+//===- gccld.cpp - LLVM 'ld' compatible linker ----------------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file was developed by the LLVM research group and is distributed under
+// the University of Illinois Open Source License. See LICENSE.TXT for details.
+//
//===----------------------------------------------------------------------===//
-// LLVM 'GCCLD' UTILITY
//
// This utility is intended to be compatible with GCC, and follows standard
-// system ld conventions. As such, the default output file is ./a.out.
+// system 'ld' conventions. As such, the default output file is ./a.out.
// Additionally, this program outputs a shell script that is used to invoke LLI
// to execute the program. In this manner, the generated executable (a.out for
// example), is directly executable, whereas the bytecode file actually lives in
//
// Note that if someone (or a script) deletes the executable program generated,
// the .bc file will be left around. Considering that this is a temporary hack,
-// I'm not to worried about this.
+// I'm not too worried about this.
//
//===----------------------------------------------------------------------===//
-#include "llvm/Transforms/Utils/Linker.h"
+#include "gccld.h"
#include "llvm/Module.h"
#include "llvm/PassManager.h"
#include "llvm/Bytecode/Reader.h"
#include "llvm/Bytecode/WriteBytecodePass.h"
+#include "llvm/Target/TargetData.h"
#include "llvm/Transforms/IPO.h"
-#include "llvm/Transforms/IPO/GlobalDCE.h"
-#include "llvm/Transforms/IPO/Internalize.h"
#include "llvm/Transforms/Scalar.h"
+#include "llvm/Transforms/Utils/Linker.h"
#include "Support/CommandLine.h"
+#include "Support/FileUtilities.h"
#include "Support/Signals.h"
+#include "Support/SystemUtils.h"
#include <fstream>
#include <memory>
-#include <algorithm>
-#include <sys/types.h> // For FileExists
-#include <sys/stat.h>
-using std::cerr;
-static cl::list<string>
-InputFilenames(cl::Positional, cl::desc("<input bytecode files>"),
- cl::OneOrMore);
+namespace {
+ cl::list<std::string>
+ InputFilenames(cl::Positional, cl::desc("<input bytecode files>"),
+ cl::OneOrMore);
-static cl::opt<string>
-OutputFilename("o", cl::desc("Override output filename"), cl::init("a.out"),
- cl::value_desc("filename"));
+ cl::opt<std::string>
+ OutputFilename("o", cl::desc("Override output filename"), cl::init("a.out"),
+ cl::value_desc("filename"));
-static cl::opt<bool>
-Verbose("v", cl::desc("Print information about actions taken"));
+ cl::opt<bool>
+ Verbose("v", cl::desc("Print information about actions taken"));
+
+ cl::list<std::string>
+ LibPaths("L", cl::desc("Specify a library search path"), cl::Prefix,
+ cl::value_desc("directory"));
+
+ cl::list<std::string>
+ Libraries("l", cl::desc("Specify libraries to link to"), cl::Prefix,
+ cl::value_desc("library prefix"));
+
+ cl::opt<bool>
+ Strip("s", cl::desc("Strip symbol info from executable"));
+
+ cl::opt<bool>
+ NoInternalize("disable-internalize",
+ cl::desc("Do not mark all symbols as internal"));
+ static cl::alias
+ ExportDynamic("export-dynamic", cl::desc("Alias for -disable-internalize"),
+ cl::aliasopt(NoInternalize));
+
+ cl::opt<bool>
+ LinkAsLibrary("link-as-library", cl::desc("Link the .bc files together as a"
+ " library, not an executable"));
+
+ cl::opt<bool>
+ Native("native",
+ cl::desc("Generate a native binary instead of a shell script"));
+
+ // Compatibility options that are ignored but supported by LD
+ cl::opt<std::string>
+ CO3("soname", cl::Hidden, cl::desc("Compatibility option: ignored"));
+ cl::opt<std::string>
+ CO4("version-script", cl::Hidden, cl::desc("Compatibility option: ignored"));
+ cl::opt<bool>
+ CO5("eh-frame-hdr", cl::Hidden, cl::desc("Compatibility option: ignored"));
+ cl::opt<bool>
+ CO6("r", cl::Hidden, cl::desc("Compatibility option: ignored"));
+}
-static cl::list<string>
-LibPaths("L", cl::desc("Specify a library search path"), cl::Prefix,
- cl::value_desc("directory"));
+//
+// Function: PrintAndReturn ()
+//
+// Description:
+// Prints a message (usually error message) to standard error (stderr) and
+// returns a value usable for an exit status.
