1 //===-- llvm-bcanalyzer.cpp - Byte Code Analyzer --------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by Reid Spencer and is distributed under the
6 // University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This tool may be invoked in the following manner:
11 // llvm-bcanalyzer [options] - Read LLVM bytecode from stdin
12 // llvm-bcanalyzer [options] x.bc - Read LLVM bytecode from the x.bc file
15 // --help - Output information about command line switches
16 // --nodetails - Don't print out detailed informaton about individual
17 // blocks and functions
18 // --dump - Dump low-level bytecode structure in readable format
20 // This tool provides analytical information about a bytecode file. It is
21 // intended as an aid to developers of bytecode reading and writing software. It
22 // produces on std::out a summary of the bytecode file that shows various
23 // statistics about the contents of the file. By default this information is
24 // detailed and contains information about individual bytecode blocks and the
25 // functions in the module. To avoid this more detailed output, use the
26 // -nodetails option to limit the output to just module level information.
27 // The tool is also able to print a bytecode file in a straight forward text
28 // format that shows the containment and relationships of the information in
29 // the bytecode file (-dump option).
31 //===----------------------------------------------------------------------===//
33 #include "llvm/Analysis/Verifier.h"
34 #include "llvm/Bitcode/BitstreamReader.h"
35 #include "llvm/Bitcode/LLVMBitCodes.h"
36 #include "llvm/Bytecode/Analyzer.h"
37 #include "llvm/Support/CommandLine.h"
38 #include "llvm/Support/Compressor.h"
39 #include "llvm/Support/ManagedStatic.h"
40 #include "llvm/Support/MemoryBuffer.h"
41 #include "llvm/System/Signals.h"
46 static cl::opt<std::string>
47 InputFilename(cl::Positional, cl::desc("<input bytecode>"), cl::init("-"));
49 static cl::opt<std::string>
50 OutputFilename("-o", cl::init("-"), cl::desc("<output file>"));
52 static cl::opt<bool> NoDetails("nodetails", cl::desc("Skip detailed output"));
53 static cl::opt<bool> Dump("dump", cl::desc("Dump low level bytecode trace"));
54 static cl::opt<bool> Verify("verify", cl::desc("Progressively verify module"));
56 //===----------------------------------------------------------------------===//
57 // Bitcode specific analysis.
58 //===----------------------------------------------------------------------===//
60 static cl::opt<bool> Bitcode("bitcode", cl::desc("Read a bitcode file"));
63 NonSymbolic("non-symbolic",
64 cl::desc("Emit numberic info in dump even if"
65 " symbolic info is available"));
67 /// CurStreamType - If we can sniff the flavor of this stream, we can produce
75 /// GetBlockName - Return a symbolic block name if known, otherwise return
77 static const char *GetBlockName(unsigned BlockID) {
78 if (CurStreamType != LLVMIRBitstream) return 0;
82 case bitc::MODULE_BLOCK_ID: return "MODULE_BLOCK";
83 case bitc::TYPE_BLOCK_ID: return "TYPE_BLOCK";
84 case bitc::CONSTANTS_BLOCK_ID: return "CONSTANTS_BLOCK";
85 case bitc::FUNCTION_BLOCK_ID: return "FUNCTION_BLOCK";
86 case bitc::TYPE_SYMTAB_BLOCK_ID: return "TYPE_SYMTAB";
87 case bitc::VALUE_SYMTAB_BLOCK_ID: return "VALUE_SYMTAB";
91 /// GetCodeName - Return a symbolic code name if known, otherwise return
93 static const char *GetCodeName(unsigned CodeID, unsigned BlockID) {
94 if (CurStreamType != LLVMIRBitstream) return 0;
98 case bitc::MODULE_BLOCK_ID:
101 case bitc::MODULE_CODE_VERSION: return "VERSION";
102 case bitc::MODULE_CODE_TRIPLE: return "TRIPLE";
