1This is gprof.info, produced by makeinfo version 6.8 from gprof.texi. 2 3This file documents the gprof profiler of the GNU system. 4 5 Copyright (C) 1988-2022 Free Software Foundation, Inc. 6 7 Permission is granted to copy, distribute and/or modify this document 8under the terms of the GNU Free Documentation License, Version 1.3 or 9any later version published by the Free Software Foundation; with no 10Invariant Sections, with no Front-Cover Texts, and with no Back-Cover 11Texts. A copy of the license is included in the section entitled "GNU 12Free Documentation License". 13 14INFO-DIR-SECTION Software development 15START-INFO-DIR-ENTRY 16* gprof: (gprof). Profiling your program's execution 17END-INFO-DIR-ENTRY 18 19 20File: gprof.info, Node: Top, Next: Introduction, Up: (dir) 21 22Profiling a Program: Where Does It Spend Its Time? 23************************************************** 24 25This manual describes the GNU profiler, 'gprof', and how you can use it 26to determine which parts of a program are taking most of the execution 27time. We assume that you know how to write, compile, and execute 28programs. GNU 'gprof' was written by Jay Fenlason. 29 30 This manual is for 'gprof' (GNU Binutils) version 2.39. 31 32 This document is distributed under the terms of the GNU Free 33Documentation License version 1.3. A copy of the license is included in 34the section entitled "GNU Free Documentation License". 35 36* Menu: 37 38* Introduction:: What profiling means, and why it is useful. 39 40* Compiling:: How to compile your program for profiling. 41* Executing:: Executing your program to generate profile data 42* Invoking:: How to run 'gprof', and its options 43 44* Output:: Interpreting 'gprof''s output 45 46* Inaccuracy:: Potential problems you should be aware of 47* How do I?:: Answers to common questions 48* Incompatibilities:: (between GNU 'gprof' and Unix 'gprof'.) 49* Details:: Details of how profiling is done 50* GNU Free Documentation License:: GNU Free Documentation License 51 52 53File: gprof.info, Node: Introduction, Next: Compiling, Prev: Top, Up: Top 54 551 Introduction to Profiling 56*************************** 57 58Profiling allows you to learn where your program spent its time and 59which functions called which other functions while it was executing. 60This information can show you which pieces of your program are slower 61than you expected, and might be candidates for rewriting to make your 62program execute faster. It can also tell you which functions are being 63called more or less often than you expected. This may help you spot 64bugs that had otherwise been unnoticed. 65 66 Since the profiler uses information collected during the actual 67execution of your program, it can be used on programs that are too large 68or too complex to analyze by reading the source. However, how your 69program is run will affect the information that shows up in the profile 70data. If you don't use some feature of your program while it is being 71profiled, no profile information will be generated for that feature. 72 73 Profiling has several steps: 74 75 * You must compile and link your program with profiling enabled. 76 *Note Compiling a Program for Profiling: Compiling. 77 78 * You must execute your program to generate a profile data file. 79 *Note Executing the Program: Executing. 80 81 * You must run 'gprof' to analyze the profile data. *Note 'gprof' 82 Command Summary: Invoking. 83 84 The next three chapters explain these steps in greater detail. 85 86 Several forms of output are available from the analysis. 87 88 The "flat profile" shows how much time your program spent in each 89function, and how many times that function was called. If you simply 90want to know which functions burn most of the cycles, it is stated 91concisely here. *Note The Flat Profile: Flat Profile. 92 93 The "call graph" shows, for each function, which functions called it, 94which other functions it called, and how many times. There is also an 95estimate of how much time was spent in the subroutines of each function. 96This can suggest places where you might try to eliminate function calls 97that use a lot of time. *Note The Call Graph: Call Graph. 98 99 The "annotated source" listing is a copy of the program's source 100code, labeled with the number of times each line of the program was 101executed. *Note The Annotated Source Listing: Annotated Source. 102 103 To better understand how profiling works, you may wish to read a 104description of its implementation. *Note Implementation of Profiling: 105Implementation. 106 107 108File: gprof.info, Node: Compiling, Next: Executing, Prev: Introduction, Up: Top 109 1102 Compiling a Program for Profiling 111*********************************** 112 113The first step in generating profile information for your program is to 114compile and link it with profiling enabled. 115 116 To compile a source file for profiling, specify the '-pg' option when 117you run the compiler. (This is in addition to the options you normally 118use.) 119 120 To link the program for profiling, if you use a compiler such as 'cc' 121to do the linking, simply specify '-pg' in addition to your usual 122options. The same option, '-pg', alters either compilation or linking 123to do what is necessary for profiling. Here are examples: 124 125 cc -g -c myprog.c utils.c -pg 126 cc -o myprog myprog.o utils.o -pg 127 128 The '-pg' option also works with a command that both compiles and 129links: 130 131 cc -o myprog myprog.c utils.c -g -pg 132 133 Note: The '-pg' option must be part of your compilation options as 134well as your link options. If it is not then no call-graph data will be 135gathered and when you run 'gprof' you will get an error message like 136this: 137 138 gprof: gmon.out file is missing call-graph data 139 140 If you add the '-Q' switch to suppress the printing of the call graph 141data you will still be able to see the time samples: 142 143 Flat profile: 144 145 Each sample counts as 0.01 seconds. 146 % cumulative self self total 147 time seconds seconds calls Ts/call Ts/call name 148 44.12 0.07 0.07 zazLoop 149 35.29 0.14 0.06 main 150 20.59 0.17 0.04 bazMillion 151 152 If you run the linker 'ld' directly instead of through a compiler 153such as 'cc', you may have to specify a profiling startup file 'gcrt0.o' 154as the first input file instead of the usual startup file 'crt0.o'. In 155addition, you would probably want to specify the profiling C library, 156'libc_p.a', by writing '-lc_p' instead of the usual '-lc'. This is not 157absolutely necessary, but doing this gives you number-of-calls 158information for standard library functions such as 'read' and 'open'. 159For example: 160 161 ld -o myprog /lib/gcrt0.o myprog.o utils.o -lc_p 162 163 If you are running the program on a system which supports shared 164libraries you may run into problems with the profiling support code in a 165shared library being called before that library has been fully 166initialised. This is usually detected by the program encountering a 167segmentation fault as soon as it is run. The solution is to link 168against a static version of the library containing the profiling support 169code, which for 'gcc' users can be done via the '-static' or 170'-static-libgcc' command-line option. For example: 171 172 gcc -g -pg -static-libgcc myprog.c utils.c -o myprog 173 174 If you compile only some of the modules of the program with '-pg', 175you can still profile the program, but you won't get complete 176information about the modules that were compiled without '-pg'. The 177only information you get for the functions in those modules is the total 178time spent in them; there is no record of how many times they were 179called, or from where. This will not affect the flat profile (except 180that the 'calls' field for the functions will be blank), but will 181greatly reduce the usefulness of the call graph. 182 183 If you wish to perform line-by-line profiling you should use the 184'gcov' tool instead of 'gprof'. See that tool's manual or info pages 185for more details of how to do this. 186 187 Note, older versions of 'gcc' produce line-by-line profiling 188information that works with 'gprof' rather than 'gcov' so there is still 189support for displaying this kind of information in 'gprof'. *Note 190Line-by-line Profiling: Line-by-line. 191 192 It also worth noting that 'gcc' implements a '-finstrument-functions' 193command-line option which will insert calls to special user supplied 194instrumentation routines at the entry and exit of every function in 195their program. This can be used to implement an alternative profiling 196scheme. 197 198 199File: gprof.info, Node: Executing, Next: Invoking, Prev: Compiling, Up: Top 200 2013 Executing the Program 202*********************** 203 204Once the program is compiled for profiling, you must run it in order to 205generate the information that 'gprof' needs. Simply run the program as 206usual, using the normal arguments, file names, etc. The program should 207run normally, producing the same output as usual. It will, however, run 208somewhat slower than normal because of the time spent collecting and 209writing the profile data. 210 211 The way you run the program--the arguments and input that you give 212it--may have a dramatic effect on what the profile information shows. 213The profile data will describe the parts of the program that were 214activated for the particular input you use. For example, if the first 215command you give to your program is to quit, the profile data will show 216the time used in initialization and in cleanup, but not much else. 217 218 Your program will write the profile data into a file called 219'gmon.out' just before exiting. If there is already a file called 220'gmon.out', its contents are overwritten. You can rename the file 221afterwards if you are concerned that it may be overwritten. If your 222system libc allows you may be able to write the profile data under a 223different name. Set the GMON_OUT_PREFIX environment variable; this name 224will be appended with the PID of the running program. 225 226 In order to write the 'gmon.out' file properly, your program must 227exit normally: by returning from 'main' or by calling 'exit'. Calling 228the low-level function '_exit' does not write the profile data, and 229neither does abnormal termination due to an unhandled signal. 230 231 The 'gmon.out' file is written in the program's _current working 232directory_ at the time it exits. This means that if your program calls 233'chdir', the 'gmon.out' file will be left in the last directory your 234program 'chdir''d to. If you don't have permission to write in this 235directory, the file is not written, and you will get an error message. 236 237 Older versions of the GNU profiling library may also write a file 238called 'bb.out'. This file, if present, contains an human-readable 239listing of the basic-block execution counts. Unfortunately, the 240appearance of a human-readable 'bb.out' means the basic-block counts 241didn't get written into 'gmon.out'. The Perl script 'bbconv.pl', 242included with the 'gprof' source distribution, will convert a 'bb.out' 243file into a format readable by 'gprof'. Invoke it like this: 244 245 bbconv.pl < bb.out > BH-DATA 246 247 This translates the information in 'bb.out' into a form that 'gprof' 248can understand. But you still need to tell 'gprof' about the existence 249of this translated information. To do that, include BB-DATA on the 250'gprof' command line, _along with 'gmon.out'_, like this: 251 252 gprof OPTIONS EXECUTABLE-FILE gmon.out BB-DATA [YET-MORE-PROFILE-DATA-FILES...] [> OUTFILE] 253 254 255File: gprof.info, Node: Invoking, Next: Output, Prev: Executing, Up: Top 256 2574 'gprof' Command Summary 258************************* 259 260After you have a profile data file 'gmon.out', you can run 'gprof' to 261interpret the information in it. The 'gprof' program prints a flat 262profile and a call graph on standard output. Typically you would 263redirect the output of 'gprof' into a file with '>'. 264 265 You run 'gprof' like this: 266 267 gprof OPTIONS [EXECUTABLE-FILE [PROFILE-DATA-FILES...]] [> OUTFILE] 268 269Here square-brackets indicate optional arguments. 270 271 If you omit the executable file name, the file 'a.out' is used. If 272you give no profile data file name, the file 'gmon.out' is used. If any 273file is not in the proper format, or if the profile data file does not 274appear to belong to the executable file, an error message is printed. 275 276 You can give more than one profile data file by entering all their 277names after the executable file name; then the statistics in all the 278data files are summed together. 279 280 The order of these options does not matter. 281 282* Menu: 283 284* Output Options:: Controlling 'gprof''s output style 285* Analysis Options:: Controlling how 'gprof' analyzes its data 286* Miscellaneous Options:: 287* Deprecated Options:: Options you no longer need to use, but which 288 have been retained for compatibility 289* Symspecs:: Specifying functions to include or exclude 290 291 292File: gprof.info, Node: Output Options, Next: Analysis Options, Up: Invoking 293 2944.1 Output Options 295================== 296 297These options specify which of several output formats 'gprof' should 298produce. 299 300 Many of these options take an optional "symspec" to specify functions 301to be included or excluded. These options can be specified multiple 302times, with different symspecs, to include or exclude sets of symbols. 303*Note Symspecs: Symspecs. 304 305 Specifying any of these options overrides the default ('-p -q'), 306which prints a flat profile and call graph analysis for all functions. 307 308'-A[SYMSPEC]' 309'--annotated-source[=SYMSPEC]' 310 The '-A' option causes 'gprof' to print annotated source code. If 311 SYMSPEC is specified, print output only for matching symbols. 312 *Note The Annotated Source Listing: Annotated Source. 