+//
+// Inputs:
+// progname - The name of the program (i.e. argv[0]).
+// Message - The message to print to standard error.
+// Extra - Extra information to print between the program name and thei
+// message. It is optional.
+//
+// Outputs:
+// None.
+//
+// Return value:
+// Returns a value that can be used as the exit status (i.e. for exit()).
+//
+int
+PrintAndReturn (const char *progname,
+ const std::string &Message,
+ const std::string &Extra)
+{
+ std::cerr << progname << Extra << ": " << Message << "\n";
+ return 1;
+}
-static cl::list<string>
-Libraries("l", cl::desc("Specify libraries to link to"), cl::Prefix,
- cl::value_desc("library prefix"));
+//
+//
+// Function: CopyEnv()
+//
+// Description:
+// This function takes an array of environment variables and makes a
+// copy of it. This copy can then be manipulated any way the caller likes
+// without affecting the process's real environment.
+//
+// Inputs:
+// envp - An array of C strings containing an environment.
+//
+// Outputs:
+// None.
+//
+// Return value:
+// NULL - An error occurred.
+//
+// Otherwise, a pointer to a new array of C strings is returned. Every string
+// in the array is a duplicate of the one in the original array (i.e. we do
+// not copy the char *'s from one array to another).
+//
+char ** CopyEnv(char ** const envp) {
+ // Count the number of entries in the old list;
+ unsigned entries; // The number of entries in the old environment list
+ for (entries = 0; envp[entries] != NULL; entries++)
+ {
+ ;
+ }
-static cl::opt<bool>
-Strip("s", cl::desc("Strip symbol info from executable"));
+ // Add one more entry for the NULL pointer that ends the list.
+ ++entries;
+ // If there are no entries at all, just return NULL.
+ if (entries == 0)
+ return NULL;
-// FileExists - Return true if the specified string is an openable file...
-static inline bool FileExists(const std::string &FN) {
- struct stat StatBuf;
- return stat(FN.c_str(), &StatBuf) != -1;
-}
+ // Allocate a new environment list.
+ char **newenv;
+ if ((newenv = new (char *) [entries]) == NULL)
+ return NULL;
-// LoadFile - Read the specified bytecode file in and return it. This routine
-// searches the link path for the specified file to try to find it...
-//
-static inline std::auto_ptr<Module> LoadFile(const std::string &FN) {
- std::string Filename = FN;
- std::string ErrorMessage;
+ // Make a copy of the list. Don't forget the NULL that ends the list.
+ entries = 0;
+ while (envp[entries] != NULL) {
+ newenv[entries] = new char[strlen (envp[entries]) + 1];
+ strcpy (newenv[entries], envp[entries]);
+ ++entries;
+ }
+ newenv[entries] = NULL;
- unsigned NextLibPathIdx = 0;
- bool FoundAFile = false;
+ return newenv;
+}
- while (1) {
- if (Verbose) cerr << "Loading '" << Filename << "'\n";
- if (FileExists(Filename)) FoundAFile = true;
- Module *Result = ParseBytecodeFile(Filename, &ErrorMessage);
- if (Result) return std::auto_ptr<Module>(Result); // Load successful!
- if (Verbose) {
- cerr << "Error opening bytecode file: '" << Filename << "'";
- if (ErrorMessage.size()) cerr << ": " << ErrorMessage;
- cerr << "\n";
- }
-
- if (NextLibPathIdx == LibPaths.size()) break;
- Filename = LibPaths[NextLibPathIdx++] + "/" + FN;
+//
+// Function: RemoveEnv()
+//
+// Description:
+// Remove the specified environment variable from the environment array.
+//
+// Inputs:
+// name - The name of the variable to remove. It cannot be NULL.
+// envp - The array of environment variables. It cannot be NULL.
+//
+// Outputs:
+// envp - The pointer to the specified variable name is removed.
+//
+// Return value:
+// None.
