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"
47 static cl::opt<std::string>
48 InputFilename(cl::Positional, cl::desc("<input bytecode>"), cl::init("-"));
50 static cl::opt<std::string>
51 OutputFilename("-o", cl::init("-"), cl::desc("<output file>"));
53 static cl::opt<bool> NoDetails("nodetails", cl::desc("Skip detailed output"));
54 static cl::opt<bool> Dump("dump", cl::desc("Dump low level bytecode trace"));
55 static cl::opt<bool> Verify("verify", cl::desc("Progressively verify module"));
57 //===----------------------------------------------------------------------===//
58 // Bitcode specific analysis.
59 //===----------------------------------------------------------------------===//
61 static cl::opt<bool> Bitcode("bitcode", cl::desc("Read a bitcode file"));
62 static cl::opt<bool> NoHistogram("disable-histogram",
63 cl::desc("Do not print per-code histogram"));
66 NonSymbolic("non-symbolic",
67 cl::desc("Emit numberic info in dump even if"
68 " symbolic info is available"));
70 /// CurStreamType - If we can sniff the flavor of this stream, we can produce
78 /// GetBlockName - Return a symbolic block name if known, otherwise return
80 static const char *GetBlockName(unsigned BlockID) {
81 // Standard blocks for all bitcode files.
82 if (BlockID < bitc::FIRST_APPLICATION_BLOCKID) {
83 if (BlockID == bitc::BLOCKINFO_BLOCK_ID)
84 return "BLOCKINFO_BLOCK";
88 if (CurStreamType != LLVMIRBitstream) return 0;
92 case bitc::MODULE_BLOCK_ID: return "MODULE_BLOCK";
93 case bitc::PARAMATTR_BLOCK_ID: return "PARAMATTR_BLOCK";
94 case bitc::TYPE_BLOCK_ID: return "TYPE_BLOCK";
95 case bitc::CONSTANTS_BLOCK_ID: return "CONSTANTS_BLOCK";
96 case bitc::FUNCTION_BLOCK_ID: return "FUNCTION_BLOCK";
97 case bitc::TYPE_SYMTAB_BLOCK_ID: return "TYPE_SYMTAB";
98 case bitc::VALUE_SYMTAB_BLOCK_ID: return "VALUE_SYMTAB";
102 /// GetCodeName - Return a symbolic code name if known, otherwise return
104 static const char *GetCodeName(unsigned CodeID, unsigned BlockID) {
105 // Standard blocks for all bitcode files.
106 if (BlockID < bitc::FIRST_APPLICATION_BLOCKID) {
107 if (BlockID == bitc::BLOCKINFO_BLOCK_ID) {
110 case bitc::MODULE_CODE_VERSION: return "VERSION";
116 if (CurStreamType != LLVMIRBitstream) return 0;
120 case bitc::MODULE_BLOCK_ID:
123 case bitc::MODULE_CODE_VERSION: return "VERSION";
124 case bitc::MODULE_CODE_TRIPLE: return "TRIPLE";
125 case bitc::MODULE_CODE_DATALAYOUT: return "DATALAYOUT";
126 case bitc::MODULE_CODE_ASM: return "ASM";
127 case bitc::MODULE_CODE_SECTIONNAME: return "SECTIONNAME";
128 case bitc::MODULE_CODE_DEPLIB: return "DEPLIB";
129 case bitc::MODULE_CODE_GLOBALVAR: return "GLOBALVAR";
130 case bitc::MODULE_CODE_FUNCTION: return "FUNCTION";
131 case bitc::MODULE_CODE_ALIAS: return "ALIAS";
132 case bitc::MODULE_CODE_PURGEVALS: return "PURGEVALS";
134 case bitc::PARAMATTR_BLOCK_ID:
137 case bitc::PARAMATTR_CODE_ENTRY: return "ENTRY";
139 case bitc::TYPE_BLOCK_ID:
142 case bitc::TYPE_CODE_NUMENTRY: return "NUMENTRY";
143 case bitc::TYPE_CODE_VOID: return "VOID";
