1 This is gprof.info, produced by makeinfo version 4.13 from 2 /tmp/android-8532/src/build/../binutils/binutils-2.21/gprof/gprof.texi. 3 4 INFO-DIR-SECTION Software development 5 START-INFO-DIR-ENTRY 6 * gprof: (gprof). Profiling your program's execution 7 END-INFO-DIR-ENTRY 8 9 This file documents the gprof profiler of the GNU system. 10 11 Copyright (C) 1988, 1992, 1997, 1998, 1999, 2000, 2001, 2003, 2007, 12 2008, 2009 Free Software Foundation, Inc. 13 14 Permission is granted to copy, distribute and/or modify this document 15 under the terms of the GNU Free Documentation License, Version 1.3 or 16 any later version published by the Free Software Foundation; with no 17 Invariant Sections, with no Front-Cover Texts, and with no Back-Cover 18 Texts. A copy of the license is included in the section entitled "GNU 19 Free Documentation License". 20 21 22 File: gprof.info, Node: Top, Next: Introduction, Up: (dir) 23 24 Profiling a Program: Where Does It Spend Its Time? 25 ************************************************** 26 27 This manual describes the GNU profiler, `gprof', and how you can use it 28 to determine which parts of a program are taking most of the execution 29 time. We assume that you know how to write, compile, and execute 30 programs. GNU `gprof' was written by Jay Fenlason. 31 32 This manual is for `gprof' (GNU Binutils) version 2.21. 33 34 This document is distributed under the terms of the GNU Free 35 Documentation License version 1.3. A copy of the license is included 36 in the section entitled "GNU Free Documentation License". 37 38 * Menu: 39 40 * Introduction:: What profiling means, and why it is useful. 41 42 * Compiling:: How to compile your program for profiling. 43 * Executing:: Executing your program to generate profile data 44 * Invoking:: How to run `gprof', and its options 45 46 * Output:: Interpreting `gprof''s output 47 48 * Inaccuracy:: Potential problems you should be aware of 49 * How do I?:: Answers to common questions 50 * Incompatibilities:: (between GNU `gprof' and Unix `gprof'.) 51 * Details:: Details of how profiling is done 52 * GNU Free Documentation License:: GNU Free Documentation License 53 54 55 File: gprof.info, Node: Introduction, Next: Compiling, Prev: Top, Up: Top 56 57 1 Introduction to Profiling 58 *************************** 59 60 Profiling allows you to learn where your program spent its time and 61 which functions called which other functions while it was executing. 62 This information can show you which pieces of your program are slower 63 than you expected, and might be candidates for rewriting to make your 64 program execute faster. It can also tell you which functions are being 65 called more or less often than you expected. This may help you spot 66 bugs that had otherwise been unnoticed. 67 68 Since the profiler uses information collected during the actual 69 execution of your program, it can be used on programs that are too 70 large or too complex to analyze by reading the source. However, how 71 your program is run will affect the information that shows up in the 72 profile data. If you don't use some feature of your program while it 73 is being profiled, no profile information will be generated for that 74 feature. 75 76 Profiling has several steps: 77 78 * You must compile and link your program with profiling enabled. 79 *Note Compiling a Program for Profiling: Compiling. 80 81 * You must execute your program to generate a profile data file. 82 *Note Executing the Program: Executing. 83 84 * You must run `gprof' to analyze the profile data. *Note `gprof' 85 Command Summary: Invoking. 86 87 The next three chapters explain these steps in greater detail. 88 89 Several forms of output are available from the analysis. 90 91 The "flat profile" shows how much time your program spent in each 92 function, and how many times that function was called. If you simply 93 want to know which functions burn most of the cycles, it is stated 94 concisely here. *Note The Flat Profile: Flat Profile. 95 96 The "call graph" shows, for each function, which functions called 97 it, which other functions it called, and how many times. There is also 98 an estimate of how much time was spent in the subroutines of each 99 function. This can suggest places where you might try to eliminate 100 function calls that use a lot of time. *Note The Call Graph: Call 101 Graph. 102 103 The "annotated source" listing is a copy of the program's source 104 code, labeled with the number of times each line of the program was 105 executed. *Note The Annotated Source Listing: Annotated Source. 106 107 To better understand how profiling works, you may wish to read a 108 description of its implementation. *Note Implementation of Profiling: 109 Implementation. 110 111 112 File: gprof.info, Node: Compiling, Next: Executing, Prev: Introduction, Up: Top 113 114 2 Compiling a Program for Profiling 115 *********************************** 116 117 The first step in generating profile information for your program is to 118 compile and link it with profiling enabled. 119 120 To compile a source file for profiling, specify the `-pg' option when 121 you run the compiler. (This is in addition to the options you normally 122 use.) 123 124 To link the program for profiling, if you use a compiler such as `cc' 125 to do the linking, simply specify `-pg' in addition to your usual 126 options. The same option, `-pg', alters either compilation or linking 127 to do what is necessary for profiling. Here are examples: 128 129 cc -g -c myprog.c utils.c -pg 130 cc -o myprog myprog.o utils.o -pg 131 132 The `-pg' option also works with a command that both compiles and 133 links: 134 135 cc -o myprog myprog.c utils.c -g -pg 136 137 Note: The `-pg' option must be part of your compilation options as 138 well as your link options. If it is not then no call-graph data will 139 be gathered and when you run `gprof' you will get an error message like 140 this: 141 142 gprof: gmon.out file is missing call-graph data 143 144 If you add the `-Q' switch to suppress the printing of the call 145 graph data you will still be able to see the time samples: 146 147 Flat profile: 148 149 Each sample counts as 0.01 seconds. 150 % cumulative self self total 151 time seconds seconds calls Ts/call Ts/call name 152 44.12 0.07 0.07 zazLoop 153 35.29 0.14 0.06 main 154 20.59 0.17 0.04 bazMillion 155 156 If you run the linker `ld' directly instead of through a compiler 157 such as `cc', you may have to specify a profiling startup file 158 `gcrt0.o' as the first input file instead of the usual startup file 159 `crt0.o'. In addition, you would probably want to specify the 160 profiling C library, `libc_p.a', by writing `-lc_p' instead of the 161 usual `-lc'. This is not absolutely necessary, but doing this gives 162 you number-of-calls information for standard library functions such as 163 `read' and `open'. For example: 164 165 ld -o myprog /lib/gcrt0.o myprog.o utils.o -lc_p 166 167 If you are running the program on a system which supports shared 168 libraries you may run into problems with the profiling support code in 169 a shared library being called before that library has been fully 170 initialised. This is usually detected by the program encountering a 171 segmentation fault as soon as it is run. The solution is to link 172 against a static version of the library containing the profiling 173 support code, which for `gcc' users can be done via the `-static' or 174 `-static-libgcc' command line option. For example: 175 176 gcc -g -pg -static-libgcc myprog.c utils.c -o myprog 177 178 If you compile only some of the modules of the program with `-pg', 179 you can still profile the program, but you won't get complete 180 information about the modules that were compiled without `-pg'. The 181 only information you get for the functions in those modules is the 182 total time spent in them; there is no record of how many times they 183 were called, or from where. This will not affect the flat profile 184 (except that the `calls' field for the functions will be blank), but 185 will greatly reduce the usefulness of the call graph. 186 187 If you wish to perform line-by-line profiling you should use the 188 `gcov' tool instead of `gprof'. See that tool's manual or info pages 189 for more details of how to do this. 190 191 Note, older versions of `gcc' produce line-by-line profiling 192 information that works with `gprof' rather than `gcov' so there is 193 still support for displaying this kind of information in `gprof'. *Note 194 Line-by-line Profiling: Line-by-line. 195 196 It also worth noting that `gcc' implements a 197 `-finstrument-functions' command line option which will insert calls to 198 special user supplied instrumentation routines at the entry and exit of 199 every function in their program. This can be used to implement an 200 alternative profiling scheme. 201 202 203 File: gprof.info, Node: Executing, Next: Invoking, Prev: Compiling, Up: Top 204 205 3 Executing the Program 206 *********************** 207 208 Once the program is compiled for profiling, you must run it in order to 209 generate the information that `gprof' needs. Simply run the program as 210 usual, using the normal arguments, file names, etc. The program should 211 run normally, producing the same output as usual. It will, however, run 212 somewhat slower than normal because of the time spent collecting and 213 writing the profile data. 214 215 The way you run the program--the arguments and input that you give 216 it--may have a dramatic effect on what the profile information shows. 217 The profile data will describe the parts of the program that were 218 activated for the particular input you use. For example, if the first 219 command you give to your program is to quit, the profile data will show 220 the time used in initialization and in cleanup, but not much else. 221 222 Your program will write the profile data into a file called 223 `gmon.out' just before exiting. If there is already a file called 224 `gmon.out', its contents are overwritten. There is currently no way to 225 tell the program to write the profile data under a different name, but 226 you can rename the file afterwards if you are concerned that it may be 227 overwritten. 228 229 In order to write the `gmon.out' file properly, your program must 230 exit normally: by returning from `main' or by calling `exit'. Calling 231 the low-level function `_exit' does not write the profile data, and 232 neither does abnormal termination due to an unhandled signal. 233 234 The `gmon.out' file is written in the program's _current working 235 directory_ at the time it exits. This means that if your program calls 236 `chdir', the `gmon.out' file will be left in the last directory your 237 program `chdir''d to. If you don't have permission to write in this 238 directory, the file is not written, and you will get an error message. 239 240 Older versions of the GNU profiling library may also write a file 241 called `bb.out'. This file, if present, contains an human-readable 242 listing of the basic-block execution counts. Unfortunately, the 243 appearance of a human-readable `bb.out' means the basic-block counts 244 didn't get written into `gmon.out'. The Perl script `bbconv.pl', 245 included with the `gprof' source distribution, will convert a `bb.out' 246 file into a format readable by `gprof'. Invoke it like this: 247 248 bbconv.pl < bb.out > BH-DATA 249 250 This translates the information in `bb.out' into a form that `gprof' 251 can understand. But you still need to tell `gprof' about the existence 252 of this translated information. To do that, include BB-DATA on the 253 `gprof' command line, _along with `gmon.out'_, like this: 254 255 gprof OPTIONS EXECUTABLE-FILE gmon.out BB-DATA [YET-MORE-PROFILE-DATA-FILES...] [> OUTFILE] 256 257 258 File: gprof.info, Node: Invoking, Next: Output, Prev: Executing, Up: Top 259 260 4 `gprof' Command Summary 261 ************************* 262 263 After you have a profile data file `gmon.out', you can run `gprof' to 264 interpret the information in it. The `gprof' program prints a flat 265 profile and a call graph on standard output. Typically you would 266 redirect the output of `gprof' into a file with `>'. 267 268 You run `gprof' like this: 269 270 gprof OPTIONS [EXECUTABLE-FILE [PROFILE-DATA-FILES...]] [> OUTFILE] 271 272 Here square-brackets indicate optional arguments. 273 274 If you omit the executable file name, the file `a.out' is used. If 275 you give no profile data file name, the file `gmon.out' is used. If 276 any file is not in the proper format, or if the profile data file does 277 not appear to belong to the executable file, an error message is 278 printed. 279 280 You can give more than one profile data file by entering all their 281 names after the executable file name; then the statistics in all the 282 data files are summed together. 283 284 The order of these options does not matter. 285 286 * Menu: 287 288 * Output Options:: Controlling `gprof''s output style 289 * Analysis Options:: Controlling how `gprof' analyzes its data 290 * Miscellaneous Options:: 291 * Deprecated Options:: Options you no longer need to use, but which 292 have been retained for compatibility 293 * Symspecs:: Specifying functions to include or exclude 294 295 296 File: gprof.info, Node: Output Options, Next: Analysis Options, Up: Invoking 297 298 4.1 Output Options 299 ================== 300 301 These options specify which of several output formats `gprof' should 302 produce. 303 304 Many of these options take an optional "symspec" to specify 305 functions to be included or excluded. These options can be specified 306 multiple times, with different symspecs, to include or exclude sets of 307 symbols. *Note Symspecs: Symspecs. 308 309 Specifying any of these options overrides the default (`-p -q'), 310 which prints a flat profile and call graph analysis for all functions. 311 312 `-A[SYMSPEC]' 313 `--annotated-source[=SYMSPEC]' 314 The `-A' option causes `gprof' to print annotated source code. If 315 SYMSPEC is specified, print output only for matching symbols. 