1 This is gprof.info, produced by makeinfo version 4.8 from
2 /home/dougkwan/android-tool-src/toolchain/android-toolchain/binutils-2.19/gprof/gprof.texi.
5 * gprof: (gprof). Profiling your program's execution
8 This file documents the gprof profiler of the GNU system.
10 Copyright (C) 1988, 92, 97, 98, 99, 2000, 2001, 2003, 2007 Free
11 Software Foundation, Inc.
13 Permission is granted to copy, distribute and/or modify this document
14 under the terms of the GNU Free Documentation License, Version 1.1 or
15 any later version published by the Free Software Foundation; with no
16 Invariant Sections, with no Front-Cover Texts, and with no Back-Cover
17 Texts. A copy of the license is included in the section entitled "GNU
18 Free Documentation License".
21 File: gprof.info, Node: Top, Next: Introduction, Up: (dir)
23 Profiling a Program: Where Does It Spend Its Time?
24 **************************************************
26 This manual describes the GNU profiler, `gprof', and how you can use it
27 to determine which parts of a program are taking most of the execution
28 time. We assume that you know how to write, compile, and execute
29 programs. GNU `gprof' was written by Jay Fenlason.
31 This manual is for `gprof' (GNU Binutils) version 2.19.
33 This document is distributed under the terms of the GNU Free
34 Documentation License. A copy of the license is included in the
35 section entitled "GNU Free Documentation License".
39 * Introduction:: What profiling means, and why it is useful.
41 * Compiling:: How to compile your program for profiling.
42 * Executing:: Executing your program to generate profile data
43 * Invoking:: How to run `gprof', and its options
45 * Output:: Interpreting `gprof''s output
47 * Inaccuracy:: Potential problems you should be aware of
48 * How do I?:: Answers to common questions
49 * Incompatibilities:: (between GNU `gprof' and Unix `gprof'.)
50 * Details:: Details of how profiling is done
51 * GNU Free Documentation License:: GNU Free Documentation License
54 File: gprof.info, Node: Introduction, Next: Compiling, Prev: Top, Up: Top
56 1 Introduction to Profiling
57 ***************************
59 Profiling allows you to learn where your program spent its time and
60 which functions called which other functions while it was executing.
61 This information can show you which pieces of your program are slower
62 than you expected, and might be candidates for rewriting to make your
63 program execute faster. It can also tell you which functions are being
64 called more or less often than you expected. This may help you spot
65 bugs that had otherwise been unnoticed.
67 Since the profiler uses information collected during the actual
68 execution of your program, it can be used on programs that are too
69 large or too complex to analyze by reading the source. However, how
70 your program is run will affect the information that shows up in the
71 profile data. If you don't use some feature of your program while it
72 is being profiled, no profile information will be generated for that
75 Profiling has several steps:
77 * You must compile and link your program with profiling enabled.
78 *Note Compiling a Program for Profiling: Compiling.
80 * You must execute your program to generate a profile data file.
81 *Note Executing the Program: Executing.
83 * You must run `gprof' to analyze the profile data. *Note `gprof'
84 Command Summary: Invoking.
86 The next three chapters explain these steps in greater detail.
88 Several forms of output are available from the analysis.
90 The "flat profile" shows how much time your program spent in each
91 function, and how many times that function was called. If you simply
92 want to know which functions burn most of the cycles, it is stated
93 concisely here. *Note The Flat Profile: Flat Profile.
95 The "call graph" shows, for each function, which functions called
96 it, which other functions it called, and how many times. There is also
97 an estimate of how much time was spent in the subroutines of each
98 function. This can suggest places where you might try to eliminate
99 function calls that use a lot of time. *Note The Call Graph: Call
102 The "annotated source" listing is a copy of the program's source
103 code, labeled with the number of times each line of the program was
104 executed. *Note The Annotated Source Listing: Annotated Source.
106 To better understand how profiling works, you may wish to read a
107 description of its implementation. *Note Implementation of Profiling:
111 File: gprof.info, Node: Compiling, Next: Executing, Prev: Introduction, Up: Top
113 2 Compiling a Program for Profiling
114 ***********************************
116 The first step in generating profile information for your program is to
117 compile and link it with profiling enabled.
119 To compile a source file for profiling, specify the `-pg' option when
120 you run the compiler. (This is in addition to the options you normally
123 To link the program for profiling, if you use a compiler such as `cc'
124 to do the linking, simply specify `-pg' in addition to your usual
125 options. The same option, `-pg', alters either compilation or linking
126 to do what is necessary for profiling. Here are examples:
128 cc -g -c myprog.c utils.c -pg
129 cc -o myprog myprog.o utils.o -pg
131 The `-pg' option also works with a command that both compiles and
134 cc -o myprog myprog.c utils.c -g -pg
136 Note: The `-pg' option must be part of your compilation options as
137 well as your link options. If it is not then no call-graph data will
138 be gathered and when you run `gprof' you will get an error message like
141 gprof: gmon.out file is missing call-graph data
143 If you add the `-Q' switch to suppress the printing of the call
144 graph data you will still be able to see the time samples:
148 Each sample counts as 0.01 seconds.
149 % cumulative self self total
150 time seconds seconds calls Ts/call Ts/call name
151 44.12 0.07 0.07 zazLoop
153 20.59 0.17 0.04 bazMillion
155 If you run the linker `ld' directly instead of through a compiler
156 such as `cc', you may have to specify a profiling startup file
157 `gcrt0.o' as the first input file instead of the usual startup file
158 `crt0.o'. In addition, you would probably want to specify the
159 profiling C library, `libc_p.a', by writing `-lc_p' instead of the
160 usual `-lc'. This is not absolutely necessary, but doing this gives
161 you number-of-calls information for standard library functions such as
162 `read' and `open'. For example:
164 ld -o myprog /lib/gcrt0.o myprog.o utils.o -lc_p
166 If you compile only some of the modules of the program with `-pg',
167 you can still profile the program, but you won't get complete
168 information about the modules that were compiled without `-pg'. The
169 only information you get for the functions in those modules is the
170 total time spent in them; there is no record of how many times they
171 were called, or from where. This will not affect the flat profile
172 (except that the `calls' field for the functions will be blank), but
173 will greatly reduce the usefulness of the call graph.
175 If you wish to perform line-by-line profiling you should use the
176 `gcov' tool instead of `gprof'. See that tool's manual or info pages
177 for more details of how to do this.
179 Note, older versions of `gcc' produce line-by-line profiling
180 information that works with `gprof' rather than `gcov' so there is
181 still support for displaying this kind of information in `gprof'. *Note
182 Line-by-line Profiling: Line-by-line.
184 It also worth noting that `gcc' implements a
185 `-finstrument-functions' command line option which will insert calls to
186 special user supplied instrumentation routines at the entry and exit of
187 every function in their program. This can be used to implement an
188 alternative profiling scheme.
191 File: gprof.info, Node: Executing, Next: Invoking, Prev: Compiling, Up: Top
193 3 Executing the Program
194 ***********************
196 Once the program is compiled for profiling, you must run it in order to
197 generate the information that `gprof' needs. Simply run the program as
198 usual, using the normal arguments, file names, etc. The program should
199 run normally, producing the same output as usual. It will, however, run
200 somewhat slower than normal because of the time spent collecting and
201 writing the profile data.
203 The way you run the program--the arguments and input that you give
204 it--may have a dramatic effect on what the profile information shows.
205 The profile data will describe the parts of the program that were
206 activated for the particular input you use. For example, if the first
207 command you give to your program is to quit, the profile data will show
208 the time used in initialization and in cleanup, but not much else.
210 Your program will write the profile data into a file called
211 `gmon.out' just before exiting. If there is already a file called
212 `gmon.out', its contents are overwritten. There is currently no way to
213 tell the program to write the profile data under a different name, but
214 you can rename the file afterwards if you are concerned that it may be
217 In order to write the `gmon.out' file properly, your program must
218 exit normally: by returning from `main' or by calling `exit'. Calling
219 the low-level function `_exit' does not write the profile data, and
220 neither does abnormal termination due to an unhandled signal.
222 The `gmon.out' file is written in the program's _current working
223 directory_ at the time it exits. This means that if your program calls
224 `chdir', the `gmon.out' file will be left in the last directory your
225 program `chdir''d to. If you don't have permission to write in this
226 directory, the file is not written, and you will get an error message.
228 Older versions of the GNU profiling library may also write a file
229 called `bb.out'. This file, if present, contains an human-readable
230 listing of the basic-block execution counts. Unfortunately, the
231 appearance of a human-readable `bb.out' means the basic-block counts
232 didn't get written into `gmon.out'. The Perl script `bbconv.pl',
233 included with the `gprof' source distribution, will convert a `bb.out'
234 file into a format readable by `gprof'. Invoke it like this:
236 bbconv.pl < bb.out > BH-DATA
238 This translates the information in `bb.out' into a form that `gprof'
239 can understand. But you still need to tell `gprof' about the existence
240 of this translated information. To do that, include BB-DATA on the
241 `gprof' command line, _along with `gmon.out'_, like this:
243 gprof OPTIONS EXECUTABLE-FILE gmon.out BB-DATA [YET-MORE-PROFILE-DATA-FILES...] [> OUTFILE]
246 File: gprof.info, Node: Invoking, Next: Output, Prev: Executing, Up: Top
248 4 `gprof' Command Summary
249 *************************
251 After you have a profile data file `gmon.out', you can run `gprof' to
252 interpret the information in it. The `gprof' program prints a flat
253 profile and a call graph on standard output. Typically you would
254 redirect the output of `gprof' into a file with `>'.
256 You run `gprof' like this:
258 gprof OPTIONS [EXECUTABLE-FILE [PROFILE-DATA-FILES...]] [> OUTFILE]
260 Here square-brackets indicate optional arguments.
262 If you omit the executable file name, the file `a.out' is used. If
263 you give no profile data file name, the file `gmon.out' is used. If
264 any file is not in the proper format, or if the profile data file does
265 not appear to belong to the executable file, an error message is
268 You can give more than one profile data file by entering all their
269 names after the executable file name; then the statistics in all the
270 data files are summed together.
272 The order of these options does not matter.
276 * Output Options:: Controlling `gprof''s output style
277 * Analysis Options:: Controlling how `gprof' analyzes its data
278 * Miscellaneous Options::
279 * Deprecated Options:: Options you no longer need to use, but which
280 have been retained for compatibility
281 * Symspecs:: Specifying functions to include or exclude
284 File: gprof.info, Node: Output Options, Next: Analysis Options, Up: Invoking
289 These options specify which of several output formats `gprof' should
292 Many of these options take an optional "symspec" to specify
293 functions to be included or excluded. These options can be specified
294 multiple times, with different symspecs, to include or exclude sets of
295 symbols. *Note Symspecs: Symspecs.
297 Specifying any of these options overrides the default (`-p -q'),
298 which prints a flat profile and call graph analysis for all functions.
