, 0, 0, 0, &ErrMsg);
274 /// GenerateCFile - generates a C source file from the specified bitcode file.
275 static int GenerateCFile(const std::string &OutputFile,
276 const std::string &InputFile,
277 const sys::Path &llc,
278 std::string& ErrMsg) {
279 // Run LLC to convert the bitcode file into C.
280 std::vector<const char*> args;
281 args.push_back(llc.c_str());
282 args.push_back("-march=c");
283 args.push_back("-f");
284 args.push_back("-o");
285 args.push_back(OutputFile.c_str());
286 args.push_back(InputFile.c_str());
290 cout << "Generating C Source With: \n";
294 return sys::Program::ExecuteAndWait(llc, &args[0], 0, 0, 0, 0, &ErrMsg);
297 /// GenerateNative - generates a native object file from the
298 /// specified bitcode file.
301 /// InputFilename - The name of the input bitcode file.
302 /// OutputFilename - The name of the file to generate.
303 /// NativeLinkItems - The native libraries, files, code with which to link
304 /// LibPaths - The list of directories in which to find libraries.
305 /// FrameworksPaths - The list of directories in which to find frameworks.
306 /// Frameworks - The list of frameworks (dynamic libraries)
307 /// gcc - The pathname to use for GGC.
308 /// envp - A copy of the process's current environment.
313 /// Returns non-zero value on error.
315 static int GenerateNative(const std::string &OutputFilename,
316 const std::string &InputFilename,
317 const Linker::ItemList &LinkItems,
318 const sys::Path &gcc, char ** const envp,
319 std::string& ErrMsg) {
320 // Remove these environment variables from the environment of the
321 // programs that we will execute. It appears that GCC sets these
322 // environment variables so that the programs it uses can configure
323 // themselves identically.
325 // However, when we invoke GCC below, we want it to use its normal
326 // configuration. Hence, we must sanitize its environment.
327 char ** clean_env = CopyEnv(envp);
328 if (clean_env == NULL)
330 RemoveEnv("LIBRARY_PATH", clean_env);
331 RemoveEnv("COLLECT_GCC_OPTIONS", clean_env);
332 RemoveEnv("GCC_EXEC_PREFIX", clean_env);
333 RemoveEnv("COMPILER_PATH", clean_env);
334 RemoveEnv("COLLECT_GCC", clean_env);
337 // Run GCC to assemble and link the program into native code.
340 // We can't just assemble and link the file with the system assembler
341 // and linker because we don't know where to put the _start symbol.
342 // GCC mysteriously knows how to do it.
343 std::vector<std::string> args;
344 args.push_back(gcc.c_str());
345 args.push_back("-fno-strict-aliasing");
346 args.push_back("-O3");
347 args.push_back("-o");
348 args.push_back(OutputFilename);
349 args.push_back(InputFilename);
351 // Add in the library and framework paths
352 for (unsigned index = 0; index < LibPaths.size(); index++) {
353 args.push_back("-L" + LibPaths[index]);
355 for (unsigned index = 0; index < FrameworkPaths.size(); index++) {
356 args.push_back("-F" + FrameworkPaths[index]);
359 // Add the requested options
360 for (unsigned index = 0; index < XLinker.size(); index++)
361 args.push_back(XLinker[index]);
363 // Add in the libraries to link.
364 for (unsigned index = 0; index < LinkItems.size(); index++)
365 if (LinkItems[index].first != "crtend") {
366 if (LinkItems[index].second)
367 args.push_back("-l" + LinkItems[index].first);
369 args.push_back(LinkItems[index].first);
372 // Add in frameworks to link.
373 for (unsigned index = 0; index < Frameworks.size(); index++) {
374 args.push_back("-framework");
375 args.push_back(Frameworks[index]);
378 // Now that "args" owns all the std::strings for the arguments, call the c_str
379 // method to get the underlying string array. We do this game so that the
380 // std::string array is guaranteed to outlive the const char* array.
381 std::vector<const char *> Args;
382 for (unsigned i = 0, e = args.size(); i != e; ++i)
383 Args.push_back(args[i].c_str());
387 cout << "Generating Native Executable With:\n";
391 // Run the compiler to assembly and link together the program.
392 int R = sys::Program::ExecuteAndWait(
393 gcc, &Args[0], (const char**)clean_env, 0, 0, 0, &ErrMsg);
398 /// EmitShellScript - Output the wrapper file that invokes the JIT on the LLVM
399 /// bitcode file for the program.
400 static void EmitShellScript(char **argv) {
402 cout << "Emitting Shell Script\n";
403 #if defined(_WIN32) || defined(__CYGWIN__)
404 // Windows doesn't support #!/bin/sh style shell scripts in .exe files. To
405 // support windows systems, we copy the llvm-stub.exe executable from the
406 // build tree to the destination file.
408 sys::Path llvmstub = FindExecutable("llvm-stub.exe", argv[0]);
409 if (llvmstub.isEmpty())
410 PrintAndExit("Could not find llvm-stub.exe executable!");
412 if (0 != sys::CopyFile(sys::Path(OutputFilename), llvmstub, &ErrMsg))
413 PrintAndExit(ErrMsg);
418 // Output the script to start the program...