144 case bitc::TYPE_CODE_FLOAT: return "FLOAT";
145 case bitc::TYPE_CODE_DOUBLE: return "DOUBLE";
146 case bitc::TYPE_CODE_LABEL: return "LABEL";
147 case bitc::TYPE_CODE_OPAQUE: return "OPAQUE";
148 case bitc::TYPE_CODE_INTEGER: return "INTEGER";
149 case bitc::TYPE_CODE_POINTER: return "POINTER";
150 case bitc::TYPE_CODE_FUNCTION: return "FUNCTION";
151 case bitc::TYPE_CODE_STRUCT: return "STRUCT";
152 case bitc::TYPE_CODE_ARRAY: return "ARRAY";
153 case bitc::TYPE_CODE_VECTOR: return "VECTOR";
156 case bitc::CONSTANTS_BLOCK_ID:
159 case bitc::CST_CODE_SETTYPE: return "SETTYPE";
160 case bitc::CST_CODE_NULL: return "NULL";
161 case bitc::CST_CODE_UNDEF: return "UNDEF";
162 case bitc::CST_CODE_INTEGER: return "INTEGER";
163 case bitc::CST_CODE_WIDE_INTEGER: return "WIDE_INTEGER";
164 case bitc::CST_CODE_FLOAT: return "FLOAT";
165 case bitc::CST_CODE_AGGREGATE: return "AGGREGATE";
166 case bitc::CST_CODE_CE_BINOP: return "CE_BINOP";
167 case bitc::CST_CODE_CE_CAST: return "CE_CAST";
168 case bitc::CST_CODE_CE_GEP: return "CE_GEP";
169 case bitc::CST_CODE_CE_SELECT: return "CE_SELECT";
170 case bitc::CST_CODE_CE_EXTRACTELT: return "CE_EXTRACTELT";
171 case bitc::CST_CODE_CE_INSERTELT: return "CE_INSERTELT";
172 case bitc::CST_CODE_CE_SHUFFLEVEC: return "CE_SHUFFLEVEC";
173 case bitc::CST_CODE_CE_CMP: return "CE_CMP";
175 case bitc::FUNCTION_BLOCK_ID:
178 case bitc::FUNC_CODE_DECLAREBLOCKS: return "DECLAREBLOCKS";
180 case bitc::FUNC_CODE_INST_BINOP: return "INST_BINOP";
181 case bitc::FUNC_CODE_INST_CAST: return "INST_CAST";
182 case bitc::FUNC_CODE_INST_GEP: return "INST_GEP";
183 case bitc::FUNC_CODE_INST_SELECT: return "INST_SELECT";
184 case bitc::FUNC_CODE_INST_EXTRACTELT: return "INST_EXTRACTELT";
185 case bitc::FUNC_CODE_INST_INSERTELT: return "INST_INSERTELT";
186 case bitc::FUNC_CODE_INST_SHUFFLEVEC: return "INST_SHUFFLEVEC";
187 case bitc::FUNC_CODE_INST_CMP: return "INST_CMP";
189 case bitc::FUNC_CODE_INST_RET: return "INST_RET";
190 case bitc::FUNC_CODE_INST_BR: return "INST_BR";
191 case bitc::FUNC_CODE_INST_SWITCH: return "INST_SWITCH";
192 case bitc::FUNC_CODE_INST_INVOKE: return "INST_INVOKE";
193 case bitc::FUNC_CODE_INST_UNWIND: return "INST_UNWIND";
194 case bitc::FUNC_CODE_INST_UNREACHABLE: return "INST_UNREACHABLE";
196 case bitc::FUNC_CODE_INST_PHI: return "INST_PHI";
197 case bitc::FUNC_CODE_INST_MALLOC: return "INST_MALLOC";
198 case bitc::FUNC_CODE_INST_FREE: return "INST_FREE";
199 case bitc::FUNC_CODE_INST_ALLOCA: return "INST_ALLOCA";
200 case bitc::FUNC_CODE_INST_LOAD: return "INST_LOAD";
201 case bitc::FUNC_CODE_INST_STORE: return "INST_STORE";
202 case bitc::FUNC_CODE_INST_CALL: return "INST_CALL";
203 case bitc::FUNC_CODE_INST_VAARG: return "INST_VAARG";
205 case bitc::TYPE_SYMTAB_BLOCK_ID:
208 case bitc::TST_CODE_ENTRY: return "ENTRY";
210 case bitc::VALUE_SYMTAB_BLOCK_ID:
213 case bitc::VST_CODE_ENTRY: return "ENTRY";
214 case bitc::VST_CODE_BBENTRY: return "BBENTRY";
220 struct PerBlockIDStats {
221 /// NumInstances - This the number of times this block ID has been seen.
