1 //===-- llvm-bcanalyzer.cpp - Bitcode Analyzer --------------------------===//
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 tool may be invoked in the following manner:
11 // llvm-bcanalyzer [options] - Read LLVM bitcode from stdin
12 // llvm-bcanalyzer [options] x.bc - Read LLVM bitcode from the x.bc file
15 // --help - Output information about command line switches
16 // --dump - Dump low-level bitcode structure in readable format
18 // This tool provides analytical information about a bitcode file. It is
19 // intended as an aid to developers of bitcode reading and writing software. It
20 // produces on std::out a summary of the bitcode file that shows various
21 // statistics about the contents of the file. By default this information is
22 // detailed and contains information about individual bitcode blocks and the
23 // functions in the module.
24 // The tool is also able to print a bitcode file in a straight forward text
25 // format that shows the containment and relationships of the information in
26 // the bitcode file (-dump option).
28 //===----------------------------------------------------------------------===//
30 #include "llvm/Analysis/Verifier.h"
31 #include "llvm/Bitcode/BitstreamReader.h"
32 #include "llvm/Bitcode/LLVMBitCodes.h"
33 #include "llvm/Support/CommandLine.h"
34 #include "llvm/Support/ManagedStatic.h"
35 #include "llvm/Support/MemoryBuffer.h"
36 #include "llvm/System/Signals.h"
43 static cl::opt<std::string>
44 InputFilename(cl::Positional, cl::desc("<input bitcode>"), cl::init("-"));
46 static cl::opt<std::string>
47 OutputFilename("-o", cl::init("-"), cl::desc("<output file>"));
49 static cl::opt<bool> Dump("dump", cl::desc("Dump low level bitcode trace"));
51 //===----------------------------------------------------------------------===//
52 // Bitcode specific analysis.
53 //===----------------------------------------------------------------------===//
55 static cl::opt<bool> NoHistogram("disable-histogram",
56 cl::desc("Do not print per-code histogram"));
59 NonSymbolic("non-symbolic",
60 cl::desc("Emit numberic info in dump even if"
61 " symbolic info is available"));
63 /// CurStreamType - If we can sniff the flavor of this stream, we can produce
71 /// GetBlockName - Return a symbolic block name if known, otherwise return
73 static const char *GetBlockName(unsigned BlockID) {
74 // Standard blocks for all bitcode files.
75 if (BlockID < bitc::FIRST_APPLICATION_BLOCKID) {
76 if (BlockID == bitc::BLOCKINFO_BLOCK_ID)
77 return "BLOCKINFO_BLOCK";
81 if (CurStreamType != LLVMIRBitstream) return 0;
85 case bitc::MODULE_BLOCK_ID: return "MODULE_BLOCK";
86 case bitc::PARAMATTR_BLOCK_ID: return "PARAMATTR_BLOCK";
87 case bitc::TYPE_BLOCK_ID: return "TYPE_BLOCK";
88 case bitc::CONSTANTS_BLOCK_ID: return "CONSTANTS_BLOCK";
89 case bitc::FUNCTION_BLOCK_ID: return "FUNCTION_BLOCK";
90 case bitc::TYPE_SYMTAB_BLOCK_ID: return "TYPE_SYMTAB";
91 case bitc::VALUE_SYMTAB_BLOCK_ID: return "VALUE_SYMTAB";
95 /// GetCodeName - Return a symbolic code name if known, otherwise return
97 static const char *GetCodeName(unsigned CodeID, unsigned BlockID) {
98 // Standard blocks for all bitcode files.