301 `--annotated-source[=SYMSPEC]'
302 The `-A' option causes `gprof' to print annotated source code. If
303 SYMSPEC is specified, print output only for matching symbols.
304 *Note The Annotated Source Listing: Annotated Source.
308 If the `-b' option is given, `gprof' doesn't print the verbose
309 blurbs that try to explain the meaning of all of the fields in the
310 tables. This is useful if you intend to print out the output, or
311 are tired of seeing the blurbs.
314 `--exec-counts[=SYMSPEC]'
315 The `-C' option causes `gprof' to print a tally of functions and
316 the number of times each was called. If SYMSPEC is specified,
317 print tally only for matching symbols.
319 If the profile data file contains basic-block count records,
320 specifying the `-l' option, along with `-C', will cause basic-block
321 execution counts to be tallied and displayed.
325 The `-i' option causes `gprof' to display summary information
326 about the profile data file(s) and then exit. The number of
327 histogram, call graph, and basic-block count records is displayed.
330 `--directory-path=DIRS'
331 The `-I' option specifies a list of search directories in which to
332 find source files. Environment variable GPROF_PATH can also be
333 used to convey this information. Used mostly for annotated source
337 `--no-annotated-source[=SYMSPEC]'
338 The `-J' option causes `gprof' not to print annotated source code.
339 If SYMSPEC is specified, `gprof' prints annotated source, but
340 excludes matching symbols.
344 Normally, source filenames are printed with the path component
345 suppressed. The `-L' option causes `gprof' to print the full
346 pathname of source filenames, which is determined from symbolic
347 debugging information in the image file and is relative to the
348 directory in which the compiler was invoked.
351 `--flat-profile[=SYMSPEC]'
352 The `-p' option causes `gprof' to print a flat profile. If
353 SYMSPEC is specified, print flat profile only for matching symbols.
354 *Note The Flat Profile: Flat Profile.
357 `--no-flat-profile[=SYMSPEC]'
358 The `-P' option causes `gprof' to suppress printing a flat profile.
359 If SYMSPEC is specified, `gprof' prints a flat profile, but
360 excludes matching symbols.
364 The `-q' option causes `gprof' to print the call graph analysis.
365 If SYMSPEC is specified, print call graph only for matching symbols
366 and their children. *Note The Call Graph: Call Graph.
369 `--no-graph[=SYMSPEC]'
370 The `-Q' option causes `gprof' to suppress printing the call graph.
371 If SYMSPEC is specified, `gprof' prints a call graph, but excludes
376 The `-t' option causes the NUM most active source lines in each
377 source file to be listed when source annotation is enabled. The
382 This option affects annotated source output only. Normally,
383 `gprof' prints annotated source files to standard-output. If this
384 option is specified, annotated source for a file named
385 `path/FILENAME' is generated in the file `FILENAME-ann'. If the
386 underlying file system would truncate `FILENAME-ann' so that it
387 overwrites the original `FILENAME', `gprof' generates annotated
388 source in the file `FILENAME.ann' instead (if the original file
389 name has an extension, that extension is _replaced_ with `.ann').
392 `--no-exec-counts[=SYMSPEC]'
393 The `-Z' option causes `gprof' not to print a tally of functions
394 and the number of times each was called. If SYMSPEC is specified,
395 print tally, but exclude matching symbols.
398 `--function-ordering'
399 The `--function-ordering' option causes `gprof' to print a
400 suggested function ordering for the program based on profiling
401 data. This option suggests an ordering which may improve paging,
402 tlb and cache behavior for the program on systems which support
403 arbitrary ordering of functions in an executable.
405 The exact details of how to force the linker to place functions in
406 a particular order is system dependent and out of the scope of this
410 `--file-ordering MAP_FILE'
411 The `--file-ordering' option causes `gprof' to print a suggested
412 .o link line ordering for the program based on profiling data.
413 This option suggests an ordering which may improve paging, tlb and
414 cache behavior for the program on systems which do not support
415 arbitrary ordering of functions in an executable.
417 Use of the `-a' argument is highly recommended with this option.
419 The MAP_FILE argument is a pathname to a file which provides
420 function name to object file mappings. The format of the file is
421 similar to the output of the program `nm'.
423 c-parse.o:00000000 T yyparse
424 c-parse.o:00000004 C yyerrflag
425 c-lang.o:00000000 T maybe_objc_method_name
426 c-lang.o:00000000 T print_lang_statistics
427 c-lang.o:00000000 T recognize_objc_keyword
428 c-decl.o:00000000 T print_lang_identifier
429 c-decl.o:00000000 T print_lang_type
432 To create a MAP_FILE with GNU `nm', type a command like `nm
433 --extern-only --defined-only -v --print-file-name program-name'.
437 The `-T' option causes `gprof' to print its output in
438 "traditional" BSD style.
442 Sets width of output lines to WIDTH. Currently only used when
443 printing the function index at the bottom of the call graph.
447 This option affects annotated source output only. By default,
448 only the lines at the beginning of a basic-block are annotated.
449 If this option is specified, every line in a basic-block is
450 annotated by repeating the annotation for the first line. This
451 behavior is similar to `tcov''s `-a'.
455 These options control whether C++ symbol names should be demangled
456 when printing output. The default is to demangle symbols. The
457 `--no-demangle' option may be used to turn off demangling.
458 Different compilers have different mangling styles. The optional
459 demangling style argument can be used to choose an appropriate
460 demangling style for your compiler.
463 File: gprof.info, Node: Analysis Options, Next: Miscellaneous Options, Prev: Output Options, Up: Invoking
470 The `-a' option causes `gprof' to suppress the printing of
471 statically declared (private) functions. (These are functions
472 whose names are not listed as global, and which are not visible
473 outside the file/function/block where they were defined.) Time
474 spent in these functions, calls to/from them, etc., will all be
475 attributed to the function that was loaded directly before it in
476 the executable file. This option affects both the flat profile
480 `--static-call-graph'
481 The `-c' option causes the call graph of the program to be
482 augmented by a heuristic which examines the text space of the
483 object file and identifies function calls in the binary machine
484 code. Since normal call graph records are only generated when
485 functions are entered, this option identifies children that could
486 have been called, but never were. Calls to functions that were
487 not compiled with profiling enabled are also identified, but only
488 if symbol table entries are present for them. Calls to dynamic
489 library routines are typically _not_ found by this option.
490 Parents or children identified via this heuristic are indicated in
491 the call graph with call counts of `0'.
494 `--ignore-non-functions'
495 The `-D' option causes `gprof' to ignore symbols which are not
496 known to be functions. This option will give more accurate
497 profile data on systems where it is supported (Solaris and HPUX for
501 The `-k' option allows you to delete from the call graph any arcs
502 from symbols matching symspec FROM to those matching symspec TO.
506 The `-l' option enables line-by-line profiling, which causes
507 histogram hits to be charged to individual source code lines,
508 instead of functions. This feature only works with programs
509 compiled by older versions of the `gcc' compiler. Newer versions
510 of `gcc' are designed to work with the `gcov' tool instead.
512 If the program was compiled with basic-block counting enabled,
513 this option will also identify how many times each line of code
514 was executed. While line-by-line profiling can help isolate where
515 in a large function a program is spending its time, it also
516 significantly increases the running time of `gprof', and magnifies
517 statistical inaccuracies. *Note Statistical Sampling Error:
522 This option affects execution count output only. Symbols that are
523 executed less than NUM times are suppressed.
527 The `-n' option causes `gprof', in its call graph analysis, to
528 only propagate times for symbols matching SYMSPEC.
532 The `-n' option causes `gprof', in its call graph analysis, not to
533 propagate times for symbols matching SYMSPEC.
536 `--display-unused-functions'
537 If you give the `-z' option, `gprof' will mention all functions in
538 the flat profile, even those that were never called, and that had
539 no time spent in them. This is useful in conjunction with the
540 `-c' option for discovering which routines were never called.
544 File: gprof.info, Node: Miscellaneous Options, Next: Deprecated Options, Prev: Analysis Options, Up: Invoking
546 4.3 Miscellaneous Options
547 =========================
551 The `-d NUM' option specifies debugging options. If NUM is not
552 specified, enable all debugging. *Note Debugging `gprof':
557 The `-h' option prints command line usage.
561 Selects the format of the profile data files. Recognized formats
562 are `auto' (the default), `bsd', `4.4bsd', `magic', and `prof'
567 The `-s' option causes `gprof' to summarize the information in the
568 profile data files it read in, and write out a profile data file
569 called `gmon.sum', which contains all the information from the
570 profile data files that `gprof' read in. The file `gmon.sum' may
571 be one of the specified input files; the effect of this is to
572 merge the data in the other input files into `gmon.sum'.
574 Eventually you can run `gprof' again without `-s' to analyze the
575 cumulative data in the file `gmon.sum'.
579 The `-v' flag causes `gprof' to print the current version number,
584 File: gprof.info, Node: Deprecated Options, Next: Symspecs, Prev: Miscellaneous Options, Up: Invoking
586 4.4 Deprecated Options
587 ======================
589 These options have been replaced with newer versions that use
593 The `-e FUNCTION' option tells `gprof' to not print information
594 about the function FUNCTION_NAME (and its children...) in the call
595 graph. The function will still be listed as a child of any
596 functions that call it, but its index number will be shown as
597 `[not printed]'. More than one `-e' option may be given; only one
598 FUNCTION_NAME may be indicated with each `-e' option.
601 The `-E FUNCTION' option works like the `-e' option, but time
602 spent in the function (and children who were not called from
603 anywhere else), will not be used to compute the
604 percentages-of-time for the call graph. More than one `-E' option
605 may be given; only one FUNCTION_NAME may be indicated with each
609 The `-f FUNCTION' option causes `gprof' to limit the call graph to
610 the function FUNCTION_NAME and its children (and their
611 children...). More than one `-f' option may be given; only one
612 FUNCTION_NAME may be indicated with each `-f' option.
615 The `-F FUNCTION' option works like the `-f' option, but only time
616 spent in the function and its children (and their children...)
617 will be used to determine total-time and percentages-of-time for
618 the call graph. More than one `-F' option may be given; only one
619 FUNCTION_NAME may be indicated with each `-F' option. The `-F'
620 option overrides the `-E' option.
623 Note that only one function can be specified with each `-e', `-E',
624 `-f' or `-F' option. To specify more than one function, use multiple
625 options. For example, this command:
627 gprof -e boring -f foo -f bar myprogram > gprof.output
629 lists in the call graph all functions that were reached from either
630 `foo' or `bar' and were not reachable from `boring'.
633 File: gprof.info, Node: Symspecs, Prev: Deprecated Options, Up: Invoking
638 Many of the output options allow functions to be included or excluded
639 using "symspecs" (symbol specifications), which observe the following
642 filename_containing_a_dot
643 | funcname_not_containing_a_dot
645 | ( [ any_filename ] `:' ( any_funcname | linenumber ) )
647 Here are some sample symspecs:
650 Selects everything in file `main.c'--the dot in the string tells
651 `gprof' to interpret the string as a filename, rather than as a
652 function name. To select a file whose name does not contain a
653 dot, a trailing colon should be specified. For example, `odd:' is
654 interpreted as the file named `odd'.