222 unsigned NumInstances;
224 /// NumBits - The total size in bits of all of these blocks.
227 /// NumSubBlocks - The total number of blocks these blocks contain.
228 unsigned NumSubBlocks;
230 /// NumAbbrevs - The total number of abbreviations.
233 /// NumRecords - The total number of records these blocks contain, and the
234 /// number that are abbreviated.
235 unsigned NumRecords, NumAbbreviatedRecords;
237 /// CodeFreq - Keep track of the number of times we see each code.
238 std::vector<unsigned> CodeFreq;
241 : NumInstances(0), NumBits(0),
242 NumSubBlocks(0), NumAbbrevs(0), NumRecords(0), NumAbbreviatedRecords(0) {}
245 static std::map<unsigned, PerBlockIDStats> BlockIDStats;
249 /// Error - All bitcode analysis errors go through this function, making this a
250 /// good place to breakpoint if debugging.
251 static bool Error(const std::string &Err) {
252 std::cerr << Err << "\n";
256 /// ParseBlock - Read a block, updating statistics, etc.
257 static bool ParseBlock(BitstreamReader &Stream, unsigned IndentLevel) {
258 std::string Indent(IndentLevel*2, ' ');
259 uint64_t BlockBitStart = Stream.GetCurrentBitNo();
260 unsigned BlockID = Stream.ReadSubBlockID();
262 // Get the statistics for this BlockID.
263 PerBlockIDStats &BlockStats = BlockIDStats[BlockID];
265 BlockStats.NumInstances++;
267 // BLOCKINFO is a special part of the stream.
268 if (BlockID == bitc::BLOCKINFO_BLOCK_ID) {
269 if (Dump) std::cerr << Indent << "<BLOCKINFO_BLOCK/>\n";
270 if (Stream.ReadBlockInfoBlock())
271 return Error("Malformed BlockInfoBlock");
272 uint64_t BlockBitEnd = Stream.GetCurrentBitNo();
273 BlockStats.NumBits += BlockBitEnd-BlockBitStart;
277 unsigned NumWords = 0;
278 if (Stream.EnterSubBlock(BlockID, &NumWords))
279 return Error("Malformed block record");
281 const char *BlockName = 0;
283 std::cerr << Indent << "<";
284 if ((BlockName = GetBlockName(BlockID)))
285 std::cerr << BlockName;
287 std::cerr << "UnknownBlock" << BlockID;
289 if (NonSymbolic && BlockName)
290 std::cerr << " BlockID=" << BlockID;
292 std::cerr << " NumWords=" << NumWords
293 << " BlockCodeSize=" << Stream.GetAbbrevIDWidth() << ">\n";
296 SmallVector<uint64_t, 64> Record;
298 // Read all the records for this block.
300 if (Stream.AtEndOfStream())
301 return Error("Premature end of bitstream");
303 // Read the code for this record.
304 unsigned AbbrevID = Stream.ReadCode();
306 case bitc::END_BLOCK: {
307 if (Stream.ReadBlockEnd())
308 return Error("Error at end of block");
309 uint64_t BlockBitEnd = Stream.GetCurrentBitNo();
310 BlockStats.NumBits += BlockBitEnd-BlockBitStart;
312 std::cerr << Indent << "</";
314 std::cerr << BlockName << ">\n";
316 std::cerr << "UnknownBlock" << BlockID << ">\n";
320 case bitc::ENTER_SUBBLOCK: {
321 uint64_t SubBlockBitStart = Stream.GetCurrentBitNo();
322 if (ParseBlock(Stream, IndentLevel+1))
324 ++BlockStats.NumSubBlocks;
325 uint64_t SubBlockBitEnd = Stream.GetCurrentBitNo();
327 // Don't include subblock sizes in the size of this block.