316 *Note The Annotated Source Listing: Annotated Source. 317 318 `-b' 319 `--brief' 320 If the `-b' option is given, `gprof' doesn't print the verbose 321 blurbs that try to explain the meaning of all of the fields in the 322 tables. This is useful if you intend to print out the output, or 323 are tired of seeing the blurbs. 324 325 `-C[SYMSPEC]' 326 `--exec-counts[=SYMSPEC]' 327 The `-C' option causes `gprof' to print a tally of functions and 328 the number of times each was called. If SYMSPEC is specified, 329 print tally only for matching symbols. 330 331 If the profile data file contains basic-block count records, 332 specifying the `-l' option, along with `-C', will cause basic-block 333 execution counts to be tallied and displayed. 334 335 `-i' 336 `--file-info' 337 The `-i' option causes `gprof' to display summary information 338 about the profile data file(s) and then exit. The number of 339 histogram, call graph, and basic-block count records is displayed. 340 341 `-I DIRS' 342 `--directory-path=DIRS' 343 The `-I' option specifies a list of search directories in which to 344 find source files. Environment variable GPROF_PATH can also be 345 used to convey this information. Used mostly for annotated source 346 output. 347 348 `-J[SYMSPEC]' 349 `--no-annotated-source[=SYMSPEC]' 350 The `-J' option causes `gprof' not to print annotated source code. 351 If SYMSPEC is specified, `gprof' prints annotated source, but 352 excludes matching symbols. 353 354 `-L' 355 `--print-path' 356 Normally, source filenames are printed with the path component 357 suppressed. The `-L' option causes `gprof' to print the full 358 pathname of source filenames, which is determined from symbolic 359 debugging information in the image file and is relative to the 360 directory in which the compiler was invoked. 361 362 `-p[SYMSPEC]' 363 `--flat-profile[=SYMSPEC]' 364 The `-p' option causes `gprof' to print a flat profile. If 365 SYMSPEC is specified, print flat profile only for matching symbols. 366 *Note The Flat Profile: Flat Profile. 367 368 `-P[SYMSPEC]' 369 `--no-flat-profile[=SYMSPEC]' 370 The `-P' option causes `gprof' to suppress printing a flat profile. 371 If SYMSPEC is specified, `gprof' prints a flat profile, but 372 excludes matching symbols. 373 374 `-q[SYMSPEC]' 375 `--graph[=SYMSPEC]' 376 The `-q' option causes `gprof' to print the call graph analysis. 377 If SYMSPEC is specified, print call graph only for matching symbols 378 and their children. *Note The Call Graph: Call Graph. 379 380 `-Q[SYMSPEC]' 381 `--no-graph[=SYMSPEC]' 382 The `-Q' option causes `gprof' to suppress printing the call graph. 383 If SYMSPEC is specified, `gprof' prints a call graph, but excludes 384 matching symbols. 385 386 `-t' 387 `--table-length=NUM' 388 The `-t' option causes the NUM most active source lines in each 389 source file to be listed when source annotation is enabled. The 390 default is 10. 391 392 `-y' 393 `--separate-files' 394 This option affects annotated source output only. Normally, 395 `gprof' prints annotated source files to standard-output. If this 396 option is specified, annotated source for a file named 397 `path/FILENAME' is generated in the file `FILENAME-ann'. If the 398 underlying file system would truncate `FILENAME-ann' so that it 399 overwrites the original `FILENAME', `gprof' generates annotated 400 source in the file `FILENAME.ann' instead (if the original file 401 name has an extension, that extension is _replaced_ with `.ann'). 402 403 `-Z[SYMSPEC]' 404 `--no-exec-counts[=SYMSPEC]' 405 The `-Z' option causes `gprof' not to print a tally of functions 406 and the number of times each was called. If SYMSPEC is specified, 407 print tally, but exclude matching symbols. 408 409 `-r' 410 `--function-ordering' 411 The `--function-ordering' option causes `gprof' to print a 412 suggested function ordering for the program based on profiling 413 data. This option suggests an ordering which may improve paging, 414 tlb and cache behavior for the program on systems which support 415 arbitrary ordering of functions in an executable. 416 417 The exact details of how to force the linker to place functions in 418 a particular order is system dependent and out of the scope of this 419 manual. 420 421 `-R MAP_FILE' 422 `--file-ordering MAP_FILE' 423 The `--file-ordering' option causes `gprof' to print a suggested 424 .o link line ordering for the program based on profiling data. 425 This option suggests an ordering which may improve paging, tlb and 426 cache behavior for the program on systems which do not support 427 arbitrary ordering of functions in an executable. 428 429 Use of the `-a' argument is highly recommended with this option. 430 431 The MAP_FILE argument is a pathname to a file which provides 432 function name to object file mappings. The format of the file is 433 similar to the output of the program `nm'. 434 435 c-parse.o:00000000 T yyparse 436 c-parse.o:00000004 C yyerrflag 437 c-lang.o:00000000 T maybe_objc_method_name 438 c-lang.o:00000000 T print_lang_statistics 439 c-lang.o:00000000 T recognize_objc_keyword 440 c-decl.o:00000000 T print_lang_identifier 441 c-decl.o:00000000 T print_lang_type 442 ... 443 444 To create a MAP_FILE with GNU `nm', type a command like `nm 445 --extern-only --defined-only -v --print-file-name program-name'. 446 447 `-T' 448 `--traditional' 449 The `-T' option causes `gprof' to print its output in 450 "traditional" BSD style. 451 452 `-w WIDTH' 453 `--width=WIDTH' 454 Sets width of output lines to WIDTH. Currently only used when 455 printing the function index at the bottom of the call graph. 456 457 `-x' 458 `--all-lines' 459 This option affects annotated source output only. By default, 460 only the lines at the beginning of a basic-block are annotated. 461 If this option is specified, every line in a basic-block is 462 annotated by repeating the annotation for the first line. This 463 behavior is similar to `tcov''s `-a'. 464 465 `--demangle[=STYLE]' 466 `--no-demangle' 467 These options control whether C++ symbol names should be demangled 468 when printing output. The default is to demangle symbols. The 469 `--no-demangle' option may be used to turn off demangling. 470 Different compilers have different mangling styles. The optional 471 demangling style argument can be used to choose an appropriate 472 demangling style for your compiler. 473 474 475 File: gprof.info, Node: Analysis Options, Next: Miscellaneous Options, Prev: Output Options, Up: Invoking 476 477 4.2 Analysis Options 478 ==================== 479 480 `-a' 481 `--no-static' 482 The `-a' option causes `gprof' to suppress the printing of 483 statically declared (private) functions. (These are functions 484 whose names are not listed as global, and which are not visible 485 outside the file/function/block where they were defined.) Time 486 spent in these functions, calls to/from them, etc., will all be 487 attributed to the function that was loaded directly before it in 488 the executable file. This option affects both the flat profile 489 and the call graph. 490 491 `-c' 492 `--static-call-graph' 493 The `-c' option causes the call graph of the program to be 494 augmented by a heuristic which examines the text space of the 495 object file and identifies function calls in the binary machine 496 code. Since normal call graph records are only generated when 497 functions are entered, this option identifies children that could 498 have been called, but never were. Calls to functions that were 499 not compiled with profiling enabled are also identified, but only 500 if symbol table entries are present for them. Calls to dynamic 501 library routines are typically _not_ found by this option. 502 Parents or children identified via this heuristic are indicated in 503 the call graph with call counts of `0'. 504 505 `-D' 506 `--ignore-non-functions' 507 The `-D' option causes `gprof' to ignore symbols which are not 508 known to be functions. This option will give more accurate 509 profile data on systems where it is supported (Solaris and HPUX for 510 example). 511 512 `-k FROM/TO' 513 The `-k' option allows you to delete from the call graph any arcs 514 from symbols matching symspec FROM to those matching symspec TO. 515 516 `-l' 517 `--line' 518 The `-l' option enables line-by-line profiling, which causes 519 histogram hits to be charged to individual source code lines, 520 instead of functions. This feature only works with programs 521 compiled by older versions of the `gcc' compiler. Newer versions 522 of `gcc' are designed to work with the `gcov' tool instead. 523 524 If the program was compiled with basic-block counting enabled, 525 this option will also identify how many times each line of code 526 was executed. While line-by-line profiling can help isolate where 527 in a large function a program is spending its time, it also 528 significantly increases the running time of `gprof', and magnifies 529 statistical inaccuracies. *Note Statistical Sampling Error: 530 Sampling Error. 531 532 `-m NUM' 533 `--min-count=NUM' 534 This option affects execution count output only. Symbols that are 535 executed less than NUM times are suppressed. 536 537 `-nSYMSPEC' 538 `--time=SYMSPEC' 539 The `-n' option causes `gprof', in its call graph analysis, to 540 only propagate times for symbols matching SYMSPEC. 541 542 `-NSYMSPEC' 543 `--no-time=SYMSPEC' 544 The `-n' option causes `gprof', in its call graph analysis, not to 545 propagate times for symbols matching SYMSPEC. 546 547 `-SFILENAME' 548 `--external-symbol-table=FILENAME' 549 The `-S' option causes `gprof' to read an external symbol table 550 file, such as `/proc/kallsyms', rather than read the symbol table 551 from the given object file (the default is `a.out'). This is useful 552 for profiling kernel modules. 553 554 `-z' 555 `--display-unused-functions' 556 If you give the `-z' option, `gprof' will mention all functions in 557 the flat profile, even those that were never called, and that had 558 no time spent in them. This is useful in conjunction with the 559 `-c' option for discovering which routines were never called. 560 561 562 563 File: gprof.info, Node: Miscellaneous Options, Next: Deprecated Options, Prev: Analysis Options, Up: Invoking 564 565 4.3 Miscellaneous Options 566 ========================= 567 568 `-d[NUM]' 569 `--debug[=NUM]' 570 The `-d NUM' option specifies debugging options. If NUM is not 571 specified, enable all debugging. *Note Debugging `gprof': 572 Debugging. 573 574 `-h' 575 `--help' 576 The `-h' option prints command line usage. 577 578 `-ONAME' 579 `--file-format=NAME' 580 Selects the format of the profile data files. Recognized formats 581 are `auto' (the default), `bsd', `4.4bsd', `magic', and `prof' 582 (not yet supported). 583 584 `-s' 585 `--sum' 586 The `-s' option causes `gprof' to summarize the information in the 587 profile data files it read in, and write out a profile data file 588 called `gmon.sum', which contains all the information from the 589 profile data files that `gprof' read in. The file `gmon.sum' may 590 be one of the specified input files; the effect of this is to 591 merge the data in the other input files into `gmon.sum'. 592 593 Eventually you can run `gprof' again without `-s' to analyze the 594 cumulative data in the file `gmon.sum'. 595 596 `-v' 597 `--version' 598 The `-v' flag causes `gprof' to print the current version number, 599 and then exit. 600 601 602 603 File: gprof.info, Node: Deprecated Options, Next: Symspecs, Prev: Miscellaneous Options, Up: Invoking 604 605 4.4 Deprecated Options 606 ====================== 607 608 These 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 615 `[not 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 620 spent in the function (and children who were not called from 621 anywhere else), will not be used to compute the 622 percentages-of-time for the call graph. More than one `-E' option 623 may be given; only one FUNCTION_NAME may be indicated with each 624 `-E' option. 625 626 `-f FUNCTION_NAME' 627 The `-f FUNCTION' option causes `gprof' to limit the call graph to 628 the function FUNCTION_NAME and its children (and their 629 children...). More than one `-f' option may be given; only one 630 FUNCTION_NAME may be indicated with each `-f' option. 631 632 `-F FUNCTION_NAME' 633 The `-F FUNCTION' option works like the `-f' option, but only time 634 spent in the function and its children (and their children...) 635 will be used to determine total-time and percentages-of-time for 636 the call graph. More than one `-F' option may be given; only one 637 FUNCTION_NAME may be indicated with each `-F' option. The `-F' 638 option overrides the `-E' option. 639 640 641 Note that only one function can be specified with each `-e', `-E', 642 `-f' or `-F' option. To specify more than one function, use multiple 643 options. For example, this command: 644 645 gprof -e boring -f foo -f bar myprogram > gprof.output 646 647 lists in the call graph all functions that were reached from either 648 `foo' or `bar' and were not reachable from `boring'. 649 650 651 File: gprof.info, Node: Symspecs, Prev: Deprecated Options, Up: Invoking 652 653 4.5 Symspecs 654 ============ 655 656 Many of the output options allow functions to be included or excluded 657 using "symspecs" (symbol specifications), which observe the following 658 syntax: 659 660 filename_containing_a_dot 661 | funcname_not_containing_a_dot 662 | linenumber 663 | ( [ any_filename ] `:' ( any_funcname | linenumber ) ) 664 665 Here are some sample symspecs: 666 667 `main.c' 668 Selects everything in file `main.c'--the dot in the string tells 669 `gprof' to interpret the string as a filename, rather than as a 670 function name. To select a file whose name does not contain a 671 dot, a trailing colon should be specified. For example, `odd:' is 672 interpreted as the file named `odd'. 673 674 `main' 675 Selects all functions named `main'. 