+//
+// Notes:
+// This is mainly done because functions to remove items from the environment
+// are not available across all platforms. In particular, Solaris does not
+// seem to have an unsetenv() function or a setenv() function (or they are
+// undocumented if they do exist).
+//
+void RemoveEnv(const char * name, char ** const envp) {
+ for (unsigned index=0; envp[index] != NULL; index++) {
+ // Find the first equals sign in the array and make it an EOS character.
+ char *p = strchr (envp[index], '=');
+ if (p == NULL)
+ continue;
+ else
+ *p = '\0';
+
+ // Compare the two strings. If they are equal, zap this string.
+ // Otherwise, restore it.
+ if (!strcmp(name, envp[index]))
+ *envp[index] = '\0';
+ else
+ *p = '=';
}
- if (FoundAFile)
- cerr << "Bytecode file '" << FN << "' corrupt! "
- << "Use 'gccld -v ...' for more info.\n";
- else
- cerr << "Could not locate bytecode file: '" << FN << "'\n";
- return std::auto_ptr<Module>();
+ return;
}
-int main(int argc, char **argv) {
+int main(int argc, char **argv, char **envp) {
cl::ParseCommandLineOptions(argc, argv, " llvm linker for GCC\n");
- unsigned BaseArg = 0;
std::string ErrorMessage;
-
- if (!Libraries.empty()) {
- // Sort libraries list...
- std::sort(Libraries.begin(), Libraries.end());
-
- // Remove duplicate libraries entries...
- Libraries.erase(unique(Libraries.begin(), Libraries.end()),
- Libraries.end());
-
- // Add all of the libraries to the end of the link line...
- for (unsigned i = 0; i < Libraries.size(); ++i)
- InputFilenames.push_back("lib" + Libraries[i] + ".bc");
+ std::auto_ptr<Module> Composite(LoadObject(InputFilenames[0], ErrorMessage));
+ if (Composite.get() == 0)
+ return PrintAndReturn(argv[0], ErrorMessage);
+
+ // We always look first in the current directory when searching for libraries.
+ LibPaths.insert(LibPaths.begin(), ".");
+
+ // If the user specified an extra search path in their environment, respect
+ // it.
+ if (char *SearchPath = getenv("LLVM_LIB_SEARCH_PATH"))
+ LibPaths.push_back(SearchPath);
+
+ // Remove any consecutive duplicates of the same library...
+ Libraries.erase(std::unique(Libraries.begin(), Libraries.end()),
+ Libraries.end());
+
+ // Link in all of the files
+ if (LinkFiles(argv[0], Composite.get(), InputFilenames, Verbose))
+ return 1; // Error already printed
+ LinkLibraries(argv[0], Composite.get(), Libraries, LibPaths, Verbose, Native);
+
+ // Link in all of the libraries next...
+
+ // Create the output file.
+ std::string RealBytecodeOutput = OutputFilename;
+ if (!LinkAsLibrary) RealBytecodeOutput += ".bc";
+ std::ofstream Out(RealBytecodeOutput.c_str());
+ if (!Out.good())
+ return PrintAndReturn(argv[0], "error opening '" + RealBytecodeOutput +
+ "' for writing!");
+
+ // Ensure that the bytecode file gets removed from the disk if we get a
+ // SIGINT signal.
+ RemoveFileOnSignal(RealBytecodeOutput);
+
+ // Generate the bytecode file.
+ if (GenerateBytecode(Composite.get(), Strip, !NoInternalize, &Out)) {
+ Out.close();
+ return PrintAndReturn(argv[0], "error generating bytcode");
}
- std::auto_ptr<Module> Composite(LoadFile(InputFilenames[BaseArg]));
- if (Composite.get() == 0) return 1;
-
- for (unsigned i = BaseArg+1; i < InputFilenames.size(); ++i) {
- std::auto_ptr<Module> M(LoadFile(InputFilenames[i]));
- if (M.get() == 0) return 1;
-
- if (Verbose) cerr << "Linking in '" << InputFilenames[i] << "'\n";
+ // Close the bytecode file.
+ Out.close();
- if (LinkModules(Composite.get(), M.get(), &ErrorMessage)) {
- cerr << "Error linking in '" << InputFilenames[i] << "': "
- << ErrorMessage << "\n";
- return 1;
+ // If we are not linking a library, generate either a native executable
+ // or a JIT shell script, depending upon what the user wants.