313 314'-b' 315'--brief' 316 If the '-b' option is given, 'gprof' doesn't print the verbose 317 blurbs that try to explain the meaning of all of the fields in the 318 tables. This is useful if you intend to print out the output, or 319 are tired of seeing the blurbs. 320 321'-C[SYMSPEC]' 322'--exec-counts[=SYMSPEC]' 323 The '-C' option causes 'gprof' to print a tally of functions and 324 the number of times each was called. If SYMSPEC is specified, 325 print tally only for matching symbols. 326 327 If the profile data file contains basic-block count records, 328 specifying the '-l' option, along with '-C', will cause basic-block 329 execution counts to be tallied and displayed. 330 331'-i' 332'--file-info' 333 The '-i' option causes 'gprof' to display summary information about 334 the profile data file(s) and then exit. The number of histogram, 335 call graph, and basic-block count records is displayed. 336 337'-I DIRS' 338'--directory-path=DIRS' 339 The '-I' option specifies a list of search directories in which to 340 find source files. Environment variable GPROF_PATH can also be 341 used to convey this information. Used mostly for annotated source 342 output. 343 344'-J[SYMSPEC]' 345'--no-annotated-source[=SYMSPEC]' 346 The '-J' option causes 'gprof' not to print annotated source code. 347 If SYMSPEC is specified, 'gprof' prints annotated source, but 348 excludes matching symbols. 349 350'-L' 351'--print-path' 352 Normally, source filenames are printed with the path component 353 suppressed. The '-L' option causes 'gprof' to print the full 354 pathname of source filenames, which is determined from symbolic 355 debugging information in the image file and is relative to the 356 directory in which the compiler was invoked. 357 358'-p[SYMSPEC]' 359'--flat-profile[=SYMSPEC]' 360 The '-p' option causes 'gprof' to print a flat profile. If SYMSPEC 361 is specified, print flat profile only for matching symbols. *Note 362 The Flat Profile: Flat Profile. 363 364'-P[SYMSPEC]' 365'--no-flat-profile[=SYMSPEC]' 366 The '-P' option causes 'gprof' to suppress printing a flat profile. 367 If SYMSPEC is specified, 'gprof' prints a flat profile, but 368 excludes matching symbols. 369 370'-q[SYMSPEC]' 371'--graph[=SYMSPEC]' 372 The '-q' option causes 'gprof' to print the call graph analysis. 373 If SYMSPEC is specified, print call graph only for matching symbols 374 and their children. *Note The Call Graph: Call Graph. 375 376'-Q[SYMSPEC]' 377'--no-graph[=SYMSPEC]' 378 The '-Q' option causes 'gprof' to suppress printing the call graph. 379 If SYMSPEC is specified, 'gprof' prints a call graph, but excludes 380 matching symbols. 381 382'-t' 383'--table-length=NUM' 384 The '-t' option causes the NUM most active source lines in each 385 source file to be listed when source annotation is enabled. The 386 default is 10. 387 388'-y' 389'--separate-files' 390 This option affects annotated source output only. Normally, 391 'gprof' prints annotated source files to standard-output. If this 392 option is specified, annotated source for a file named 393 'path/FILENAME' is generated in the file 'FILENAME-ann'. If the 394 underlying file system would truncate 'FILENAME-ann' so that it 395 overwrites the original 'FILENAME', 'gprof' generates annotated 396 source in the file 'FILENAME.ann' instead (if the original file 397 name has an extension, that extension is _replaced_ with '.ann'). 398 399'-Z[SYMSPEC]' 400'--no-exec-counts[=SYMSPEC]' 401 The '-Z' option causes 'gprof' not to print a tally of functions 402 and the number of times each was called. If SYMSPEC is specified, 403 print tally, but exclude matching symbols. 404 405'-r' 406'--function-ordering' 407 The '--function-ordering' option causes 'gprof' to print a 408 suggested function ordering for the program based on profiling 409 data. This option suggests an ordering which may improve paging, 410 tlb and cache behavior for the program on systems which support 411 arbitrary ordering of functions in an executable. 412 413 The exact details of how to force the linker to place functions in 414 a particular order is system dependent and out of the scope of this 415 manual. 416 417'-R MAP_FILE' 418'--file-ordering MAP_FILE' 419 The '--file-ordering' option causes 'gprof' to print a suggested .o 420 link line ordering for the program based on profiling data. This 421 option suggests an ordering which may improve paging, tlb and cache 422 behavior for the program on systems which do not support arbitrary 423 ordering of functions in an executable. 424 425 Use of the '-a' argument is highly recommended with this option. 426 427 The MAP_FILE argument is a pathname to a file which provides 428 function name to object file mappings. The format of the file is 429 similar to the output of the program 'nm'. 430 431 c-parse.o:00000000 T yyparse 432 c-parse.o:00000004 C yyerrflag 433 c-lang.o:00000000 T maybe_objc_method_name 434 c-lang.o:00000000 T print_lang_statistics 435 c-lang.o:00000000 T recognize_objc_keyword 436 c-decl.o:00000000 T print_lang_identifier 437 c-decl.o:00000000 T print_lang_type 438 ... 439 440 441 To create a MAP_FILE with GNU 'nm', type a command like 'nm 442 --extern-only --defined-only -v --print-file-name program-name'. 443 444'-T' 445'--traditional' 446 The '-T' option causes 'gprof' to print its output in "traditional" 447 BSD style. 448 449'-w WIDTH' 450'--width=WIDTH' 451 Sets width of output lines to WIDTH. Currently only used when 452 printing the function index at the bottom of the call graph. 453 454'-x' 455'--all-lines' 456 This option affects annotated source output only. By default, only 457 the lines at the beginning of a basic-block are annotated. If this 458 option is specified, every line in a basic-block is annotated by 459 repeating the annotation for the first line. This behavior is 460 similar to 'tcov''s '-a'. 461 462'--demangle[=STYLE]' 463'--no-demangle' 464 These options control whether C++ symbol names should be demangled 465 when printing output. The default is to demangle symbols. The 466 '--no-demangle' option may be used to turn off demangling. 467 Different compilers have different mangling styles. The optional 468 demangling style argument can be used to choose an appropriate 469 demangling style for your compiler. 470 471 472File: gprof.info, Node: Analysis Options, Next: Miscellaneous Options, Prev: Output Options, Up: Invoking 473 4744.2 Analysis Options 475==================== 476 477'-a' 478'--no-static' 479 The '-a' option causes 'gprof' to suppress the printing of 480 statically declared (private) functions. (These are functions 481 whose names are not listed as global, and which are not visible 482 outside the file/function/block where they were defined.) Time 483 spent in these functions, calls to/from them, etc., will all be 484 attributed to the function that was loaded directly before it in 485 the executable file. This option affects both the flat profile and 486 the call graph. 487 488'-c' 489'--static-call-graph' 490 The '-c' option causes the call graph of the program to be 491 augmented by a heuristic which examines the text space of the 492 object file and identifies function calls in the binary machine 493 code. Since normal call graph records are only generated when 494 functions are entered, this option identifies children that could 495 have been called, but never were. Calls to functions that were not 496 compiled with profiling enabled are also identified, but only if 497 symbol table entries are present for them. Calls to dynamic 498 library routines are typically _not_ found by this option. Parents 499 or children identified via this heuristic are indicated in the call 500 graph with call counts of '0'. 501 502'-D' 503'--ignore-non-functions' 504 The '-D' option causes 'gprof' to ignore symbols which are not 505 known to be functions. This option will give more accurate profile 506 data on systems where it is supported (Solaris and HPUX for 507 example). 508 509'-k FROM/TO' 510 The '-k' option allows you to delete from the call graph any arcs 511 from symbols matching symspec FROM to those matching symspec TO. 512 513'-l' 514'--line' 515 The '-l' option enables line-by-line profiling, which causes 516 histogram hits to be charged to individual source code lines, 517 instead of functions. This feature only works with programs 518 compiled by older versions of the 'gcc' compiler. Newer versions 519 of 'gcc' are designed to work with the 'gcov' tool instead. 520 521 If the program was compiled with basic-block counting enabled, this 522 option will also identify how many times each line of code was 523 executed. While line-by-line profiling can help isolate where in a 524 large function a program is spending its time, it also 525 significantly increases the running time of 'gprof', and magnifies 526 statistical inaccuracies. *Note Statistical Sampling Error: 527 Sampling Error. 528 529'--inline-file-names' 530 This option causes 'gprof' to print the source file after each 531 symbol in both the flat profile and the call graph. The full path 532 to the file is printed if used with the '-L' option. 533 534'-m NUM' 535'--min-count=NUM' 536 This option affects execution count output only. Symbols that are 537 executed less than NUM times are suppressed. 538 539'-nSYMSPEC' 540'--time=SYMSPEC' 541 The '-n' option causes 'gprof', in its call graph analysis, to only 542 propagate times for symbols matching SYMSPEC. 543 544'-NSYMSPEC' 545'--no-time=SYMSPEC' 546 The '-n' option causes 'gprof', in its call graph analysis, not to 547 propagate times for symbols matching SYMSPEC. 548 549'-SFILENAME' 550'--external-symbol-table=FILENAME' 551 The '-S' option causes 'gprof' to read an external symbol table 552 file, such as '/proc/kallsyms', rather than read the symbol table 553 from the given object file (the default is 'a.out'). This is 554 useful for profiling kernel modules. 555 556'-z' 557'--display-unused-functions' 558 If you give the '-z' option, 'gprof' will mention all functions in 559 the flat profile, even those that were never called, and that had 560 no time spent in them. This is useful in conjunction with the '-c' 561 option for discovering which routines were never called. 562 563 564File: gprof.info, Node: Miscellaneous Options, Next: Deprecated Options, Prev: Analysis Options, Up: Invoking 565 5664.3 Miscellaneous Options 567========================= 568 569'-d[NUM]' 570'--debug[=NUM]' 571 The '-d NUM' option specifies debugging options. If NUM is not 572 specified, enable all debugging. *Note Debugging 'gprof': 573 Debugging. 574 575'-h' 576'--help' 577 The '-h' option prints command line usage. 578 579'-ONAME' 580'--file-format=NAME' 581 Selects the format of the profile data files. Recognized formats 582 are 'auto' (the default), 'bsd', '4.4bsd', 'magic', and 'prof' (not 583 yet supported). 584 585'-s' 586'--sum' 587 The '-s' option causes 'gprof' to summarize the information in the 588 profile data files it read in, and write out a profile data file 589 called 'gmon.sum', which contains all the information from the 590 profile data files that 'gprof' read in. The file 'gmon.sum' may 591 be one of the specified input files; the effect of this is to merge 592 the data in the other input files into 'gmon.sum'. 593 594 Eventually you can run 'gprof' again without '-s' to analyze the 595 cumulative data in the file 'gmon.sum'. 596 597'-v' 598'--version' 599 The '-v' flag causes 'gprof' to print the current version number, 600 and then exit. 601 602 603File: gprof.info, Node: Deprecated Options, Next: Symspecs, Prev: Miscellaneous Options, Up: Invoking 604 6054.4 Deprecated Options 606====================== 607 608These options have been replaced with newer versions that use symspecs. 609 610'-e FUNCTION_NAME' 611 The '-e FUNCTION' option tells 'gprof' to not print information 612 about the function FUNCTION_NAME (and its children...) in the call 613 graph. The function will still be listed as a child of any 614 functions that call it, but its index number will be shown as '[not 615 printed]'. More than one '-e' option may be given; only one 616 FUNCTION_NAME may be indicated with each '-e' option. 617 618'-E FUNCTION_NAME' 619 The '-E FUNCTION' option works like the '-e' option, but time spent 620 in the function (and children who were not called from anywhere 621 else), will not be used to compute the percentages-of-time for the 622 call graph. More than one '-E' option may be given; only one 623 FUNCTION_NAME may be indicated with each '-E' option. 624 625'-f FUNCTION_NAME' 626 The '-f FUNCTION' option causes 'gprof' to limit the call graph to 627 the function FUNCTION_NAME and its children (and their 628 children...). More than one '-f' option may be given; only one 629 FUNCTION_NAME may be indicated with each '-f' option. 630 631'-F FUNCTION_NAME' 632 The '-F FUNCTION' option works like the '-f' option, but only time 633 spent in the function and its children (and their children...) will 634 be used to determine total-time and percentages-of-time for the 635 call graph. More than one '-F' option may be given; only one 636 FUNCTION_NAME may be indicated with each '-F' option. The '-F' 637 option overrides the '-E' option. 638 639 Note that only one function can be specified with each '-e', '-E', 640'-f' or '-F' option. To specify more than one function, use multiple 641options. For example, this command: 642 643 gprof -e boring -f foo -f bar myprogram > gprof.output 644 645lists in the call graph all functions that were reached from either 646'foo' or 'bar' and were not reachable from 'boring'. 647 648 649File: gprof.info, Node: Symspecs, Prev: Deprecated Options, Up: Invoking 650 6514.5 Symspecs 652============ 653 654Many of the output options allow functions to be included or excluded 655using "symspecs" (symbol specifications), which observe the following 656syntax: 657 658 filename_containing_a_dot 659 | funcname_not_containing_a_dot 660 | linenumber 661 | ( [ any_filename ] `:' ( any_funcname | linenumber ) ) 662 663 Here are some sample symspecs: 664 665'main.c' 666 Selects everything in file 'main.c'--the dot in the string tells 667 'gprof' to interpret the string as a filename, rather than as a 668 function name. To select a file whose name does not contain a dot, 669 a trailing colon should be specified. For example, 'odd:' is 670 interpreted as the file named 'odd'. 