676 677 Note that there may be multiple instances of the same function name 678 because some of the definitions may be local (i.e., static). 679 Unless a function name is unique in a program, you must use the 680 colon notation explained below to specify a function from a 681 specific source file. 682 683 Sometimes, function names contain dots. In such cases, it is 684 necessary to add a leading colon to the name. For example, 685 `:.mul' selects function `.mul'. 686 687 In some object file formats, symbols have a leading underscore. 688 `gprof' will normally not print these underscores. When you name a 689 symbol in a symspec, you should type it exactly as `gprof' prints 690 it in its output. For example, if the compiler produces a symbol 691 `_main' from your `main' function, `gprof' still prints it as 692 `main' in its output, so you should use `main' in symspecs. 693 694 `main.c:main' 695 Selects function `main' in file `main.c'. 696 697 `main.c:134' 698 Selects line 134 in file `main.c'. 699 700 701 File: gprof.info, Node: Output, Next: Inaccuracy, Prev: Invoking, Up: Top 702 703 5 Interpreting `gprof''s Output 704 ******************************* 705 706 `gprof' can produce several different output styles, the most important 707 of which are described below. The simplest output styles (file 708 information, execution count, and function and file ordering) are not 709 described here, but are documented with the respective options that 710 trigger them. *Note Output Options: Output Options. 711 712 * Menu: 713 714 * Flat Profile:: The flat profile shows how much time was spent 715 executing directly in each function. 716 * Call Graph:: The call graph shows which functions called which 717 others, and how much time each function used 718 when its subroutine calls are included. 719 * Line-by-line:: `gprof' can analyze individual source code lines 720 * Annotated Source:: The annotated source listing displays source code 721 labeled with execution counts 722 723 724 File: gprof.info, Node: Flat Profile, Next: Call Graph, Up: Output 725 726 5.1 The Flat Profile 727 ==================== 728 729 The "flat profile" shows the total amount of time your program spent 730 executing each function. Unless the `-z' option is given, functions 731 with no apparent time spent in them, and no apparent calls to them, are 732 not mentioned. Note that if a function was not compiled for profiling, 733 and didn't run long enough to show up on the program counter histogram, 734 it will be indistinguishable from a function that was never called. 735 736 This is part of a flat profile for a small program: 737 738 Flat profile: 739 740 Each sample counts as 0.01 seconds. 741 % cumulative self self total 742 time seconds seconds calls ms/call ms/call name 743 33.34 0.02 0.02 7208 0.00 0.00 open 744 16.67 0.03 0.01 244 0.04 0.12 offtime 745 16.67 0.04 0.01 8 1.25 1.25 memccpy 746 16.67 0.05 0.01 7 1.43 1.43 write 747 16.67 0.06 0.01 mcount 748 0.00 0.06 0.00 236 0.00 0.00 tzset 749 0.00 0.06 0.00 192 0.00 0.00 tolower 750 0.00 0.06 0.00 47 0.00 0.00 strlen 751 0.00 0.06 0.00 45 0.00 0.00 strchr 752 0.00 0.06 0.00 1 0.00 50.00 main 753 0.00 0.06 0.00 1 0.00 0.00 memcpy 754 0.00 0.06 0.00 1 0.00 10.11 print 755 0.00 0.06 0.00 1 0.00 0.00 profil 756 0.00 0.06 0.00 1 0.00 50.00 report 757 ... 758 759 The functions are sorted first by decreasing run-time spent in them, 760 then by decreasing number of calls, then alphabetically by name. The 761 functions `mcount' and `profil' are part of the profiling apparatus and 762 appear in every flat profile; their time gives a measure of the amount 763 of overhead due to profiling. 764 765 Just before the column headers, a statement appears indicating how 766 much time each sample counted as. This "sampling period" estimates the 767 margin of error in each of the time figures. A time figure that is not 768 much larger than this is not reliable. In this example, each sample 769 counted as 0.01 seconds, suggesting a 100 Hz sampling rate. The 770 program's total execution time was 0.06 seconds, as indicated by the 771 `cumulative seconds' field. Since each sample counted for 0.01 772 seconds, this means only six samples were taken during the run. Two of 773 the samples occurred while the program was in the `open' function, as 774 indicated by the `self seconds' field. Each of the other four samples 775 occurred one each in `offtime', `memccpy', `write', and `mcount'. 776 Since only six samples were taken, none of these values can be regarded 777 as particularly reliable. In another run, the `self seconds' field for 778 `mcount' might well be `0.00' or `0.02'. *Note Statistical Sampling 779 Error: Sampling Error, for a complete discussion. 780 781 The remaining functions in the listing (those whose `self seconds' 782 field is `0.00') didn't appear in the histogram samples at all. 783 However, the call graph indicated that they were called, so therefore 784 they are listed, sorted in decreasing order by the `calls' field. 785 Clearly some time was spent executing these functions, but the paucity 786 of histogram samples prevents any determination of how much time each 787 took. 788 789 Here is what the fields in each line mean: 790 791 `% time' 792 This is the percentage of the total execution time your program 793 spent in this function. These should all add up to 100%. 794 795 `cumulative seconds' 796 This is the cumulative total number of seconds the computer spent 797 executing this functions, plus the time spent in all the functions 798 above this one in this table. 799 800 `self seconds' 801 This is the number of seconds accounted for by this function alone. 802 The flat profile listing is sorted first by this number. 803 804 `calls' 805 This is the total number of times the function was called. If the 806 function was never called, or the number of times it was called 807 cannot be determined (probably because the function was not 808 compiled with profiling enabled), the "calls" field is blank. 809 810 `self ms/call' 811 This represents the average number of milliseconds spent in this 812 function per call, if this function is profiled. Otherwise, this 813 field is blank for this function. 814 815 `total ms/call' 816 This represents the average number of milliseconds spent in this 817 function and its descendants per call, if this function is 818 profiled. Otherwise, this field is blank for this function. This 819 is the only field in the flat profile that uses call graph 820 analysis. 821 822 `name' 823 This is the name of the function. The flat profile is sorted by 824 this field alphabetically after the "self seconds" and "calls" 825 fields are sorted. 826 827 828 File: gprof.info, Node: Call Graph, Next: Line-by-line, Prev: Flat Profile, Up: Output 829 830 5.2 The Call Graph 831 ================== 832 833 The "call graph" shows how much time was spent in each function and its 834 children. From this information, you can find functions that, while 835 they themselves may not have used much time, called other functions 836 that did use unusual amounts of time. 837 838 Here is a sample call from a small program. This call came from the 839 same `gprof' run as the flat profile example in the previous section. 840 841 granularity: each sample hit covers 2 byte(s) for 20.00% of 0.05 seconds 842 843 index % time self children called name 844 <spontaneous> 845 [1] 100.0 0.00 0.05 start [1] 846 0.00 0.05 1/1 main [2] 847 0.00 0.00 1/2 on_exit [28] 848 0.00 0.00 1/1 exit [59] 849 ----------------------------------------------- 850 0.00 0.05 1/1 start [1] 851 [2] 100.0 0.00 0.05 1 main [2] 852 0.00 0.05 1/1 report [3] 853 ----------------------------------------------- 854 0.00 0.05 1/1 main [2] 855 [3] 100.0 0.00 0.05 1 report [3] 856 0.00 0.03 8/8 timelocal [6] 857 0.00 0.01 1/1 print [9] 858 0.00 0.01 9/9 fgets [12] 859 0.00 0.00 12/34 strncmp <cycle 1> [40] 860 0.00 0.00 8/8 lookup [20] 861 0.00 0.00 1/1 fopen [21] 862 0.00 0.00 8/8 chewtime [24] 863 0.00 0.00 8/16 skipspace [44] 864 ----------------------------------------------- 865 [4] 59.8 0.01 0.02 8+472 <cycle 2 as a whole> [4] 866 0.01 0.02 244+260 offtime <cycle 2> [7] 867 0.00 0.00 236+1 tzset <cycle 2> [26] 868 ----------------------------------------------- 869 870 The lines full of dashes divide this table into "entries", one for 871 each function. Each entry has one or more lines. 872 873 In each entry, the primary line is the one that starts with an index 874 number in square brackets. The end of this line says which function 875 the entry is for. The preceding lines in the entry describe the 876 callers of this function and the following lines describe its 877 subroutines (also called "children" when we speak of the call graph). 878 879 The entries are sorted by time spent in the function and its 880 subroutines. 881 882 The internal profiling function `mcount' (*note The Flat Profile: 883 Flat Profile.) is never mentioned in the call graph. 884 885 * Menu: 886 887 * Primary:: Details of the primary line's contents. 888 * Callers:: Details of caller-lines' contents. 889 * Subroutines:: Details of subroutine-lines' contents. 890 * Cycles:: When there are cycles of recursion, 891 such as `a' calls `b' calls `a'... 892 893 894 File: gprof.info, Node: Primary, Next: Callers, Up: Call Graph 895 896 5.2.1 The Primary Line 897 ---------------------- 898 899 The "primary line" in a call graph entry is the line that describes the 900 function which the entry is about and gives the overall statistics for 901 this function. 902 903 For reference, we repeat the primary line from the entry for function 904 `report' in our main example, together with the heading line that shows 905 the names of the fields: 906 907 index % time self children called name 908 ... 909 [3] 100.0 0.00 0.05 1 report [3] 910 911 Here is what the fields in the primary line mean: 912 913 `index' 914 Entries are numbered with consecutive integers. Each function 915 therefore has an index number, which appears at the beginning of 916 its primary line. 917 918 Each cross-reference to a function, as a caller or subroutine of 919 another, gives its index number as well as its name. The index 920 number guides you if you wish to look for the entry for that 921 function. 922 923 `% time' 924 This is the percentage of the total time that was spent in this 925 function, including time spent in subroutines called from this 926 function. 927 928 The time spent in this function is counted again for the callers of 929 this function. Therefore, adding up these percentages is 930 meaningless. 931 932 `self' 933 This is the total amount of time spent in this function. This 934 should be identical to the number printed in the `seconds' field 935 for this function in the flat profile. 936 937 `children' 938 This is the total amount of time spent in the subroutine calls 939 made by this function. This should be equal to the sum of all the 940 `self' and `children' entries of the children listed directly 941 below this function. 942 943 `called' 944 This is the number of times the function was called. 945 946 If the function called itself recursively, there are two numbers, 947 separated by a `+'. The first number counts non-recursive calls, 948 and the second counts recursive calls. 949 950 In the example above, the function `report' was called once from 951 `main'. 952 953 `name' 954 This is the name of the current function. The index number is 955 repeated after it. 956 957 If the function is part of a cycle of recursion, the cycle number 958 is printed between the function's name and the index number (*note 959 How Mutually Recursive Functions Are Described: Cycles.). For 960 example, if function `gnurr' is part of cycle number one, and has 961 index number twelve, its primary line would be end like this: 962 963 gnurr <cycle 1> [12] 964 965 966 File: gprof.info, Node: Callers, Next: Subroutines, Prev: Primary, Up: Call Graph 967 968 5.2.2 Lines for a Function's Callers 969 ------------------------------------ 970 971 A function's entry has a line for each function it was called by. 972 These lines' fields correspond to the fields of the primary line, but 973 their meanings are different because of the difference in context. 974 975 For reference, we repeat two lines from the entry for the function 976 `report', the primary line and one caller-line preceding it, together 977 with the heading line that shows the names of the fields: 978 979 index % time self children called name 980 ... 981 0.00 0.05 1/1 main [2] 982 [3] 100.0 0.00 0.05 1 report [3] 983 984 Here are the meanings of the fields in the caller-line for `report' 985 called from `main': 986 987 `self' 988 An estimate of the amount of time spent in `report' itself when it 989 was called from `main'. 990 991 `children' 992 An estimate of the amount of time spent in subroutines of `report' 993 when `report' was called from `main'. 994 995 The sum of the `self' and `children' fields is an estimate of the 996 amount of time spent within calls to `report' from `main'. 997 998 `called' 999 Two numbers: the number of times `report' was called from `main', 1000 followed by the total number of non-recursive calls to `report' 1001 from all its callers. 1002 1003 `name and index number' 1004 The name of the caller of `report' to which this line applies, 1005 followed by the caller's index number. 1006 1007 Not all functions have entries in the call graph; some options to 1008 `gprof' request the omission of certain functions. When a caller 1009 has no entry of its own, it still has caller-lines in the entries 1010 of the functions it calls. 1011 1012 If the caller is part of a recursion cycle, the cycle number is 1013 printed between the name and the index number. 1014 1015 If the identity of the callers of a function cannot be determined, a 1016 dummy caller-line is printed which has `<spontaneous>' as the "caller's 1017 name" and all other fields blank. This can happen for signal handlers. 