419 std::ofstream Out2(OutputFilename.c_str());
421 PrintAndExit("error opening '" + OutputFilename + "' for writing!");
423 Out2 << "#!/bin/sh\n";
424 // Allow user to setenv LLVMINTERP if lli is not in their PATH.
425 Out2 << "lli=${LLVMINTERP-lli}\n";
426 Out2 << "exec $lli \\\n";
427 // gcc accepts -l<lib> and implicitly searches /lib and /usr/lib.
428 LibPaths.push_back("/lib");
429 LibPaths.push_back("/usr/lib");
430 LibPaths.push_back("/usr/X11R6/lib");
431 // We don't need to link in libc! In fact, /usr/lib/libc.so may not be a
432 // shared object at all! See RH 8: plain text.
433 std::vector<std::string>::iterator libc =
434 std::find(Libraries.begin(), Libraries.end(), "c");
435 if (libc != Libraries.end()) Libraries.erase(libc);
436 // List all the shared object (native) libraries this executable will need
437 // on the command line, so that we don't have to do this manually!
438 for (std::vector<std::string>::iterator i = Libraries.begin(),
439 e = Libraries.end(); i != e; ++i) {
440 sys::Path FullLibraryPath = sys::Path::FindLibrary(*i);
441 if (!FullLibraryPath.isEmpty() && FullLibraryPath.isDynamicLibrary())
442 Out2 << " -load=" << FullLibraryPath.toString() << " \\\n";
444 Out2 << " $0.bc ${1+\"$@\"}\n";
448 // BuildLinkItems -- This function generates a LinkItemList for the LinkItems
449 // linker function by combining the Files and Libraries in the order they were
450 // declared on the command line.
451 static void BuildLinkItems(
452 Linker::ItemList& Items,
453 const cl::list<std::string>& Files,
454 const cl::list<std::string>& Libraries) {
456 // Build the list of linkage items for LinkItems.
458 cl::list<std::string>::const_iterator fileIt = Files.begin();
459 cl::list<std::string>::const_iterator libIt = Libraries.begin();
461 int libPos = -1, filePos = -1;
462 while ( libIt != Libraries.end() || fileIt != Files.end() ) {
463 if (libIt != Libraries.end())
464 libPos = Libraries.getPosition(libIt - Libraries.begin());
467 if (fileIt != Files.end())
468 filePos = Files.getPosition(fileIt - Files.begin());
472 if (filePos != -1 && (libPos == -1 || filePos < libPos)) {
474 Items.push_back(std::make_pair(*fileIt++, false));
475 } else if (libPos != -1 && (filePos == -1 || libPos < filePos)) {
477 Items.push_back(std::make_pair(*libIt++, true));
482 // Rightly this should go in a header file but it just seems such a waste.
484 extern void Optimize(Module*);
487 int main(int argc, char **argv, char **envp) {
488 llvm_shutdown_obj X; // Call llvm_shutdown() on exit.
490 // Initial global variable above for convenience printing of program name.
491 progname = sys::Path(argv[0]).getBasename();
493 // Parse the command line options
494 cl::ParseCommandLineOptions(argc, argv, "llvm linker\n");
495 sys::PrintStackTraceOnErrorSignal();
497 // Construct a Linker (now that Verbose is set)
498 Linker TheLinker(progname, OutputFilename, Verbose);
500 // Keep track of the native link items (versus the bitcode items)
501 Linker::ItemList NativeLinkItems;
503 // Add library paths to the linker
504 TheLinker.addPaths(LibPaths);
505 TheLinker.addSystemPaths();
507 // Remove any consecutive duplicates of the same library...
508 Libraries.erase(std::unique(Libraries.begin(), Libraries.end()),
512 std::vector<sys::Path> Files;
513 for (unsigned i = 0; i < InputFilenames.size(); ++i )
514 Files.push_back(sys::Path(InputFilenames[i]));
515 if (TheLinker.LinkInFiles(Files))
516 return 1; // Error already printed
518 // The libraries aren't linked in but are noted as "dependent" in the
520 for (cl::list<std::string>::const_iterator I = Libraries.begin(),
521 E = Libraries.end(); I != E ; ++I) {
522 TheLinker.getModule()->addLibrary(*I);
525 // Build a list of the items from our command line
526 Linker::ItemList Items;
527 BuildLinkItems(Items, InputFilenames, Libraries);
529 // Link all the items together
530 if (TheLinker.LinkInItems(Items, NativeLinkItems) )
531 return 1; // Error already printed
534 std::auto_ptr<Module> Composite(TheLinker.releaseModule());
536 // Optimize the module
537 Optimize(Composite.get());
539 #if defined(_WIN32) || defined(__CYGWIN__)
540 if (!LinkAsLibrary) {
541 // Default to "a.exe" instead of "a.out".
542 if (OutputFilename.getNumOccurrences() == 0)
543 OutputFilename = "a.exe";
545 // If there is no suffix add an "exe" one.