671 672'main' 673 Selects all functions named 'main'. 674 675 Note that there may be multiple instances of the same function name 676 because some of the definitions may be local (i.e., static). 677 Unless a function name is unique in a program, you must use the 678 colon notation explained below to specify a function from a 679 specific source file. 680 681 Sometimes, function names contain dots. In such cases, it is 682 necessary to add a leading colon to the name. For example, ':.mul' 683 selects function '.mul'. 684 685 In some object file formats, symbols have a leading underscore. 686 'gprof' will normally not print these underscores. When you name a 687 symbol in a symspec, you should type it exactly as 'gprof' prints 688 it in its output. For example, if the compiler produces a symbol 689 '_main' from your 'main' function, 'gprof' still prints it as 690 'main' in its output, so you should use 'main' in symspecs. 691 692'main.c:main' 693 Selects function 'main' in file 'main.c'. 694 695'main.c:134' 696 Selects line 134 in file 'main.c'. 697 698 699File: gprof.info, Node: Output, Next: Inaccuracy, Prev: Invoking, Up: Top 700 7015 Interpreting 'gprof''s Output 702******************************* 703 704'gprof' can produce several different output styles, the most important 705of which are described below. The simplest output styles (file 706information, execution count, and function and file ordering) are not 707described here, but are documented with the respective options that 708trigger them. *Note Output Options: Output Options. 709 710* Menu: 711 712* Flat Profile:: The flat profile shows how much time was spent 713 executing directly in each function. 714* Call Graph:: The call graph shows which functions called which 715 others, and how much time each function used 716 when its subroutine calls are included. 717* Line-by-line:: 'gprof' can analyze individual source code lines 718* Annotated Source:: The annotated source listing displays source code 719 labeled with execution counts 720 721 722File: gprof.info, Node: Flat Profile, Next: Call Graph, Up: Output 723 7245.1 The Flat Profile 725==================== 726 727The "flat profile" shows the total amount of time your program spent 728executing each function. Unless the '-z' option is given, functions 729with no apparent time spent in them, and no apparent calls to them, are 730not mentioned. Note that if a function was not compiled for profiling, 731and didn't run long enough to show up on the program counter histogram, 732it will be indistinguishable from a function that was never called. 733 734 This is part of a flat profile for a small program: 735 736 Flat profile: 737 738 Each sample counts as 0.01 seconds. 739 % cumulative self self total 740 time seconds seconds calls ms/call ms/call name 741 33.34 0.02 0.02 7208 0.00 0.00 open 742 16.67 0.03 0.01 244 0.04 0.12 offtime 743 16.67 0.04 0.01 8 1.25 1.25 memccpy 744 16.67 0.05 0.01 7 1.43 1.43 write 745 16.67 0.06 0.01 mcount 746 0.00 0.06 0.00 236 0.00 0.00 tzset 747 0.00 0.06 0.00 192 0.00 0.00 tolower 748 0.00 0.06 0.00 47 0.00 0.00 strlen 749 0.00 0.06 0.00 45 0.00 0.00 strchr 750 0.00 0.06 0.00 1 0.00 50.00 main 751 0.00 0.06 0.00 1 0.00 0.00 memcpy 752 0.00 0.06 0.00 1 0.00 10.11 print 753 0.00 0.06 0.00 1 0.00 0.00 profil 754 0.00 0.06 0.00 1 0.00 50.00 report 755 ... 756 757The functions are sorted first by decreasing run-time spent in them, 758then by decreasing number of calls, then alphabetically by name. The 759functions 'mcount' and 'profil' are part of the profiling apparatus and 760appear in every flat profile; their time gives a measure of the amount 761of overhead due to profiling. 762 763 Just before the column headers, a statement appears indicating how 764much time each sample counted as. This "sampling period" estimates the 765margin of error in each of the time figures. A time figure that is not 766much larger than this is not reliable. In this example, each sample 767counted as 0.01 seconds, suggesting a 100 Hz sampling rate. The 768program's total execution time was 0.06 seconds, as indicated by the 769'cumulative seconds' field. Since each sample counted for 0.01 seconds, 770this means only six samples were taken during the run. Two of the 771samples occurred while the program was in the 'open' function, as 772indicated by the 'self seconds' field. Each of the other four samples 773occurred one each in 'offtime', 'memccpy', 'write', and 'mcount'. Since 774only six samples were taken, none of these values can be regarded as 775particularly reliable. In another run, the 'self seconds' field for 776'mcount' might well be '0.00' or '0.02'. *Note Statistical Sampling 777Error: Sampling Error, for a complete discussion. 778 779 The remaining functions in the listing (those whose 'self seconds' 780field is '0.00') didn't appear in the histogram samples at all. 781However, the call graph indicated that they were called, so therefore 782they are listed, sorted in decreasing order by the 'calls' field. 783Clearly some time was spent executing these functions, but the paucity 784of histogram samples prevents any determination of how much time each 785took. 786 787 Here is what the fields in each line mean: 788 789'% time' 790 This is the percentage of the total execution time your program 791 spent in this function. These should all add up to 100%. 792 793'cumulative seconds' 794 This is the cumulative total number of seconds the computer spent 795 executing this functions, plus the time spent in all the functions 796 above this one in this table. 797 798'self seconds' 799 This is the number of seconds accounted for by this function alone. 800 The flat profile listing is sorted first by this number. 801 802'calls' 803 This is the total number of times the function was called. If the 804 function was never called, or the number of times it was called 805 cannot be determined (probably because the function was not 806 compiled with profiling enabled), the "calls" field is blank. 807 808'self ms/call' 809 This represents the average number of milliseconds spent in this 810 function per call, if this function is profiled. Otherwise, this 811 field is blank for this function. 812 813'total ms/call' 814 This represents the average number of milliseconds spent in this 815 function and its descendants per call, if this function is 816 profiled. Otherwise, this field is blank for this function. This 817 is the only field in the flat profile that uses call graph 818 analysis. 819 820'name' 821 This is the name of the function. The flat profile is sorted by 822 this field alphabetically after the "self seconds" and "calls" 823 fields are sorted. 824 825 826File: gprof.info, Node: Call Graph, Next: Line-by-line, Prev: Flat Profile, Up: Output 827 8285.2 The Call Graph 829================== 830 831The "call graph" shows how much time was spent in each function and its 832children. From this information, you can find functions that, while 833they themselves may not have used much time, called other functions that 834did use unusual amounts of time. 835 836 Here is a sample call from a small program. This call came from the 837same 'gprof' run as the flat profile example in the previous section. 838 839 granularity: each sample hit covers 2 byte(s) for 20.00% of 0.05 seconds 840 841 index % time self children called name 842 <spontaneous> 843 [1] 100.0 0.00 0.05 start [1] 844 0.00 0.05 1/1 main [2] 845 0.00 0.00 1/2 on_exit [28] 846 0.00 0.00 1/1 exit [59] 847 ----------------------------------------------- 848 0.00 0.05 1/1 start [1] 849 [2] 100.0 0.00 0.05 1 main [2] 850 0.00 0.05 1/1 report [3] 851 ----------------------------------------------- 852 0.00 0.05 1/1 main [2] 853 [3] 100.0 0.00 0.05 1 report [3] 854 0.00 0.03 8/8 timelocal [6] 855 0.00 0.01 1/1 print [9] 856 0.00 0.01 9/9 fgets [12] 857 0.00 0.00 12/34 strncmp <cycle 1> [40] 858 0.00 0.00 8/8 lookup [20] 859 0.00 0.00 1/1 fopen [21] 860 0.00 0.00 8/8 chewtime [24] 861 0.00 0.00 8/16 skipspace [44] 862 ----------------------------------------------- 863 [4] 59.8 0.01 0.02 8+472 <cycle 2 as a whole> [4] 864 0.01 0.02 244+260 offtime <cycle 2> [7] 865 0.00 0.00 236+1 tzset <cycle 2> [26] 866 ----------------------------------------------- 867 868 The lines full of dashes divide this table into "entries", one for 869each function. Each entry has one or more lines. 870 871 In each entry, the primary line is the one that starts with an index 872number in square brackets. The end of this line says which function the 873entry is for. The preceding lines in the entry describe the callers of 874this function and the following lines describe its subroutines (also 875called "children" when we speak of the call graph). 876 877 The entries are sorted by time spent in the function and its 878subroutines. 879 880 The internal profiling function 'mcount' (*note The Flat Profile: 881Flat Profile.) is never mentioned in the call graph. 882 883* Menu: 884 885* Primary:: Details of the primary line's contents. 886* Callers:: Details of caller-lines' contents. 887* Subroutines:: Details of subroutine-lines' contents. 888* Cycles:: When there are cycles of recursion, 889 such as 'a' calls 'b' calls 'a'... 890 891 892File: gprof.info, Node: Primary, Next: Callers, Up: Call Graph 893 8945.2.1 The Primary Line 895---------------------- 896 897The "primary line" in a call graph entry is the line that describes the 898function which the entry is about and gives the overall statistics for 899this function. 900 901 For reference, we repeat the primary line from the entry for function 902'report' in our main example, together with the heading line that shows 903the names of the fields: 904 905 index % time self children called name 906 ... 907 [3] 100.0 0.00 0.05 1 report [3] 908 909 Here is what the fields in the primary line mean: 910 911'index' 912 Entries are numbered with consecutive integers. Each function 913 therefore has an index number, which appears at the beginning of 914 its primary line. 915 916 Each cross-reference to a function, as a caller or subroutine of 917 another, gives its index number as well as its name. The index 918 number guides you if you wish to look for the entry for that 919 function. 920 921'% time' 922 This is the percentage of the total time that was spent in this 923 function, including time spent in subroutines called from this 924 function. 925 926 The time spent in this function is counted again for the callers of 927 this function. Therefore, adding up these percentages is 928 meaningless. 929 930'self' 931 This is the total amount of time spent in this function. This 932 should be identical to the number printed in the 'seconds' field 933 for this function in the flat profile. 934 935'children' 936 This is the total amount of time spent in the subroutine calls made 937 by this function. This should be equal to the sum of all the 938 'self' and 'children' entries of the children listed directly below 939 this function. 940 941'called' 942 This is the number of times the function was called. 943 944 If the function called itself recursively, there are two numbers, 945 separated by a '+'. The first number counts non-recursive calls, 946 and the second counts recursive calls. 947 948 In the example above, the function 'report' was called once from 949 'main'. 950 951'name' 952 This is the name of the current function. The index number is 953 repeated after it. 954 955 If the function is part of a cycle of recursion, the cycle number 956 is printed between the function's name and the index number (*note 957 How Mutually Recursive Functions Are Described: Cycles.). For 958 example, if function 'gnurr' is part of cycle number one, and has 959 index number twelve, its primary line would be end like this: 960 961 gnurr <cycle 1> [12] 962 963 964File: gprof.info, Node: Callers, Next: Subroutines, Prev: Primary, Up: Call Graph 965 9665.2.2 Lines for a Function's Callers 967------------------------------------ 968 969A function's entry has a line for each function it was called by. These 970lines' fields correspond to the fields of the primary line, but their 971meanings are different because of the difference in context. 972 973 For reference, we repeat two lines from the entry for the function 974'report', the primary line and one caller-line preceding it, together 975with the heading line that shows the names of the fields: 976 977 index % time self children called name 978 ... 979 0.00 0.05 1/1 main [2] 980 [3] 100.0 0.00 0.05 1 report [3] 981 982 Here are the meanings of the fields in the caller-line for 'report' 983called from 'main': 984 985'self' 986 An estimate of the amount of time spent in 'report' itself when it 987 was called from 'main'. 988 989'children' 990 An estimate of the amount of time spent in subroutines of 'report' 991 when 'report' was called from 'main'. 992 993 The sum of the 'self' and 'children' fields is an estimate of the 994 amount of time spent within calls to 'report' from 'main'. 995 996'called' 997 Two numbers: the number of times 'report' was called from 'main', 998 followed by the total number of non-recursive calls to 'report' 999 from all its callers. 1000 1001'name and index number' 1002 The name of the caller of 'report' to which this line applies, 1003 followed by the caller's index number. 1004 1005 Not all functions have entries in the call graph; some options to 1006 'gprof' request the omission of certain functions. When a caller 1007 has no entry of its own, it still has caller-lines in the entries 1008 of the functions it calls. 1009 1010 If the caller is part of a recursion cycle, the cycle number is 1011 printed between the name and the index number. 