328 BlockBitStart += SubBlockBitEnd-SubBlockBitStart;
331 case bitc::DEFINE_ABBREV:
332 Stream.ReadAbbrevRecord();
333 ++BlockStats.NumAbbrevs;
336 ++BlockStats.NumRecords;
337 if (AbbrevID != bitc::UNABBREV_RECORD)
338 ++BlockStats.NumAbbreviatedRecords;
341 unsigned Code = Stream.ReadRecord(AbbrevID, Record);
343 // Increment the # occurrences of this code.
344 if (BlockStats.CodeFreq.size() <= Code)
345 BlockStats.CodeFreq.resize(Code+1);
346 BlockStats.CodeFreq[Code]++;
349 std::cerr << Indent << " <";
350 if (const char *CodeName = GetCodeName(Code, BlockID))
351 std::cerr << CodeName;
353 std::cerr << "UnknownCode" << Code;
354 if (NonSymbolic && GetCodeName(Code, BlockID))
355 std::cerr << " codeid=" << Code;
356 if (AbbrevID != bitc::UNABBREV_RECORD)
357 std::cerr << " abbrevid=" << AbbrevID;
359 for (unsigned i = 0, e = Record.size(); i != e; ++i)
360 std::cerr << " op" << i << "=" << (int64_t)Record[i];
370 static void PrintSize(double Bits) {
371 std::cerr << Bits << "b/" << Bits/8 << "B/" << Bits/32 << "W";
375 /// AnalyzeBitcode - Analyze the bitcode file specified by InputFilename.
376 static int AnalyzeBitcode() {
377 // Read the input file.
378 MemoryBuffer *Buffer;
379 if (InputFilename == "-")
380 Buffer = MemoryBuffer::getSTDIN();
382 Buffer = MemoryBuffer::getFile(&InputFilename[0], InputFilename.size());
385 return Error("Error reading '" + InputFilename + "'.");
387 if (Buffer->getBufferSize() & 3)
388 return Error("Bitcode stream should be a multiple of 4 bytes in length");
390 unsigned char *BufPtr = (unsigned char *)Buffer->getBufferStart();
391 BitstreamReader Stream(BufPtr, BufPtr+Buffer->getBufferSize());
394 // Read the stream signature.
396 Signature[0] = Stream.Read(8);
397 Signature[1] = Stream.Read(8);
398 Signature[2] = Stream.Read(4);
399 Signature[3] = Stream.Read(4);
400 Signature[4] = Stream.Read(4);
401 Signature[5] = Stream.Read(4);
403 // Autodetect the file contents, if it is one we know.
404 CurStreamType = UnknownBitstream;
405 if (Signature[0] == 'B' && Signature[1] == 'C' &&
406 Signature[2] == 0x0 && Signature[3] == 0xC &&
407 Signature[4] == 0xE && Signature[5] == 0xD)
408 CurStreamType = LLVMIRBitstream;
410 unsigned NumTopBlocks = 0;
412 // Parse the top-level structure. We only allow blocks at the top-level.
413 while (!Stream.AtEndOfStream()) {
414 unsigned Code = Stream.ReadCode();
415 if (Code != bitc::ENTER_SUBBLOCK)
416 return Error("Invalid record at top-level");
418 if (ParseBlock(Stream, 0))
423 if (Dump) std::cerr << "\n\n";
425 uint64_t BufferSizeBits = Buffer->getBufferSize()*8;
426 // Print a summary of the read file.
427 std::cerr << "Summary of " << InputFilename << ":\n";
428 std::cerr << " Total size: ";
429 PrintSize(BufferSizeBits);
431 std::cerr << " Stream type: ";
432 switch (CurStreamType) {
433 default: assert(0 && "Unknown bitstream type");
434 case UnknownBitstream: std::cerr << "unknown\n"; break;
435 case LLVMIRBitstream: std::cerr << "LLVM IR\n"; break;
437 std::cerr << " # Toplevel Blocks: " << NumTopBlocks << "\n";
440 // Emit per-block stats.