1018 1019 1020 File: gprof.info, Node: Subroutines, Next: Cycles, Prev: Callers, Up: Call Graph 1021 1022 5.2.3 Lines for a Function's Subroutines 1023 ---------------------------------------- 1024 1025 A function's entry has a line for each of its subroutines--in other 1026 words, a line for each other function that it called. These lines' 1027 fields correspond to the fields of the primary line, but their meanings 1028 are different because of the difference in context. 1029 1030 For reference, we repeat two lines from the entry for the function 1031 `main', the primary line and a line for a subroutine, together with the 1032 heading line that shows the names of the fields: 1033 1034 index % time self children called name 1035 ... 1036 [2] 100.0 0.00 0.05 1 main [2] 1037 0.00 0.05 1/1 report [3] 1038 1039 Here are the meanings of the fields in the subroutine-line for `main' 1040 calling `report': 1041 1042 `self' 1043 An estimate of the amount of time spent directly within `report' 1044 when `report' was called from `main'. 1045 1046 `children' 1047 An estimate of the amount of time spent in subroutines of `report' 1048 when `report' was called from `main'. 1049 1050 The sum of the `self' and `children' fields is an estimate of the 1051 total time spent in calls to `report' from `main'. 1052 1053 `called' 1054 Two numbers, the number of calls to `report' from `main' followed 1055 by the total number of non-recursive calls to `report'. This 1056 ratio is used to determine how much of `report''s `self' and 1057 `children' time gets credited to `main'. *Note Estimating 1058 `children' Times: Assumptions. 1059 1060 `name' 1061 The name of the subroutine of `main' to which this line applies, 1062 followed by the subroutine's index number. 1063 1064 If the caller is part of a recursion cycle, the cycle number is 1065 printed between the name and the index number. 1066 1067 1068 File: gprof.info, Node: Cycles, Prev: Subroutines, Up: Call Graph 1069 1070 5.2.4 How Mutually Recursive Functions Are Described 1071 ---------------------------------------------------- 1072 1073 The graph may be complicated by the presence of "cycles of recursion" 1074 in the call graph. A cycle exists if a function calls another function 1075 that (directly or indirectly) calls (or appears to call) the original 1076 function. For example: if `a' calls `b', and `b' calls `a', then `a' 1077 and `b' form a cycle. 1078 1079 Whenever there are call paths both ways between a pair of functions, 1080 they belong to the same cycle. If `a' and `b' call each other and `b' 1081 and `c' call each other, all three make one cycle. Note that even if 1082 `b' only calls `a' if it was not called from `a', `gprof' cannot 1083 determine this, so `a' and `b' are still considered a cycle. 1084 1085 The cycles are numbered with consecutive integers. When a function 1086 belongs to a cycle, each time the function name appears in the call 1087 graph it is followed by `<cycle NUMBER>'. 1088 1089 The reason cycles matter is that they make the time values in the 1090 call graph paradoxical. The "time spent in children" of `a' should 1091 include the time spent in its subroutine `b' and in `b''s 1092 subroutines--but one of `b''s subroutines is `a'! How much of `a''s 1093 time should be included in the children of `a', when `a' is indirectly 1094 recursive? 1095 1096 The way `gprof' resolves this paradox is by creating a single entry 1097 for the cycle as a whole. The primary line of this entry describes the 1098 total time spent directly in the functions of the cycle. The 1099 "subroutines" of the cycle are the individual functions of the cycle, 1100 and all other functions that were called directly by them. The 1101 "callers" of the cycle are the functions, outside the cycle, that 1102 called functions in the cycle. 1103 1104 Here is an example portion of a call graph which shows a cycle 1105 containing functions `a' and `b'. The cycle was entered by a call to 1106 `a' from `main'; both `a' and `b' called `c'. 1107 1108 index % time self children called name 1109 ---------------------------------------- 1110 1.77 0 1/1 main [2] 1111 [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3] 1112 1.02 0 3 b <cycle 1> [4] 1113 0.75 0 2 a <cycle 1> [5] 1114 ---------------------------------------- 1115 3 a <cycle 1> [5] 1116 [4] 52.85 1.02 0 0 b <cycle 1> [4] 1117 2 a <cycle 1> [5] 1118 0 0 3/6 c [6] 1119 ---------------------------------------- 1120 1.77 0 1/1 main [2] 1121 2 b <cycle 1> [4] 1122 [5] 38.86 0.75 0 1 a <cycle 1> [5] 1123 3 b <cycle 1> [4] 1124 0 0 3/6 c [6] 1125 ---------------------------------------- 1126 1127 (The entire call graph for this program contains in addition an entry 1128 for `main', which calls `a', and an entry for `c', with callers `a' and 1129 `b'.) 1130 1131 index % time self children called name 1132 <spontaneous> 1133 [1] 100.00 0 1.93 0 start [1] 1134 0.16 1.77 1/1 main [2] 1135 ---------------------------------------- 1136 0.16 1.77 1/1 start [1] 1137 [2] 100.00 0.16 1.77 1 main [2] 1138 1.77 0 1/1 a <cycle 1> [5] 1139 ---------------------------------------- 1140 1.77 0 1/1 main [2] 1141 [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3] 1142 1.02 0 3 b <cycle 1> [4] 1143 0.75 0 2 a <cycle 1> [5] 1144 0 0 6/6 c [6] 1145 ---------------------------------------- 1146 3 a <cycle 1> [5] 1147 [4] 52.85 1.02 0 0 b <cycle 1> [4] 1148 2 a <cycle 1> [5] 1149 0 0 3/6 c [6] 1150 ---------------------------------------- 1151 1.77 0 1/1 main [2] 1152 2 b <cycle 1> [4] 1153 [5] 38.86 0.75 0 1 a <cycle 1> [5] 1154 3 b <cycle 1> [4] 1155 0 0 3/6 c [6] 1156 ---------------------------------------- 1157 0 0 3/6 b <cycle 1> [4] 1158 0 0 3/6 a <cycle 1> [5] 1159 [6] 0.00 0 0 6 c [6] 1160 ---------------------------------------- 1161 1162 The `self' field of the cycle's primary line is the total time spent 1163 in all the functions of the cycle. It equals the sum of the `self' 1164 fields for the individual functions in the cycle, found in the entry in 1165 the subroutine lines for these functions. 1166 1167 The `children' fields of the cycle's primary line and subroutine 1168 lines count only subroutines outside the cycle. Even though `a' calls 1169 `b', the time spent in those calls to `b' is not counted in `a''s 1170 `children' time. Thus, we do not encounter the problem of what to do 1171 when the time in those calls to `b' includes indirect recursive calls 1172 back to `a'. 1173 1174 The `children' field of a caller-line in the cycle's entry estimates 1175 the amount of time spent _in the whole cycle_, and its other 1176 subroutines, on the times when that caller called a function in the 1177 cycle. 1178 1179 The `called' field in the primary line for the cycle has two numbers: 1180 first, the number of times functions in the cycle were called by 1181 functions outside the cycle; second, the number of times they were 1182 called by functions in the cycle (including times when a function in 1183 the cycle calls itself). This is a generalization of the usual split 1184 into non-recursive and recursive calls. 1185 1186 The `called' field of a subroutine-line for a cycle member in the 1187 cycle's entry says how many time that function was called from 1188 functions in the cycle. The total of all these is the second number in 1189 the primary line's `called' field. 1190 1191 In the individual entry for a function in a cycle, the other 1192 functions in the same cycle can appear as subroutines and as callers. 1193 These lines show how many times each function in the cycle called or 1194 was called from each other function in the cycle. The `self' and 1195 `children' fields in these lines are blank because of the difficulty of 1196 defining meanings for them when recursion is going on. 1197 1198 1199 File: gprof.info, Node: Line-by-line, Next: Annotated Source, Prev: Call Graph, Up: Output 1200 1201 5.3 Line-by-line Profiling 1202 ========================== 1203 1204 `gprof''s `-l' option causes the program to perform "line-by-line" 1205 profiling. In this mode, histogram samples are assigned not to 1206 functions, but to individual lines of source code. This only works 1207 with programs compiled with older versions of the `gcc' compiler. 1208 Newer versions of `gcc' use a different program - `gcov' - to display 1209 line-by-line profiling information. 1210 1211 With the older versions of `gcc' the program usually has to be 1212 compiled with a `-g' option, in addition to `-pg', in order to generate 1213 debugging symbols for tracking source code lines. Note, in much older 1214 versions of `gcc' the program had to be compiled with the `-a' command 1215 line option as well. 1216 1217 The flat profile is the most useful output table in line-by-line 1218 mode. The call graph isn't as useful as normal, since the current 1219 version of `gprof' does not propagate call graph arcs from source code 1220 lines to the enclosing function. The call graph does, however, show 1221 each line of code that called each function, along with a count. 1222 1223 Here is a section of `gprof''s output, without line-by-line 1224 profiling. Note that `ct_init' accounted for four histogram hits, and 1225 13327 calls to `init_block'. 1226 1227 Flat profile: 1228 1229 Each sample counts as 0.01 seconds. 1230 % cumulative self self total 1231 time seconds seconds calls us/call us/call name 1232 30.77 0.13 0.04 6335 6.31 6.31 ct_init 1233 1234 1235 Call graph (explanation follows) 1236 1237 1238 granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds 1239 1240 index % time self children called name 1241 1242 0.00 0.00 1/13496 name_too_long 1243 0.00 0.00 40/13496 deflate 1244 0.00 0.00 128/13496 deflate_fast 1245 0.00 0.00 13327/13496 ct_init 1246 [7] 0.0 0.00 0.00 13496 init_block 1247 1248 Now let's look at some of `gprof''s output from the same program run, 1249 this time with line-by-line profiling enabled. Note that `ct_init''s 1250 four histogram hits are broken down into four lines of source code--one 1251 hit occurred on each of lines 349, 351, 382 and 385. In the call graph, 1252 note how `ct_init''s 13327 calls to `init_block' are broken down into 1253 one call from line 396, 3071 calls from line 384, 3730 calls from line 1254 385, and 6525 calls from 387. 1255 1256 Flat profile: 1257 1258 Each sample counts as 0.01 seconds. 1259 % cumulative self 1260 time seconds seconds calls name 1261 7.69 0.10 0.01 ct_init (trees.c:349) 1262 7.69 0.11 0.01 ct_init (trees.c:351) 1263 7.69 0.12 0.01 ct_init (trees.c:382) 1264 7.69 0.13 0.01 ct_init (trees.c:385) 1265 1266 1267 Call graph (explanation follows) 1268 1269 1270 granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds 1271 1272 % time self children called name 1273 1274 0.00 0.00 1/13496 name_too_long (gzip.c:1440) 1275 0.00 0.00 1/13496 deflate (deflate.c:763) 1276 0.00 0.00 1/13496 ct_init (trees.c:396) 1277 0.00 0.00 2/13496 deflate (deflate.c:727) 1278 0.00 0.00 4/13496 deflate (deflate.c:686) 1279 0.00 0.00 5/13496 deflate (deflate.c:675) 1280 0.00 0.00 12/13496 deflate (deflate.c:679) 1281 0.00 0.00 16/13496 deflate (deflate.c:730) 1282 0.00 0.00 128/13496 deflate_fast (deflate.c:654) 1283 0.00 0.00 3071/13496 ct_init (trees.c:384) 1284 0.00 0.00 3730/13496 ct_init (trees.c:385) 1285 0.00 0.00 6525/13496 ct_init (trees.c:387) 1286 [6] 0.0 0.00 0.00 13496 init_block (trees.c:408) 1287 1288 1289 File: gprof.info, Node: Annotated Source, Prev: Line-by-line, Up: Output 1290 1291 5.4 The Annotated Source Listing 1292 ================================ 1293 1294 `gprof''s `-A' option triggers an annotated source listing, which lists 1295 the program's source code, each function labeled with the number of 1296 times 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' 1300 augments your program with basic-block counting code, in addition to 1301 function counting code. This enables `gprof' to determine how many 1302 times each line of code was executed. With newer versions of `gcc' 1303 support for displaying basic-block counts is provided by the `gcov' 1304 program. 1305 1306 For example, consider the following function, taken from gzip, with 1307 line 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 `updcrc' has at least five basic-blocks. One is the function 1330 itself. The `if' statement on line 9 generates two more basic-blocks, 1331 one for each branch of the `if'. A fourth basic-block results from the 1332 `if' on line 13, and the contents of the `do' loop form the fifth 1333 basic-block. The compiler may also generate additional basic-blocks to 1334 handle various special cases. 1335 1336 A program augmented for basic-block counting can be analyzed with 1337 `gprof -l -A'. The `-x' option is also helpful, to ensure that each 1338 line of code is labeled at least once. Here is `updcrc''s annotated 1339 source listing for a sample `gzip' run: 1340 1341 ulg updcrc(s, n) 1342 uch *s; 1343 unsigned n; 1344 2 ->{ 1345 register ulg c; 1346 1347 static ulg crc = (ulg)0xffffffffL; 1348 1349 2 -> if (s == NULL) { 1350 1 -> c = 0xffffffffL; 1351 1 -> } else { 1352 1 -> c = crc; 1353 1 -> if (n) do { 1354 26312 -> c = crc_32_tab[...]; 1355 26312,1,26311 -> } while (--n); 1356 } 1357 2 -> crc = c; 1358 2 -> return c ^ 0xffffffffL; 1359 2 ->} 1360 1361 In this example, the function was called twice, passing once through 1362 each branch of the `if' statement. The body of the `do' loop was 1363 executed a total of 26312 times. Note how the `while' statement is 1364 annotated. It began execution 26312 times, once for each iteration 1365 through the loop. One of those times (the last time) it exited, while 1366 it branched back to the beginning of the loop 26311 times. 