657 Selects all functions named `main'.
659 Note that there may be multiple instances of the same function name
660 because some of the definitions may be local (i.e., static).
661 Unless a function name is unique in a program, you must use the
662 colon notation explained below to specify a function from a
663 specific source file.
665 Sometimes, function names contain dots. In such cases, it is
666 necessary to add a leading colon to the name. For example,
667 `:.mul' selects function `.mul'.
669 In some object file formats, symbols have a leading underscore.
670 `gprof' will normally not print these underscores. When you name a
671 symbol in a symspec, you should type it exactly as `gprof' prints
672 it in its output. For example, if the compiler produces a symbol
673 `_main' from your `main' function, `gprof' still prints it as
674 `main' in its output, so you should use `main' in symspecs.
677 Selects function `main' in file `main.c'.
680 Selects line 134 in file `main.c'.
683 File: gprof.info, Node: Output, Next: Inaccuracy, Prev: Invoking, Up: Top
685 5 Interpreting `gprof''s Output
686 *******************************
688 `gprof' can produce several different output styles, the most important
689 of which are described below. The simplest output styles (file
690 information, execution count, and function and file ordering) are not
691 described here, but are documented with the respective options that
692 trigger them. *Note Output Options: Output Options.
696 * Flat Profile:: The flat profile shows how much time was spent
697 executing directly in each function.
698 * Call Graph:: The call graph shows which functions called which
699 others, and how much time each function used
700 when its subroutine calls are included.
701 * Line-by-line:: `gprof' can analyze individual source code lines
702 * Annotated Source:: The annotated source listing displays source code
703 labeled with execution counts
706 File: gprof.info, Node: Flat Profile, Next: Call Graph, Up: Output
711 The "flat profile" shows the total amount of time your program spent
712 executing each function. Unless the `-z' option is given, functions
713 with no apparent time spent in them, and no apparent calls to them, are
714 not mentioned. Note that if a function was not compiled for profiling,
715 and didn't run long enough to show up on the program counter histogram,
716 it will be indistinguishable from a function that was never called.
718 This is part of a flat profile for a small program:
722 Each sample counts as 0.01 seconds.
723 % cumulative self self total
724 time seconds seconds calls ms/call ms/call name
725 33.34 0.02 0.02 7208 0.00 0.00 open
726 16.67 0.03 0.01 244 0.04 0.12 offtime
727 16.67 0.04 0.01 8 1.25 1.25 memccpy
728 16.67 0.05 0.01 7 1.43 1.43 write
729 16.67 0.06 0.01 mcount
730 0.00 0.06 0.00 236 0.00 0.00 tzset
731 0.00 0.06 0.00 192 0.00 0.00 tolower
732 0.00 0.06 0.00 47 0.00 0.00 strlen
733 0.00 0.06 0.00 45 0.00 0.00 strchr
734 0.00 0.06 0.00 1 0.00 50.00 main
735 0.00 0.06 0.00 1 0.00 0.00 memcpy
736 0.00 0.06 0.00 1 0.00 10.11 print
737 0.00 0.06 0.00 1 0.00 0.00 profil
738 0.00 0.06 0.00 1 0.00 50.00 report
741 The functions are sorted first by decreasing run-time spent in them,
742 then by decreasing number of calls, then alphabetically by name. The
743 functions `mcount' and `profil' are part of the profiling apparatus and
744 appear in every flat profile; their time gives a measure of the amount
745 of overhead due to profiling.
747 Just before the column headers, a statement appears indicating how
748 much time each sample counted as. This "sampling period" estimates the
749 margin of error in each of the time figures. A time figure that is not
750 much larger than this is not reliable. In this example, each sample
751 counted as 0.01 seconds, suggesting a 100 Hz sampling rate. The
752 program's total execution time was 0.06 seconds, as indicated by the
753 `cumulative seconds' field. Since each sample counted for 0.01
754 seconds, this means only six samples were taken during the run. Two of
755 the samples occurred while the program was in the `open' function, as
756 indicated by the `self seconds' field. Each of the other four samples
757 occurred one each in `offtime', `memccpy', `write', and `mcount'.
758 Since only six samples were taken, none of these values can be regarded
759 as particularly reliable. In another run, the `self seconds' field for
760 `mcount' might well be `0.00' or `0.02'. *Note Statistical Sampling
761 Error: Sampling Error, for a complete discussion.
763 The remaining functions in the listing (those whose `self seconds'
764 field is `0.00') didn't appear in the histogram samples at all.
765 However, the call graph indicated that they were called, so therefore
766 they are listed, sorted in decreasing order by the `calls' field.
767 Clearly some time was spent executing these functions, but the paucity
768 of histogram samples prevents any determination of how much time each
771 Here is what the fields in each line mean:
774 This is the percentage of the total execution time your program
775 spent in this function. These should all add up to 100%.
778 This is the cumulative total number of seconds the computer spent
779 executing this functions, plus the time spent in all the functions
780 above this one in this table.
783 This is the number of seconds accounted for by this function alone.
784 The flat profile listing is sorted first by this number.
787 This is the total number of times the function was called. If the
788 function was never called, or the number of times it was called
789 cannot be determined (probably because the function was not
790 compiled with profiling enabled), the "calls" field is blank.
793 This represents the average number of milliseconds spent in this
794 function per call, if this function is profiled. Otherwise, this
795 field is blank for this function.
798 This represents the average number of milliseconds spent in this
799 function and its descendants per call, if this function is
800 profiled. Otherwise, this field is blank for this function. This
801 is the only field in the flat profile that uses call graph
805 This is the name of the function. The flat profile is sorted by
806 this field alphabetically after the "self seconds" and "calls"
810 File: gprof.info, Node: Call Graph, Next: Line-by-line, Prev: Flat Profile, Up: Output
815 The "call graph" shows how much time was spent in each function and its
816 children. From this information, you can find functions that, while
817 they themselves may not have used much time, called other functions
818 that did use unusual amounts of time.
820 Here is a sample call from a small program. This call came from the
821 same `gprof' run as the flat profile example in the previous section.
823 granularity: each sample hit covers 2 byte(s) for 20.00% of 0.05 seconds
825 index % time self children called name
827 [1] 100.0 0.00 0.05 start [1]
828 0.00 0.05 1/1 main [2]
829 0.00 0.00 1/2 on_exit [28]
830 0.00 0.00 1/1 exit [59]
831 -----------------------------------------------
832 0.00 0.05 1/1 start [1]
833 [2] 100.0 0.00 0.05 1 main [2]
834 0.00 0.05 1/1 report [3]
835 -----------------------------------------------
836 0.00 0.05 1/1 main [2]
837 [3] 100.0 0.00 0.05 1 report [3]
838 0.00 0.03 8/8 timelocal [6]
839 0.00 0.01 1/1 print [9]
840 0.00 0.01 9/9 fgets [12]
841 0.00 0.00 12/34 strncmp <cycle 1> [40]
842 0.00 0.00 8/8 lookup [20]
843 0.00 0.00 1/1 fopen [21]
844 0.00 0.00 8/8 chewtime [24]
845 0.00 0.00 8/16 skipspace [44]
846 -----------------------------------------------
847 [4] 59.8 0.01 0.02 8+472 <cycle 2 as a whole> [4]
848 0.01 0.02 244+260 offtime <cycle 2> [7]
849 0.00 0.00 236+1 tzset <cycle 2> [26]
850 -----------------------------------------------
852 The lines full of dashes divide this table into "entries", one for
853 each function. Each entry has one or more lines.
855 In each entry, the primary line is the one that starts with an index
856 number in square brackets. The end of this line says which function
857 the entry is for. The preceding lines in the entry describe the
858 callers of this function and the following lines describe its
859 subroutines (also called "children" when we speak of the call graph).
861 The entries are sorted by time spent in the function and its
864 The internal profiling function `mcount' (*note The Flat Profile:
865 Flat Profile.) is never mentioned in the call graph.
869 * Primary:: Details of the primary line's contents.
870 * Callers:: Details of caller-lines' contents.
871 * Subroutines:: Details of subroutine-lines' contents.
872 * Cycles:: When there are cycles of recursion,
873 such as `a' calls `b' calls `a'...
876 File: gprof.info, Node: Primary, Next: Callers, Up: Call Graph
878 5.2.1 The Primary Line
879 ----------------------
881 The "primary line" in a call graph entry is the line that describes the
882 function which the entry is about and gives the overall statistics for
885 For reference, we repeat the primary line from the entry for function
886 `report' in our main example, together with the heading line that shows
887 the names of the fields:
889 index % time self children called name
891 [3] 100.0 0.00 0.05 1 report [3]
893 Here is what the fields in the primary line mean:
896 Entries are numbered with consecutive integers. Each function
897 therefore has an index number, which appears at the beginning of
900 Each cross-reference to a function, as a caller or subroutine of
901 another, gives its index number as well as its name. The index
902 number guides you if you wish to look for the entry for that
906 This is the percentage of the total time that was spent in this
907 function, including time spent in subroutines called from this
910 The time spent in this function is counted again for the callers of
911 this function. Therefore, adding up these percentages is
915 This is the total amount of time spent in this function. This
916 should be identical to the number printed in the `seconds' field
917 for this function in the flat profile.
920 This is the total amount of time spent in the subroutine calls
921 made by this function. This should be equal to the sum of all the
922 `self' and `children' entries of the children listed directly
926 This is the number of times the function was called.
928 If the function called itself recursively, there are two numbers,
929 separated by a `+'. The first number counts non-recursive calls,
930 and the second counts recursive calls.
932 In the example above, the function `report' was called once from
936 This is the name of the current function. The index number is
939 If the function is part of a cycle of recursion, the cycle number
940 is printed between the function's name and the index number (*note
941 How Mutually Recursive Functions Are Described: Cycles.). For
942 example, if function `gnurr' is part of cycle number one, and has
943 index number twelve, its primary line would be end like this:
948 File: gprof.info, Node: Callers, Next: Subroutines, Prev: Primary, Up: Call Graph
950 5.2.2 Lines for a Function's Callers
951 ------------------------------------
953 A function's entry has a line for each function it was called by.
954 These lines' fields correspond to the fields of the primary line, but
955 their meanings are different because of the difference in context.
957 For reference, we repeat two lines from the entry for the function
958 `report', the primary line and one caller-line preceding it, together
959 with the heading line that shows the names of the fields:
961 index % time self children called name
963 0.00 0.05 1/1 main [2]
964 [3] 100.0 0.00 0.05 1 report [3]
966 Here are the meanings of the fields in the caller-line for `report'
970 An estimate of the amount of time spent in `report' itself when it
971 was called from `main'.
974 An estimate of the amount of time spent in subroutines of `report'
975 when `report' was called from `main'.
977 The sum of the `self' and `children' fields is an estimate of the
978 amount of time spent within calls to `report' from `main'.
981 Two numbers: the number of times `report' was called from `main',
982 followed by the total number of non-recursive calls to `report'
983 from all its callers.