1012 1013 If the identity of the callers of a function cannot be determined, a 1014dummy caller-line is printed which has '<spontaneous>' as the "caller's 1015name" and all other fields blank. This can happen for signal handlers. 1016 1017 1018File: gprof.info, Node: Subroutines, Next: Cycles, Prev: Callers, Up: Call Graph 1019 10205.2.3 Lines for a Function's Subroutines 1021---------------------------------------- 1022 1023A function's entry has a line for each of its subroutines--in other 1024words, a line for each other function that it called. These lines' 1025fields correspond to the fields of the primary line, but their meanings 1026are different because of the difference in context. 1027 1028 For reference, we repeat two lines from the entry for the function 1029'main', the primary line and a line for a subroutine, together with the 1030heading line that shows the names of the fields: 1031 1032 index % time self children called name 1033 ... 1034 [2] 100.0 0.00 0.05 1 main [2] 1035 0.00 0.05 1/1 report [3] 1036 1037 Here are the meanings of the fields in the subroutine-line for 'main' 1038calling 'report': 1039 1040'self' 1041 An estimate of the amount of time spent directly within 'report' 1042 when 'report' was called from 'main'. 1043 1044'children' 1045 An estimate of the amount of time spent in subroutines of 'report' 1046 when 'report' was called from 'main'. 1047 1048 The sum of the 'self' and 'children' fields is an estimate of the 1049 total time spent in calls to 'report' from 'main'. 1050 1051'called' 1052 Two numbers, the number of calls to 'report' from 'main' followed 1053 by the total number of non-recursive calls to 'report'. This ratio 1054 is used to determine how much of 'report''s 'self' and 'children' 1055 time gets credited to 'main'. *Note Estimating 'children' Times: 1056 Assumptions. 1057 1058'name' 1059 The name of the subroutine of 'main' to which this line applies, 1060 followed by the subroutine's index number. 1061 1062 If the caller is part of a recursion cycle, the cycle number is 1063 printed between the name and the index number. 1064 1065 1066File: gprof.info, Node: Cycles, Prev: Subroutines, Up: Call Graph 1067 10685.2.4 How Mutually Recursive Functions Are Described 1069---------------------------------------------------- 1070 1071The graph may be complicated by the presence of "cycles of recursion" in 1072the call graph. A cycle exists if a function calls another function 1073that (directly or indirectly) calls (or appears to call) the original 1074function. For example: if 'a' calls 'b', and 'b' calls 'a', then 'a' 1075and 'b' form a cycle. 1076 1077 Whenever there are call paths both ways between a pair of functions, 1078they belong to the same cycle. If 'a' and 'b' call each other and 'b' 1079and 'c' call each other, all three make one cycle. Note that even if 1080'b' only calls 'a' if it was not called from 'a', 'gprof' cannot 1081determine this, so 'a' and 'b' are still considered a cycle. 1082 1083 The cycles are numbered with consecutive integers. When a function 1084belongs to a cycle, each time the function name appears in the call 1085graph it is followed by '<cycle NUMBER>'. 1086 1087 The reason cycles matter is that they make the time values in the 1088call graph paradoxical. The "time spent in children" of 'a' should 1089include the time spent in its subroutine 'b' and in 'b''s 1090subroutines--but one of 'b''s subroutines is 'a'! How much of 'a''s 1091time should be included in the children of 'a', when 'a' is indirectly 1092recursive? 1093 1094 The way 'gprof' resolves this paradox is by creating a single entry 1095for the cycle as a whole. The primary line of this entry describes the 1096total time spent directly in the functions of the cycle. The 1097"subroutines" of the cycle are the individual functions of the cycle, 1098and all other functions that were called directly by them. The 1099"callers" of the cycle are the functions, outside the cycle, that called 1100functions in the cycle. 1101 1102 Here is an example portion of a call graph which shows a cycle 1103containing functions 'a' and 'b'. The cycle was entered by a call to 1104'a' from 'main'; both 'a' and 'b' called 'c'. 1105 1106 index % time self children called name 1107 ---------------------------------------- 1108 1.77 0 1/1 main [2] 1109 [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3] 1110 1.02 0 3 b <cycle 1> [4] 1111 0.75 0 2 a <cycle 1> [5] 1112 ---------------------------------------- 1113 3 a <cycle 1> [5] 1114 [4] 52.85 1.02 0 0 b <cycle 1> [4] 1115 2 a <cycle 1> [5] 1116 0 0 3/6 c [6] 1117 ---------------------------------------- 1118 1.77 0 1/1 main [2] 1119 2 b <cycle 1> [4] 1120 [5] 38.86 0.75 0 1 a <cycle 1> [5] 1121 3 b <cycle 1> [4] 1122 0 0 3/6 c [6] 1123 ---------------------------------------- 1124 1125(The entire call graph for this program contains in addition an entry 1126for 'main', which calls 'a', and an entry for 'c', with callers 'a' and 1127'b'.) 1128 1129 index % time self children called name 1130 <spontaneous> 1131 [1] 100.00 0 1.93 0 start [1] 1132 0.16 1.77 1/1 main [2] 1133 ---------------------------------------- 1134 0.16 1.77 1/1 start [1] 1135 [2] 100.00 0.16 1.77 1 main [2] 1136 1.77 0 1/1 a <cycle 1> [5] 1137 ---------------------------------------- 1138 1.77 0 1/1 main [2] 1139 [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3] 1140 1.02 0 3 b <cycle 1> [4] 1141 0.75 0 2 a <cycle 1> [5] 1142 0 0 6/6 c [6] 1143 ---------------------------------------- 1144 3 a <cycle 1> [5] 1145 [4] 52.85 1.02 0 0 b <cycle 1> [4] 1146 2 a <cycle 1> [5] 1147 0 0 3/6 c [6] 1148 ---------------------------------------- 1149 1.77 0 1/1 main [2] 1150 2 b <cycle 1> [4] 1151 [5] 38.86 0.75 0 1 a <cycle 1> [5] 1152 3 b <cycle 1> [4] 1153 0 0 3/6 c [6] 1154 ---------------------------------------- 1155 0 0 3/6 b <cycle 1> [4] 1156 0 0 3/6 a <cycle 1> [5] 1157 [6] 0.00 0 0 6 c [6] 1158 ---------------------------------------- 1159 1160 The 'self' field of the cycle's primary line is the total time spent 1161in all the functions of the cycle. It equals the sum of the 'self' 1162fields for the individual functions in the cycle, found in the entry in 1163the subroutine lines for these functions. 1164 1165 The 'children' fields of the cycle's primary line and subroutine 1166lines count only subroutines outside the cycle. Even though 'a' calls 1167'b', the time spent in those calls to 'b' is not counted in 'a''s 1168'children' time. Thus, we do not encounter the problem of what to do 1169when the time in those calls to 'b' includes indirect recursive calls 1170back to 'a'. 1171 1172 The 'children' field of a caller-line in the cycle's entry estimates 1173the amount of time spent _in the whole cycle_, and its other 1174subroutines, on the times when that caller called a function in the 1175cycle. 1176 1177 The 'called' field in the primary line for the cycle has two numbers: 1178first, the number of times functions in the cycle were called by 1179functions outside the cycle; second, the number of times they were 1180called by functions in the cycle (including times when a function in the 1181cycle calls itself). This is a generalization of the usual split into 1182non-recursive and recursive calls. 1183 1184 The 'called' field of a subroutine-line for a cycle member in the 1185cycle's entry says how many time that function was called from functions 1186in the cycle. The total of all these is the second number in the 1187primary line's 'called' field. 1188 1189 In the individual entry for a function in a cycle, the other 1190functions in the same cycle can appear as subroutines and as callers. 1191These lines show how many times each function in the cycle called or was 1192called from each other function in the cycle. The 'self' and 'children' 1193fields in these lines are blank because of the difficulty of defining 1194meanings for them when recursion is going on. 1195 1196 1197File: gprof.info, Node: Line-by-line, Next: Annotated Source, Prev: Call Graph, Up: Output 1198 11995.3 Line-by-line Profiling 1200========================== 1201 1202'gprof''s '-l' option causes the program to perform "line-by-line" 1203profiling. In this mode, histogram samples are assigned not to 1204functions, but to individual lines of source code. This only works with 1205programs compiled with older versions of the 'gcc' compiler. Newer 1206versions of 'gcc' use a different program - 'gcov' - to display 1207line-by-line profiling information. 1208 1209 With the older versions of 'gcc' the program usually has to be 1210compiled with a '-g' option, in addition to '-pg', in order to generate 1211debugging symbols for tracking source code lines. Note, in much older 1212versions of 'gcc' the program had to be compiled with the '-a' 1213command-line option as well. 1214 1215 The flat profile is the most useful output table in line-by-line 1216mode. The call graph isn't as useful as normal, since the current 1217version of 'gprof' does not propagate call graph arcs from source code 1218lines to the enclosing function. The call graph does, however, show 1219each line of code that called each function, along with a count. 1220 1221 Here is a section of 'gprof''s output, without line-by-line 1222profiling. Note that 'ct_init' accounted for four histogram hits, and 122313327 calls to 'init_block'. 1224 1225 Flat profile: 1226 1227 Each sample counts as 0.01 seconds. 1228 % cumulative self self total 1229 time seconds seconds calls us/call us/call name 1230 30.77 0.13 0.04 6335 6.31 6.31 ct_init 1231 1232 1233 Call graph (explanation follows) 1234 1235 1236 granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds 1237 1238 index % time self children called name 1239 1240 0.00 0.00 1/13496 name_too_long 1241 0.00 0.00 40/13496 deflate 1242 0.00 0.00 128/13496 deflate_fast 1243 0.00 0.00 13327/13496 ct_init 1244 [7] 0.0 0.00 0.00 13496 init_block 1245 1246 1247 Now let's look at some of 'gprof''s output from the same program run, 1248this time with line-by-line profiling enabled. Note that 'ct_init''s 1249four histogram hits are broken down into four lines of source code--one 1250hit occurred on each of lines 349, 351, 382 and 385. In the call graph, 1251note how 'ct_init''s 13327 calls to 'init_block' are broken down into 1252one call from line 396, 3071 calls from line 384, 3730 calls from line 1253385, and 6525 calls from 387. 1254 1255 Flat profile: 1256 1257 Each sample counts as 0.01 seconds. 1258 % cumulative self 1259 time seconds seconds calls name 1260 7.69 0.10 0.01 ct_init (trees.c:349) 1261 7.69 0.11 0.01 ct_init (trees.c:351) 1262 7.69 0.12 0.01 ct_init (trees.c:382) 1263 7.69 0.13 0.01 ct_init (trees.c:385) 1264 1265 1266 Call graph (explanation follows) 1267 1268 1269 granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds 1270 1271 % time self children called name 1272 1273 0.00 0.00 1/13496 name_too_long (gzip.c:1440) 1274 0.00 0.00 1/13496 deflate (deflate.c:763) 1275 0.00 0.00 1/13496 ct_init (trees.c:396) 1276 0.00 0.00 2/13496 deflate (deflate.c:727) 1277 0.00 0.00 4/13496 deflate (deflate.c:686) 1278 0.00 0.00 5/13496 deflate (deflate.c:675) 1279 0.00 0.00 12/13496 deflate (deflate.c:679) 1280 0.00 0.00 16/13496 deflate (deflate.c:730) 1281 0.00 0.00 128/13496 deflate_fast (deflate.c:654) 1282 0.00 0.00 3071/13496 ct_init (trees.c:384) 1283 0.00 0.00 3730/13496 ct_init (trees.c:385) 1284 0.00 0.00 6525/13496 ct_init (trees.c:387) 1285 [6] 0.0 0.00 0.00 13496 init_block (trees.c:408) 1286 1287 1288 1289File: gprof.info, Node: Annotated Source, Prev: Line-by-line, Up: Output 1290 12915.4 The Annotated Source Listing 1292================================ 1293 1294'gprof''s '-A' option triggers an annotated source listing, which lists 1295the program's source code, each function labeled with the number of 1296times it was called. You may also need to specify the '-I' option, if 1297'gprof' can't find the source code files. 1298 1299 With older versions of 'gcc' compiling with 'gcc ... -g -pg -a' 1300augments your program with basic-block counting code, in addition to 1301function counting code. This enables 'gprof' to determine how many 1302times each line of code was executed. With newer versions of 'gcc' 1303support for displaying basic-block counts is provided by the 'gcov' 1304program. 1305 1306 For example, consider the following function, taken from gzip, with 1307line numbers added: 1308 1309 1 ulg updcrc(s, n) 1310 2 uch *s; 1311 3 unsigned n; 1312 4 { 1313 5 register ulg c; 1314 6 1315 7 static ulg crc = (ulg)0xffffffffL; 1316 8 1317 9 if (s == NULL) { 1318 10 c = 0xffffffffL; 1319 11 } else { 1320 12 c = crc; 1321 13 if (n) do { 1322 14 c = crc_32_tab[...]; 1323 15 } while (--n); 1324 16 } 1325 17 crc = c; 1326 18 return c ^ 0xffffffffL; 1327 19 } 1328 1329 1330 'updcrc' has at least five basic-blocks. One is the function itself. 1331The 'if' statement on line 9 generates two more basic-blocks, one for 1332each branch of the 'if'. A fourth basic-block results from the 'if' on 1333line 13, and the contents of the 'do' loop form the fifth basic-block. 1334The compiler may also generate additional basic-blocks to handle various 1335special cases. 1336 1337 A program augmented for basic-block counting can be analyzed with 1338'gprof -l -A'. The '-x' option is also helpful, to ensure that each 1339line of code is labeled at least once. Here is 'updcrc''s annotated 1340source listing for a sample 'gzip' run: 1341 1342 ulg updcrc(s, n) 1343 uch *s; 1344 unsigned n; 1345 2 ->{ 1346 register ulg c; 1347 1348 static ulg crc = (ulg)0xffffffffL; 1349 1350 2 -> if (s == NULL) { 1351 1 -> c = 0xffffffffL; 1352 1 -> } else { 1353 1 -> c = crc; 1354 1 -> if (n) do { 1355 26312 -> c = crc_32_tab[...]; 1356 26312,1,26311 -> } while (--n); 1357 } 1358 2 -> crc = c; 1359 2 -> return c ^ 0xffffffffL; 1360 2 ->} 1361 1362 In this example, the function was called twice, passing once through 1363each branch of the 'if' statement. The body of the 'do' loop was 1364executed a total of 26312 times. Note how the 'while' statement is 1365annotated. It began execution 26312 times, once for each iteration 1366through the loop. One of those times (the last time) it exited, while 1367it branched back to the beginning of the loop 26311 times. 