546 sys::Path ExeFile( OutputFilename );
547 if (ExeFile.getSuffix() == "") {
548 ExeFile.appendSuffix("exe");
549 OutputFilename = ExeFile.toString();
554 // Generate the bitcode for the optimized module.
555 std::string RealBitcodeOutput = OutputFilename;
557 if (!LinkAsLibrary) RealBitcodeOutput += ".bc";
558 GenerateBitcode(Composite.get(), RealBitcodeOutput);
560 // If we are not linking a library, generate either a native executable
561 // or a JIT shell script, depending upon what the user wants.
562 if (!LinkAsLibrary) {
563 // If the user wants to run a post-link optimization, run it now.
564 if (!PostLinkOpts.empty()) {
565 std::vector<std::string> opts = PostLinkOpts;
566 for (std::vector<std::string>::iterator I = opts.begin(),
567 E = opts.end(); I != E; ++I) {
569 if (!prog.canExecute()) {
570 prog = sys::Program::FindProgramByName(*I);
572 PrintAndExit(std::string("Optimization program '") + *I +
573 "' is not found or not executable.");
575 // Get the program arguments
576 sys::Path tmp_output("opt_result");
578 if (tmp_output.createTemporaryFileOnDisk(true, &ErrMsg))
579 PrintAndExit(ErrMsg);
582 args[0] = I->c_str();
583 args[1] = RealBitcodeOutput.c_str();
584 args[2] = tmp_output.c_str();
586 if (0 == sys::Program::ExecuteAndWait(prog, args, 0,0,0,0, &ErrMsg)) {
587 if (tmp_output.isBitcodeFile() || tmp_output.isBitcodeFile()) {
588 sys::Path target(RealBitcodeOutput);
589 target.eraseFromDisk();
590 if (tmp_output.renamePathOnDisk(target, &ErrMsg))
591 PrintAndExit(ErrMsg, 2);
593 PrintAndExit("Post-link optimization output is not bitcode");
595 PrintAndExit(ErrMsg);
600 // If the user wants to generate a native executable, compile it from the
603 // Otherwise, create a script that will run the bitcode through the JIT.
605 // Name of the Assembly Language output file
606 sys::Path AssemblyFile ( OutputFilename);
607 AssemblyFile.appendSuffix("s");
609 // Mark the output files for removal if we get an interrupt.
610 sys::RemoveFileOnSignal(AssemblyFile);
611 sys::RemoveFileOnSignal(sys::Path(OutputFilename));
613 // Determine the locations of the llc and gcc programs.
614 sys::Path llc = FindExecutable("llc", argv[0]);
616 PrintAndExit("Failed to find llc");
618 sys::Path gcc = FindExecutable("gcc", argv[0]);
620 PrintAndExit("Failed to find gcc");
622 // Generate an assembly language file for the bitcode.
624 if (0 != GenerateAssembly(AssemblyFile.toString(), RealBitcodeOutput,
626 PrintAndExit(ErrMsg);
628 if (0 != GenerateNative(OutputFilename, AssemblyFile.toString(),
629 NativeLinkItems, gcc, envp, ErrMsg))
630 PrintAndExit(ErrMsg);
632 // Remove the assembly language file.
633 AssemblyFile.eraseFromDisk();
634 } else if (NativeCBE) {
635 sys::Path CFile (OutputFilename);
636 CFile.appendSuffix("cbe.c");
638 // Mark the output files for removal if we get an interrupt.
639 sys::RemoveFileOnSignal(CFile);
640 sys::RemoveFileOnSignal(sys::Path(OutputFilename));
642 // Determine the locations of the llc and gcc programs.
643 sys::Path llc = FindExecutable("llc", argv[0]);
645 PrintAndExit("Failed to find llc");
647 sys::Path gcc = FindExecutable("gcc", argv[0]);
649 PrintAndExit("Failed to find gcc");
651 // Generate an assembly language file for the bitcode.
653 if (0 != GenerateCFile(
654 CFile.toString(), RealBitcodeOutput, llc, ErrMsg))
655 PrintAndExit(ErrMsg);
657 if (0 != GenerateNative(OutputFilename, CFile.toString(),
658 NativeLinkItems, gcc, envp, ErrMsg))
659 PrintAndExit(ErrMsg);
661 // Remove the assembly language file.
662 CFile.eraseFromDisk();
665 EmitShellScript(argv);
668 // Make the script executable...
670 if (sys::Path(OutputFilename).makeExecutableOnDisk(&ErrMsg))
671 PrintAndExit(ErrMsg);
673 // Make the bitcode file readable and directly executable in LLEE as well
674 if (sys::Path(RealBitcodeOutput).makeExecutableOnDisk(&ErrMsg))
675 PrintAndExit(ErrMsg);
677 if (sys::Path(RealBitcodeOutput).makeReadableOnDisk(&ErrMsg))
678 PrintAndExit(ErrMsg);
680 } catch (const std::string& msg) {
683 PrintAndExit("Unexpected unknown exception occurred.", 2);