985 `name and index number'
986 The name of the caller of `report' to which this line applies,
987 followed by the caller's index number.
989 Not all functions have entries in the call graph; some options to
990 `gprof' request the omission of certain functions. When a caller
991 has no entry of its own, it still has caller-lines in the entries
992 of the functions it calls.
994 If the caller is part of a recursion cycle, the cycle number is
995 printed between the name and the index number.
997 If the identity of the callers of a function cannot be determined, a
998 dummy caller-line is printed which has `<spontaneous>' as the "caller's
999 name" and all other fields blank. This can happen for signal handlers.
1002 File: gprof.info, Node: Subroutines, Next: Cycles, Prev: Callers, Up: Call Graph
1004 5.2.3 Lines for a Function's Subroutines
1005 ----------------------------------------
1007 A function's entry has a line for each of its subroutines--in other
1008 words, a line for each other function that it called. These lines'
1009 fields correspond to the fields of the primary line, but their meanings
1010 are different because of the difference in context.
1012 For reference, we repeat two lines from the entry for the function
1013 `main', the primary line and a line for a subroutine, together with the
1014 heading line that shows the names of the fields:
1016 index % time self children called name
1018 [2] 100.0 0.00 0.05 1 main [2]
1019 0.00 0.05 1/1 report [3]
1021 Here are the meanings of the fields in the subroutine-line for `main'
1025 An estimate of the amount of time spent directly within `report'
1026 when `report' was called from `main'.
1029 An estimate of the amount of time spent in subroutines of `report'
1030 when `report' was called from `main'.
1032 The sum of the `self' and `children' fields is an estimate of the
1033 total time spent in calls to `report' from `main'.
1036 Two numbers, the number of calls to `report' from `main' followed
1037 by the total number of non-recursive calls to `report'. This
1038 ratio is used to determine how much of `report''s `self' and
1039 `children' time gets credited to `main'. *Note Estimating
1040 `children' Times: Assumptions.
1043 The name of the subroutine of `main' to which this line applies,
1044 followed by the subroutine's index number.
1046 If the caller is part of a recursion cycle, the cycle number is
1047 printed between the name and the index number.
1050 File: gprof.info, Node: Cycles, Prev: Subroutines, Up: Call Graph
1052 5.2.4 How Mutually Recursive Functions Are Described
1053 ----------------------------------------------------
1055 The graph may be complicated by the presence of "cycles of recursion"
1056 in the call graph. A cycle exists if a function calls another function
1057 that (directly or indirectly) calls (or appears to call) the original
1058 function. For example: if `a' calls `b', and `b' calls `a', then `a'
1059 and `b' form a cycle.
1061 Whenever there are call paths both ways between a pair of functions,
1062 they belong to the same cycle. If `a' and `b' call each other and `b'
1063 and `c' call each other, all three make one cycle. Note that even if
1064 `b' only calls `a' if it was not called from `a', `gprof' cannot
1065 determine this, so `a' and `b' are still considered a cycle.
1067 The cycles are numbered with consecutive integers. When a function
1068 belongs to a cycle, each time the function name appears in the call
1069 graph it is followed by `<cycle NUMBER>'.
1071 The reason cycles matter is that they make the time values in the
1072 call graph paradoxical. The "time spent in children" of `a' should
1073 include the time spent in its subroutine `b' and in `b''s
1074 subroutines--but one of `b''s subroutines is `a'! How much of `a''s
1075 time should be included in the children of `a', when `a' is indirectly
1078 The way `gprof' resolves this paradox is by creating a single entry
1079 for the cycle as a whole. The primary line of this entry describes the
1080 total time spent directly in the functions of the cycle. The
1081 "subroutines" of the cycle are the individual functions of the cycle,
1082 and all other functions that were called directly by them. The
1083 "callers" of the cycle are the functions, outside the cycle, that
1084 called functions in the cycle.
1086 Here is an example portion of a call graph which shows a cycle
1087 containing functions `a' and `b'. The cycle was entered by a call to
1088 `a' from `main'; both `a' and `b' called `c'.
1090 index % time self children called name
1091 ----------------------------------------
1093 [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3]
1094 1.02 0 3 b <cycle 1> [4]
1095 0.75 0 2 a <cycle 1> [5]
1096 ----------------------------------------
1098 [4] 52.85 1.02 0 0 b <cycle 1> [4]
1101 ----------------------------------------
1104 [5] 38.86 0.75 0 1 a <cycle 1> [5]
1107 ----------------------------------------
1109 (The entire call graph for this program contains in addition an entry
1110 for `main', which calls `a', and an entry for `c', with callers `a' and
1113 index % time self children called name
1115 [1] 100.00 0 1.93 0 start [1]
1116 0.16 1.77 1/1 main [2]
1117 ----------------------------------------
1118 0.16 1.77 1/1 start [1]
1119 [2] 100.00 0.16 1.77 1 main [2]
1120 1.77 0 1/1 a <cycle 1> [5]
1121 ----------------------------------------
1123 [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3]
1124 1.02 0 3 b <cycle 1> [4]
1125 0.75 0 2 a <cycle 1> [5]
1127 ----------------------------------------
1129 [4] 52.85 1.02 0 0 b <cycle 1> [4]
1132 ----------------------------------------
1135 [5] 38.86 0.75 0 1 a <cycle 1> [5]
1138 ----------------------------------------
1139 0 0 3/6 b <cycle 1> [4]
1140 0 0 3/6 a <cycle 1> [5]
1141 [6] 0.00 0 0 6 c [6]
1142 ----------------------------------------
1144 The `self' field of the cycle's primary line is the total time spent
1145 in all the functions of the cycle. It equals the sum of the `self'
1146 fields for the individual functions in the cycle, found in the entry in
1147 the subroutine lines for these functions.
1149 The `children' fields of the cycle's primary line and subroutine
1150 lines count only subroutines outside the cycle. Even though `a' calls
1151 `b', the time spent in those calls to `b' is not counted in `a''s
1152 `children' time. Thus, we do not encounter the problem of what to do
1153 when the time in those calls to `b' includes indirect recursive calls
1156 The `children' field of a caller-line in the cycle's entry estimates
1157 the amount of time spent _in the whole cycle_, and its other
1158 subroutines, on the times when that caller called a function in the
1161 The `called' field in the primary line for the cycle has two numbers:
1162 first, the number of times functions in the cycle were called by
1163 functions outside the cycle; second, the number of times they were
1164 called by functions in the cycle (including times when a function in
1165 the cycle calls itself). This is a generalization of the usual split
1166 into non-recursive and recursive calls.
1168 The `called' field of a subroutine-line for a cycle member in the
1169 cycle's entry says how many time that function was called from
1170 functions in the cycle. The total of all these is the second number in
1171 the primary line's `called' field.
1173 In the individual entry for a function in a cycle, the other
1174 functions in the same cycle can appear as subroutines and as callers.
1175 These lines show how many times each function in the cycle called or
1176 was called from each other function in the cycle. The `self' and
1177 `children' fields in these lines are blank because of the difficulty of
1178 defining meanings for them when recursion is going on.
1181 File: gprof.info, Node: Line-by-line, Next: Annotated Source, Prev: Call Graph, Up: Output
1183 5.3 Line-by-line Profiling
1184 ==========================
1186 `gprof''s `-l' option causes the program to perform "line-by-line"
1187 profiling. In this mode, histogram samples are assigned not to
1188 functions, but to individual lines of source code. This only works
1189 with programs compiled with older versions of the `gcc' compiler.
1190 Newer versions of `gcc' use a different program - `gcov' - to display
1191 line-by-line profiling information.
1193 With the older versions of `gcc' the program usually has to be
1194 compiled with a `-g' option, in addition to `-pg', in order to generate
1195 debugging symbols for tracking source code lines. Note, in much older
1196 versions of `gcc' the program had to be compiled with the `-a' command
1197 line option as well.
1199 The flat profile is the most useful output table in line-by-line
1200 mode. The call graph isn't as useful as normal, since the current
1201 version of `gprof' does not propagate call graph arcs from source code
1202 lines to the enclosing function. The call graph does, however, show
1203 each line of code that called each function, along with a count.
1205 Here is a section of `gprof''s output, without line-by-line
1206 profiling. Note that `ct_init' accounted for four histogram hits, and
1207 13327 calls to `init_block'.
1211 Each sample counts as 0.01 seconds.
1212 % cumulative self self total
1213 time seconds seconds calls us/call us/call name
1214 30.77 0.13 0.04 6335 6.31 6.31 ct_init
1217 Call graph (explanation follows)
1220 granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds
1222 index % time self children called name
1224 0.00 0.00 1/13496 name_too_long
1225 0.00 0.00 40/13496 deflate
1226 0.00 0.00 128/13496 deflate_fast
1227 0.00 0.00 13327/13496 ct_init
1228 [7] 0.0 0.00 0.00 13496 init_block
1230 Now let's look at some of `gprof''s output from the same program run,
1231 this time with line-by-line profiling enabled. Note that `ct_init''s
1232 four histogram hits are broken down into four lines of source code--one
1233 hit occurred on each of lines 349, 351, 382 and 385. In the call graph,
1234 note how `ct_init''s 13327 calls to `init_block' are broken down into
1235 one call from line 396, 3071 calls from line 384, 3730 calls from line
1236 385, and 6525 calls from 387.
1240 Each sample counts as 0.01 seconds.
1242 time seconds seconds calls name
1243 7.69 0.10 0.01 ct_init (trees.c:349)
1244 7.69 0.11 0.01 ct_init (trees.c:351)
1245 7.69 0.12 0.01 ct_init (trees.c:382)
1246 7.69 0.13 0.01 ct_init (trees.c:385)
1249 Call graph (explanation follows)
1252 granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds
1254 % time self children called name
1256 0.00 0.00 1/13496 name_too_long (gzip.c:1440)
1257 0.00 0.00 1/13496 deflate (deflate.c:763)
1258 0.00 0.00 1/13496 ct_init (trees.c:396)
1259 0.00 0.00 2/13496 deflate (deflate.c:727)
1260 0.00 0.00 4/13496 deflate (deflate.c:686)
1261 0.00 0.00 5/13496 deflate (deflate.c:675)
1262 0.00 0.00 12/13496 deflate (deflate.c:679)
1263 0.00 0.00 16/13496 deflate (deflate.c:730)
1264 0.00 0.00 128/13496 deflate_fast (deflate.c:654)
1265 0.00 0.00 3071/13496 ct_init (trees.c:384)
1266 0.00 0.00 3730/13496 ct_init (trees.c:385)
1267 0.00 0.00 6525/13496 ct_init (trees.c:387)
1268 [6] 0.0 0.00 0.00 13496 init_block (trees.c:408)
1271 File: gprof.info, Node: Annotated Source, Prev: Line-by-line, Up: Output
1273 5.4 The Annotated Source Listing
1274 ================================
1276 `gprof''s `-A' option triggers an annotated source listing, which lists
1277 the program's source code, each function labeled with the number of
1278 times it was called. You may also need to specify the `-I' option, if
1279 `gprof' can't find the source code files.