1367 1368 1369 File: gprof.info, Node: Inaccuracy, Next: How do I?, Prev: Output, Up: Top 1370 1371 6 Inaccuracy of `gprof' Output 1372 ****************************** 1373 1374 * Menu: 1375 1376 * Sampling Error:: Statistical margins of error 1377 * Assumptions:: Estimating children times 1378 1379 1380 File: gprof.info, Node: Sampling Error, Next: Assumptions, Up: Inaccuracy 1381 1382 6.1 Statistical Sampling Error 1383 ============================== 1384 1385 The run-time figures that `gprof' gives you are based on a sampling 1386 process, so they are subject to statistical inaccuracy. If a function 1387 runs only a small amount of time, so that on the average the sampling 1388 process ought to catch that function in the act only once, there is a 1389 pretty good chance it will actually find that function zero times, or 1390 twice. 1391 1392 By contrast, the number-of-calls and basic-block figures are derived 1393 by counting, not sampling. They are completely accurate and will not 1394 vary from run to run if your program is deterministic and single 1395 threaded. In multi-threaded applications, or single threaded 1396 applications that link with multi-threaded libraries, the counts are 1397 only deterministic if the counting function is thread-safe. (Note: 1398 beware that the mcount counting function in glibc is _not_ 1399 thread-safe). *Note Implementation of Profiling: Implementation. 1400 1401 The "sampling period" that is printed at the beginning of the flat 1402 profile says how often samples are taken. The rule of thumb is that a 1403 run-time figure is accurate if it is considerably bigger than the 1404 sampling period. 1405 1406 The actual amount of error can be predicted. For N samples, the 1407 _expected_ error is the square-root of N. For example, if the sampling 1408 period is 0.01 seconds and `foo''s run-time is 1 second, N is 100 1409 samples (1 second/0.01 seconds), sqrt(N) is 10 samples, so the expected 1410 error in `foo''s run-time is 0.1 seconds (10*0.01 seconds), or ten 1411 percent of the observed value. Again, if the sampling period is 0.01 1412 seconds and `bar''s run-time is 100 seconds, N is 10000 samples, 1413 sqrt(N) is 100 samples, so the expected error in `bar''s run-time is 1 1414 second, or one percent of the observed value. It is likely to vary 1415 this much _on the average_ from one profiling run to the next. 1416 (_Sometimes_ it will vary more.) 1417 1418 This does not mean that a small run-time figure is devoid of 1419 information. If the program's _total_ run-time is large, a small 1420 run-time for one function does tell you that that function used an 1421 insignificant fraction of the whole program's time. Usually this means 1422 it is not worth optimizing. 1423 1424 One way to get more accuracy is to give your program more (but 1425 similar) input data so it will take longer. Another way is to combine 1426 the data from several runs, using the `-s' option of `gprof'. Here is 1427 how: 1428 1429 1. Run your program once. 1430 1431 2. Issue the command `mv gmon.out gmon.sum'. 1432 1433 3. Run your program again, the same as before. 1434 1435 4. Merge the new data in `gmon.out' into `gmon.sum' with this command: 1436 1437 gprof -s EXECUTABLE-FILE gmon.out gmon.sum 1438 1439 5. Repeat the last two steps as often as you wish. 1440 1441 6. Analyze the cumulative data using this command: 1442 1443 gprof EXECUTABLE-FILE gmon.sum > OUTPUT-FILE 1444 1445 1446 File: gprof.info, Node: Assumptions, Prev: Sampling Error, Up: Inaccuracy 1447 1448 6.2 Estimating `children' Times 1449 =============================== 1450 1451 Some of the figures in the call graph are estimates--for example, the 1452 `children' time values and all the time figures in caller and 1453 subroutine lines. 1454 1455 There is no direct information about these measurements in the 1456 profile data itself. Instead, `gprof' estimates them by making an 1457 assumption about your program that might or might not be true. 1458 1459 The assumption made is that the average time spent in each call to 1460 any function `foo' is not correlated with who called `foo'. If `foo' 1461 used 5 seconds in all, and 2/5 of the calls to `foo' came from `a', 1462 then `foo' contributes 2 seconds to `a''s `children' time, by 1463 assumption. 1464 1465 This assumption is usually true enough, but for some programs it is 1466 far from true. Suppose that `foo' returns very quickly when its 1467 argument is zero; suppose that `a' always passes zero as an argument, 1468 while other callers of `foo' pass other arguments. In this program, 1469 all the time spent in `foo' is in the calls from callers other than `a'. 1470 But `gprof' has no way of knowing this; it will blindly and incorrectly 1471 charge 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 1474 this assumption is no longer needed, if we can figure out how. For the 1475 novice, the estimated figures are usually more useful than misleading. 1476 1477 1478 File: gprof.info, Node: How do I?, Next: Incompatibilities, Prev: Inaccuracy, Up: Top 1479 1480 7 Answers to Common Questions 1481 ***************************** 1482 1483 How can I get more exact information about hot spots in my program? 1484 Looking at the per-line call counts only tells part of the story. 1485 Because `gprof' can only report call times and counts by function, 1486 the best way to get finer-grained information on where the program 1487 is spending its time is to re-factor large functions into sequences 1488 of calls to smaller ones. Beware however that this can introduce 1489 artificial hot spots since compiling with `-pg' adds a significant 1490 overhead to function calls. An alternative solution is to use a 1491 non-intrusive profiler, e.g. oprofile. 1492 1493 How do I find which lines in my program were executed the most times? 1494 Use the `gcov' program. 1495 1496 How do I find which lines in my program called a particular function? 1497 Use `gprof -l' and lookup the function in the call graph. The 1498 callers will be broken down by function and line number. 1499 1500 How do I analyze a program that runs for less than a second? 1501 Try using a shell script like this one: 1502 1503 for i in `seq 1 100`; do 1504 fastprog 1505 mv gmon.out gmon.out.$i 1506 done 1507 1508 gprof -s fastprog gmon.out.* 1509 1510 gprof fastprog gmon.sum 1511 1512 If your program is completely deterministic, all the call counts 1513 will be simple multiples of 100 (i.e., a function called once in 1514 each run will appear with a call count of 100). 1515 1516 1517 1518 File: gprof.info, Node: Incompatibilities, Next: Details, Prev: How do I?, Up: Top 1519 1520 8 Incompatibilities with Unix `gprof' 1521 ************************************* 1522 1523 GNU `gprof' and Berkeley Unix `gprof' use the same data file 1524 `gmon.out', and provide essentially the same information. But there 1525 are a few differences. 1526 1527 * GNU `gprof' uses a new, generalized file format with support for 1528 basic-block execution counts and non-realtime histograms. A magic 1529 cookie and version number allows `gprof' to easily identify new 1530 style files. Old BSD-style files can still be read. *Note 1531 Profiling Data File Format: File Format. 1532 1533 * For a recursive function, Unix `gprof' lists the function as a 1534 parent and as a child, with a `calls' field that lists the number 1535 of recursive calls. GNU `gprof' omits these lines and puts the 1536 number of recursive calls in the primary line. 1537 1538 * When a function is suppressed from the call graph with `-e', GNU 1539 `gprof' still lists it as a subroutine of functions that call it. 1540 1541 * GNU `gprof' accepts the `-k' with its argument in the form 1542 `from/to', instead of `from to'. 1543 1544 * In the annotated source listing, if there are multiple basic 1545 blocks on the same line, GNU `gprof' prints all of their counts, 1546 separated by commas. 1547 1548 * The blurbs, field widths, and output formats are different. GNU 1549 `gprof' prints blurbs after the tables, so that you can see the 1550 tables without skipping the blurbs. 1551 1552 1553 File: gprof.info, Node: Details, Next: GNU Free Documentation License, Prev: Incompatibilities, Up: Top 1554 1555 9 Details of Profiling 1556 ********************** 1557 1558 * Menu: 1559 1560 * Implementation:: How a program collects profiling information 1561 * File Format:: Format of `gmon.out' files 1562 * Internals:: `gprof''s internal operation 1563 * Debugging:: Using `gprof''s `-d' option 1564 1565 1566 File: gprof.info, Node: Implementation, Next: File Format, Up: Details 1567 1568 9.1 Implementation of Profiling 1569 =============================== 1570 1571 Profiling works by changing how every function in your program is 1572 compiled so that when it is called, it will stash away some information 1573 about where it was called from. From this, the profiler can figure out 1574 what function called it, and can count how many times it was called. 1575 This change is made by the compiler when your program is compiled with 1576 the `-pg' option, which causes every function to call `mcount' (or 1577 `_mcount', or `__mcount', depending on the OS and compiler) as one of 1578 its first operations. 1579 1580 The `mcount' routine, included in the profiling library, is 1581 responsible for recording in an in-memory call graph table both its 1582 parent routine (the child) and its parent's parent. This is typically 1583 done by examining the stack frame to find both the address of the 1584 child, and the return address in the original parent. Since this is a 1585 very machine-dependent operation, `mcount' itself is typically a short 1586 assembly-language stub routine that extracts the required information, 1587 and then calls `__mcount_internal' (a normal C function) with two 1588 arguments--`frompc' and `selfpc'. `__mcount_internal' is responsible 1589 for maintaining the in-memory call graph, which records `frompc', 1590 `selfpc', and the number of times each of these call arcs was traversed. 1591 1592 GCC Version 2 provides a magical function 1593 (`__builtin_return_address'), which allows a generic `mcount' function 1594 to extract the required information from the stack frame. However, on 1595 some architectures, most notably the SPARC, using this builtin can be 1596 very computationally expensive, and an assembly language version of 1597 `mcount' is used for performance reasons. 1598 1599 Number-of-calls information for library routines is collected by 1600 using a special version of the C library. The programs in it are the 1601 same as in the usual C library, but they were compiled with `-pg'. If 1602 you link your program with `gcc ... -pg', it automatically uses the 1603 profiling version of the library. 1604 1605 Profiling also involves watching your program as it runs, and 1606 keeping a histogram of where the program counter happens to be every 1607 now and then. Typically the program counter is looked at around 100 1608 times per second of run time, but the exact frequency may vary from 1609 system to system. 1610 1611 This is done is one of two ways. Most UNIX-like operating systems 1612 provide a `profil()' system call, which registers a memory array with 1613 the kernel, along with a scale factor that determines how the program's 1614 address space maps into the array. Typical scaling values cause every 1615 2 to 8 bytes of address space to map into a single array slot. On 1616 every tick of the system clock (assuming the profiled program is 1617 running), the value of the program counter is examined and the 1618 corresponding slot in the memory array is incremented. Since this is 1619 done in the kernel, which had to interrupt the process anyway to handle 1620 the clock interrupt, very little additional system overhead is required. 1621 1622 However, some operating systems, most notably Linux 2.0 (and 1623 earlier), do not provide a `profil()' system call. On such a system, 1624 arrangements are made for the kernel to periodically deliver a signal 1625 to the process (typically via `setitimer()'), which then performs the 1626 same operation of examining the program counter and incrementing a slot 1627 in the memory array. Since this method requires a signal to be 1628 delivered to user space every time a sample is taken, it uses 1629 considerably more overhead than kernel-based profiling. Also, due to 1630 the added delay required to deliver the signal, this method is less 1631 accurate as well. 1632 1633 A special startup routine allocates memory for the histogram and 1634 either calls `profil()' or sets up a clock signal handler. This 1635 routine (`monstartup') can be invoked in several ways. On Linux 1636 systems, a special profiling startup file `gcrt0.o', which invokes 1637 `monstartup' before `main', is used instead of the default `crt0.o'. 1638 Use of this special startup file is one of the effects of using `gcc 1639 ... -pg' to link. On SPARC systems, no special startup files are used. 1640 Rather, the `mcount' routine, when it is invoked for the first time 1641 (typically when `main' is called), calls `monstartup'. 1642 1643 If the compiler's `-a' option was used, basic-block counting is also 1644 enabled. Each object file is then compiled with a static array of 1645 counts, initially zero. In the executable code, every time a new 1646 basic-block begins (i.e., when an `if' statement appears), an extra 1647 instruction is inserted to increment the corresponding count in the 1648 array. At compile time, a paired array was constructed that recorded 1649 the starting address of each basic-block. Taken together, the two 1650 arrays record the starting address of every basic-block, along with the 1651 number of times it was executed. 1652 1653 The profiling library also includes a function (`mcleanup') which is 1654 typically registered using `atexit()' to be called as the program 1655 exits, and is responsible for writing the file `gmon.out'. Profiling 1656 is turned off, various headers are output, and the histogram is 1657 written, followed by the call-graph arcs and the basic-block counts. 