1368 1369 1370File: gprof.info, Node: Inaccuracy, Next: How do I?, Prev: Output, Up: Top 1371 13726 Inaccuracy of 'gprof' Output 1373****************************** 1374 1375* Menu: 1376 1377* Sampling Error:: Statistical margins of error 1378* Assumptions:: Estimating children times 1379 1380 1381File: gprof.info, Node: Sampling Error, Next: Assumptions, Up: Inaccuracy 1382 13836.1 Statistical Sampling Error 1384============================== 1385 1386The run-time figures that 'gprof' gives you are based on a sampling 1387process, so they are subject to statistical inaccuracy. If a function 1388runs only a small amount of time, so that on the average the sampling 1389process ought to catch that function in the act only once, there is a 1390pretty good chance it will actually find that function zero times, or 1391twice. 1392 1393 By contrast, the number-of-calls and basic-block figures are derived 1394by counting, not sampling. They are completely accurate and will not 1395vary from run to run if your program is deterministic and single 1396threaded. In multi-threaded applications, or single threaded 1397applications that link with multi-threaded libraries, the counts are 1398only deterministic if the counting function is thread-safe. (Note: 1399beware that the mcount counting function in glibc is _not_ thread-safe). 1400*Note Implementation of Profiling: Implementation. 1401 1402 The "sampling period" that is printed at the beginning of the flat 1403profile says how often samples are taken. The rule of thumb is that a 1404run-time figure is accurate if it is considerably bigger than the 1405sampling period. 1406 1407 The actual amount of error can be predicted. For N samples, the 1408_expected_ error is the square-root of N. For example, if the sampling 1409period is 0.01 seconds and 'foo''s run-time is 1 second, N is 100 1410samples (1 second/0.01 seconds), sqrt(N) is 10 samples, so the expected 1411error in 'foo''s run-time is 0.1 seconds (10*0.01 seconds), or ten 1412percent of the observed value. Again, if the sampling period is 0.01 1413seconds and 'bar''s run-time is 100 seconds, N is 10000 samples, sqrt(N) 1414is 100 samples, so the expected error in 'bar''s run-time is 1 second, 1415or one percent of the observed value. It is likely to vary this much 1416_on the average_ from one profiling run to the next. (_Sometimes_ it 1417will vary more.) 1418 1419 This does not mean that a small run-time figure is devoid of 1420information. If the program's _total_ run-time is large, a small 1421run-time for one function does tell you that that function used an 1422insignificant fraction of the whole program's time. Usually this means 1423it is not worth optimizing. 1424 1425 One way to get more accuracy is to give your program more (but 1426similar) input data so it will take longer. Another way is to combine 1427the data from several runs, using the '-s' option of 'gprof'. Here is 1428how: 1429 1430 1. Run your program once. 1431 1432 2. Issue the command 'mv gmon.out gmon.sum'. 1433 1434 3. Run your program again, the same as before. 1435 1436 4. Merge the new data in 'gmon.out' into 'gmon.sum' with this command: 1437 1438 gprof -s EXECUTABLE-FILE gmon.out gmon.sum 1439 1440 5. Repeat the last two steps as often as you wish. 1441 1442 6. Analyze the cumulative data using this command: 1443 1444 gprof EXECUTABLE-FILE gmon.sum > OUTPUT-FILE 1445 1446 1447File: gprof.info, Node: Assumptions, Prev: Sampling Error, Up: Inaccuracy 1448 14496.2 Estimating 'children' Times 1450=============================== 1451 1452Some of the figures in the call graph are estimates--for example, the 1453'children' time values and all the time figures in caller and subroutine 1454lines. 1455 1456 There is no direct information about these measurements in the 1457profile data itself. Instead, 'gprof' estimates them by making an 1458assumption about your program that might or might not be true. 1459 1460 The assumption made is that the average time spent in each call to 1461any function 'foo' is not correlated with who called 'foo'. If 'foo' 1462used 5 seconds in all, and 2/5 of the calls to 'foo' came from 'a', then 1463'foo' contributes 2 seconds to 'a''s 'children' time, by assumption. 1464 1465 This assumption is usually true enough, but for some programs it is 1466far from true. Suppose that 'foo' returns very quickly when its 1467argument is zero; suppose that 'a' always passes zero as an argument, 1468while other callers of 'foo' pass other arguments. In this program, all 1469the time spent in 'foo' is in the calls from callers other than 'a'. 1470But 'gprof' has no way of knowing this; it will blindly and incorrectly 1471charge 2 seconds of time in 'foo' to the children of 'a'. 1472 1473 We hope some day to put more complete data into 'gmon.out', so that 1474this assumption is no longer needed, if we can figure out how. For the 1475novice, the estimated figures are usually more useful than misleading. 1476 1477 1478File: gprof.info, Node: How do I?, Next: Incompatibilities, Prev: Inaccuracy, Up: Top 1479 14807 Answers to Common Questions 1481***************************** 1482 1483How can I get more exact information about hot spots in my program? 1484 1485 Looking at the per-line call counts only tells part of the story. 1486 Because 'gprof' can only report call times and counts by function, 1487 the best way to get finer-grained information on where the program 1488 is spending its time is to re-factor large functions into sequences 1489 of calls to smaller ones. Beware however that this can introduce 1490 artificial hot spots since compiling with '-pg' adds a significant 1491 overhead to function calls. An alternative solution is to use a 1492 non-intrusive profiler, e.g. oprofile. 1493 1494How do I find which lines in my program were executed the most times? 1495 1496 Use the 'gcov' program. 1497 1498How do I find which lines in my program called a particular function? 1499 1500 Use 'gprof -l' and lookup the function in the call graph. The 1501 callers will be broken down by function and line number. 1502 1503How do I analyze a program that runs for less than a second? 1504 1505 Try using a shell script like this one: 1506 1507 for i in `seq 1 100`; do 1508 fastprog 1509 mv gmon.out gmon.out.$i 1510 done 1511 1512 gprof -s fastprog gmon.out.* 1513 1514 gprof fastprog gmon.sum 1515 1516 If your program is completely deterministic, all the call counts 1517 will be simple multiples of 100 (i.e., a function called once in 1518 each run will appear with a call count of 100). 1519 1520 1521File: gprof.info, Node: Incompatibilities, Next: Details, Prev: How do I?, Up: Top 1522 15238 Incompatibilities with Unix 'gprof' 1524************************************* 1525 1526GNU 'gprof' and Berkeley Unix 'gprof' use the same data file 'gmon.out', 1527and provide essentially the same information. But there are a few 1528differences. 1529 1530 * GNU 'gprof' uses a new, generalized file format with support for 1531 basic-block execution counts and non-realtime histograms. A magic 1532 cookie and version number allows 'gprof' to easily identify new 1533 style files. Old BSD-style files can still be read. *Note 1534 Profiling Data File Format: File Format. 1535 1536 * For a recursive function, Unix 'gprof' lists the function as a 1537 parent and as a child, with a 'calls' field that lists the number 1538 of recursive calls. GNU 'gprof' omits these lines and puts the 1539 number of recursive calls in the primary line. 1540 1541 * When a function is suppressed from the call graph with '-e', GNU 1542 'gprof' still lists it as a subroutine of functions that call it. 1543 1544 * GNU 'gprof' accepts the '-k' with its argument in the form 1545 'from/to', instead of 'from to'. 1546 1547 * In the annotated source listing, if there are multiple basic blocks 1548 on the same line, GNU 'gprof' prints all of their counts, separated 1549 by commas. 1550 1551 * The blurbs, field widths, and output formats are different. GNU 1552 'gprof' prints blurbs after the tables, so that you can see the 1553 tables without skipping the blurbs. 1554 1555 1556File: gprof.info, Node: Details, Next: GNU Free Documentation License, Prev: Incompatibilities, Up: Top 1557 15589 Details of Profiling 1559********************** 1560 1561* Menu: 1562 1563* Implementation:: How a program collects profiling information 1564* File Format:: Format of 'gmon.out' files 1565* Internals:: 'gprof''s internal operation 1566* Debugging:: Using 'gprof''s '-d' option 1567 1568 1569File: gprof.info, Node: Implementation, Next: File Format, Up: Details 1570 15719.1 Implementation of Profiling 1572=============================== 1573 1574Profiling works by changing how every function in your program is 1575compiled so that when it is called, it will stash away some information 1576about where it was called from. From this, the profiler can figure out 1577what function called it, and can count how many times it was called. 1578This change is made by the compiler when your program is compiled with 1579the '-pg' option, which causes every function to call 'mcount' (or 1580'_mcount', or '__mcount', depending on the OS and compiler) as one of 1581its first operations. 1582 1583 The 'mcount' routine, included in the profiling library, is 1584responsible for recording in an in-memory call graph table both its 1585parent routine (the child) and its parent's parent. This is typically 1586done by examining the stack frame to find both the address of the child, 1587and the return address in the original parent. Since this is a very 1588machine-dependent operation, 'mcount' itself is typically a short 1589assembly-language stub routine that extracts the required information, 1590and then calls '__mcount_internal' (a normal C function) with two 1591arguments--'frompc' and 'selfpc'. '__mcount_internal' is responsible 1592for maintaining the in-memory call graph, which records 'frompc', 1593'selfpc', and the number of times each of these call arcs was traversed. 1594 1595 GCC Version 2 provides a magical function 1596('__builtin_return_address'), which allows a generic 'mcount' function 1597to extract the required information from the stack frame. However, on 1598some architectures, most notably the SPARC, using this builtin can be 1599very computationally expensive, and an assembly language version of 1600'mcount' is used for performance reasons. 1601 1602 Number-of-calls information for library routines is collected by 1603using a special version of the C library. The programs in it are the 1604same as in the usual C library, but they were compiled with '-pg'. If 1605you link your program with 'gcc ... -pg', it automatically uses the 1606profiling version of the library. 1607 1608 Profiling also involves watching your program as it runs, and keeping 1609a histogram of where the program counter happens to be every now and 1610then. Typically the program counter is looked at around 100 times per 1611second of run time, but the exact frequency may vary from system to 1612system. 1613 1614 This is done is one of two ways. Most UNIX-like operating systems 1615provide a 'profil()' system call, which registers a memory array with 1616the kernel, along with a scale factor that determines how the program's 1617address space maps into the array. Typical scaling values cause every 2 1618to 8 bytes of address space to map into a single array slot. On every 1619tick of the system clock (assuming the profiled program is running), the 1620value of the program counter is examined and the corresponding slot in 1621the memory array is incremented. Since this is done in the kernel, 1622which had to interrupt the process anyway to handle the clock interrupt, 1623very little additional system overhead is required. 1624 1625 However, some operating systems, most notably Linux 2.0 (and 1626earlier), do not provide a 'profil()' system call. On such a system, 1627arrangements are made for the kernel to periodically deliver a signal to 1628the process (typically via 'setitimer()'), which then performs the same 1629operation of examining the program counter and incrementing a slot in 1630the memory array. Since this method requires a signal to be delivered 1631to user space every time a sample is taken, it uses considerably more 1632overhead than kernel-based profiling. Also, due to the added delay 1633required to deliver the signal, this method is less accurate as well. 1634 1635 A special startup routine allocates memory for the histogram and 1636either calls 'profil()' or sets up a clock signal handler. This routine 1637('monstartup') can be invoked in several ways. On Linux systems, a 1638special profiling startup file 'gcrt0.o', which invokes 'monstartup' 1639before 'main', is used instead of the default 'crt0.o'. Use of this 1640special startup file is one of the effects of using 'gcc ... -pg' to 1641link. On SPARC systems, no special startup files are used. Rather, the 1642'mcount' routine, when it is invoked for the first time (typically when 1643'main' is called), calls 'monstartup'. 1644 1645 If the compiler's '-a' option was used, basic-block counting is also 1646enabled. Each object file is then compiled with a static array of 1647counts, initially zero. In the executable code, every time a new 1648basic-block begins (i.e., when an 'if' statement appears), an extra 1649instruction is inserted to increment the corresponding count in the 1650array. At compile time, a paired array was constructed that recorded 1651the starting address of each basic-block. Taken together, the two 1652arrays record the starting address of every basic-block, along with the 1653number of times it was executed. 1654 1655 The profiling library also includes a function ('mcleanup') which is 1656typically registered using 'atexit()' to be called as the program exits, 1657and is responsible for writing the file 'gmon.out'. Profiling is turned 1658off, various headers are output, and the histogram is written, followed 1659by the call-graph arcs and the basic-block counts. 