1281 With older versions of `gcc' compiling with `gcc ... -g -pg -a'
1282 augments your program with basic-block counting code, in addition to
1283 function counting code. This enables `gprof' to determine how many
1284 times each line of code was executed. With newer versions of `gcc'
1285 support for displaying basic-block counts is provided by the `gcov'
1288 For example, consider the following function, taken from gzip, with
1297 7 static ulg crc = (ulg)0xffffffffL;
1304 14 c = crc_32_tab[...];
1308 18 return c ^ 0xffffffffL;
1311 `updcrc' has at least five basic-blocks. One is the function
1312 itself. The `if' statement on line 9 generates two more basic-blocks,
1313 one for each branch of the `if'. A fourth basic-block results from the
1314 `if' on line 13, and the contents of the `do' loop form the fifth
1315 basic-block. The compiler may also generate additional basic-blocks to
1316 handle various special cases.
1318 A program augmented for basic-block counting can be analyzed with
1319 `gprof -l -A'. The `-x' option is also helpful, to ensure that each
1320 line of code is labeled at least once. Here is `updcrc''s annotated
1321 source listing for a sample `gzip' run:
1329 static ulg crc = (ulg)0xffffffffL;
1331 2 -> if (s == NULL) {
1332 1 -> c = 0xffffffffL;
1336 26312 -> c = crc_32_tab[...];
1337 26312,1,26311 -> } while (--n);
1340 2 -> return c ^ 0xffffffffL;
1343 In this example, the function was called twice, passing once through
1344 each branch of the `if' statement. The body of the `do' loop was
1345 executed a total of 26312 times. Note how the `while' statement is
1346 annotated. It began execution 26312 times, once for each iteration
1347 through the loop. One of those times (the last time) it exited, while
1348 it branched back to the beginning of the loop 26311 times.
1351 File: gprof.info, Node: Inaccuracy, Next: How do I?, Prev: Output, Up: Top
1353 6 Inaccuracy of `gprof' Output
1354 ******************************
1358 * Sampling Error:: Statistical margins of error
1359 * Assumptions:: Estimating children times
1362 File: gprof.info, Node: Sampling Error, Next: Assumptions, Up: Inaccuracy
1364 6.1 Statistical Sampling Error
1365 ==============================
1367 The run-time figures that `gprof' gives you are based on a sampling
1368 process, so they are subject to statistical inaccuracy. If a function
1369 runs only a small amount of time, so that on the average the sampling
1370 process ought to catch that function in the act only once, there is a
1371 pretty good chance it will actually find that function zero times, or
1374 By contrast, the number-of-calls and basic-block figures are derived
1375 by counting, not sampling. They are completely accurate and will not
1376 vary from run to run if your program is deterministic.
1378 The "sampling period" that is printed at the beginning of the flat
1379 profile says how often samples are taken. The rule of thumb is that a
1380 run-time figure is accurate if it is considerably bigger than the
1383 The actual amount of error can be predicted. For N samples, the
1384 _expected_ error is the square-root of N. For example, if the sampling
1385 period is 0.01 seconds and `foo''s run-time is 1 second, N is 100
1386 samples (1 second/0.01 seconds), sqrt(N) is 10 samples, so the expected
1387 error in `foo''s run-time is 0.1 seconds (10*0.01 seconds), or ten
1388 percent of the observed value. Again, if the sampling period is 0.01
1389 seconds and `bar''s run-time is 100 seconds, N is 10000 samples,
1390 sqrt(N) is 100 samples, so the expected error in `bar''s run-time is 1
1391 second, or one percent of the observed value. It is likely to vary
1392 this much _on the average_ from one profiling run to the next.
1393 (_Sometimes_ it will vary more.)
1395 This does not mean that a small run-time figure is devoid of
1396 information. If the program's _total_ run-time is large, a small
1397 run-time for one function does tell you that that function used an
1398 insignificant fraction of the whole program's time. Usually this means
1399 it is not worth optimizing.
1401 One way to get more accuracy is to give your program more (but
1402 similar) input data so it will take longer. Another way is to combine
1403 the data from several runs, using the `-s' option of `gprof'. Here is
1406 1. Run your program once.
1408 2. Issue the command `mv gmon.out gmon.sum'.
1410 3. Run your program again, the same as before.
1412 4. Merge the new data in `gmon.out' into `gmon.sum' with this command:
1414 gprof -s EXECUTABLE-FILE gmon.out gmon.sum
1416 5. Repeat the last two steps as often as you wish.
1418 6. Analyze the cumulative data using this command:
1420 gprof EXECUTABLE-FILE gmon.sum > OUTPUT-FILE
1423 File: gprof.info, Node: Assumptions, Prev: Sampling Error, Up: Inaccuracy
1425 6.2 Estimating `children' Times
1426 ===============================
1428 Some of the figures in the call graph are estimates--for example, the
1429 `children' time values and all the time figures in caller and
1432 There is no direct information about these measurements in the
1433 profile data itself. Instead, `gprof' estimates them by making an
1434 assumption about your program that might or might not be true.
1436 The assumption made is that the average time spent in each call to
1437 any function `foo' is not correlated with who called `foo'. If `foo'
1438 used 5 seconds in all, and 2/5 of the calls to `foo' came from `a',
1439 then `foo' contributes 2 seconds to `a''s `children' time, by
1442 This assumption is usually true enough, but for some programs it is
1443 far from true. Suppose that `foo' returns very quickly when its
1444 argument is zero; suppose that `a' always passes zero as an argument,
1445 while other callers of `foo' pass other arguments. In this program,
1446 all the time spent in `foo' is in the calls from callers other than `a'.
1447 But `gprof' has no way of knowing this; it will blindly and incorrectly
1448 charge 2 seconds of time in `foo' to the children of `a'.
1450 We hope some day to put more complete data into `gmon.out', so that
1451 this assumption is no longer needed, if we can figure out how. For the
1452 novice, the estimated figures are usually more useful than misleading.
1455 File: gprof.info, Node: How do I?, Next: Incompatibilities, Prev: Inaccuracy, Up: Top
1457 7 Answers to Common Questions
1458 *****************************
1460 How can I get more exact information about hot spots in my program?
1461 Looking at the per-line call counts only tells part of the story.
1462 Because `gprof' can only report call times and counts by function,
1463 the best way to get finer-grained information on where the program
1464 is spending its time is to re-factor large functions into sequences
1465 of calls to smaller ones. Beware however that this can introduce
1466 artificial hot spots since compiling with `-pg' adds a significant
1467 overhead to function calls. An alternative solution is to use a
1468 non-intrusive profiler, e.g. oprofile.
1470 How do I find which lines in my program were executed the most times?
1471 Use the `gcov' program.
1473 How do I find which lines in my program called a particular function?
1474 Use `gprof -l' and lookup the function in the call graph. The
1475 callers will be broken down by function and line number.
1477 How do I analyze a program that runs for less than a second?
1478 Try using a shell script like this one:
1480 for i in `seq 1 100`; do
1482 mv gmon.out gmon.out.$i
1485 gprof -s fastprog gmon.out.*
1487 gprof fastprog gmon.sum
1489 If your program is completely deterministic, all the call counts
1490 will be simple multiples of 100 (i.e., a function called once in
1491 each run will appear with a call count of 100).
1495 File: gprof.info, Node: Incompatibilities, Next: Details, Prev: How do I?, Up: Top
1497 8 Incompatibilities with Unix `gprof'
1498 *************************************
1500 GNU `gprof' and Berkeley Unix `gprof' use the same data file
1501 `gmon.out', and provide essentially the same information. But there
1502 are a few differences.
1504 * GNU `gprof' uses a new, generalized file format with support for
1505 basic-block execution counts and non-realtime histograms. A magic
1506 cookie and version number allows `gprof' to easily identify new
1507 style files. Old BSD-style files can still be read. *Note
1508 Profiling Data File Format: File Format.
1510 * For a recursive function, Unix `gprof' lists the function as a
1511 parent and as a child, with a `calls' field that lists the number
1512 of recursive calls. GNU `gprof' omits these lines and puts the
1513 number of recursive calls in the primary line.
1515 * When a function is suppressed from the call graph with `-e', GNU
1516 `gprof' still lists it as a subroutine of functions that call it.
1518 * GNU `gprof' accepts the `-k' with its argument in the form
1519 `from/to', instead of `from to'.
1521 * In the annotated source listing, if there are multiple basic
1522 blocks on the same line, GNU `gprof' prints all of their counts,
1523 separated by commas.
1525 * The blurbs, field widths, and output formats are different. GNU
1526 `gprof' prints blurbs after the tables, so that you can see the
1527 tables without skipping the blurbs.
1530 File: gprof.info, Node: Details, Next: GNU Free Documentation License, Prev: Incompatibilities, Up: Top
1532 9 Details of Profiling
1533 **********************
1537 * Implementation:: How a program collects profiling information
1538 * File Format:: Format of `gmon.out' files
1539 * Internals:: `gprof''s internal operation
1540 * Debugging:: Using `gprof''s `-d' option
1543 File: gprof.info, Node: Implementation, Next: File Format, Up: Details
1545 9.1 Implementation of Profiling
1546 ===============================
1548 Profiling works by changing how every function in your program is
1549 compiled so that when it is called, it will stash away some information
1550 about where it was called from. From this, the profiler can figure out
1551 what function called it, and can count how many times it was called.
1552 This change is made by the compiler when your program is compiled with
1553 the `-pg' option, which causes every function to call `mcount' (or
1554 `_mcount', or `__mcount', depending on the OS and compiler) as one of
1555 its first operations.
1557 The `mcount' routine, included in the profiling library, is
1558 responsible for recording in an in-memory call graph table both its
1559 parent routine (the child) and its parent's parent. This is typically
1560 done by examining the stack frame to find both the address of the
1561 child, and the return address in the original parent. Since this is a
1562 very machine-dependent operation, `mcount' itself is typically a short
1563 assembly-language stub routine that extracts the required information,
1564 and then calls `__mcount_internal' (a normal C function) with two
1565 arguments--`frompc' and `selfpc'. `__mcount_internal' is responsible
1566 for maintaining the in-memory call graph, which records `frompc',
1567 `selfpc', and the number of times each of these call arcs was traversed.
1569 GCC Version 2 provides a magical function
1570 (`__builtin_return_address'), which allows a generic `mcount' function
1571 to extract the required information from the stack frame. However, on
1572 some architectures, most notably the SPARC, using this builtin can be
1573 very computationally expensive, and an assembly language version of
1574 `mcount' is used for performance reasons.
1576 Number-of-calls information for library routines is collected by
1577 using a special version of the C library. The programs in it are the
1578 same as in the usual C library, but they were compiled with `-pg'. If
1579 you link your program with `gcc ... -pg', it automatically uses the
1580 profiling version of the library.