99 if (BlockID < bitc::FIRST_APPLICATION_BLOCKID) {
100 if (BlockID == bitc::BLOCKINFO_BLOCK_ID) {
103 case bitc::MODULE_CODE_VERSION: return "VERSION";
109 if (CurStreamType != LLVMIRBitstream) return 0;
113 case bitc::MODULE_BLOCK_ID:
116 case bitc::MODULE_CODE_VERSION: return "VERSION";
117 case bitc::MODULE_CODE_TRIPLE: return "TRIPLE";
118 case bitc::MODULE_CODE_DATALAYOUT: return "DATALAYOUT";
119 case bitc::MODULE_CODE_ASM: return "ASM";
120 case bitc::MODULE_CODE_SECTIONNAME: return "SECTIONNAME";
121 case bitc::MODULE_CODE_DEPLIB: return "DEPLIB";
122 case bitc::MODULE_CODE_GLOBALVAR: return "GLOBALVAR";
123 case bitc::MODULE_CODE_FUNCTION: return "FUNCTION";
124 case bitc::MODULE_CODE_ALIAS: return "ALIAS";
125 case bitc::MODULE_CODE_PURGEVALS: return "PURGEVALS";
126 case bitc::MODULE_CODE_GCNAME: return "GCNAME";
128 case bitc::PARAMATTR_BLOCK_ID:
131 case bitc::PARAMATTR_CODE_ENTRY: return "ENTRY";
133 case bitc::TYPE_BLOCK_ID:
136 case bitc::TYPE_CODE_NUMENTRY: return "NUMENTRY";
137 case bitc::TYPE_CODE_VOID: return "VOID";
138 case bitc::TYPE_CODE_FLOAT: return "FLOAT";
139 case bitc::TYPE_CODE_DOUBLE: return "DOUBLE";
140 case bitc::TYPE_CODE_LABEL: return "LABEL";
141 case bitc::TYPE_CODE_OPAQUE: return "OPAQUE";
142 case bitc::TYPE_CODE_INTEGER: return "INTEGER";
143 case bitc::TYPE_CODE_POINTER: return "POINTER";
144 case bitc::TYPE_CODE_FUNCTION: return "FUNCTION";
145 case bitc::TYPE_CODE_STRUCT: return "STRUCT";
146 case bitc::TYPE_CODE_ARRAY: return "ARRAY";
147 case bitc::TYPE_CODE_VECTOR: return "VECTOR";
148 case bitc::TYPE_CODE_X86_FP80: return "X86_FP80";
149 case bitc::TYPE_CODE_FP128: return "FP128";
150 case bitc::TYPE_CODE_PPC_FP128: return "PPC_FP128";
153 case bitc::CONSTANTS_BLOCK_ID:
156 case bitc::CST_CODE_SETTYPE: return "SETTYPE";
157 case bitc::CST_CODE_NULL: return "NULL";
158 case bitc::CST_CODE_UNDEF: return "UNDEF";
159 case bitc::CST_CODE_INTEGER: return "INTEGER";
160 case bitc::CST_CODE_WIDE_INTEGER: return "WIDE_INTEGER";
161 case bitc::CST_CODE_FLOAT: return "FLOAT";
162 case bitc::CST_CODE_AGGREGATE: return "AGGREGATE";
163 case bitc::CST_CODE_STRING: return "STRING";
164 case bitc::CST_CODE_CSTRING: return "CSTRING";
165 case bitc::CST_CODE_CE_BINOP: return "CE_BINOP";
166 case bitc::CST_CODE_CE_CAST: return "CE_CAST";
167 case bitc::CST_CODE_CE_GEP: return "CE_GEP";
168 case bitc::CST_CODE_CE_SELECT: return "CE_SELECT";
169 case bitc::CST_CODE_CE_EXTRACTELT: return "CE_EXTRACTELT";
170 case bitc::CST_CODE_CE_INSERTELT: return "CE_INSERTELT";
171 case bitc::CST_CODE_CE_SHUFFLEVEC: return "CE_SHUFFLEVEC";
172 case bitc::CST_CODE_CE_CMP: return "CE_CMP";
173 case bitc::CST_CODE_INLINEASM: return "INLINEASM";
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";
204 case bitc::FUNC_CODE_INST_STORE2: return "INST_STORE2";
205 case bitc::FUNC_CODE_INST_GETRESULT: return "INST_GETRESULT";
206 case bitc::FUNC_CODE_INST_EXTRACTVAL: return "INST_EXTRACTVAL";
207 case bitc::FUNC_CODE_INST_INSERTVAL: return "INST_INSERTVAL";
208 case bitc::FUNC_CODE_INST_CMP2: return "INST_CMP2";
209 case bitc::FUNC_CODE_INST_VSELECT: return "INST_VSELECT";
211 case bitc::TYPE_SYMTAB_BLOCK_ID:
214 case bitc::TST_CODE_ENTRY: return "ENTRY";
216 case bitc::VALUE_SYMTAB_BLOCK_ID:
219 case bitc::VST_CODE_ENTRY: return "ENTRY";
220 case bitc::VST_CODE_BBENTRY: return "BBENTRY";
226 struct PerBlockIDStats {
227 /// NumInstances - This the number of times this block ID has been seen.