1660 1661 The output from 'gprof' gives no indication of parts of your program 1662that are limited by I/O or swapping bandwidth. This is because samples 1663of the program counter are taken at fixed intervals of the program's run 1664time. Therefore, the time measurements in 'gprof' output say nothing 1665about time that your program was not running. For example, a part of 1666the program that creates so much data that it cannot all fit in physical 1667memory at once may run very slowly due to thrashing, but 'gprof' will 1668say it uses little time. On the other hand, sampling by run time has 1669the advantage that the amount of load due to other users won't directly 1670affect the output you get. 1671 1672 1673File: gprof.info, Node: File Format, Next: Internals, Prev: Implementation, Up: Details 1674 16759.2 Profiling Data File Format 1676============================== 1677 1678The old BSD-derived file format used for profile data does not contain a 1679magic cookie that allows one to check whether a data file really is a 1680'gprof' file. Furthermore, it does not provide a version number, thus 1681rendering changes to the file format almost impossible. GNU 'gprof' 1682uses a new file format that provides these features. For backward 1683compatibility, GNU 'gprof' continues to support the old BSD-derived 1684format, but not all features are supported with it. For example, 1685basic-block execution counts cannot be accommodated by the old file 1686format. 1687 1688 The new file format is defined in header file 'gmon_out.h'. It 1689consists of a header containing the magic cookie and a version number, 1690as well as some spare bytes available for future extensions. All data 1691in a profile data file is in the native format of the target for which 1692the profile was collected. GNU 'gprof' adapts automatically to the 1693byte-order in use. 1694 1695 In the new file format, the header is followed by a sequence of 1696records. Currently, there are three different record types: histogram 1697records, call-graph arc records, and basic-block execution count 1698records. Each file can contain any number of each record type. When 1699reading a file, GNU 'gprof' will ensure records of the same type are 1700compatible with each other and compute the union of all records. For 1701example, for basic-block execution counts, the union is simply the sum 1702of all execution counts for each basic-block. 1703 17049.2.1 Histogram Records 1705----------------------- 1706 1707Histogram records consist of a header that is followed by an array of 1708bins. The header contains the text-segment range that the histogram 1709spans, the size of the histogram in bytes (unlike in the old BSD format, 1710this does not include the size of the header), the rate of the profiling 1711clock, and the physical dimension that the bin counts represent after 1712being scaled by the profiling clock rate. The physical dimension is 1713specified in two parts: a long name of up to 15 characters and a single 1714character abbreviation. For example, a histogram representing real-time 1715would specify the long name as "seconds" and the abbreviation as "s". 1716This feature is useful for architectures that support performance 1717monitor hardware (which, fortunately, is becoming increasingly common). 1718For example, under DEC OSF/1, the "uprofile" command can be used to 1719produce a histogram of, say, instruction cache misses. In this case, 1720the dimension in the histogram header could be set to "i-cache misses" 1721and the abbreviation could be set to "1" (because it is simply a count, 1722not a physical dimension). Also, the profiling rate would have to be 1723set to 1 in this case. 1724 1725 Histogram bins are 16-bit numbers and each bin represent an equal 1726amount of text-space. For example, if the text-segment is one thousand 1727bytes long and if there are ten bins in the histogram, each bin 1728represents one hundred bytes. 1729 17309.2.2 Call-Graph Records 1731------------------------ 1732 1733Call-graph records have a format that is identical to the one used in 1734the BSD-derived file format. It consists of an arc in the call graph 1735and a count indicating the number of times the arc was traversed during 1736program execution. Arcs are specified by a pair of addresses: the first 1737must be within caller's function and the second must be within the 1738callee's function. When performing profiling at the function level, 1739these addresses can point anywhere within the respective function. 1740However, when profiling at the line-level, it is better if the addresses 1741are as close to the call-site/entry-point as possible. This will ensure 1742that the line-level call-graph is able to identify exactly which line of 1743source code performed calls to a function. 1744 17459.2.3 Basic-Block Execution Count Records 1746----------------------------------------- 1747 1748Basic-block execution count records consist of a header followed by a 1749sequence of address/count pairs. The header simply specifies the length 1750of the sequence. In an address/count pair, the address identifies a 1751basic-block and the count specifies the number of times that basic-block 1752was executed. Any address within the basic-address can be used. 1753 1754 1755File: gprof.info, Node: Internals, Next: Debugging, Prev: File Format, Up: Details 1756 17579.3 'gprof''s Internal Operation 1758================================ 1759 1760Like most programs, 'gprof' begins by processing its options. During 1761this stage, it may building its symspec list ('sym_ids.c:sym_id_add'), 1762if options are specified which use symspecs. 'gprof' maintains a single 1763linked list of symspecs, which will eventually get turned into 12 symbol 1764tables, organized into six include/exclude pairs--one pair each for the 1765flat profile (INCL_FLAT/EXCL_FLAT), the call graph arcs 1766(INCL_ARCS/EXCL_ARCS), printing in the call graph 1767(INCL_GRAPH/EXCL_GRAPH), timing propagation in the call graph 1768(INCL_TIME/EXCL_TIME), the annotated source listing 1769(INCL_ANNO/EXCL_ANNO), and the execution count listing 1770(INCL_EXEC/EXCL_EXEC). 1771 1772 After option processing, 'gprof' finishes building the symspec list 1773by adding all the symspecs in 'default_excluded_list' to the exclude 1774lists EXCL_TIME and EXCL_GRAPH, and if line-by-line profiling is 1775specified, EXCL_FLAT as well. These default excludes are not added to 1776EXCL_ANNO, EXCL_ARCS, and EXCL_EXEC. 1777 1778 Next, the BFD library is called to open the object file, verify that 1779it is an object file, and read its symbol table ('core.c:core_init'), 1780using 'bfd_canonicalize_symtab' after mallocing an appropriately sized 1781array of symbols. At this point, function mappings are read (if the 1782'--file-ordering' option has been specified), and the core text space is 1783read into memory (if the '-c' option was given). 1784 1785 'gprof''s own symbol table, an array of Sym structures, is now built. 1786This is done in one of two ways, by one of two routines, depending on 1787whether line-by-line profiling ('-l' option) has been enabled. For 1788normal profiling, the BFD canonical symbol table is scanned. For 1789line-by-line profiling, every text space address is examined, and a new 1790symbol table entry gets created every time the line number changes. In 1791either case, two passes are made through the symbol table--one to count 1792the size of the symbol table required, and the other to actually read 1793the symbols. In between the two passes, a single array of type 'Sym' is 1794created of the appropriate length. Finally, 'symtab.c:symtab_finalize' 1795is called to sort the symbol table and remove duplicate entries (entries 1796with the same memory address). 1797 1798 The symbol table must be a contiguous array for two reasons. First, 1799the 'qsort' library function (which sorts an array) will be used to sort 1800the symbol table. Also, the symbol lookup routine 1801('symtab.c:sym_lookup'), which finds symbols based on memory address, 1802uses a binary search algorithm which requires the symbol table to be a 1803sorted array. Function symbols are indicated with an 'is_func' flag. 1804Line number symbols have no special flags set. Additionally, a symbol 1805can have an 'is_static' flag to indicate that it is a local symbol. 1806 1807 With the symbol table read, the symspecs can now be translated into 1808Syms ('sym_ids.c:sym_id_parse'). Remember that a single symspec can 1809match multiple symbols. An array of symbol tables ('syms') is created, 1810each entry of which is a symbol table of Syms to be included or excluded 1811from a particular listing. The master symbol table and the symspecs are 1812examined by nested loops, and every symbol that matches a symspec is 1813inserted into the appropriate syms table. This is done twice, once to 1814count the size of each required symbol table, and again to build the 1815tables, which have been malloced between passes. From now on, to 1816determine whether a symbol is on an include or exclude symspec list, 1817'gprof' simply uses its standard symbol lookup routine on the 1818appropriate table in the 'syms' array. 1819 1820 Now the profile data file(s) themselves are read 1821('gmon_io.c:gmon_out_read'), first by checking for a new-style 1822'gmon.out' header, then assuming this is an old-style BSD 'gmon.out' if 1823the magic number test failed. 1824 1825 New-style histogram records are read by 'hist.c:hist_read_rec'. For 1826the first histogram record, allocate a memory array to hold all the 1827bins, and read them in. When multiple profile data files (or files with 1828multiple histogram records) are read, the memory ranges of each pair of 1829histogram records must be either equal, or non-overlapping. For each 1830pair of histogram records, the resolution (memory region size divided by 1831the number of bins) must be the same. The time unit must be the same 1832for all histogram records. If the above containts are met, all 1833histograms for the same memory range are merged. 1834 1835 As each call graph record is read ('call_graph.c:cg_read_rec'), the 1836parent and child addresses are matched to symbol table entries, and a 1837call graph arc is created by 'cg_arcs.c:arc_add', unless the arc fails a 1838symspec check against INCL_ARCS/EXCL_ARCS. As each arc is added, a 1839linked list is maintained of the parent's child arcs, and of the child's 1840parent arcs. Both the child's call count and the arc's call count are 1841incremented by the record's call count. 1842 1843 Basic-block records are read ('basic_blocks.c:bb_read_rec'), but only 1844if line-by-line profiling has been selected. Each basic-block address 1845is matched to a corresponding line symbol in the symbol table, and an 1846entry made in the symbol's bb_addr and bb_calls arrays. Again, if 1847multiple basic-block records are present for the same address, the call 1848counts are cumulative. 1849 1850 A gmon.sum file is dumped, if requested ('gmon_io.c:gmon_out_write'). 1851 1852 If histograms were present in the data files, assign them to symbols 1853('hist.c:hist_assign_samples') by iterating over all the sample bins and 1854assigning them to symbols. Since the symbol table is sorted in order of 1855ascending memory addresses, we can simple follow along in the symbol 1856table as we make our pass over the sample bins. This step includes a 1857symspec check against INCL_FLAT/EXCL_FLAT. Depending on the histogram 1858scale factor, a sample bin may span multiple symbols, in which case a 1859fraction of the sample count is allocated to each symbol, proportional 1860to the degree of overlap. This effect is rare for normal profiling, but 1861overlaps are more common during line-by-line profiling, and can cause 1862each of two adjacent lines to be credited with half a hit, for example. 1863 1864 If call graph data is present, 'cg_arcs.c:cg_assemble' is called. 1865First, if '-c' was specified, a machine-dependent routine ('find_call') 1866scans through each symbol's machine code, looking for subroutine call 1867instructions, and adding them to the call graph with a zero call count. 1868A topological sort is performed by depth-first numbering all the symbols 1869('cg_dfn.c:cg_dfn'), so that children are always numbered less than 1870their parents, then making a array of pointers into the symbol table and 1871sorting it into numerical order, which is reverse topological order 1872(children appear before parents). Cycles are also detected at this 1873point, all members of which are assigned the same topological number. 1874Two passes are now made through this sorted array of symbol pointers. 1875The first pass, from end to beginning (parents to children), computes 1876the fraction of child time to propagate to each parent and a print flag. 1877The print flag reflects symspec handling of INCL_GRAPH/EXCL_GRAPH, with 1878a parent's include or exclude (print or no print) property being 1879propagated to its children, unless they themselves explicitly appear in 1880INCL_GRAPH or EXCL_GRAPH. A second pass, from beginning to end (children 1881to parents) actually propagates the timings along the call graph, 1882subject to a check against INCL_TIME/EXCL_TIME. With the print flag, 1883fractions, and timings now stored in the symbol structures, the 1884topological sort array is now discarded, and a new array of pointers is 1885assembled, this time sorted by propagated time. 1886 1887 Finally, print the various outputs the user requested, which is now 1888fairly straightforward. The call graph ('cg_print.c:cg_print') and flat 1889profile ('hist.c:hist_print') are regurgitations of values already 1890computed. The annotated source listing 1891('basic_blocks.c:print_annotated_source') uses basic-block information, 1892if present, to label each line of code with call counts, otherwise only 1893the function call counts are presented. 