1658 1659 The output from `gprof' gives no indication of parts of your program 1660 that are limited by I/O or swapping bandwidth. This is because samples 1661 of the program counter are taken at fixed intervals of the program's 1662 run time. Therefore, the time measurements in `gprof' output say 1663 nothing about time that your program was not running. For example, a 1664 part of the program that creates so much data that it cannot all fit in 1665 physical memory at once may run very slowly due to thrashing, but 1666 `gprof' will say it uses little time. On the other hand, sampling by 1667 run time has the advantage that the amount of load due to other users 1668 won't directly affect the output you get. 1669 1670 1671 File: gprof.info, Node: File Format, Next: Internals, Prev: Implementation, Up: Details 1672 1673 9.2 Profiling Data File Format 1674 ============================== 1675 1676 The old BSD-derived file format used for profile data does not contain a 1677 magic cookie that allows to check whether a data file really is a 1678 `gprof' file. Furthermore, it does not provide a version number, thus 1679 rendering changes to the file format almost impossible. GNU `gprof' 1680 uses a new file format that provides these features. For backward 1681 compatibility, GNU `gprof' continues to support the old BSD-derived 1682 format, but not all features are supported with it. For example, 1683 basic-block execution counts cannot be accommodated by the old file 1684 format. 1685 1686 The new file format is defined in header file `gmon_out.h'. It 1687 consists of a header containing the magic cookie and a version number, 1688 as well as some spare bytes available for future extensions. All data 1689 in a profile data file is in the native format of the target for which 1690 the profile was collected. GNU `gprof' adapts automatically to the 1691 byte-order in use. 1692 1693 In the new file format, the header is followed by a sequence of 1694 records. Currently, there are three different record types: histogram 1695 records, call-graph arc records, and basic-block execution count 1696 records. Each file can contain any number of each record type. When 1697 reading a file, GNU `gprof' will ensure records of the same type are 1698 compatible with each other and compute the union of all records. For 1699 example, for basic-block execution counts, the union is simply the sum 1700 of all execution counts for each basic-block. 1701 1702 9.2.1 Histogram Records 1703 ----------------------- 1704 1705 Histogram records consist of a header that is followed by an array of 1706 bins. The header contains the text-segment range that the histogram 1707 spans, the size of the histogram in bytes (unlike in the old BSD 1708 format, this does not include the size of the header), the rate of the 1709 profiling clock, and the physical dimension that the bin counts 1710 represent after being scaled by the profiling clock rate. The physical 1711 dimension is specified in two parts: a long name of up to 15 characters 1712 and a single character abbreviation. For example, a histogram 1713 representing real-time would specify the long name as "seconds" and the 1714 abbreviation as "s". This feature is useful for architectures that 1715 support performance monitor hardware (which, fortunately, is becoming 1716 increasingly common). For example, under DEC OSF/1, the "uprofile" 1717 command can be used to produce a histogram of, say, instruction cache 1718 misses. In this case, the dimension in the histogram header could be 1719 set to "i-cache misses" and the abbreviation could be set to "1" 1720 (because it is simply a count, not a physical dimension). Also, the 1721 profiling rate would have to be set to 1 in this case. 1722 1723 Histogram bins are 16-bit numbers and each bin represent an equal 1724 amount of text-space. For example, if the text-segment is one thousand 1725 bytes long and if there are ten bins in the histogram, each bin 1726 represents one hundred bytes. 1727 1728 9.2.2 Call-Graph Records 1729 ------------------------ 1730 1731 Call-graph records have a format that is identical to the one used in 1732 the BSD-derived file format. It consists of an arc in the call graph 1733 and a count indicating the number of times the arc was traversed during 1734 program execution. Arcs are specified by a pair of addresses: the 1735 first must be within caller's function and the second must be within 1736 the callee's function. When performing profiling at the function 1737 level, these addresses can point anywhere within the respective 1738 function. However, when profiling at the line-level, it is better if 1739 the addresses are as close to the call-site/entry-point as possible. 1740 This will ensure that the line-level call-graph is able to identify 1741 exactly which line of source code performed calls to a function. 1742 1743 9.2.3 Basic-Block Execution Count Records 1744 ----------------------------------------- 1745 1746 Basic-block execution count records consist of a header followed by a 1747 sequence of address/count pairs. The header simply specifies the 1748 length of the sequence. In an address/count pair, the address 1749 identifies a basic-block and the count specifies the number of times 1750 that basic-block was executed. Any address within the basic-address can 1751 be used. 1752 1753 1754 File: gprof.info, Node: Internals, Next: Debugging, Prev: File Format, Up: Details 1755 1756 9.3 `gprof''s Internal Operation 1757 ================================ 1758 1759 Like most programs, `gprof' begins by processing its options. During 1760 this stage, it may building its symspec list (`sym_ids.c:sym_id_add'), 1761 if options are specified which use symspecs. `gprof' maintains a 1762 single linked list of symspecs, which will eventually get turned into 1763 12 symbol tables, organized into six include/exclude pairs--one pair 1764 each for the flat profile (INCL_FLAT/EXCL_FLAT), the call graph arcs 1765 (INCL_ARCS/EXCL_ARCS), printing in the call graph 1766 (INCL_GRAPH/EXCL_GRAPH), timing propagation in the call graph 1767 (INCL_TIME/EXCL_TIME), the annotated source listing 1768 (INCL_ANNO/EXCL_ANNO), and the execution count listing 1769 (INCL_EXEC/EXCL_EXEC). 1770 1771 After option processing, `gprof' finishes building the symspec list 1772 by adding all the symspecs in `default_excluded_list' to the exclude 1773 lists EXCL_TIME and EXCL_GRAPH, and if line-by-line profiling is 1774 specified, EXCL_FLAT as well. These default excludes are not added to 1775 EXCL_ANNO, EXCL_ARCS, and EXCL_EXEC. 1776 1777 Next, the BFD library is called to open the object file, verify that 1778 it is an object file, and read its symbol table (`core.c:core_init'), 1779 using `bfd_canonicalize_symtab' after mallocing an appropriately sized 1780 array of symbols. At this point, function mappings are read (if the 1781 `--file-ordering' option has been specified), and the core text space 1782 is read into memory (if the `-c' option was given). 1783 1784 `gprof''s own symbol table, an array of Sym structures, is now built. 1785 This is done in one of two ways, by one of two routines, depending on 1786 whether line-by-line profiling (`-l' option) has been enabled. For 1787 normal profiling, the BFD canonical symbol table is scanned. For 1788 line-by-line profiling, every text space address is examined, and a new 1789 symbol table entry gets created every time the line number changes. In 1790 either case, two passes are made through the symbol table--one to count 1791 the size of the symbol table required, and the other to actually read 1792 the symbols. In between the two passes, a single array of type `Sym' 1793 is created of the appropriate length. Finally, 1794 `symtab.c:symtab_finalize' is called to sort the symbol table and 1795 remove duplicate entries (entries with the same memory address). 1796 1797 The symbol table must be a contiguous array for two reasons. First, 1798 the `qsort' library function (which sorts an array) will be used to 1799 sort the symbol table. Also, the symbol lookup routine 1800 (`symtab.c:sym_lookup'), which finds symbols based on memory address, 1801 uses a binary search algorithm which requires the symbol table to be a 1802 sorted array. Function symbols are indicated with an `is_func' flag. 1803 Line number symbols have no special flags set. Additionally, a symbol 1804 can have an `is_static' flag to indicate that it is a local symbol. 1805 1806 With the symbol table read, the symspecs can now be translated into 1807 Syms (`sym_ids.c:sym_id_parse'). Remember that a single symspec can 1808 match multiple symbols. An array of symbol tables (`syms') is created, 1809 each entry of which is a symbol table of Syms to be included or 1810 excluded from a particular listing. The master symbol table and the 1811 symspecs are examined by nested loops, and every symbol that matches a 1812 symspec is inserted into the appropriate syms table. This is done 1813 twice, once to count the size of each required symbol table, and again 1814 to build the tables, which have been malloced between passes. From now 1815 on, to determine whether a symbol is on an include or exclude symspec 1816 list, `gprof' simply uses its standard symbol lookup routine on the 1817 appropriate table in the `syms' array. 1818 1819 Now the profile data file(s) themselves are read 1820 (`gmon_io.c:gmon_out_read'), first by checking for a new-style 1821 `gmon.out' header, then assuming this is an old-style BSD `gmon.out' if 1822 the magic number test failed. 1823 1824 New-style histogram records are read by `hist.c:hist_read_rec'. For 1825 the first histogram record, allocate a memory array to hold all the 1826 bins, and read them in. When multiple profile data files (or files 1827 with multiple histogram records) are read, the memory ranges of each 1828 pair of histogram records must be either equal, or non-overlapping. 1829 For each pair of histogram records, the resolution (memory region size 1830 divided by the number of bins) must be the same. The time unit must be 1831 the same for all histogram records. If the above containts are met, all 1832 histograms for the same memory range are merged. 1833 1834 As each call graph record is read (`call_graph.c:cg_read_rec'), the 1835 parent and child addresses are matched to symbol table entries, and a 1836 call graph arc is created by `cg_arcs.c:arc_add', unless the arc fails 1837 a symspec check against INCL_ARCS/EXCL_ARCS. As each arc is added, a 1838 linked list is maintained of the parent's child arcs, and of the child's 1839 parent arcs. Both the child's call count and the arc's call count are 1840 incremented by the record's call count. 1841 1842 Basic-block records are read (`basic_blocks.c:bb_read_rec'), but 1843 only if line-by-line profiling has been selected. Each basic-block 1844 address is matched to a corresponding line symbol in the symbol table, 1845 and an entry made in the symbol's bb_addr and bb_calls arrays. Again, 1846 if multiple basic-block records are present for the same address, the 1847 call counts are cumulative. 1848 1849 A gmon.sum file is dumped, if requested (`gmon_io.c:gmon_out_write'). 1850 1851 If histograms were present in the data files, assign them to symbols 1852 (`hist.c:hist_assign_samples') by iterating over all the sample bins 1853 and assigning them to symbols. Since the symbol table is sorted in 1854 order of ascending memory addresses, we can simple follow along in the 1855 symbol table as we make our pass over the sample bins. This step 1856 includes a symspec check against INCL_FLAT/EXCL_FLAT. Depending on the 1857 histogram scale factor, a sample bin may span multiple symbols, in 1858 which case a fraction of the sample count is allocated to each symbol, 1859 proportional to the degree of overlap. This effect is rare for normal 1860 profiling, but overlaps are more common during line-by-line profiling, 1861 and can cause each of two adjacent lines to be credited with half a 1862 hit, for example. 1863 1864 If call graph data is present, `cg_arcs.c:cg_assemble' is called. 1865 First, if `-c' was specified, a machine-dependent routine (`find_call') 1866 scans through each symbol's machine code, looking for subroutine call 1867 instructions, and adding them to the call graph with a zero call count. 1868 A topological sort is performed by depth-first numbering all the 1869 symbols (`cg_dfn.c:cg_dfn'), so that children are always numbered less 1870 than their parents, then making a array of pointers into the symbol 1871 table and sorting it into numerical order, which is reverse topological 1872 order (children appear before parents). Cycles are also detected at 1873 this point, all members of which are assigned the same topological 1874 number. Two passes are now made through this sorted array of symbol 1875 pointers. The first pass, from end to beginning (parents to children), 1876 computes the fraction of child time to propagate to each parent and a 1877 print flag. The print flag reflects symspec handling of 1878 INCL_GRAPH/EXCL_GRAPH, with a parent's include or exclude (print or no 1879 print) property being propagated to its children, unless they 1880 themselves explicitly appear in INCL_GRAPH or EXCL_GRAPH. A second 1881 pass, from beginning to end (children to parents) actually propagates 1882 the timings along the call graph, subject to a check against 1883 INCL_TIME/EXCL_TIME. With the print flag, fractions, and timings now 1884 stored in the symbol structures, the topological sort array is now 1885 discarded, and a new array of pointers is assembled, this time sorted 1886 by propagated time. 1887 1888 Finally, print the various outputs the user requested, which is now 1889 fairly straightforward. The call graph (`cg_print.c:cg_print') and 1890 flat profile (`hist.c:hist_print') are regurgitations of values already 1891 computed. The annotated source listing 1892 (`basic_blocks.c:print_annotated_source') uses basic-block information, 1893 if present, to label each line of code with call counts, otherwise only 1894 the function call counts are presented. 