1582 Profiling also involves watching your program as it runs, and
1583 keeping a histogram of where the program counter happens to be every
1584 now and then. Typically the program counter is looked at around 100
1585 times per second of run time, but the exact frequency may vary from
1588 This is done is one of two ways. Most UNIX-like operating systems
1589 provide a `profil()' system call, which registers a memory array with
1590 the kernel, along with a scale factor that determines how the program's
1591 address space maps into the array. Typical scaling values cause every
1592 2 to 8 bytes of address space to map into a single array slot. On
1593 every tick of the system clock (assuming the profiled program is
1594 running), the value of the program counter is examined and the
1595 corresponding slot in the memory array is incremented. Since this is
1596 done in the kernel, which had to interrupt the process anyway to handle
1597 the clock interrupt, very little additional system overhead is required.
1599 However, some operating systems, most notably Linux 2.0 (and
1600 earlier), do not provide a `profil()' system call. On such a system,
1601 arrangements are made for the kernel to periodically deliver a signal
1602 to the process (typically via `setitimer()'), which then performs the
1603 same operation of examining the program counter and incrementing a slot
1604 in the memory array. Since this method requires a signal to be
1605 delivered to user space every time a sample is taken, it uses
1606 considerably more overhead than kernel-based profiling. Also, due to
1607 the added delay required to deliver the signal, this method is less
1610 A special startup routine allocates memory for the histogram and
1611 either calls `profil()' or sets up a clock signal handler. This
1612 routine (`monstartup') can be invoked in several ways. On Linux
1613 systems, a special profiling startup file `gcrt0.o', which invokes
1614 `monstartup' before `main', is used instead of the default `crt0.o'.
1615 Use of this special startup file is one of the effects of using `gcc
1616 ... -pg' to link. On SPARC systems, no special startup files are used.
1617 Rather, the `mcount' routine, when it is invoked for the first time
1618 (typically when `main' is called), calls `monstartup'.
1620 If the compiler's `-a' option was used, basic-block counting is also
1621 enabled. Each object file is then compiled with a static array of
1622 counts, initially zero. In the executable code, every time a new
1623 basic-block begins (i.e., when an `if' statement appears), an extra
1624 instruction is inserted to increment the corresponding count in the
1625 array. At compile time, a paired array was constructed that recorded
1626 the starting address of each basic-block. Taken together, the two
1627 arrays record the starting address of every basic-block, along with the
1628 number of times it was executed.
1630 The profiling library also includes a function (`mcleanup') which is
1631 typically registered using `atexit()' to be called as the program
1632 exits, and is responsible for writing the file `gmon.out'. Profiling
1633 is turned off, various headers are output, and the histogram is
1634 written, followed by the call-graph arcs and the basic-block counts.
1636 The output from `gprof' gives no indication of parts of your program
1637 that are limited by I/O or swapping bandwidth. This is because samples
1638 of the program counter are taken at fixed intervals of the program's
1639 run time. Therefore, the time measurements in `gprof' output say
1640 nothing about time that your program was not running. For example, a
1641 part of the program that creates so much data that it cannot all fit in
1642 physical memory at once may run very slowly due to thrashing, but
1643 `gprof' will say it uses little time. On the other hand, sampling by
1644 run time has the advantage that the amount of load due to other users
1645 won't directly affect the output you get.
1648 File: gprof.info, Node: File Format, Next: Internals, Prev: Implementation, Up: Details
1650 9.2 Profiling Data File Format
1651 ==============================
1653 The old BSD-derived file format used for profile data does not contain a
1654 magic cookie that allows to check whether a data file really is a
1655 `gprof' file. Furthermore, it does not provide a version number, thus
1656 rendering changes to the file format almost impossible. GNU `gprof'
1657 uses a new file format that provides these features. For backward
1658 compatibility, GNU `gprof' continues to support the old BSD-derived
1659 format, but not all features are supported with it. For example,
1660 basic-block execution counts cannot be accommodated by the old file
1663 The new file format is defined in header file `gmon_out.h'. It
1664 consists of a header containing the magic cookie and a version number,
1665 as well as some spare bytes available for future extensions. All data
1666 in a profile data file is in the native format of the target for which
1667 the profile was collected. GNU `gprof' adapts automatically to the
1670 In the new file format, the header is followed by a sequence of
1671 records. Currently, there are three different record types: histogram
1672 records, call-graph arc records, and basic-block execution count
1673 records. Each file can contain any number of each record type. When
1674 reading a file, GNU `gprof' will ensure records of the same type are
1675 compatible with each other and compute the union of all records. For
1676 example, for basic-block execution counts, the union is simply the sum
1677 of all execution counts for each basic-block.
1679 9.2.1 Histogram Records
1680 -----------------------
1682 Histogram records consist of a header that is followed by an array of
1683 bins. The header contains the text-segment range that the histogram
1684 spans, the size of the histogram in bytes (unlike in the old BSD
1685 format, this does not include the size of the header), the rate of the
1686 profiling clock, and the physical dimension that the bin counts
1687 represent after being scaled by the profiling clock rate. The physical
1688 dimension is specified in two parts: a long name of up to 15 characters
1689 and a single character abbreviation. For example, a histogram
1690 representing real-time would specify the long name as "seconds" and the
1691 abbreviation as "s". This feature is useful for architectures that
1692 support performance monitor hardware (which, fortunately, is becoming
1693 increasingly common). For example, under DEC OSF/1, the "uprofile"
1694 command can be used to produce a histogram of, say, instruction cache
1695 misses. In this case, the dimension in the histogram header could be
1696 set to "i-cache misses" and the abbreviation could be set to "1"
1697 (because it is simply a count, not a physical dimension). Also, the
1698 profiling rate would have to be set to 1 in this case.
1700 Histogram bins are 16-bit numbers and each bin represent an equal
1701 amount of text-space. For example, if the text-segment is one thousand
1702 bytes long and if there are ten bins in the histogram, each bin
1703 represents one hundred bytes.
1705 9.2.2 Call-Graph Records
1706 ------------------------
1708 Call-graph records have a format that is identical to the one used in
1709 the BSD-derived file format. It consists of an arc in the call graph
1710 and a count indicating the number of times the arc was traversed during
1711 program execution. Arcs are specified by a pair of addresses: the
1712 first must be within caller's function and the second must be within
1713 the callee's function. When performing profiling at the function
1714 level, these addresses can point anywhere within the respective
1715 function. However, when profiling at the line-level, it is better if
1716 the addresses are as close to the call-site/entry-point as possible.
1717 This will ensure that the line-level call-graph is able to identify
1718 exactly which line of source code performed calls to a function.
1720 9.2.3 Basic-Block Execution Count Records
1721 -----------------------------------------
1723 Basic-block execution count records consist of a header followed by a
1724 sequence of address/count pairs. The header simply specifies the
1725 length of the sequence. In an address/count pair, the address
1726 identifies a basic-block and the count specifies the number of times
1727 that basic-block was executed. Any address within the basic-address can
1731 File: gprof.info, Node: Internals, Next: Debugging, Prev: File Format, Up: Details
1733 9.3 `gprof''s Internal Operation
1734 ================================
1736 Like most programs, `gprof' begins by processing its options. During
1737 this stage, it may building its symspec list (`sym_ids.c:sym_id_add'),
1738 if options are specified which use symspecs. `gprof' maintains a
1739 single linked list of symspecs, which will eventually get turned into
1740 12 symbol tables, organized into six include/exclude pairs--one pair
1741 each for the flat profile (INCL_FLAT/EXCL_FLAT), the call graph arcs
1742 (INCL_ARCS/EXCL_ARCS), printing in the call graph
1743 (INCL_GRAPH/EXCL_GRAPH), timing propagation in the call graph
1744 (INCL_TIME/EXCL_TIME), the annotated source listing
1745 (INCL_ANNO/EXCL_ANNO), and the execution count listing
1746 (INCL_EXEC/EXCL_EXEC).
1748 After option processing, `gprof' finishes building the symspec list
1749 by adding all the symspecs in `default_excluded_list' to the exclude
1750 lists EXCL_TIME and EXCL_GRAPH, and if line-by-line profiling is
1751 specified, EXCL_FLAT as well. These default excludes are not added to
1752 EXCL_ANNO, EXCL_ARCS, and EXCL_EXEC.
1754 Next, the BFD library is called to open the object file, verify that
1755 it is an object file, and read its symbol table (`core.c:core_init'),
1756 using `bfd_canonicalize_symtab' after mallocing an appropriately sized
1757 array of symbols. At this point, function mappings are read (if the
1758 `--file-ordering' option has been specified), and the core text space
1759 is read into memory (if the `-c' option was given).
1761 `gprof''s own symbol table, an array of Sym structures, is now built.
1762 This is done in one of two ways, by one of two routines, depending on
1763 whether line-by-line profiling (`-l' option) has been enabled. For
1764 normal profiling, the BFD canonical symbol table is scanned. For
1765 line-by-line profiling, every text space address is examined, and a new
1766 symbol table entry gets created every time the line number changes. In
1767 either case, two passes are made through the symbol table--one to count
1768 the size of the symbol table required, and the other to actually read
1769 the symbols. In between the two passes, a single array of type `Sym'
1770 is created of the appropriate length. Finally,
1771 `symtab.c:symtab_finalize' is called to sort the symbol table and
1772 remove duplicate entries (entries with the same memory address).
1774 The symbol table must be a contiguous array for two reasons. First,
1775 the `qsort' library function (which sorts an array) will be used to
1776 sort the symbol table. Also, the symbol lookup routine
1777 (`symtab.c:sym_lookup'), which finds symbols based on memory address,
1778 uses a binary search algorithm which requires the symbol table to be a
1779 sorted array. Function symbols are indicated with an `is_func' flag.
1780 Line number symbols have no special flags set. Additionally, a symbol
1781 can have an `is_static' flag to indicate that it is a local symbol.
1783 With the symbol table read, the symspecs can now be translated into
1784 Syms (`sym_ids.c:sym_id_parse'). Remember that a single symspec can
1785 match multiple symbols. An array of symbol tables (`syms') is created,
1786 each entry of which is a symbol table of Syms to be included or
1787 excluded from a particular listing. The master symbol table and the
1788 symspecs are examined by nested loops, and every symbol that matches a
1789 symspec is inserted into the appropriate syms table. This is done
1790 twice, once to count the size of each required symbol table, and again
1791 to build the tables, which have been malloced between passes. From now
1792 on, to determine whether a symbol is on an include or exclude symspec
1793 list, `gprof' simply uses its standard symbol lookup routine on the
1794 appropriate table in the `syms' array.
1796 Now the profile data file(s) themselves are read
1797 (`gmon_io.c:gmon_out_read'), first by checking for a new-style
1798 `gmon.out' header, then assuming this is an old-style BSD `gmon.out' if
1799 the magic number test failed.
1801 New-style histogram records are read by `hist.c:hist_read_rec'. For
1802 the first histogram record, allocate a memory array to hold all the
1803 bins, and read them in. When multiple profile data files (or files
1804 with multiple histogram records) are read, the memory ranges of each
1805 pair of histogram records must be either equal, or non-overlapping.