+ if (!LinkAsLibrary) {
+ // If the user wants to generate a native executable, compile it from the
+ // bytecode file.
+ //
+ // Otherwise, create a script that will run the bytecode through the JIT.
+ if (Native) {
+ // Name of the Assembly Language output file
+ std::string AssemblyFile = OutputFilename + ".s";
+
+ // Mark the output files for removal if we get an interrupt.
+ RemoveFileOnSignal(AssemblyFile);
+ RemoveFileOnSignal(OutputFilename);
+
+ // Determine the locations of the llc and gcc programs.
+ std::string llc = FindExecutable("llc", argv[0]);
+ std::string gcc = FindExecutable("gcc", argv[0]);
+ if (llc.empty())
+ return PrintAndReturn(argv[0], "Failed to find llc");
+
+ if (gcc.empty())
+ return PrintAndReturn(argv[0], "Failed to find gcc");
+
+ // Generate an assembly language file for the bytecode.
+ if (Verbose) std::cout << "Generating Assembly Code\n";
+ GenerateAssembly(AssemblyFile, RealBytecodeOutput, llc, envp);
+ if (Verbose) std::cout << "Generating Native Code\n";
+ GenerateNative(OutputFilename, AssemblyFile, Libraries, LibPaths,
+ gcc, envp);
+
+ // Remove the assembly language file.
+ removeFile (AssemblyFile);
+ } else {
+ // Output the script to start the program...
+ std::ofstream Out2(OutputFilename.c_str());
+ if (!Out2.good())
+ return PrintAndReturn(argv[0], "error opening '" + OutputFilename +
+ "' for writing!");
+ Out2 << "#!/bin/sh\nlli $0.bc $*\n";
+ Out2.close();
}
- }
-
- // In addition to just linking the input from GCC, we also want to spiff it up
- // a little bit. Do this now.
- //
- PassManager Passes;
-
- // Linking modules together can lead to duplicated global constants, only keep
- // one copy of each constant...
- //
- Passes.add(createConstantMergePass());
-
- // If the -s command line option was specified, strip the symbols out of the
- // resulting program to make it smaller. -s is a GCC option that we are
- // supporting.
- //
- if (Strip)
- Passes.add(createSymbolStrippingPass());
-
- // Often if the programmer does not specify proper prototypes for the
- // functions they are calling, they end up calling a vararg version of the
- // function that does not get a body filled in (the real function has typed
- // arguments). This pass merges the two functions.
- //
- Passes.add(createFunctionResolvingPass());
-
- // Now that composite has been compiled, scan through the module, looking for
- // a main function. If main is defined, mark all other functions internal.
- //
- Passes.add(createInternalizePass());
-
- // Now that we have optimized the program, discard unreachable functions...
- //
- Passes.add(createGlobalDCEPass());
-
- // Add the pass that writes bytecode to the output file...
- std::ofstream Out((OutputFilename+".bc").c_str());
- if (!Out.good()) {
- cerr << "Error opening '" << OutputFilename << ".bc' for writing!\n";
- return 1;
- }
- Passes.add(new WriteBytecodePass(&Out)); // Write bytecode to file...
-
- // Make sure that the Out file gets unlink'd from the disk if we get a SIGINT
- RemoveFileOnSignal(OutputFilename+".bc");
-
- // Run our queue of passes all at once now, efficiently.
- Passes.run(*Composite.get());
- Out.close();
+
+ // Make the script executable...
+ MakeFileExecutable(OutputFilename);
- // Output the script to start the program...
- std::ofstream Out2(OutputFilename.c_str());
- if (!Out2.good()) {
- cerr << "Error opening '" << OutputFilename << "' for writing!\n";
- return 1;
+ // Make the bytecode file readable and directly executable in LLEE as well
+ MakeFileExecutable(RealBytecodeOutput);
+ MakeFileReadable(RealBytecodeOutput);
}
- Out2 << "#!/bin/sh\nlli -q $0.bc $*\n";
- Out2.close();
-
- // Make the script executable...
- chmod(OutputFilename.c_str(), 0755);
return 0;
}