228 unsigned NumInstances;
230 /// NumBits - The total size in bits of all of these blocks.
233 /// NumSubBlocks - The total number of blocks these blocks contain.
234 unsigned NumSubBlocks;
236 /// NumAbbrevs - The total number of abbreviations.
239 /// NumRecords - The total number of records these blocks contain, and the
240 /// number that are abbreviated.
241 unsigned NumRecords, NumAbbreviatedRecords;
243 /// CodeFreq - Keep track of the number of times we see each code.
244 std::vector<unsigned> CodeFreq;
247 : NumInstances(0), NumBits(0),
248 NumSubBlocks(0), NumAbbrevs(0), NumRecords(0), NumAbbreviatedRecords(0) {}
251 static std::map<unsigned, PerBlockIDStats> BlockIDStats;
255 /// Error - All bitcode analysis errors go through this function, making this a
256 /// good place to breakpoint if debugging.
257 static bool Error(const std::string &Err) {
258 std::cerr << Err << "\n";
262 /// ParseBlock - Read a block, updating statistics, etc.
263 static bool ParseBlock(BitstreamReader &Stream, unsigned IndentLevel) {
264 std::string Indent(IndentLevel*2, ' ');
265 uint64_t BlockBitStart = Stream.GetCurrentBitNo();
266 unsigned BlockID = Stream.ReadSubBlockID();
268 // Get the statistics for this BlockID.
269 PerBlockIDStats &BlockStats = BlockIDStats[BlockID];
271 BlockStats.NumInstances++;
273 // BLOCKINFO is a special part of the stream.
274 if (BlockID == bitc::BLOCKINFO_BLOCK_ID) {
275 if (Dump) std::cerr << Indent << "<BLOCKINFO_BLOCK/>\n";
276 if (Stream.ReadBlockInfoBlock())
277 return Error("Malformed BlockInfoBlock");
278 uint64_t BlockBitEnd = Stream.GetCurrentBitNo();
279 BlockStats.NumBits += BlockBitEnd-BlockBitStart;
283 unsigned NumWords = 0;
284 if (Stream.EnterSubBlock(BlockID, &NumWords))
285 return Error("Malformed block record");
287 const char *BlockName = 0;
289 std::cerr << Indent << "<";
290 if ((BlockName = GetBlockName(BlockID)))
291 std::cerr << BlockName;
293 std::cerr << "UnknownBlock" << BlockID;
295 if (NonSymbolic && BlockName)
296 std::cerr << " BlockID=" << BlockID;
298 std::cerr << " NumWords=" << NumWords
299 << " BlockCodeSize=" << Stream.GetAbbrevIDWidth() << ">\n";
302 SmallVector<uint64_t, 64> Record;
304 // Read all the records for this block.
306 if (Stream.AtEndOfStream())
307 return Error("Premature end of bitstream");
309 // Read the code for this record.