1806 For each pair of histogram records, the resolution (memory region size
1807 divided by the number of bins) must be the same. The time unit must be
1808 the same for all histogram records. If the above containts are met, all
1809 histograms for the same memory range are merged.
1811 As each call graph record is read (`call_graph.c:cg_read_rec'), the
1812 parent and child addresses are matched to symbol table entries, and a
1813 call graph arc is created by `cg_arcs.c:arc_add', unless the arc fails
1814 a symspec check against INCL_ARCS/EXCL_ARCS. As each arc is added, a
1815 linked list is maintained of the parent's child arcs, and of the child's
1816 parent arcs. Both the child's call count and the arc's call count are
1817 incremented by the record's call count.
1819 Basic-block records are read (`basic_blocks.c:bb_read_rec'), but
1820 only if line-by-line profiling has been selected. Each basic-block
1821 address is matched to a corresponding line symbol in the symbol table,
1822 and an entry made in the symbol's bb_addr and bb_calls arrays. Again,
1823 if multiple basic-block records are present for the same address, the
1824 call counts are cumulative.
1826 A gmon.sum file is dumped, if requested (`gmon_io.c:gmon_out_write').
1828 If histograms were present in the data files, assign them to symbols
1829 (`hist.c:hist_assign_samples') by iterating over all the sample bins
1830 and assigning them to symbols. Since the symbol table is sorted in
1831 order of ascending memory addresses, we can simple follow along in the
1832 symbol table as we make our pass over the sample bins. This step
1833 includes a symspec check against INCL_FLAT/EXCL_FLAT. Depending on the
1834 histogram scale factor, a sample bin may span multiple symbols, in
1835 which case a fraction of the sample count is allocated to each symbol,
1836 proportional to the degree of overlap. This effect is rare for normal
1837 profiling, but overlaps are more common during line-by-line profiling,
1838 and can cause each of two adjacent lines to be credited with half a
1841 If call graph data is present, `cg_arcs.c:cg_assemble' is called.
1842 First, if `-c' was specified, a machine-dependent routine (`find_call')
1843 scans through each symbol's machine code, looking for subroutine call
1844 instructions, and adding them to the call graph with a zero call count.
1845 A topological sort is performed by depth-first numbering all the
1846 symbols (`cg_dfn.c:cg_dfn'), so that children are always numbered less
1847 than their parents, then making a array of pointers into the symbol
1848 table and sorting it into numerical order, which is reverse topological
1849 order (children appear before parents). Cycles are also detected at
1850 this point, all members of which are assigned the same topological
1851 number. Two passes are now made through this sorted array of symbol
1852 pointers. The first pass, from end to beginning (parents to children),
1853 computes the fraction of child time to propagate to each parent and a
1854 print flag. The print flag reflects symspec handling of
1855 INCL_GRAPH/EXCL_GRAPH, with a parent's include or exclude (print or no
1856 print) property being propagated to its children, unless they
1857 themselves explicitly appear in INCL_GRAPH or EXCL_GRAPH. A second
1858 pass, from beginning to end (children to parents) actually propagates
1859 the timings along the call graph, subject to a check against
1860 INCL_TIME/EXCL_TIME. With the print flag, fractions, and timings now
1861 stored in the symbol structures, the topological sort array is now
1862 discarded, and a new array of pointers is assembled, this time sorted
1865 Finally, print the various outputs the user requested, which is now
1866 fairly straightforward. The call graph (`cg_print.c:cg_print') and
1867 flat profile (`hist.c:hist_print') are regurgitations of values already
1868 computed. The annotated source listing
1869 (`basic_blocks.c:print_annotated_source') uses basic-block information,
1870 if present, to label each line of code with call counts, otherwise only
1871 the function call counts are presented.
1873 The function ordering code is marginally well documented in the
1874 source code itself (`cg_print.c'). Basically, the functions with the
1875 most use and the most parents are placed first, followed by other
1876 functions with the most use, followed by lower use functions, followed
1877 by unused functions at the end.
1880 File: gprof.info, Node: Debugging, Prev: Internals, Up: Details
1882 9.4 Debugging `gprof'
1883 =====================
1885 If `gprof' was compiled with debugging enabled, the `-d' option
1886 triggers debugging output (to stdout) which can be helpful in
1887 understanding its operation. The debugging number specified is
1888 interpreted as a sum of the following options:
1890 2 - Topological sort
1891 Monitor depth-first numbering of symbols during call graph analysis
1894 Shows symbols as they are identified as cycle heads
1897 As the call graph arcs are read, show each arc and how the total
1898 calls to each function are tallied
1900 32 - Call graph arc sorting
1901 Details sorting individual parents/children within each call graph
1904 64 - Reading histogram and call graph records
1905 Shows address ranges of histograms as they are read, and each call
1909 Reading, classifying, and sorting the symbol table from the object
1910 file. For line-by-line profiling (`-l' option), also shows line
1911 numbers being assigned to memory addresses.
1913 256 - Static call graph
1914 Trace operation of `-c' option
1916 512 - Symbol table and arc table lookups
1917 Detail operation of lookup routines
1919 1024 - Call graph propagation
1920 Shows how function times are propagated along the call graph
1923 Shows basic-block records as they are read from profile data (only
1924 meaningful with `-l' option)
1927 Shows symspec-to-symbol pattern matching operation
1929 8192 - Annotate source
1930 Tracks operation of `-A' option
1933 File: gprof.info, Node: GNU Free Documentation License, Prev: Details, Up: Top
1935 Appendix A GNU Free Documentation License
1936 *****************************************
1938 Version 1.1, March 2000
1940 Copyright (C) 2000, 2003 Free Software Foundation, Inc.
1941 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
1943 Everyone is permitted to copy and distribute verbatim copies
1944 of this license document, but changing it is not allowed.
1949 The purpose of this License is to make a manual, textbook, or other
1950 written document "free" in the sense of freedom: to assure everyone
1951 the effective freedom to copy and redistribute it, with or without
1952 modifying it, either commercially or noncommercially. Secondarily,
1953 this License preserves for the author and publisher a way to get
1954 credit for their work, while not being considered responsible for
1955 modifications made by others.
1957 This License is a kind of "copyleft", which means that derivative
1958 works of the document must themselves be free in the same sense.
1959 It complements the GNU General Public License, which is a copyleft
1960 license designed for free software.
1962 We have designed this License in order to use it for manuals for
1963 free software, because free software needs free documentation: a
1964 free program should come with manuals providing the same freedoms
1965 that the software does. But this License is not limited to
1966 software manuals; it can be used for any textual work, regardless
1967 of subject matter or whether it is published as a printed book.
1968 We recommend this License principally for works whose purpose is
1969 instruction or reference.
1972 1. APPLICABILITY AND DEFINITIONS
1974 This License applies to any manual or other work that contains a
1975 notice placed by the copyright holder saying it can be distributed
1976 under the terms of this License. The "Document", below, refers to
1977 any such manual or work. Any member of the public is a licensee,
1978 and is addressed as "you."
1980 A "Modified Version" of the Document means any work containing the
1981 Document or a portion of it, either copied verbatim, or with
1982 modifications and/or translated into another language.
1984 A "Secondary Section" is a named appendix or a front-matter
1985 section of the Document that deals exclusively with the
1986 relationship of the publishers or authors of the Document to the
1987 Document's overall subject (or to related matters) and contains
1988 nothing that could fall directly within that overall subject.
1989 (For example, if the Document is in part a textbook of
1990 mathematics, a Secondary Section may not explain any mathematics.)
1991 The relationship could be a matter of historical connection with
1992 the subject or with related matters, or of legal, commercial,
1993 philosophical, ethical or political position regarding them.
1995 The "Invariant Sections" are certain Secondary Sections whose
1996 titles are designated, as being those of Invariant Sections, in
1997 the notice that says that the Document is released under this
2000 The "Cover Texts" are certain short passages of text that are
2001 listed, as Front-Cover Texts or Back-Cover Texts, in the notice
2002 that says that the Document is released under this License.
2004 A "Transparent" copy of the Document means a machine-readable copy,
2005 represented in a format whose specification is available to the
2006 general public, whose contents can be viewed and edited directly
2007 and straightforwardly with generic text editors or (for images
2008 composed of pixels) generic paint programs or (for drawings) some
2009 widely available drawing editor, and that is suitable for input to
2010 text formatters or for automatic translation to a variety of
2011 formats suitable for input to text formatters. A copy made in an
2012 otherwise Transparent file format whose markup has been designed
2013 to thwart or discourage subsequent modification by readers is not
2014 Transparent. A copy that is not "Transparent" is called "Opaque."
2016 Examples of suitable formats for Transparent copies include plain
2017 ASCII without markup, Texinfo input format, LaTeX input format,
2018 SGML or XML using a publicly available DTD, and
2019 standard-conforming simple HTML designed for human modification.
2020 Opaque formats include PostScript, PDF, proprietary formats that
2021 can be read and edited only by proprietary word processors, SGML
2022 or XML for which the DTD and/or processing tools are not generally
2023 available, and the machine-generated HTML produced by some word
2024 processors for output purposes only.
2026 The "Title Page" means, for a printed book, the title page itself,
2027 plus such following pages as are needed to hold, legibly, the
2028 material this License requires to appear in the title page. For
2029 works in formats which do not have any title page as such, "Title
2030 Page" means the text near the most prominent appearance of the
2031 work's title, preceding the beginning of the body of the text.
2035 You may copy and distribute the Document in any medium, either
2036 commercially or noncommercially, provided that this License, the
2037 copyright notices, and the license notice saying this License
2038 applies to the Document are reproduced in all copies, and that you
2039 add no other conditions whatsoever to those of this License. You
2040 may not use technical measures to obstruct or control the reading
2041 or further copying of the copies you make or distribute. However,
2042 you may accept compensation in exchange for copies. If you
2043 distribute a large enough number of copies you must also follow
2044 the conditions in section 3.
2046 You may also lend copies, under the same conditions stated above,
2047 and you may publicly display copies.
2049 3. COPYING IN QUANTITY
2051 If you publish printed copies of the Document numbering more than
2052 100, and the Document's license notice requires Cover Texts, you
2053 must enclose the copies in covers that carry, clearly and legibly,
2054 all these Cover Texts: Front-Cover Texts on the front cover, and
2055 Back-Cover Texts on the back cover. Both covers must also clearly
2056 and legibly identify you as the publisher of these copies. The
2057 front cover must present the full title with all words of the
2058 title equally prominent and visible. You may add other material
2059 on the covers in addition. Copying with changes limited to the
2060 covers, as long as they preserve the title of the Document and
2061 satisfy these conditions, can be treated as verbatim copying in
2064 If the required texts for either cover are too voluminous to fit
2065 legibly, you should put the first ones listed (as many as fit
2066 reasonably) on the actual cover, and continue the rest onto
2069 If you publish or distribute Opaque copies of the Document
2070 numbering more than 100, you must either include a
2071 machine-readable Transparent copy along with each Opaque copy, or
2072 state in or with each Opaque copy a publicly-accessible
2073 computer-network location containing a complete Transparent copy
2074 of the Document, free of added material, which the general
2075 network-using public has access to download anonymously at no
2076 charge using public-standard network protocols. If you use the
2077 latter option, you must take reasonably prudent steps, when you
2078 begin distribution of Opaque copies in quantity, to ensure that
2079 this Transparent copy will remain thus accessible at the stated
2080 location until at least one year after the last time you
2081 distribute an Opaque copy (directly or through your agents or
2082 retailers) of that edition to the public.