103 case bitc::MODULE_CODE_DATALAYOUT: return "DATALAYOUT";
104 case bitc::MODULE_CODE_ASM: return "ASM";
105 case bitc::MODULE_CODE_SECTIONNAME: return "SECTIONNAME";
106 case bitc::MODULE_CODE_DEPLIB: return "DEPLIB";
107 case bitc::MODULE_CODE_GLOBALVAR: return "GLOBALVAR";
108 case bitc::MODULE_CODE_FUNCTION: return "FUNCTION";
109 case bitc::MODULE_CODE_ALIAS: return "ALIAS";
110 case bitc::MODULE_CODE_PURGEVALS: return "PURGEVALS";
112 case bitc::TYPE_BLOCK_ID:
115 case bitc::TYPE_CODE_NUMENTRY: return "NUMENTRY";
116 case bitc::TYPE_CODE_META: return "META";
117 case bitc::TYPE_CODE_VOID: return "VOID";
118 case bitc::TYPE_CODE_FLOAT: return "FLOAT";
119 case bitc::TYPE_CODE_DOUBLE: return "DOUBLE";
120 case bitc::TYPE_CODE_LABEL: return "LABEL";
121 case bitc::TYPE_CODE_OPAQUE: return "OPAQUE";
122 case bitc::TYPE_CODE_INTEGER: return "INTEGER";
123 case bitc::TYPE_CODE_POINTER: return "POINTER";
124 case bitc::TYPE_CODE_FUNCTION: return "FUNCTION";
125 case bitc::TYPE_CODE_STRUCT: return "STRUCT";
126 case bitc::TYPE_CODE_ARRAY: return "ARRAY";
127 case bitc::TYPE_CODE_VECTOR: return "VECTOR";
130 case bitc::CONSTANTS_BLOCK_ID:
133 case bitc::CST_CODE_SETTYPE: return "SETTYPE";
134 case bitc::CST_CODE_NULL: return "NULL";
135 case bitc::CST_CODE_UNDEF: return "UNDEF";
136 case bitc::CST_CODE_INTEGER: return "INTEGER";
137 case bitc::CST_CODE_WIDE_INTEGER: return "WIDE_INTEGER";
138 case bitc::CST_CODE_FLOAT: return "FLOAT";
139 case bitc::CST_CODE_AGGREGATE: return "AGGREGATE";
140 case bitc::CST_CODE_CE_BINOP: return "CE_BINOP";
141 case bitc::CST_CODE_CE_CAST: return "CE_CAST";
142 case bitc::CST_CODE_CE_GEP: return "CE_GEP";
143 case bitc::CST_CODE_CE_SELECT: return "CE_SELECT";
144 case bitc::CST_CODE_CE_EXTRACTELT: return "CE_EXTRACTELT";
145 case bitc::CST_CODE_CE_INSERTELT: return "CE_INSERTELT";
146 case bitc::CST_CODE_CE_SHUFFLEVEC: return "CE_SHUFFLEVEC";
147 case bitc::CST_CODE_CE_CMP: return "CE_CMP";
149 case bitc::FUNCTION_BLOCK_ID:
152 case bitc::FUNC_CODE_DECLAREBLOCKS: return "DECLAREBLOCKS";
154 case bitc::FUNC_CODE_INST_BINOP: return "INST_BINOP";
155 case bitc::FUNC_CODE_INST_CAST: return "INST_CAST";
156 case bitc::FUNC_CODE_INST_GEP: return "INST_GEP";
157 case bitc::FUNC_CODE_INST_SELECT: return "INST_SELECT";
158 case bitc::FUNC_CODE_INST_EXTRACTELT: return "INST_EXTRACTELT";
159 case bitc::FUNC_CODE_INST_INSERTELT: return "INST_INSERTELT";
160 case bitc::FUNC_CODE_INST_SHUFFLEVEC: return "INST_SHUFFLEVEC";
161 case bitc::FUNC_CODE_INST_CMP: return "INST_CMP";
163 case bitc::FUNC_CODE_INST_RET: return "INST_RET";
164 case bitc::FUNC_CODE_INST_BR: return "INST_BR";
165 case bitc::FUNC_CODE_INST_SWITCH: return "INST_SWITCH";
166 case bitc::FUNC_CODE_INST_INVOKE: return "INST_INVOKE";
167 case bitc::FUNC_CODE_INST_UNWIND: return "INST_UNWIND";
168 case bitc::FUNC_CODE_INST_UNREACHABLE: return "INST_UNREACHABLE";
170 case bitc::FUNC_CODE_INST_MALLOC: return "INST_MALLOC";
171 case bitc::FUNC_CODE_INST_FREE: return "INST_FREE";
172 case bitc::FUNC_CODE_INST_ALLOCA: return "INST_ALLOCA";
173 case bitc::FUNC_CODE_INST_LOAD: return "INST_LOAD";
174 case bitc::FUNC_CODE_INST_STORE: return "INST_STORE";
175 case bitc::FUNC_CODE_INST_CALL: return "INST_CALL";
176 case bitc::FUNC_CODE_INST_VAARG: return "INST_VAARG";
178 case bitc::TYPE_SYMTAB_BLOCK_ID:
181 case bitc::TST_CODE_ENTRY: return "ENTRY";
183 case bitc::VALUE_SYMTAB_BLOCK_ID:
186 case bitc::VST_CODE_ENTRY: return "ENTRY";
192 struct PerBlockIDStats {
193 /// NumInstances - This the number of times this block ID has been seen.
194 unsigned NumInstances;
196 /// NumBits - The total size in bits of all of these blocks.