441 std::cerr << "Per-block Summary:\n";
442 for (std::map<unsigned, PerBlockIDStats>::iterator I = BlockIDStats.begin(),
443 E = BlockIDStats.end(); I != E; ++I) {
444 std::cerr << " Block ID #" << I->first;
445 if (const char *BlockName = GetBlockName(I->first))
446 std::cerr << " (" << BlockName << ")";
449 const PerBlockIDStats &Stats = I->second;
450 std::cerr << " Num Instances: " << Stats.NumInstances << "\n";
451 std::cerr << " Total Size: ";
452 PrintSize(Stats.NumBits);
454 std::cerr << " % of file: "
455 << Stats.NumBits/(double)BufferSizeBits*100 << "\n";
456 if (Stats.NumInstances > 1) {
457 std::cerr << " Average Size: ";
458 PrintSize(Stats.NumBits/(double)Stats.NumInstances);
460 std::cerr << " Tot/Avg SubBlocks: " << Stats.NumSubBlocks << "/"
461 << Stats.NumSubBlocks/(double)Stats.NumInstances << "\n";
462 std::cerr << " Tot/Avg Abbrevs: " << Stats.NumAbbrevs << "/"
463 << Stats.NumAbbrevs/(double)Stats.NumInstances << "\n";
464 std::cerr << " Tot/Avg Records: " << Stats.NumRecords << "/"
465 << Stats.NumRecords/(double)Stats.NumInstances << "\n";
467 std::cerr << " Num SubBlocks: " << Stats.NumSubBlocks << "\n";
468 std::cerr << " Num Abbrevs: " << Stats.NumAbbrevs << "\n";
469 std::cerr << " Num Records: " << Stats.NumRecords << "\n";
471 if (Stats.NumRecords)
472 std::cerr << " % Abbrev Recs: " << (Stats.NumAbbreviatedRecords/
473 (double)Stats.NumRecords)*100 << "\n";
476 // Print a histogram of the codes we see.
477 if (!NoHistogram && !Stats.CodeFreq.empty()) {
478 std::vector<std::pair<unsigned, unsigned> > FreqPairs; // <freq,code>
479 for (unsigned i = 0, e = Stats.CodeFreq.size(); i != e; ++i)
480 if (unsigned Freq = Stats.CodeFreq[i])
481 FreqPairs.push_back(std::make_pair(Freq, i));
482 std::stable_sort(FreqPairs.begin(), FreqPairs.end());
483 std::reverse(FreqPairs.begin(), FreqPairs.end());
485 std::cerr << "\tCode Histogram:\n";
486 for (unsigned i = 0, e = FreqPairs.size(); i != e; ++i) {
487 std::cerr << "\t\t" << FreqPairs[i].first << "\t";
488 if (const char *CodeName = GetCodeName(FreqPairs[i].second, I->first))
489 std::cerr << CodeName << "\n";
491 std::cerr << "UnknownCode" << FreqPairs[i].second << "\n";
501 //===----------------------------------------------------------------------===//
502 // Bytecode specific analysis.
503 //===----------------------------------------------------------------------===//
505 int main(int argc, char **argv) {
506 llvm_shutdown_obj X; // Call llvm_shutdown() on exit.
507 cl::ParseCommandLineOptions(argc, argv, " llvm-bcanalyzer file analyzer\n");
509 sys::PrintStackTraceOnErrorSignal();
512 return AnalyzeBitcode();
515 std::ostream *Out = &std::cout; // Default to printing to stdout...
516 std::string ErrorMessage;
517 BytecodeAnalysis bca;
519 /// Determine what to generate
520 bca.detailedResults = !NoDetails;
521 bca.progressiveVerify = Verify;
523 /// Analyze the bytecode file
524 Module* M = AnalyzeBytecodeFile(InputFilename, bca,
525 Compressor::decompressToNewBuffer,
526 &ErrorMessage, (Dump?Out:0));
528 // All that bcanalyzer does is write the gathered statistics to the output
529 PrintBytecodeAnalysis(bca,*Out);
532 std::string verificationMsg;
533 if (verifyModule(*M, ReturnStatusAction, &verificationMsg))
534 std::cerr << "Final Verification Message: " << verificationMsg << "\n";
537 if (Out != &std::cout) {
538 ((std::ofstream*)Out)->close();
542 } catch (const std::string& msg) {
543 std::cerr << argv[0] << ": " << msg << "\n";
545 std::cerr << argv[0] << ": Unexpected unknown exception occurred.\n";