1894 1895 The function ordering code is marginally well documented in the 1896source code itself ('cg_print.c'). Basically, the functions with the 1897most use and the most parents are placed first, followed by other 1898functions with the most use, followed by lower use functions, followed 1899by unused functions at the end. 1900 1901 1902File: gprof.info, Node: Debugging, Prev: Internals, Up: Details 1903 19049.4 Debugging 'gprof' 1905===================== 1906 1907If 'gprof' was compiled with debugging enabled, the '-d' option triggers 1908debugging output (to stdout) which can be helpful in understanding its 1909operation. The debugging number specified is interpreted as a sum of 1910the following options: 1911 19122 - Topological sort 1913 Monitor depth-first numbering of symbols during call graph analysis 19144 - Cycles 1915 Shows symbols as they are identified as cycle heads 191616 - Tallying 1917 As the call graph arcs are read, show each arc and how the total 1918 calls to each function are tallied 191932 - Call graph arc sorting 1920 Details sorting individual parents/children within each call graph 1921 entry 192264 - Reading histogram and call graph records 1923 Shows address ranges of histograms as they are read, and each call 1924 graph arc 1925128 - Symbol table 1926 Reading, classifying, and sorting the symbol table from the object 1927 file. For line-by-line profiling ('-l' option), also shows line 1928 numbers being assigned to memory addresses. 1929256 - Static call graph 1930 Trace operation of '-c' option 1931512 - Symbol table and arc table lookups 1932 Detail operation of lookup routines 19331024 - Call graph propagation 1934 Shows how function times are propagated along the call graph 19352048 - Basic-blocks 1936 Shows basic-block records as they are read from profile data (only 1937 meaningful with '-l' option) 19384096 - Symspecs 1939 Shows symspec-to-symbol pattern matching operation 19408192 - Annotate source 1941 Tracks operation of '-A' option 1942 1943 1944File: gprof.info, Node: GNU Free Documentation License, Prev: Details, Up: Top 1945 1946Appendix A GNU Free Documentation License 1947***************************************** 1948 1949 Version 1.3, 3 November 2008 1950 1951 Copyright (C) 2000, 2001, 2002, 2007, 2008 Free Software Foundation, Inc. 1952 <http://fsf.org/> 1953 1954 Everyone is permitted to copy and distribute verbatim copies 1955 of this license document, but changing it is not allowed. 1956 1957 0. PREAMBLE 1958 1959 The purpose of this License is to make a manual, textbook, or other 1960 functional and useful document "free" in the sense of freedom: to 1961 assure everyone the effective freedom to copy and redistribute it, 1962 with or without modifying it, either commercially or 1963 noncommercially. Secondarily, this License preserves for the 1964 author and publisher a way to get credit for their work, while not 1965 being considered responsible for modifications made by others. 1966 1967 This License is a kind of "copyleft", which means that derivative 1968 works of the document must themselves be free in the same sense. 1969 It complements the GNU General Public License, which is a copyleft 1970 license designed for free software. 1971 1972 We have designed this License in order to use it for manuals for 1973 free software, because free software needs free documentation: a 1974 free program should come with manuals providing the same freedoms 1975 that the software does. But this License is not limited to 1976 software manuals; it can be used for any textual work, regardless 1977 of subject matter or whether it is published as a printed book. We 1978 recommend this License principally for works whose purpose is 1979 instruction or reference. 1980 1981 1. APPLICABILITY AND DEFINITIONS 1982 1983 This License applies to any manual or other work, in any medium, 1984 that contains a notice placed by the copyright holder saying it can 1985 be distributed under the terms of this License. Such a notice 1986 grants a world-wide, royalty-free license, unlimited in duration, 1987 to use that work under the conditions stated herein. The 1988 "Document", below, refers to any such manual or work. Any member 1989 of the public is a licensee, and is addressed as "you". You accept 1990 the license if you copy, modify or distribute the work in a way 1991 requiring permission under copyright law. 1992 1993 A "Modified Version" of the Document means any work containing the 1994 Document or a portion of it, either copied verbatim, or with 1995 modifications and/or translated into another language. 1996 1997 A "Secondary Section" is a named appendix or a front-matter section 1998 of the Document that deals exclusively with the relationship of the 1999 publishers or authors of the Document to the Document's overall 2000 subject (or to related matters) and contains nothing that could 2001 fall directly within that overall subject. (Thus, if the Document 2002 is in part a textbook of mathematics, a Secondary Section may not 2003 explain any mathematics.) The relationship could be a matter of 2004 historical connection with the subject or with related matters, or 2005 of legal, commercial, philosophical, ethical or political position 2006 regarding them. 2007 2008 The "Invariant Sections" are certain Secondary Sections whose 2009 titles are designated, as being those of Invariant Sections, in the 2010 notice that says that the Document is released under this License. 2011 If a section does not fit the above definition of Secondary then it 2012 is not allowed to be designated as Invariant. The Document may 2013 contain zero Invariant Sections. If the Document does not identify 2014 any Invariant Sections then there are none. 2015 2016 The "Cover Texts" are certain short passages of text that are 2017 listed, as Front-Cover Texts or Back-Cover Texts, in the notice 2018 that says that the Document is released under this License. A 2019 Front-Cover Text may be at most 5 words, and a Back-Cover Text may 2020 be at most 25 words. 2021 2022 A "Transparent" copy of the Document means a machine-readable copy, 2023 represented in a format whose specification is available to the 2024 general public, that is suitable for revising the document 2025 straightforwardly with generic text editors or (for images composed 2026 of pixels) generic paint programs or (for drawings) some widely 2027 available drawing editor, and that is suitable for input to text 2028 formatters or for automatic translation to a variety of formats 2029 suitable for input to text formatters. A copy made in an otherwise 2030 Transparent file format whose markup, or absence of markup, has 2031 been arranged to thwart or discourage subsequent modification by 2032 readers is not Transparent. An image format is not Transparent if 2033 used for any substantial amount of text. A copy that is not 2034 "Transparent" is called "Opaque". 2035 2036 Examples of suitable formats for Transparent copies include plain 2037 ASCII without markup, Texinfo input format, LaTeX input format, 2038 SGML or XML using a publicly available DTD, and standard-conforming 2039 simple HTML, PostScript or PDF designed for human modification. 2040 Examples of transparent image formats include PNG, XCF and JPG. 2041 Opaque formats include proprietary formats that can be read and 2042 edited only by proprietary word processors, SGML or XML for which 2043 the DTD and/or processing tools are not generally available, and 2044 the machine-generated HTML, PostScript or PDF produced by some word 2045 processors for output purposes only. 2046 2047 The "Title Page" means, for a printed book, the title page itself, 2048 plus such following pages as are needed to hold, legibly, the 2049 material this License requires to appear in the title page. For 2050 works in formats which do not have any title page as such, "Title 2051 Page" means the text near the most prominent appearance of the 2052 work's title, preceding the beginning of the body of the text. 2053 2054 The "publisher" means any person or entity that distributes copies 2055 of the Document to the public. 2056 2057 A section "Entitled XYZ" means a named subunit of the Document 2058 whose title either is precisely XYZ or contains XYZ in parentheses 2059 following text that translates XYZ in another language. (Here XYZ 2060 stands for a specific section name mentioned below, such as 2061 "Acknowledgements", "Dedications", "Endorsements", or "History".) 2062 To "Preserve the Title" of such a section when you modify the 2063 Document means that it remains a section "Entitled XYZ" according 2064 to this definition. 2065 2066 The Document may include Warranty Disclaimers next to the notice 2067 which states that this License applies to the Document. These 2068 Warranty Disclaimers are considered to be included by reference in 2069 this License, but only as regards disclaiming warranties: any other 2070 implication that these Warranty Disclaimers may have is void and 2071 has no effect on the meaning of this License. 2072 2073 2. VERBATIM COPYING 2074 2075 You may copy and distribute the Document in any medium, either 2076 commercially or noncommercially, provided that this License, the 2077 copyright notices, and the license notice saying this License 2078 applies to the Document are reproduced in all copies, and that you 2079 add no other conditions whatsoever to those of this License. You 2080 may not use technical measures to obstruct or control the reading 2081 or further copying of the copies you make or distribute. However, 2082 you may accept compensation in exchange for copies. If you 2083 distribute a large enough number of copies you must also follow the 2084 conditions in section 3. 2085 2086 You may also lend copies, under the same conditions stated above, 2087 and you may publicly display copies. 2088 2089 3. COPYING IN QUANTITY 2090 2091 If you publish printed copies (or copies in media that commonly 2092 have printed covers) of the Document, numbering more than 100, and 2093 the Document's license notice requires Cover Texts, you must 2094 enclose the copies in covers that carry, clearly and legibly, all 2095 these Cover Texts: Front-Cover Texts on the front cover, and 2096 Back-Cover Texts on the back cover. Both covers must also clearly 2097 and legibly identify you as the publisher of these copies. The 2098 front cover must present the full title with all words of the title 2099 equally prominent and visible. You may add other material on the 2100 covers in addition. Copying with changes limited to the covers, as 2101 long as they preserve the title of the Document and satisfy these 2102 conditions, can be treated as verbatim copying in other respects. 2103 2104 If the required texts for either cover are too voluminous to fit 2105 legibly, you should put the first ones listed (as many as fit 2106 reasonably) on the actual cover, and continue the rest onto 2107 adjacent pages. 2108 2109 If you publish or distribute Opaque copies of the Document 2110 numbering more than 100, you must either include a machine-readable 2111 Transparent copy along with each Opaque copy, or state in or with 2112 each Opaque copy a computer-network location from which the general 2113 network-using public has access to download using public-standard 2114 network protocols a complete Transparent copy of the Document, free 2115 of added material. If you use the latter option, you must take 2116 reasonably prudent steps, when you begin distribution of Opaque 2117 copies in quantity, to ensure that this Transparent copy will 2118 remain thus accessible at the stated location until at least one 2119 year after the last time you distribute an Opaque copy (directly or 2120 through your agents or retailers) of that edition to the public. 2121 2122 It is requested, but not required, that you contact the authors of 2123 the Document well before redistributing any large number of copies, 2124 to give them a chance to provide you with an updated version of the 2125 Document. 2126 2127 4. MODIFICATIONS 2128 2129 You may copy and distribute a Modified Version of the Document 2130 under the conditions of sections 2 and 3 above, provided that you 2131 release the Modified Version under precisely this License, with the 2132 Modified Version filling the role of the Document, thus licensing 2133 distribution and modification of the Modified Version to whoever 2134 possesses a copy of it. In addition, you must do these things in 2135 the Modified Version: 2136 2137 A. Use in the Title Page (and on the covers, if any) a title 2138 distinct from that of the Document, and from those of previous 2139 versions (which should, if there were any, be listed in the 2140 History section of the Document). You may use the same title 2141 as a previous version if the original publisher of that 2142 version gives permission. 2143 2144 B. List on the Title Page, as authors, one or more persons or 2145 entities responsible for authorship of the modifications in 2146 the Modified Version, together with at least five of the 2147 principal authors of the Document (all of its principal 2148 authors, if it has fewer than five), unless they release you 2149 from this requirement. 2150 2151 C. State on the Title page the name of the publisher of the 2152 Modified Version, as the publisher. 2153 2154 D. Preserve all the copyright notices of the Document. 2155 2156 E. Add an appropriate copyright notice for your modifications 2157 adjacent to the other copyright notices. 2158 2159 F. Include, immediately after the copyright notices, a license 2160 notice giving the public permission to use the Modified 2161 Version under the terms of this License, in the form shown in 2162 the Addendum below. 2163 2164 G. Preserve in that license notice the full lists of Invariant 2165 Sections and required Cover Texts given in the Document's 2166 license notice. 2167 2168 H. Include an unaltered copy of this License. 