1895 1896 The function ordering code is marginally well documented in the 1897 source code itself (`cg_print.c'). Basically, the functions with the 1898 most use and the most parents are placed first, followed by other 1899 functions with the most use, followed by lower use functions, followed 1900 by unused functions at the end. 1901 1902 1903 File: gprof.info, Node: Debugging, Prev: Internals, Up: Details 1904 1905 9.4 Debugging `gprof' 1906 ===================== 1907 1908 If `gprof' was compiled with debugging enabled, the `-d' option 1909 triggers debugging output (to stdout) which can be helpful in 1910 understanding its operation. The debugging number specified is 1911 interpreted as a sum of the following options: 1912 1913 2 - Topological sort 1914 Monitor depth-first numbering of symbols during call graph analysis 1915 1916 4 - Cycles 1917 Shows symbols as they are identified as cycle heads 1918 1919 16 - Tallying 1920 As the call graph arcs are read, show each arc and how the total 1921 calls to each function are tallied 1922 1923 32 - Call graph arc sorting 1924 Details sorting individual parents/children within each call graph 1925 entry 1926 1927 64 - Reading histogram and call graph records 1928 Shows address ranges of histograms as they are read, and each call 1929 graph arc 1930 1931 128 - Symbol table 1932 Reading, classifying, and sorting the symbol table from the object 1933 file. For line-by-line profiling (`-l' option), also shows line 1934 numbers being assigned to memory addresses. 1935 1936 256 - Static call graph 1937 Trace operation of `-c' option 1938 1939 512 - Symbol table and arc table lookups 1940 Detail operation of lookup routines 1941 1942 1024 - Call graph propagation 1943 Shows how function times are propagated along the call graph 1944 1945 2048 - Basic-blocks 1946 Shows basic-block records as they are read from profile data (only 1947 meaningful with `-l' option) 1948 1949 4096 - Symspecs 1950 Shows symspec-to-symbol pattern matching operation 1951 1952 8192 - Annotate source 1953 Tracks operation of `-A' option 1954 1955 1956 File: gprof.info, Node: GNU Free Documentation License, Prev: Details, Up: Top 1957 1958 Appendix A GNU Free Documentation License 1959 ***************************************** 1960 1961 Version 1.3, 3 November 2008 1962 1963 Copyright (C) 2000, 2001, 2002, 2007, 2008 Free Software Foundation, Inc. 1964 `http://fsf.org/' 1965 1966 Everyone is permitted to copy and distribute verbatim copies 1967 of this license document, but changing it is not allowed. 1968 1969 0. PREAMBLE 1970 1971 The purpose of this License is to make a manual, textbook, or other 1972 functional and useful document "free" in the sense of freedom: to 1973 assure everyone the effective freedom to copy and redistribute it, 1974 with or without modifying it, either commercially or 1975 noncommercially. Secondarily, this License preserves for the 1976 author and publisher a way to get credit for their work, while not 1977 being considered responsible for modifications made by others. 1978 1979 This License is a kind of "copyleft", which means that derivative 1980 works of the document must themselves be free in the same sense. 1981 It complements the GNU General Public License, which is a copyleft 1982 license designed for free software. 1983 1984 We have designed this License in order to use it for manuals for 1985 free software, because free software needs free documentation: a 1986 free program should come with manuals providing the same freedoms 1987 that the software does. But this License is not limited to 1988 software manuals; it can be used for any textual work, regardless 1989 of subject matter or whether it is published as a printed book. 1990 We recommend this License principally for works whose purpose is 1991 instruction or reference. 1992 1993 1. APPLICABILITY AND DEFINITIONS 1994 1995 This License applies to any manual or other work, in any medium, 1996 that contains a notice placed by the copyright holder saying it 1997 can be distributed under the terms of this License. Such a notice 1998 grants a world-wide, royalty-free license, unlimited in duration, 1999 to use that work under the conditions stated herein. The 2000 "Document", below, refers to any such manual or work. Any member 2001 of the public is a licensee, and is addressed as "you". You 2002 accept the license if you copy, modify or distribute the work in a 2003 way requiring permission under copyright law. 2004 2005 A "Modified Version" of the Document means any work containing the 2006 Document or a portion of it, either copied verbatim, or with 2007 modifications and/or translated into another language. 2008 2009 A "Secondary Section" is a named appendix or a front-matter section 2010 of the Document that deals exclusively with the relationship of the 2011 publishers or authors of the Document to the Document's overall 2012 subject (or to related matters) and contains nothing that could 2013 fall directly within that overall subject. (Thus, if the Document 2014 is in part a textbook of mathematics, a Secondary Section may not 2015 explain any mathematics.) The relationship could be a matter of 2016 historical connection with the subject or with related matters, or 2017 of legal, commercial, philosophical, ethical or political position 2018 regarding them. 2019 2020 The "Invariant Sections" are certain Secondary Sections whose 2021 titles are designated, as being those of Invariant Sections, in 2022 the notice that says that the Document is released under this 2023 License. If a section does not fit the above definition of 2024 Secondary then it is not allowed to be designated as Invariant. 2025 The Document may contain zero Invariant Sections. If the Document 2026 does not identify any Invariant Sections then there are none. 2027 2028 The "Cover Texts" are certain short passages of text that are 2029 listed, as Front-Cover Texts or Back-Cover Texts, in the notice 2030 that says that the Document is released under this License. A 2031 Front-Cover Text may be at most 5 words, and a Back-Cover Text may 2032 be at most 25 words. 2033 2034 A "Transparent" copy of the Document means a machine-readable copy, 2035 represented in a format whose specification is available to the 2036 general public, that is suitable for revising the document 2037 straightforwardly with generic text editors or (for images 2038 composed of pixels) generic paint programs or (for drawings) some 2039 widely available drawing editor, and that is suitable for input to 2040 text formatters or for automatic translation to a variety of 2041 formats suitable for input to text formatters. A copy made in an 2042 otherwise Transparent file format whose markup, or absence of 2043 markup, has been arranged to thwart or discourage subsequent 2044 modification by readers is not Transparent. An image format is 2045 not Transparent if used for any substantial amount of text. A 2046 copy that is not "Transparent" is called "Opaque". 2047 2048 Examples of suitable formats for Transparent copies include plain 2049 ASCII without markup, Texinfo input format, LaTeX input format, 2050 SGML or XML using a publicly available DTD, and 2051 standard-conforming simple HTML, PostScript or PDF designed for 2052 human modification. Examples of transparent image formats include 2053 PNG, XCF and JPG. Opaque formats include proprietary formats that 2054 can be read and edited only by proprietary word processors, SGML or 2055 XML for which the DTD and/or processing tools are not generally 2056 available, and the machine-generated HTML, PostScript or PDF 2057 produced by some word processors for output purposes only. 2058 2059 The "Title Page" means, for a printed book, the title page itself, 2060 plus such following pages as are needed to hold, legibly, the 2061 material this License requires to appear in the title page. For 2062 works in formats which do not have any title page as such, "Title 2063 Page" means the text near the most prominent appearance of the 2064 work's title, preceding the beginning of the body of the text. 2065 2066 The "publisher" means any person or entity that distributes copies 2067 of the Document to the public. 2068 2069 A section "Entitled XYZ" means a named subunit of the Document 2070 whose title either is precisely XYZ or contains XYZ in parentheses 2071 following text that translates XYZ in another language. (Here XYZ 2072 stands for a specific section name mentioned below, such as 2073 "Acknowledgements", "Dedications", "Endorsements", or "History".) 2074 To "Preserve the Title" of such a section when you modify the 2075 Document means that it remains a section "Entitled XYZ" according 2076 to this definition. 2077 2078 The Document may include Warranty Disclaimers next to the notice 2079 which states that this License applies to the Document. These 2080 Warranty Disclaimers are considered to be included by reference in 2081 this License, but only as regards disclaiming warranties: any other 2082 implication that these Warranty Disclaimers may have is void and 2083 has no effect on the meaning of this License. 2084 2085 2. VERBATIM COPYING 2086 2087 You may copy and distribute the Document in any medium, either 2088 commercially or noncommercially, provided that this License, the 2089 copyright notices, and the license notice saying this License 2090 applies to the Document are reproduced in all copies, and that you 2091 add no other conditions whatsoever to those of this License. You 2092 may not use technical measures to obstruct or control the reading 2093 or further copying of the copies you make or distribute. However, 2094 you may accept compensation in exchange for copies. If you 2095 distribute a large enough number of copies you must also follow 2096 the conditions in section 3. 2097 2098 You may also lend copies, under the same conditions stated above, 2099 and you may publicly display copies. 2100 2101 3. COPYING IN QUANTITY 2102 2103 If you publish printed copies (or copies in media that commonly 2104 have printed covers) of the Document, numbering more than 100, and 2105 the Document's license notice requires Cover Texts, you must 2106 enclose the copies in covers that carry, clearly and legibly, all 2107 these Cover Texts: Front-Cover Texts on the front cover, and 2108 Back-Cover Texts on the back cover. Both covers must also clearly 2109 and legibly identify you as the publisher of these copies. The 2110 front cover must present the full title with all words of the 2111 title equally prominent and visible. You may add other material 2112 on the covers in addition. Copying with changes limited to the 2113 covers, as long as they preserve the title of the Document and 2114 satisfy these conditions, can be treated as verbatim copying in 2115 other respects. 2116 2117 If the required texts for either cover are too voluminous to fit 2118 legibly, you should put the first ones listed (as many as fit 2119 reasonably) on the actual cover, and continue the rest onto 2120 adjacent pages. 2121 2122 If you publish or distribute Opaque copies of the Document 2123 numbering more than 100, you must either include a 2124 machine-readable Transparent copy along with each Opaque copy, or 2125 state in or with each Opaque copy a computer-network location from 2126 which the general network-using public has access to download 2127 using public-standard network protocols a complete Transparent 2128 copy of the Document, free of added material. If you use the 2129 latter option, you must take reasonably prudent steps, when you 2130 begin distribution of Opaque copies in quantity, to ensure that 2131 this Transparent copy will remain thus accessible at the stated 2132 location until at least one year after the last time you 2133 distribute an Opaque copy (directly or through your agents or 2134 retailers) of that edition to the public. 2135 2136 It is requested, but not required, that you contact the authors of 2137 the Document well before redistributing any large number of 2138 copies, to give them a chance to provide you with an updated 2139 version of the Document. 2140 2141 4. MODIFICATIONS 2142 2143 You may copy and distribute a Modified Version of the Document 2144 under the conditions of sections 2 and 3 above, provided that you 2145 release the Modified Version under precisely this License, with 2146 the Modified Version filling the role of the Document, thus 2147 licensing distribution and modification of the Modified Version to 2148 whoever possesses a copy of it. In addition, you must do these 2149 things in the Modified Version: 2150 2151 A. Use in the Title Page (and on the covers, if any) a title 2152 distinct from that of the Document, and from those of 2153 previous versions (which should, if there were any, be listed 2154 in the History section of the Document). You may use the 2155 same title as a previous version if the original publisher of 2156 that version gives permission. 2157 2158 B. List on the Title Page, as authors, one or more persons or 2159 entities responsible for authorship of the modifications in 2160 the Modified Version, together with at least five of the 2161 principal authors of the Document (all of its principal 2162 authors, if it has fewer than five), unless they release you 2163 from this requirement. 2164 2165 C. State on the Title page the name of the publisher of the 2166 Modified Version, as the publisher. 2167 2168 D. Preserve all the copyright notices of the Document. 2169 2170 E. Add an appropriate copyright notice for your modifications 2171 adjacent to the other copyright notices. 2172 2173 F. Include, immediately after the copyright notices, a license 2174 notice giving the public permission to use the Modified 2175 Version under the terms of this License, in the form shown in 2176 the Addendum below. 2177 2178 G. Preserve in that license notice the full lists of Invariant 2179 Sections and required Cover Texts given in the Document's 2180 license notice. 2181 2182 H. Include an unaltered copy of this License. 