310 unsigned AbbrevID = Stream.ReadCode();
312 case bitc::END_BLOCK: {
313 if (Stream.ReadBlockEnd())
314 return Error("Error at end of block");
315 uint64_t BlockBitEnd = Stream.GetCurrentBitNo();
316 BlockStats.NumBits += BlockBitEnd-BlockBitStart;
318 std::cerr << Indent << "</";
320 std::cerr << BlockName << ">\n";
322 std::cerr << "UnknownBlock" << BlockID << ">\n";
326 case bitc::ENTER_SUBBLOCK: {
327 uint64_t SubBlockBitStart = Stream.GetCurrentBitNo();
328 if (ParseBlock(Stream, IndentLevel+1))
330 ++BlockStats.NumSubBlocks;
331 uint64_t SubBlockBitEnd = Stream.GetCurrentBitNo();
333 // Don't include subblock sizes in the size of this block.
334 BlockBitStart += SubBlockBitEnd-SubBlockBitStart;
337 case bitc::DEFINE_ABBREV:
338 Stream.ReadAbbrevRecord();
339 ++BlockStats.NumAbbrevs;
342 ++BlockStats.NumRecords;
343 if (AbbrevID != bitc::UNABBREV_RECORD)
344 ++BlockStats.NumAbbreviatedRecords;
347 unsigned Code = Stream.ReadRecord(AbbrevID, Record);
349 // Increment the # occurrences of this code.
350 if (BlockStats.CodeFreq.size() <= Code)
351 BlockStats.CodeFreq.resize(Code+1);
352 BlockStats.CodeFreq[Code]++;
355 std::cerr << Indent << " <";
356 if (const char *CodeName = GetCodeName(Code, BlockID))
357 std::cerr << CodeName;
359 std::cerr << "UnknownCode" << Code;
360 if (NonSymbolic && GetCodeName(Code, BlockID))
361 std::cerr << " codeid=" << Code;
362 if (AbbrevID != bitc::UNABBREV_RECORD)
363 std::cerr << " abbrevid=" << AbbrevID;
365 for (unsigned i = 0, e = Record.size(); i != e; ++i)
366 std::cerr << " op" << i << "=" << (int64_t)Record[i];
376 static void PrintSize(double Bits) {
377 std::cerr << Bits << "b/" << Bits/8 << "B/" << Bits/32 << "W";
381 /// AnalyzeBitcode - Analyze the bitcode file specified by InputFilename.
382 static int AnalyzeBitcode() {
383 // Read the input file.
384 MemoryBuffer *Buffer = MemoryBuffer::getFileOrSTDIN(InputFilename.c_str());
387 return Error("Error reading '" + InputFilename + "'.");
389 if (Buffer->getBufferSize() & 3)
390 return Error("Bitcode stream should be a multiple of 4 bytes in length");
392 unsigned char *BufPtr = (unsigned char *)Buffer->getBufferStart();
393 BitstreamReader Stream(BufPtr, BufPtr+Buffer->getBufferSize());
396 // Read the stream signature.
398 Signature[0] = Stream.Read(8);
399 Signature[1] = Stream.Read(8);
400 Signature[2] = Stream.Read(4);
401 Signature[3] = Stream.Read(4);
402 Signature[4] = Stream.Read(4);
403 Signature[5] = Stream.Read(4);
405 // Autodetect the file contents, if it is one we know.
406 CurStreamType = UnknownBitstream;
407 if (Signature[0] == 'B' && Signature[1] == 'C' &&
408 Signature[2] == 0x0 && Signature[3] == 0xC &&
409 Signature[4] == 0xE && Signature[5] == 0xD)
410 CurStreamType = LLVMIRBitstream;
412 unsigned NumTopBlocks = 0;
414 // Parse the top-level structure. We only allow blocks at the top-level.
415 while (!Stream.AtEndOfStream()) {
416 unsigned Code = Stream.ReadCode();
417 if (Code != bitc::ENTER_SUBBLOCK)
418 return Error("Invalid record at top-level");
420 if (ParseBlock(Stream, 0))
425 if (Dump) std::cerr << "\n\n";
427 uint64_t BufferSizeBits = Buffer->getBufferSize()*8;
428 // Print a summary of the read file.
