1 //===- llvm-extract.cpp - LLVM function extraction utility ----------------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This utility changes the input module to only contain a single function, 11 // which is primarily used for debugging transformations. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #include "llvm/IR/LLVMContext.h" 16 #include "llvm/ADT/SetVector.h" 17 #include "llvm/ADT/SmallPtrSet.h" 18 #include "llvm/Assembly/PrintModulePass.h" 19 #include "llvm/Bitcode/ReaderWriter.h" 20 #include "llvm/IR/DataLayout.h" 21 #include "llvm/IR/Module.h" 22 #include "llvm/PassManager.h" 23 #include "llvm/Support/CommandLine.h" 24 #include "llvm/Support/IRReader.h" 25 #include "llvm/Support/ManagedStatic.h" 26 #include "llvm/Support/PrettyStackTrace.h" 27 #include "llvm/Support/Regex.h" 28 #include "llvm/Support/Signals.h" 29 #include "llvm/Support/SystemUtils.h" 30 #include "llvm/Support/ToolOutputFile.h" 31 #include "llvm/Transforms/IPO.h" 32 #include <memory> 33 using namespace llvm; 34 35 // InputFilename - The filename to read from. 36 static cl::opt<std::string> 37 InputFilename(cl::Positional, cl::desc("<input bitcode file>"), 38 cl::init("-"), cl::value_desc("filename")); 39 40 static cl::opt<std::string> 41 OutputFilename("o", cl::desc("Specify output filename"), 42 cl::value_desc("filename"), cl::init("-")); 43 44 static cl::opt<bool> 45 Force("f", cl::desc("Enable binary output on terminals")); 46 47 static cl::opt<bool> 48 DeleteFn("delete", cl::desc("Delete specified Globals from Module")); 49 50 // ExtractFuncs - The functions to extract from the module. 51 static cl::list<std::string> 52 ExtractFuncs("func", cl::desc("Specify function to extract"), 53 cl::ZeroOrMore, cl::value_desc("function")); 54 55 // ExtractRegExpFuncs - The functions, matched via regular expression, to 56 // extract from the module. 57 static cl::list<std::string> 58 ExtractRegExpFuncs("rfunc", cl::desc("Specify function(s) to extract using a " 59 "regular expression"), 60 cl::ZeroOrMore, cl::value_desc("rfunction")); 61 62 // ExtractAlias - The alias to extract from the module. 63 static cl::list<std::string> 64 ExtractAliases("alias", cl::desc("Specify alias to extract"), 65 cl::ZeroOrMore, cl::value_desc("alias")); 66 67 68 // ExtractRegExpAliases - The aliases, matched via regular expression, to 69 // extract from the module. 70 static cl::list<std::string> 71 ExtractRegExpAliases("ralias", cl::desc("Specify alias(es) to extract using a " 72 "regular expression"), 73 cl::ZeroOrMore, cl::value_desc("ralias")); 74 75 // ExtractGlobals - The globals to extract from the module. 76 static cl::list<std::string> 77 ExtractGlobals("glob", cl::desc("Specify global to extract"), 78 cl::ZeroOrMore, cl::value_desc("global")); 79 80 // ExtractRegExpGlobals - The globals, matched via regular expression, to 81 // extract from the module... 82 static cl::list<std::string> 83 ExtractRegExpGlobals("rglob", cl::desc("Specify global(s) to extract using a " 84 "regular expression"), 85 cl::ZeroOrMore, cl::value_desc("rglobal")); 86 87 static cl::opt<bool> 88 OutputAssembly("S", 89 cl::desc("Write output as LLVM assembly"), cl::Hidden); 90 91 int main(int argc, char **argv) { 92 // Print a stack trace if we signal out. 93 sys::PrintStackTraceOnErrorSignal(); 94 PrettyStackTraceProgram X(argc, argv); 95 96 LLVMContext &Context = getGlobalContext(); 97 llvm_shutdown_obj Y; // Call llvm_shutdown() on exit. 98 cl::ParseCommandLineOptions(argc, argv, "llvm extractor\n"); 99 100 // Use lazy loading, since we only care about selected global values. 101 SMDiagnostic Err; 102 std::auto_ptr<Module> M; 103 M.reset(getLazyIRFileModule(InputFilename, Err, Context)); 104 105 if (M.get() == 0) { 106 Err.print(argv[0], errs()); 107 return 1; 108 } 109 110 // Use SetVector to avoid duplicates. 111 SetVector<GlobalValue *> GVs; 112 113 // Figure out which aliases we should extract. 114 for (size_t i = 0, e = ExtractAliases.size(); i != e; ++i) { 115 GlobalAlias *GA = M->getNamedAlias(ExtractAliases[i]); 116 if (!GA) { 117 errs() << argv[0] << ": program doesn't contain alias named '" 118 << ExtractAliases[i] << "'!\n"; 119 return 1; 120 } 121 GVs.insert(GA); 122 } 123 124 // Extract aliases via regular expression matching. 125 for (size_t i = 0, e = ExtractRegExpAliases.size(); i != e; ++i) { 126 std::string Error; 127 Regex RegEx(ExtractRegExpAliases[i]); 128 if (!RegEx.isValid(Error)) { 129 errs() << argv[0] << ": '" << ExtractRegExpAliases[i] << "' " 130 "invalid regex: " << Error; 131 } 132 bool match = false; 133 for (Module::alias_iterator GA = M->alias_begin(), E = M->alias_end(); 134 GA != E; GA++) { 135 if (RegEx.match(GA->getName())) { 136 GVs.insert(&*GA); 137 match = true; 138 } 139 } 140 if (!match) { 141 errs() << argv[0] << ": program doesn't contain global named '" 142 << ExtractRegExpAliases[i] << "'!