199 /// NumSubBlocks - The total number of blocks these blocks contain.
200 unsigned NumSubBlocks;
202 /// NumAbbrevs - The total number of abbreviations.
205 /// NumRecords - The total number of records these blocks contain, and the
206 /// number that are abbreviated.
207 unsigned NumRecords, NumAbbreviatedRecords;
210 : NumInstances(0), NumBits(0),
211 NumSubBlocks(0), NumAbbrevs(0), NumRecords(0), NumAbbreviatedRecords(0) {}
214 static std::map<unsigned, PerBlockIDStats> BlockIDStats;
218 /// Error - All bitcode analysis errors go through this function, making this a
219 /// good place to breakpoint if debugging.
220 static bool Error(const std::string &Err) {
221 std::cerr << Err << "\n";
225 /// ParseBlock - Read a block, updating statistics, etc.
226 static bool ParseBlock(BitstreamReader &Stream, unsigned IndentLevel) {
227 uint64_t BlockBitStart = Stream.GetCurrentBitNo();
228 unsigned BlockID = Stream.ReadSubBlockID();
230 // Get the statistics for this BlockID.
231 PerBlockIDStats &BlockStats = BlockIDStats[BlockID];
233 BlockStats.NumInstances++;
235 unsigned NumWords = 0;
236 if (Stream.EnterSubBlock(&NumWords))
237 return Error("Malformed block record");
239 std::string Indent(IndentLevel*2, ' ');
240 const char *BlockName = 0;
242 std::cerr << Indent << "<";
243 if ((BlockName = GetBlockName(BlockID)))
244 std::cerr << BlockName;
246 std::cerr << "UnknownBlock" << BlockID;
248 if (NonSymbolic && BlockName)
249 std::cerr << " BlockID=" << BlockID;
251 std::cerr << " NumWords=" << NumWords
252 << " BlockCodeSize=" << Stream.GetAbbrevIDWidth() << ">\n";
255 SmallVector<uint64_t, 64> Record;
257 // Read all the records for this block.
259 if (Stream.AtEndOfStream())
260 return Error("Premature end of bitstream");
262 // Read the code for this record.
263 unsigned AbbrevID = Stream.ReadCode();
265 case bitc::END_BLOCK: {
266 if (Stream.ReadBlockEnd())
267 return Error("Error at end of block");
268 uint64_t BlockBitEnd = Stream.GetCurrentBitNo();
269 BlockStats.NumBits += BlockBitEnd-BlockBitStart;
271 std::cerr << Indent << "</";
273 std::cerr << BlockName << ">\n";
275 std::cerr << "UnknownBlock" << BlockID << ">\n";
279 case bitc::ENTER_SUBBLOCK:
280 if (ParseBlock(Stream, IndentLevel+1))
282 ++BlockStats.NumSubBlocks;
284 case bitc::DEFINE_ABBREV:
285 Stream.ReadAbbrevRecord();
286 ++BlockStats.NumAbbrevs;
289 ++BlockStats.NumRecords;
290 if (AbbrevID != bitc::UNABBREV_RECORD)
291 ++BlockStats.NumAbbreviatedRecords;
294 unsigned Code = Stream.ReadRecord(AbbrevID, Record);
295 // TODO: Compute per-blockid/code stats.
298 std::cerr << Indent << " <";
299 if (const char *CodeName = GetCodeName(Code, BlockID))
300 std::cerr << CodeName;
302 std::cerr << "UnknownCode" << Code;
303 if (NonSymbolic && GetCodeName(Code, BlockID))
304 std::cerr << " codeid=" << Code;
305 if (AbbrevID != bitc::UNABBREV_RECORD)
306 std::cerr << " abbrevid=" << AbbrevID;
308 for (unsigned i = 0, e = Record.size(); i != e; ++i)
309 std::cerr << " op" << i << "=" << (int64_t)Record[i];
319 static void PrintSize(double Bits) {
320 std::cerr << Bits << "b/" << Bits/8 << "B/" << Bits/32 << "W";
324 /// AnalyzeBitcode - Analyze the bitcode file specified by InputFilename.
325 static int AnalyzeBitcode() {
326 // Read the input file.
327 MemoryBuffer *Buffer;
328 if (InputFilename == "-")
329 Buffer = MemoryBuffer::getSTDIN();
331 Buffer = MemoryBuffer::getFile(&InputFilename[0], InputFilename.size());
334 return Error("Error reading '" + InputFilename + "'.");
336 if (Buffer->getBufferSize() & 3)
337 return Error("Bitcode stream should be a multiple of 4 bytes in length");
339 unsigned char *BufPtr = (unsigned char *)Buffer->getBufferStart();
340 BitstreamReader Stream(BufPtr, BufPtr+Buffer->getBufferSize());
343 // Read the stream signature.