429 std::cerr << "Summary of " << InputFilename << ":\n";
430 std::cerr << " Total size: ";
431 PrintSize(BufferSizeBits);
433 std::cerr << " Stream type: ";
434 switch (CurStreamType) {
435 default: assert(0 && "Unknown bitstream type");
436 case UnknownBitstream: std::cerr << "unknown\n"; break;
437 case LLVMIRBitstream: std::cerr << "LLVM IR\n"; break;
439 std::cerr << " # Toplevel Blocks: " << NumTopBlocks << "\n";
442 // Emit per-block stats.
443 std::cerr << "Per-block Summary:\n";
444 for (std::map<unsigned, PerBlockIDStats>::iterator I = BlockIDStats.begin(),
445 E = BlockIDStats.end(); I != E; ++I) {
446 std::cerr << " Block ID #" << I->first;
447 if (const char *BlockName = GetBlockName(I->first))
448 std::cerr << " (" << BlockName << ")";
451 const PerBlockIDStats &Stats = I->second;
452 std::cerr << " Num Instances: " << Stats.NumInstances << "\n";
453 std::cerr << " Total Size: ";
454 PrintSize(Stats.NumBits);
456 std::cerr << " % of file: "
457 << Stats.NumBits/(double)BufferSizeBits*100 << "\n";
458 if (Stats.NumInstances > 1) {
459 std::cerr << " Average Size: ";
460 PrintSize(Stats.NumBits/(double)Stats.NumInstances);
462 std::cerr << " Tot/Avg SubBlocks: " << Stats.NumSubBlocks << "/"
463 << Stats.NumSubBlocks/(double)Stats.NumInstances << "\n";
464 std::cerr << " Tot/Avg Abbrevs: " << Stats.NumAbbrevs << "/"
465 << Stats.NumAbbrevs/(double)Stats.NumInstances << "\n";
466 std::cerr << " Tot/Avg Records: " << Stats.NumRecords << "/"
467 << Stats.NumRecords/(double)Stats.NumInstances << "\n";
469 std::cerr << " Num SubBlocks: " << Stats.NumSubBlocks << "\n";
470 std::cerr << " Num Abbrevs: " << Stats.NumAbbrevs << "\n";
471 std::cerr << " Num Records: " << Stats.NumRecords << "\n";
473 if (Stats.NumRecords)
474 std::cerr << " % Abbrev Recs: " << (Stats.NumAbbreviatedRecords/
475 (double)Stats.NumRecords)*100 << "\n";
478 // Print a histogram of the codes we see.
479 if (!NoHistogram && !Stats.CodeFreq.empty()) {
480 std::vector<std::pair<unsigned, unsigned> > FreqPairs; // <freq,code>
481 for (unsigned i = 0, e = Stats.CodeFreq.size(); i != e; ++i)
482 if (unsigned Freq = Stats.CodeFreq[i])
483 FreqPairs.push_back(std::make_pair(Freq, i));
484 std::stable_sort(FreqPairs.begin(), FreqPairs.end());
485 std::reverse(FreqPairs.begin(), FreqPairs.end());
487 std::cerr << "\tCode Histogram:\n";
488 for (unsigned i = 0, e = FreqPairs.size(); i != e; ++i) {
489 std::cerr << "\t\t" << FreqPairs[i].first << "\t";
490 if (const char *CodeName = GetCodeName(FreqPairs[i].second, I->first))
491 std::cerr << CodeName << "\n";
493 std::cerr << "UnknownCode" << FreqPairs[i].second << "\n";
503 int main(int argc, char **argv) {
504 llvm_shutdown_obj X; // Call llvm_shutdown() on exit.
505 cl::ParseCommandLineOptions(argc, argv, "llvm-bcanalyzer file analyzer\n");
507 sys::PrintStackTraceOnErrorSignal();
509 return AnalyzeBitcode();