2169 2170 I. Preserve the section Entitled "History", Preserve its Title, 2171 and add to it an item stating at least the title, year, new 2172 authors, and publisher of the Modified Version as given on the 2173 Title Page. If there is no section Entitled "History" in the 2174 Document, create one stating the title, year, authors, and 2175 publisher of the Document as given on its Title Page, then add 2176 an item describing the Modified Version as stated in the 2177 previous sentence. 2178 2179 J. Preserve the network location, if any, given in the Document 2180 for public access to a Transparent copy of the Document, and 2181 likewise the network locations given in the Document for 2182 previous versions it was based on. These may be placed in the 2183 "History" section. You may omit a network location for a work 2184 that was published at least four years before the Document 2185 itself, or if the original publisher of the version it refers 2186 to gives permission. 2187 2188 K. For any section Entitled "Acknowledgements" or "Dedications", 2189 Preserve the Title of the section, and preserve in the section 2190 all the substance and tone of each of the contributor 2191 acknowledgements and/or dedications given therein. 2192 2193 L. Preserve all the Invariant Sections of the Document, unaltered 2194 in their text and in their titles. Section numbers or the 2195 equivalent are not considered part of the section titles. 2196 2197 M. Delete any section Entitled "Endorsements". Such a section 2198 may not be included in the Modified Version. 2199 2200 N. Do not retitle any existing section to be Entitled 2201 "Endorsements" or to conflict in title with any Invariant 2202 Section. 2203 2204 O. Preserve any Warranty Disclaimers. 2205 2206 If the Modified Version includes new front-matter sections or 2207 appendices that qualify as Secondary Sections and contain no 2208 material copied from the Document, you may at your option designate 2209 some or all of these sections as invariant. To do this, add their 2210 titles to the list of Invariant Sections in the Modified Version's 2211 license notice. These titles must be distinct from any other 2212 section titles. 2213 2214 You may add a section Entitled "Endorsements", provided it contains 2215 nothing but endorsements of your Modified Version by various 2216 parties--for example, statements of peer review or that the text 2217 has been approved by an organization as the authoritative 2218 definition of a standard. 2219 2220 You may add a passage of up to five words as a Front-Cover Text, 2221 and a passage of up to 25 words as a Back-Cover Text, to the end of 2222 the list of Cover Texts in the Modified Version. Only one passage 2223 of Front-Cover Text and one of Back-Cover Text may be added by (or 2224 through arrangements made by) any one entity. If the Document 2225 already includes a cover text for the same cover, previously added 2226 by you or by arrangement made by the same entity you are acting on 2227 behalf of, you may not add another; but you may replace the old 2228 one, on explicit permission from the previous publisher that added 2229 the old one. 2230 2231 The author(s) and publisher(s) of the Document do not by this 2232 License give permission to use their names for publicity for or to 2233 assert or imply endorsement of any Modified Version. 2234 2235 5. COMBINING DOCUMENTS 2236 2237 You may combine the Document with other documents released under 2238 this License, under the terms defined in section 4 above for 2239 modified versions, provided that you include in the combination all 2240 of the Invariant Sections of all of the original documents, 2241 unmodified, and list them all as Invariant Sections of your 2242 combined work in its license notice, and that you preserve all 2243 their Warranty Disclaimers. 2244 2245 The combined work need only contain one copy of this License, and 2246 multiple identical Invariant Sections may be replaced with a single 2247 copy. If there are multiple Invariant Sections with the same name 2248 but different contents, make the title of each such section unique 2249 by adding at the end of it, in parentheses, the name of the 2250 original author or publisher of that section if known, or else a 2251 unique number. Make the same adjustment to the section titles in 2252 the list of Invariant Sections in the license notice of the 2253 combined work. 2254 2255 In the combination, you must combine any sections Entitled 2256 "History" in the various original documents, forming one section 2257 Entitled "History"; likewise combine any sections Entitled 2258 "Acknowledgements", and any sections Entitled "Dedications". You 2259 must delete all sections Entitled "Endorsements." 2260 2261 6. COLLECTIONS OF DOCUMENTS 2262 2263 You may make a collection consisting of the Document and other 2264 documents released under this License, and replace the individual 2265 copies of this License in the various documents with a single copy 2266 that is included in the collection, provided that you follow the 2267 rules of this License for verbatim copying of each of the documents 2268 in all other respects. 2269 2270 You may extract a single document from such a collection, and 2271 distribute it individually under this License, provided you insert 2272 a copy of this License into the extracted document, and follow this 2273 License in all other respects regarding verbatim copying of that 2274 document. 2275 2276 7. AGGREGATION WITH INDEPENDENT WORKS 2277 2278 A compilation of the Document or its derivatives with other 2279 separate and independent documents or works, in or on a volume of a 2280 storage or distribution medium, is called an "aggregate" if the 2281 copyright resulting from the compilation is not used to limit the 2282 legal rights of the compilation's users beyond what the individual 2283 works permit. When the Document is included in an aggregate, this 2284 License does not apply to the other works in the aggregate which 2285 are not themselves derivative works of the Document. 2286 2287 If the Cover Text requirement of section 3 is applicable to these 2288 copies of the Document, then if the Document is less than one half 2289 of the entire aggregate, the Document's Cover Texts may be placed 2290 on covers that bracket the Document within the aggregate, or the 2291 electronic equivalent of covers if the Document is in electronic 2292 form. Otherwise they must appear on printed covers that bracket 2293 the whole aggregate. 2294 2295 8. TRANSLATION 2296 2297 Translation is considered a kind of modification, so you may 2298 distribute translations of the Document under the terms of section 2299 4. Replacing Invariant Sections with translations requires special 2300 permission from their copyright holders, but you may include 2301 translations of some or all Invariant Sections in addition to the 2302 original versions of these Invariant Sections. You may include a 2303 translation of this License, and all the license notices in the 2304 Document, and any Warranty Disclaimers, provided that you also 2305 include the original English version of this License and the 2306 original versions of those notices and disclaimers. In case of a 2307 disagreement between the translation and the original version of 2308 this License or a notice or disclaimer, the original version will 2309 prevail. 2310 2311 If a section in the Document is Entitled "Acknowledgements", 2312 "Dedications", or "History", the requirement (section 4) to 2313 Preserve its Title (section 1) will typically require changing the 2314 actual title. 2315 2316 9. TERMINATION 2317 2318 You may not copy, modify, sublicense, or distribute the Document 2319 except as expressly provided under this License. Any attempt 2320 otherwise to copy, modify, sublicense, or distribute it is void, 2321 and will automatically terminate your rights under this License. 2322 2323 However, if you cease all violation of this License, then your 2324 license from a particular copyright holder is reinstated (a) 2325 provisionally, unless and until the copyright holder explicitly and 2326 finally terminates your license, and (b) permanently, if the 2327 copyright holder fails to notify you of the violation by some 2328 reasonable means prior to 60 days after the cessation. 2329 2330 Moreover, your license from a particular copyright holder is 2331 reinstated permanently if the copyright holder notifies you of the 2332 violation by some reasonable means, this is the first time you have 2333 received notice of violation of this License (for any work) from 2334 that copyright holder, and you cure the violation prior to 30 days 2335 after your receipt of the notice. 2336 2337 Termination of your rights under this section does not terminate 2338 the licenses of parties who have received copies or rights from you 2339 under this License. If your rights have been terminated and not 2340 permanently reinstated, receipt of a copy of some or all of the 2341 same material does not give you any rights to use it. 2342 2343 10. FUTURE REVISIONS OF THIS LICENSE 2344 2345 The Free Software Foundation may publish new, revised versions of 2346 the GNU Free Documentation License from time to time. Such new 2347 versions will be similar in spirit to the present version, but may 2348 differ in detail to address new problems or concerns. See 2349 <http://www.gnu.org/copyleft/>. 2350 2351 Each version of the License is given a distinguishing version 2352 number. If the Document specifies that a particular numbered 2353 version of this License "or any later version" applies to it, you 2354 have the option of following the terms and conditions either of 2355 that specified version or of any later version that has been 2356 published (not as a draft) by the Free Software Foundation. If the 2357 Document does not specify a version number of this License, you may 2358 choose any version ever published (not as a draft) by the Free 2359 Software Foundation. If the Document specifies that a proxy can 2360 decide which future versions of this License can be used, that 2361 proxy's public statement of acceptance of a version permanently 2362 authorizes you to choose that version for the Document. 2363 2364 11. RELICENSING 2365 2366 "Massive Multiauthor Collaboration Site" (or "MMC Site") means any 2367 World Wide Web server that publishes copyrightable works and also 2368 provides prominent facilities for anybody to edit those works. A 2369 public wiki that anybody can edit is an example of such a server. 2370 A "Massive Multiauthor Collaboration" (or "MMC") contained in the 2371 site means any set of copyrightable works thus published on the MMC 2372 site. 2373 2374 "CC-BY-SA" means the Creative Commons Attribution-Share Alike 3.0 2375 license published by Creative Commons Corporation, a not-for-profit 2376 corporation with a principal place of business in San Francisco, 2377 California, as well as future copyleft versions of that license 2378 published by that same organization. 2379 2380 "Incorporate" means to publish or republish a Document, in whole or 2381 in part, as part of another Document. 2382 2383 An MMC is "eligible for relicensing" if it is licensed under this 2384 License, and if all works that were first published under this 2385 License somewhere other than this MMC, and subsequently 2386 incorporated in whole or in part into the MMC, (1) had no cover 2387 texts or invariant sections, and (2) were thus incorporated prior 2388 to November 1, 2008. 2389 2390 The operator of an MMC Site may republish an MMC contained in the 2391 site under CC-BY-SA on the same site at any time before August 1, 2392 2009, provided the MMC is eligible for relicensing. 2393 2394ADDENDUM: How to use this License for your documents 2395==================================================== 2396 2397To use this License in a document you have written, include a copy of 2398the License in the document and put the following copyright and license 2399notices just after the title page: 2400 2401 Copyright (C) YEAR YOUR NAME. 2402 Permission is granted to copy, distribute and/or modify this document 2403 under the terms of the GNU Free Documentation License, Version 1.3 2404 or any later version published by the Free Software Foundation; 2405 with no Invariant Sections, no Front-Cover Texts, and no Back-Cover 2406 Texts. A copy of the license is included in the section entitled ``GNU 2407 Free Documentation License''. 2408 2409 If you have Invariant Sections, Front-Cover Texts and Back-Cover 2410Texts, replace the "with...Texts." line with this: 2411 2412 with the Invariant Sections being LIST THEIR TITLES, with 2413 the Front-Cover Texts being LIST, and with the Back-Cover Texts 2414 being LIST. 2415 2416 If you have Invariant Sections without Cover Texts, or some other 2417combination of the three, merge those two alternatives to suit the 2418situation. 2419 2420 If your document contains nontrivial examples of program code, we 2421recommend releasing these examples in parallel under your choice of free 2422software license, such as the GNU General Public License, to permit 2423their use in free software. 2424 2425 2426 2427Tag Table: 2428Node: Top719 2429Node: Introduction2042 2430Node: Compiling4533 2431Node: Executing8589 2432Node: Invoking11482 2433Node: Output Options12897 2434Node: Analysis Options19989 2435Node: Miscellaneous Options23909 2436Node: Deprecated Options25163 2437Node: Symspecs27226 2438Node: Output29052 2439Node: Flat Profile30092 2440Node: Call Graph35045 2441Node: Primary38277 2442Node: Callers40865 2443Node: Subroutines42983 2444Node: Cycles44824 2445Node: Line-by-line51601 2446Node: Annotated Source55677 2447Node: Inaccuracy58675 2448Node: Sampling Error58933 2449Node: Assumptions61837 2450Node: How do I?63307 2451Node: Incompatibilities64864 2452Node: Details66358 2453Node: Implementation66751 2454Node: File Format72650 2455Node: Internals76942 2456Node: Debugging85432 2457Node: GNU Free Documentation License87022 2458 2459End Tag Table 2460 2461 2462Local Variables: 2463coding: utf-8 2464End: 2465