345 Signature[0] = Stream.Read(8);
346 Signature[1] = Stream.Read(8);
347 Signature[2] = Stream.Read(4);
348 Signature[3] = Stream.Read(4);
349 Signature[4] = Stream.Read(4);
350 Signature[5] = Stream.Read(4);
352 // Autodetect the file contents, if it is one we know.
353 CurStreamType = UnknownBitstream;
354 if (Signature[0] == 'B' && Signature[1] == 'C' &&
355 Signature[2] == 0x0 && Signature[3] == 0xC &&
356 Signature[4] == 0xE && Signature[5] == 0xD)
357 CurStreamType = LLVMIRBitstream;
359 unsigned NumTopBlocks = 0;
361 // Parse the top-level structure. We only allow blocks at the top-level.
362 while (!Stream.AtEndOfStream()) {
363 unsigned Code = Stream.ReadCode();
364 if (Code != bitc::ENTER_SUBBLOCK)
365 return Error("Invalid record at top-level");
367 if (ParseBlock(Stream, 0))
372 if (Dump) std::cerr << "\n\n";
374 // Print a summary of the read file.
375 std::cerr << "Summary of " << InputFilename << ":\n";
376 std::cerr << " Total size: ";
377 PrintSize(Buffer->getBufferSize()*8);
379 std::cerr << " Stream type: ";
380 switch (CurStreamType) {
381 default: assert(0 && "Unknown bitstream type");
382 case UnknownBitstream: std::cerr << "unknown\n"; break;
383 case LLVMIRBitstream: std::cerr << "LLVM IR\n"; break;
385 std::cerr << " # Toplevel Blocks: " << NumTopBlocks << "\n";
388 // Emit per-block stats.
389 std::cerr << "Per-block Summary:\n";
390 for (std::map<unsigned, PerBlockIDStats>::iterator I = BlockIDStats.begin(),
391 E = BlockIDStats.end(); I != E; ++I) {
392 std::cerr << " Block ID #" << I->first;
393 if (const char *BlockName = GetBlockName(I->first))
394 std::cerr << " (" << BlockName << ")";
397 const PerBlockIDStats &Stats = I->second;
398 std::cerr << " Num Instances: " << Stats.NumInstances << "\n";
399 std::cerr << " Total Size: ";
400 PrintSize(Stats.NumBits);
402 std::cerr << " Average Size: ";
403 PrintSize(Stats.NumBits/(double)Stats.NumInstances);
405 std::cerr << " Tot/Avg SubBlocks: " << Stats.NumSubBlocks << "/"
406 << Stats.NumSubBlocks/(double)Stats.NumInstances << "\n";
407 std::cerr << " Tot/Avg Abbrevs: " << Stats.NumAbbrevs << "/"
408 << Stats.NumAbbrevs/(double)Stats.NumInstances << "\n";
409 std::cerr << " Tot/Avg Records: " << Stats.NumRecords << "/"
410 << Stats.NumRecords/(double)Stats.NumInstances << "\n";
411 std::cerr << " % Abbrev Recs: " << (Stats.NumAbbreviatedRecords/
412 (double)Stats.NumRecords)*100 << "\n";
419 //===----------------------------------------------------------------------===//
420 // Bytecode specific analysis.
421 //===----------------------------------------------------------------------===//
423 int main(int argc, char **argv) {
424 llvm_shutdown_obj X; // Call llvm_shutdown() on exit.
425 cl::ParseCommandLineOptions(argc, argv, " llvm-bcanalyzer file analyzer\n");
427 sys::PrintStackTraceOnErrorSignal();
430 return AnalyzeBitcode();
433 std::ostream *Out = &std::cout; // Default to printing to stdout...
434 std::string ErrorMessage;
435 BytecodeAnalysis bca;
437 /// Determine what to generate
438 bca.detailedResults = !NoDetails;
439 bca.progressiveVerify = Verify;
441 /// Analyze the bytecode file
442 Module* M = AnalyzeBytecodeFile(InputFilename, bca,
443 Compressor::decompressToNewBuffer,
444 &ErrorMessage, (Dump?Out:0));
446 // All that bcanalyzer does is write the gathered statistics to the output
447 PrintBytecodeAnalysis(bca,*Out);
450 std::string verificationMsg;
451 if (verifyModule(*M, ReturnStatusAction, &verificationMsg))
452 std::cerr << "Final Verification Message: " << verificationMsg << "\n";
455 if (Out != &std::cout) {
456 ((std::ofstream*)Out)->close();
460 } catch (const std::string& msg) {
461 std::cerr << argv[0] << ": " << msg << "\n";
463 std::cerr << argv[0] << ": Unexpected unknown exception occurred.\n";