\n"; 143 return 1; 144 } 145 } 146 147 // Figure out which globals we should extract. 148 for (size_t i = 0, e = ExtractGlobals.size(); i != e; ++i) { 149 GlobalValue *GV = M->getNamedGlobal(ExtractGlobals[i]); 150 if (!GV) { 151 errs() << argv[0] << ": program doesn't contain global named '" 152 << ExtractGlobals[i] << "'!\n"; 153 return 1; 154 } 155 GVs.insert(GV); 156 } 157 158 // Extract globals via regular expression matching. 159 for (size_t i = 0, e = ExtractRegExpGlobals.size(); i != e; ++i) { 160 std::string Error; 161 Regex RegEx(ExtractRegExpGlobals[i]); 162 if (!RegEx.isValid(Error)) { 163 errs() << argv[0] << ": '" << ExtractRegExpGlobals[i] << "' " 164 "invalid regex: " << Error; 165 } 166 bool match = false; 167 for (Module::global_iterator GV = M->global_begin(), 168 E = M->global_end(); GV != E; GV++) { 169 if (RegEx.match(GV->getName())) { 170 GVs.insert(&*GV); 171 match = true; 172 } 173 } 174 if (!match) { 175 errs() << argv[0] << ": program doesn't contain global named '" 176 << ExtractRegExpGlobals[i] << "'!\n"; 177 return 1; 178 } 179 } 180 181 // Figure out which functions we should extract. 182 for (size_t i = 0, e = ExtractFuncs.size(); i != e; ++i) { 183 GlobalValue *GV = M->getFunction(ExtractFuncs[i]); 184 if (!GV) { 185 errs() << argv[0] << ": program doesn't contain function named '" 186 << ExtractFuncs[i] << "'!\n"; 187 return 1; 188 } 189 GVs.insert(GV); 190 } 191 // Extract functions via regular expression matching. 192 for (size_t i = 0, e = ExtractRegExpFuncs.size(); i != e; ++i) { 193 std::string Error; 194 StringRef RegExStr = ExtractRegExpFuncs[i]; 195 Regex RegEx(RegExStr); 196 if (!RegEx.isValid(Error)) { 197 errs() << argv[0] << ": '" << ExtractRegExpFuncs[i] << "' " 198 "invalid regex: " << Error; 199 } 200 bool match = false; 201 for (Module::iterator F = M->begin(), E = M->end(); F != E; 202 F++) { 203 if (RegEx.match(F->getName())) { 204 GVs.insert(&*F); 205 match = true; 206 } 207 } 208 if (!match) { 209 errs() << argv[0] << ": program doesn't contain global named '" 210 << ExtractRegExpFuncs[i] << "'!\n"; 211 return 1; 212 } 213 } 214 215 // Materialize requisite global values. 216 if (!DeleteFn) 217 for (size_t i = 0, e = GVs.size(); i != e; ++i) { 218 GlobalValue *GV = GVs[i]; 219 if (GV->isMaterializable()) { 220 std::string ErrInfo; 221 if (GV->Materialize(&ErrInfo)) { 222 errs() << argv[0] << ": error reading input: " << ErrInfo << "\n"; 223 return 1; 224 } 225 } 226 } 227 else { 228 // Deleting. Materialize every GV that's *not* in GVs. 229 SmallPtrSet<GlobalValue *, 8> GVSet(GVs.begin(), GVs.end()); 230 for (Module::global_iterator I = M->global_begin(), E = M->global_end(); 231 I != E; ++I) { 232 GlobalVariable *G = I; 233 if (!GVSet.count(G) && G->isMaterializable()) { 234 std::string ErrInfo; 235 if (G->Materialize(&ErrInfo)) { 236 errs() << argv[0] << ": error reading input: " << ErrInfo << "\n"; 237 return 1; 238 } 239 } 240 } 241 for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) { 242 Function *F = I; 243 if (!GVSet.count(F) && F->isMaterializable()) { 244 std::string ErrInfo; 245 if (F->Materialize(&ErrInfo)) { 246 errs() << argv[0] << ": error reading input: " << ErrInfo << "\n"; 247 return 1; 248 } 249 } 250 } 251 } 252 253 // In addition to deleting all other functions, we also want to spiff it 254 // up a little bit. Do this now. 255 PassManager Passes; 256 Passes.add(new DataLayout(M.get())); // Use correct DataLayout 257 258 std::vector<GlobalValue*> Gvs(GVs.begin(), GVs.end()); 259 260 Passes.add(createGVExtractionPass(Gvs, DeleteFn)); 261 if (!DeleteFn) 262 Passes.add(createGlobalDCEPass()); // Delete unreachable globals 263 Passes.add(createStripDeadDebugInfoPass()); // Remove dead debug info 264 Passes.add(createStripDeadPrototypesPass()); // Remove dead func decls 265 266 std::string ErrorInfo; 267 tool_output_file Out(OutputFilename.c_str(), ErrorInfo, 268 raw_fd_ostream::F_Binary); 269 if (!ErrorInfo.empty()) { 270 errs() << ErrorInfo << '\n'; 271 return 1; 272 } 273 274 if (OutputAssembly) 275 Passes.add(createPrintModulePass(&Out.os())); 276 else if (Force || !CheckBitcodeOutputToConsole(Out.os(), true)) 277 Passes.add(createBitcodeWriterPass(Out.os())); 278 279 Passes.run(*M.get()); 280 281 // Declare success. 282 Out.keep(); 283 284 return 0; 285 } 286