2183 2184 I. Preserve the section Entitled "History", Preserve its Title, 2185 and add to it an item stating at least the title, year, new 2186 authors, and publisher of the Modified Version as given on 2187 the Title Page. If there is no section Entitled "History" in 2188 the Document, create one stating the title, year, authors, 2189 and publisher of the Document as given on its Title Page, 2190 then add an item describing the Modified Version as stated in 2191 the previous sentence. 2192 2193 J. Preserve the network location, if any, given in the Document 2194 for public access to a Transparent copy of the Document, and 2195 likewise the network locations given in the Document for 2196 previous versions it was based on. These may be placed in 2197 the "History" section. You may omit a network location for a 2198 work that was published at least four years before the 2199 Document itself, or if the original publisher of the version 2200 it refers to gives permission. 2201 2202 K. For any section Entitled "Acknowledgements" or "Dedications", 2203 Preserve the Title of the section, and preserve in the 2204 section all the substance and tone of each of the contributor 2205 acknowledgements and/or dedications given therein. 2206 2207 L. Preserve all the Invariant Sections of the Document, 2208 unaltered in their text and in their titles. Section numbers 2209 or the equivalent are not considered part of the section 2210 titles. 2211 2212 M. Delete any section Entitled "Endorsements". Such a section 2213 may not be included in the Modified Version. 2214 2215 N. Do not retitle any existing section to be Entitled 2216 "Endorsements" or to conflict in title with any Invariant 2217 Section. 2218 2219 O. Preserve any Warranty Disclaimers. 2220 2221 If the Modified Version includes new front-matter sections or 2222 appendices that qualify as Secondary Sections and contain no 2223 material copied from the Document, you may at your option 2224 designate some or all of these sections as invariant. To do this, 2225 add their titles to the list of Invariant Sections in the Modified 2226 Version's license notice. These titles must be distinct from any 2227 other section titles. 2228 2229 You may add a section Entitled "Endorsements", provided it contains 2230 nothing but endorsements of your Modified Version by various 2231 parties--for example, statements of peer review or that the text 2232 has been approved by an organization as the authoritative 2233 definition of a standard. 2234 2235 You may add a passage of up to five words as a Front-Cover Text, 2236 and a passage of up to 25 words as a Back-Cover Text, to the end 2237 of the list of Cover Texts in the Modified Version. Only one 2238 passage of Front-Cover Text and one of Back-Cover Text may be 2239 added by (or through arrangements made by) any one entity. If the 2240 Document already includes a cover text for the same cover, 2241 previously added by you or by arrangement made by the same entity 2242 you are acting on behalf of, you may not add another; but you may 2243 replace the old one, on explicit permission from the previous 2244 publisher that added the old one. 2245 2246 The author(s) and publisher(s) of the Document do not by this 2247 License give permission to use their names for publicity for or to 2248 assert or imply endorsement of any Modified Version. 2249 2250 5. COMBINING DOCUMENTS 2251 2252 You may combine the Document with other documents released under 2253 this License, under the terms defined in section 4 above for 2254 modified versions, provided that you include in the combination 2255 all of the Invariant Sections of all of the original documents, 2256 unmodified, and list them all as Invariant Sections of your 2257 combined work in its license notice, and that you preserve all 2258 their Warranty Disclaimers. 2259 2260 The combined work need only contain one copy of this License, and 2261 multiple identical Invariant Sections may be replaced with a single 2262 copy. If there are multiple Invariant Sections with the same name 2263 but different contents, make the title of each such section unique 2264 by adding at the end of it, in parentheses, the name of the 2265 original author or publisher of that section if known, or else a 2266 unique number. Make the same adjustment to the section titles in 2267 the list of Invariant Sections in the license notice of the 2268 combined work. 2269 2270 In the combination, you must combine any sections Entitled 2271 "History" in the various original documents, forming one section 2272 Entitled "History"; likewise combine any sections Entitled 2273 "Acknowledgements", and any sections Entitled "Dedications". You 2274 must delete all sections Entitled "Endorsements." 2275 2276 6. COLLECTIONS OF DOCUMENTS 2277 2278 You may make a collection consisting of the Document and other 2279 documents released under this License, and replace the individual 2280 copies of this License in the various documents with a single copy 2281 that is included in the collection, provided that you follow the 2282 rules of this License for verbatim copying of each of the 2283 documents in all other respects. 2284 2285 You may extract a single document from such a collection, and 2286 distribute it individually under this License, provided you insert 2287 a copy of this License into the extracted document, and follow 2288 this License in all other respects regarding verbatim copying of 2289 that document. 2290 2291 7. AGGREGATION WITH INDEPENDENT WORKS 2292 2293 A compilation of the Document or its derivatives with other 2294 separate and independent documents or works, in or on a volume of 2295 a storage or distribution medium, is called an "aggregate" if the 2296 copyright resulting from the compilation is not used to limit the 2297 legal rights of the compilation's users beyond what the individual 2298 works permit. When the Document is included in an aggregate, this 2299 License does not apply to the other works in the aggregate which 2300 are not themselves derivative works of the Document. 2301 2302 If the Cover Text requirement of section 3 is applicable to these 2303 copies of the Document, then if the Document is less than one half 2304 of the entire aggregate, the Document's Cover Texts may be placed 2305 on covers that bracket the Document within the aggregate, or the 2306 electronic equivalent of covers if the Document is in electronic 2307 form. Otherwise they must appear on printed covers that bracket 2308 the whole aggregate. 2309 2310 8. TRANSLATION 2311 2312 Translation is considered a kind of modification, so you may 2313 distribute translations of the Document under the terms of section 2314 4. Replacing Invariant Sections with translations requires special 2315 permission from their copyright holders, but you may include 2316 translations of some or all Invariant Sections in addition to the 2317 original versions of these Invariant Sections. You may include a 2318 translation of this License, and all the license notices in the 2319 Document, and any Warranty Disclaimers, provided that you also 2320 include the original English version of this License and the 2321 original versions of those notices and disclaimers. In case of a 2322 disagreement between the translation and the original version of 2323 this License or a notice or disclaimer, the original version will 2324 prevail. 2325 2326 If a section in the Document is Entitled "Acknowledgements", 2327 "Dedications", or "History", the requirement (section 4) to 2328 Preserve its Title (section 1) will typically require changing the 2329 actual title. 2330 2331 9. TERMINATION 2332 2333 You may not copy, modify, sublicense, or distribute the Document 2334 except as expressly provided under this License. Any attempt 2335 otherwise to copy, modify, sublicense, or distribute it is void, 2336 and will automatically terminate your rights under this License. 2337 2338 However, if you cease all violation of this License, then your 2339 license from a particular copyright holder is reinstated (a) 2340 provisionally, unless and until the copyright holder explicitly 2341 and finally terminates your license, and (b) permanently, if the 2342 copyright holder fails to notify you of the violation by some 2343 reasonable means prior to 60 days after the cessation. 2344 2345 Moreover, your license from a particular copyright holder is 2346 reinstated permanently if the copyright holder notifies you of the 2347 violation by some reasonable means, this is the first time you have 2348 received notice of violation of this License (for any work) from 2349 that copyright holder, and you cure the violation prior to 30 days 2350 after your receipt of the notice. 2351 2352 Termination of your rights under this section does not terminate 2353 the licenses of parties who have received copies or rights from 2354 you under this License. If your rights have been terminated and 2355 not permanently reinstated, receipt of a copy of some or all of 2356 the same material does not give you any rights to use it. 2357 2358 10. FUTURE REVISIONS OF THIS LICENSE 2359 2360 The Free Software Foundation may publish new, revised versions of 2361 the GNU Free Documentation License from time to time. Such new 2362 versions will be similar in spirit to the present version, but may 2363 differ in detail to address new problems or concerns. See 2364 `http://www.gnu.org/copyleft/'. 2365 2366 Each version of the License is given a distinguishing version 2367 number. If the Document specifies that a particular numbered 2368 version of this License "or any later version" applies to it, you 2369 have the option of following the terms and conditions either of 2370 that specified version or of any later version that has been 2371 published (not as a draft) by the Free Software Foundation. If 2372 the Document does not specify a version number of this License, 2373 you may choose any version ever published (not as a draft) by the 2374 Free Software Foundation. If the Document specifies that a proxy 2375 can decide which future versions of this License can be used, that 2376 proxy's public statement of acceptance of a version permanently 2377 authorizes you to choose that version for the Document. 2378 2379 11. RELICENSING 2380 2381 "Massive Multiauthor Collaboration Site" (or "MMC Site") means any 2382 World Wide Web server that publishes copyrightable works and also 2383 provides prominent facilities for anybody to edit those works. A 2384 public wiki that anybody can edit is an example of such a server. 2385 A "Massive Multiauthor Collaboration" (or "MMC") contained in the 2386 site means any set of copyrightable works thus published on the MMC 2387 site. 2388 2389 "CC-BY-SA" means the Creative Commons Attribution-Share Alike 3.0 2390 license published by Creative Commons Corporation, a not-for-profit 2391 corporation with a principal place of business in San Francisco, 2392 California, as well as future copyleft versions of that license 2393 published by that same organization. 2394 2395 "Incorporate" means to publish or republish a Document, in whole or 2396 in part, as part of another Document. 2397 2398 An MMC is "eligible for relicensing" if it is licensed under this 2399 License, and if all works that were first published under this 2400 License somewhere other than this MMC, and subsequently 2401 incorporated in whole or in part into the MMC, (1) had no cover 2402 texts or invariant sections, and (2) were thus incorporated prior 2403 to November 1, 2008. 2404 2405 The operator of an MMC Site may republish an MMC contained in the 2406 site under CC-BY-SA on the same site at any time before August 1, 2407 2009, provided the MMC is eligible for relicensing. 2408 2409 2410 ADDENDUM: How to use this License for your documents 2411 ==================================================== 2412 2413 To use this License in a document you have written, include a copy of 2414 the License in the document and put the following copyright and license 2415 notices just after the title page: 2416 2417 Copyright (C) YEAR YOUR NAME. 2418 Permission is granted to copy, distribute and/or modify this document 2419 under the terms of the GNU Free Documentation License, Version 1.3 2420 or any later version published by the Free Software Foundation; 2421 with no Invariant Sections, no Front-Cover Texts, and no Back-Cover 2422 Texts. A copy of the license is included in the section entitled ``GNU 2423 Free Documentation License''. 2424 2425 If you have Invariant Sections, Front-Cover Texts and Back-Cover 2426 Texts, replace the "with...Texts." line with this: 2427 2428 with the Invariant Sections being LIST THEIR TITLES, with 2429 the Front-Cover Texts being LIST, and with the Back-Cover Texts 2430 being LIST. 2431 2432 If you have Invariant Sections without Cover Texts, or some other 2433 combination of the three, merge those two alternatives to suit the 2434 situation. 2435 2436 If your document contains nontrivial examples of program code, we 2437 recommend releasing these examples in parallel under your choice of 2438 free software license, such as the GNU General Public License, to 2439 permit their use in free software. 2440 2441 2442 2443 Tag Table: 2444 Node: Top838 2445 Node: Introduction2161 2446 Node: Compiling4653 2447 Node: Executing8709 2448 Node: Invoking11497 2449 Node: Output Options12912 2450 Node: Analysis Options20001 2451 Node: Miscellaneous Options23699 2452 Node: Deprecated Options24954 2453 Node: Symspecs27023 2454 Node: Output28849 2455 Node: Flat Profile29889 2456 Node: Call Graph34842 2457 Node: Primary38074 2458 Node: Callers40662 2459 Node: Subroutines42779 2460 Node: Cycles44620 2461 Node: Line-by-line51397 2462 Node: Annotated Source55470 2463 Node: Inaccuracy58469 2464 Node: Sampling Error58727 2465 Node: Assumptions61631 2466 Node: How do I?63101 2467 Node: Incompatibilities64655 2468 Node: Details66149 2469 Node: Implementation66542 2470 Node: File Format72439 2471 Node: Internals76729 2472 Node: Debugging85224 2473 Node: GNU Free Documentation License86825 2474 2475 End Tag Table 2476