2084 It is requested, but not required, that you contact the authors of
2085 the Document well before redistributing any large number of
2086 copies, to give them a chance to provide you with an updated
2087 version of the Document.
2091 You may copy and distribute a Modified Version of the Document
2092 under the conditions of sections 2 and 3 above, provided that you
2093 release the Modified Version under precisely this License, with
2094 the Modified Version filling the role of the Document, thus
2095 licensing distribution and modification of the Modified Version to
2096 whoever possesses a copy of it. In addition, you must do these
2097 things in the Modified Version:
2099 A. Use in the Title Page (and on the covers, if any) a title
2100 distinct from that of the Document, and from those of previous
2101 versions (which should, if there were any, be listed in the
2102 History section of the Document). You may use the same title
2103 as a previous version if the original publisher of that version
2105 B. List on the Title Page, as authors, one or more persons or
2106 entities responsible for authorship of the modifications in the
2107 Modified Version, together with at least five of the principal
2108 authors of the Document (all of its principal authors, if it
2109 has less than five).
2110 C. State on the Title page the name of the publisher of the
2111 Modified Version, as the publisher.
2112 D. Preserve all the copyright notices of the Document.
2113 E. Add an appropriate copyright notice for your modifications
2114 adjacent to the other copyright notices.
2115 F. Include, immediately after the copyright notices, a license
2116 notice giving the public permission to use the Modified Version
2117 under the terms of this License, in the form shown in the
2119 G. Preserve in that license notice the full lists of Invariant
2120 Sections and required Cover Texts given in the Document's
2122 H. Include an unaltered copy of this License.
2123 I. Preserve the section entitled "History", and its title, and add
2124 to it an item stating at least the title, year, new authors, and
2125 publisher of the Modified Version as given on the Title Page.
2126 If there is no section entitled "History" in the Document,
2127 create one stating the title, year, authors, and publisher of
2128 the Document as given on its Title Page, then add an item
2129 describing the Modified Version as stated in the previous
2131 J. Preserve the network location, if any, given in the Document for
2132 public access to a Transparent copy of the Document, and
2133 likewise the network locations given in the Document for
2134 previous versions it was based on. These may be placed in the
2135 "History" section. You may omit a network location for a work
2136 that was published at least four years before the Document
2137 itself, or if the original publisher of the version it refers
2138 to gives permission.
2139 K. In any section entitled "Acknowledgements" or "Dedications",
2140 preserve the section's title, and preserve in the section all the
2141 substance and tone of each of the contributor acknowledgements
2142 and/or dedications given therein.
2143 L. Preserve all the Invariant Sections of the Document,
2144 unaltered in their text and in their titles. Section numbers
2145 or the equivalent are not considered part of the section titles.
2146 M. Delete any section entitled "Endorsements." Such a section
2147 may not be included in the Modified Version.
2148 N. Do not retitle any existing section as "Endorsements" or to
2149 conflict in title with any Invariant Section.
2151 If the Modified Version includes new front-matter sections or
2152 appendices that qualify as Secondary Sections and contain no
2153 material copied from the Document, you may at your option
2154 designate some or all of these sections as invariant. To do this,
2155 add their titles to the list of Invariant Sections in the Modified
2156 Version's license notice. These titles must be distinct from any
2157 other section titles.
2159 You may add a section entitled "Endorsements", provided it contains
2160 nothing but endorsements of your Modified Version by various
2161 parties-for example, statements of peer review or that the text has
2162 been approved by an organization as the authoritative definition
2165 You may add a passage of up to five words as a Front-Cover Text,
2166 and a passage of up to 25 words as a Back-Cover Text, to the end
2167 of the list of Cover Texts in the Modified Version. Only one
2168 passage of Front-Cover Text and one of Back-Cover Text may be
2169 added by (or through arrangements made by) any one entity. If the
2170 Document already includes a cover text for the same cover,
2171 previously added by you or by arrangement made by the same entity
2172 you are acting on behalf of, you may not add another; but you may
2173 replace the old one, on explicit permission from the previous
2174 publisher that added the old one.
2176 The author(s) and publisher(s) of the Document do not by this
2177 License give permission to use their names for publicity for or to
2178 assert or imply endorsement of any Modified Version.
2180 5. COMBINING DOCUMENTS
2182 You may combine the Document with other documents released under
2183 this License, under the terms defined in section 4 above for
2184 modified versions, provided that you include in the combination
2185 all of the Invariant Sections of all of the original documents,
2186 unmodified, and list them all as Invariant Sections of your
2187 combined work in its license notice.
2189 The combined work need only contain one copy of this License, and
2190 multiple identical Invariant Sections may be replaced with a single
2191 copy. If there are multiple Invariant Sections with the same name
2192 but different contents, make the title of each such section unique
2193 by adding at the end of it, in parentheses, the name of the
2194 original author or publisher of that section if known, or else a
2195 unique number. Make the same adjustment to the section titles in
2196 the list of Invariant Sections in the license notice of the
2199 In the combination, you must combine any sections entitled
2200 "History" in the various original documents, forming one section
2201 entitled "History"; likewise combine any sections entitled
2202 "Acknowledgements", and any sections entitled "Dedications." You
2203 must delete all sections entitled "Endorsements."
2205 6. COLLECTIONS OF DOCUMENTS
2207 You may make a collection consisting of the Document and other
2208 documents released under this License, and replace the individual
2209 copies of this License in the various documents with a single copy
2210 that is included in the collection, provided that you follow the
2211 rules of this License for verbatim copying of each of the
2212 documents in all other respects.
2214 You may extract a single document from such a collection, and
2215 distribute it individually under this License, provided you insert
2216 a copy of this License into the extracted document, and follow
2217 this License in all other respects regarding verbatim copying of
2220 7. AGGREGATION WITH INDEPENDENT WORKS
2222 A compilation of the Document or its derivatives with other
2223 separate and independent documents or works, in or on a volume of
2224 a storage or distribution medium, does not as a whole count as a
2225 Modified Version of the Document, provided no compilation
2226 copyright is claimed for the compilation. Such a compilation is
2227 called an "aggregate", and this License does not apply to the
2228 other self-contained works thus compiled with the Document, on
2229 account of their being thus compiled, if they are not themselves
2230 derivative works of the Document.
2232 If the Cover Text requirement of section 3 is applicable to these
2233 copies of the Document, then if the Document is less than one
2234 quarter of the entire aggregate, the Document's Cover Texts may be
2235 placed on covers that surround only the Document within the
2236 aggregate. Otherwise they must appear on covers around the whole
2241 Translation is considered a kind of modification, so you may
2242 distribute translations of the Document under the terms of section
2243 4. Replacing Invariant Sections with translations requires special
2244 permission from their copyright holders, but you may include
2245 translations of some or all Invariant Sections in addition to the
2246 original versions of these Invariant Sections. You may include a
2247 translation of this License provided that you also include the
2248 original English version of this License. In case of a
2249 disagreement between the translation and the original English
2250 version of this License, the original English version will prevail.
2254 You may not copy, modify, sublicense, or distribute the Document
2255 except as expressly provided for under this License. Any other
2256 attempt to copy, modify, sublicense or distribute the Document is
2257 void, and will automatically terminate your rights under this
2258 License. However, parties who have received copies, or rights,
2259 from you under this License will not have their licenses
2260 terminated so long as such parties remain in full compliance.
2262 10. FUTURE REVISIONS OF THIS LICENSE
2264 The Free Software Foundation may publish new, revised versions of
2265 the GNU Free Documentation License from time to time. Such new
2266 versions will be similar in spirit to the present version, but may
2267 differ in detail to address new problems or concerns. See
2268 http://www.gnu.org/copyleft/.
2270 Each version of the License is given a distinguishing version
2271 number. If the Document specifies that a particular numbered
2272 version of this License "or any later version" applies to it, you
2273 have the option of following the terms and conditions either of
2274 that specified version or of any later version that has been
2275 published (not as a draft) by the Free Software Foundation. If
2276 the Document does not specify a version number of this License,
2277 you may choose any version ever published (not as a draft) by the
2278 Free Software Foundation.
2281 ADDENDUM: How to use this License for your documents
2282 ====================================================
2284 To use this License in a document you have written, include a copy of
2285 the License in the document and put the following copyright and license
2286 notices just after the title page:
2288 Copyright (C) YEAR YOUR NAME.
2289 Permission is granted to copy, distribute and/or modify this document
2290 under the terms of the GNU Free Documentation License, Version 1.1
2291 or any later version published by the Free Software Foundation;
2292 with the Invariant Sections being LIST THEIR TITLES, with the
2293 Front-Cover Texts being LIST, and with the Back-Cover Texts being LIST.
2294 A copy of the license is included in the section entitled "GNU
2295 Free Documentation License."
2297 If you have no Invariant Sections, write "with no Invariant Sections"
2298 instead of saying which ones are invariant. If you have no Front-Cover
2299 Texts, write "no Front-Cover Texts" instead of "Front-Cover Texts being
2300 LIST"; likewise for Back-Cover Texts.
2302 If your document contains nontrivial examples of program code, we
2303 recommend releasing these examples in parallel under your choice of
2304 free software license, such as the GNU General Public License, to
2305 permit their use in free software.
2311 Node: Introduction
\7f2110
2312 Node: Compiling
\7f4602
2313 Node: Executing
\7f8073
2314 Node: Invoking
\7f10861
2315 Node: Output Options
\7f12276
2316 Node: Analysis Options
\7f19365
2317 Node: Miscellaneous Options
\7f22766
2318 Node: Deprecated Options
\7f24021
2319 Node: Symspecs
\7f26100
2320 Node: Output
\7f27926
2321 Node: Flat Profile
\7f28966
2322 Node: Call Graph
\7f33919
2323 Node: Primary
\7f37151
2324 Node: Callers
\7f39739
2325 Node: Subroutines
\7f41856
2326 Node: Cycles
\7f43697
2327 Node: Line-by-line
\7f50474
2328 Node: Annotated Source
\7f54547
2329 Node: Inaccuracy
\7f57546
2330 Node: Sampling Error
\7f57804
2331 Node: Assumptions
\7f60374
2332 Node: How do I?
\7f61844
2333 Node: Incompatibilities
\7f63398
2334 Node: Details
\7f64892
2335 Node: Implementation
\7f65285
2336 Node: File Format
\7f71182
2337 Node: Internals
\7f75472
2338 Node: Debugging
\7f83967
2339 Node: GNU Free Documentation License
\7f85568
OSDN Git Service
