1 This is gprof.info, produced by makeinfo version 4.8 from gprof.texi.
3 INFO-DIR-SECTION Software development
5 * gprof: (gprof). Profiling your program's execution
8 This file documents the gprof profiler of the GNU system.
10 Copyright (C) 1988-2016 Free Software Foundation, Inc.
12 Permission is granted to copy, distribute and/or modify this document
13 under the terms of the GNU Free Documentation License, Version 1.3 or
14 any later version published by the Free Software Foundation; with no
15 Invariant Sections, with no Front-Cover Texts, and with no Back-Cover
16 Texts. A copy of the license is included in the section entitled "GNU
17 Free Documentation License".
20 File: gprof.info, Node: Top, Next: Introduction, Up: (dir)
22 Profiling a Program: Where Does It Spend Its Time?
23 **************************************************
25 This manual describes the GNU profiler, `gprof', and how you can use it
26 to determine which parts of a program are taking most of the execution
27 time. We assume that you know how to write, compile, and execute
28 programs. GNU `gprof' was written by Jay Fenlason.
30 This manual is for `gprof' (GNU Binutils) version 2.27.
32 This document is distributed under the terms of the GNU Free
33 Documentation License version 1.3. A copy of the license is included
34 in the section entitled "GNU Free Documentation License".
38 * Introduction:: What profiling means, and why it is useful.
40 * Compiling:: How to compile your program for profiling.
41 * Executing:: Executing your program to generate profile data
42 * Invoking:: How to run `gprof', and its options
44 * Output:: Interpreting `gprof''s output
46 * Inaccuracy:: Potential problems you should be aware of
47 * How do I?:: Answers to common questions
48 * Incompatibilities:: (between GNU `gprof' and Unix `gprof'.)
49 * Details:: Details of how profiling is done
50 * GNU Free Documentation License:: GNU Free Documentation License
53 File: gprof.info, Node: Introduction, Next: Compiling, Prev: Top, Up: Top
55 1 Introduction to Profiling
56 ***************************
58 Profiling allows you to learn where your program spent its time and
59 which functions called which other functions while it was executing.
60 This information can show you which pieces of your program are slower
61 than you expected, and might be candidates for rewriting to make your
62 program execute faster. It can also tell you which functions are being
63 called more or less often than you expected. This may help you spot
64 bugs that had otherwise been unnoticed.
66 Since the profiler uses information collected during the actual
67 execution of your program, it can be used on programs that are too
68 large or too complex to analyze by reading the source. However, how
69 your program is run will affect the information that shows up in the
70 profile data. If you don't use some feature of your program while it
71 is being profiled, no profile information will be generated for that
74 Profiling has several steps:
76 * You must compile and link your program with profiling enabled.
77 *Note Compiling a Program for Profiling: Compiling.
79 * You must execute your program to generate a profile data file.
80 *Note Executing the Program: Executing.
82 * You must run `gprof' to analyze the profile data. *Note `gprof'
83 Command Summary: Invoking.
85 The next three chapters explain these steps in greater detail.
87 Several forms of output are available from the analysis.
89 The "flat profile" shows how much time your program spent in each
90 function, and how many times that function was called. If you simply
91 want to know which functions burn most of the cycles, it is stated
92 concisely here. *Note The Flat Profile: Flat Profile.
94 The "call graph" shows, for each function, which functions called
95 it, which other functions it called, and how many times. There is also
96 an estimate of how much time was spent in the subroutines of each
97 function. This can suggest places where you might try to eliminate
98 function calls that use a lot of time. *Note The Call Graph: Call
101 The "annotated source" listing is a copy of the program's source
102 code, labeled with the number of times each line of the program was
103 executed. *Note The Annotated Source Listing: Annotated Source.
105 To better understand how profiling works, you may wish to read a
106 description of its implementation. *Note Implementation of Profiling:
110 File: gprof.info, Node: Compiling, Next: Executing, Prev: Introduction, Up: Top
112 2 Compiling a Program for Profiling
113 ***********************************
115 The first step in generating profile information for your program is to
116 compile and link it with profiling enabled.
118 To compile a source file for profiling, specify the `-pg' option when
119 you run the compiler. (This is in addition to the options you normally
122 To link the program for profiling, if you use a compiler such as `cc'
123 to do the linking, simply specify `-pg' in addition to your usual
124 options. The same option, `-pg', alters either compilation or linking
125 to do what is necessary for profiling. Here are examples:
127 cc -g -c myprog.c utils.c -pg
128 cc -o myprog myprog.o utils.o -pg
130 The `-pg' option also works with a command that both compiles and
133 cc -o myprog myprog.c utils.c -g -pg
135 Note: The `-pg' option must be part of your compilation options as
136 well as your link options. If it is not then no call-graph data will
137 be gathered and when you run `gprof' you will get an error message like
140 gprof: gmon.out file is missing call-graph data
142 If you add the `-Q' switch to suppress the printing of the call
143 graph data you will still be able to see the time samples:
147 Each sample counts as 0.01 seconds.
148 % cumulative self self total
149 time seconds seconds calls Ts/call Ts/call name
150 44.12 0.07 0.07 zazLoop
152 20.59 0.17 0.04 bazMillion
154 If you run the linker `ld' directly instead of through a compiler
155 such as `cc', you may have to specify a profiling startup file
156 `gcrt0.o' as the first input file instead of the usual startup file
157 `crt0.o'. In addition, you would probably want to specify the
158 profiling C library, `libc_p.a', by writing `-lc_p' instead of the
159 usual `-lc'. This is not absolutely necessary, but doing this gives
160 you number-of-calls information for standard library functions such as
161 `read' and `open'. For example:
163 ld -o myprog /lib/gcrt0.o myprog.o utils.o -lc_p
165 If you are running the program on a system which supports shared
166 libraries you may run into problems with the profiling support code in
167 a shared library being called before that library has been fully
168 initialised. This is usually detected by the program encountering a
169 segmentation fault as soon as it is run. The solution is to link
170 against a static version of the library containing the profiling
171 support code, which for `gcc' users can be done via the `-static' or
172 `-static-libgcc' command line option. For example:
174 gcc -g -pg -static-libgcc myprog.c utils.c -o myprog
176 If you compile only some of the modules of the program with `-pg',
177 you can still profile the program, but you won't get complete
178 information about the modules that were compiled without `-pg'. The
179 only information you get for the functions in those modules is the
180 total time spent in them; there is no record of how many times they
181 were called, or from where. This will not affect the flat profile
182 (except that the `calls' field for the functions will be blank), but
183 will greatly reduce the usefulness of the call graph.
185 If you wish to perform line-by-line profiling you should use the
186 `gcov' tool instead of `gprof'. See that tool's manual or info pages
187 for more details of how to do this.
189 Note, older versions of `gcc' produce line-by-line profiling
190 information that works with `gprof' rather than `gcov' so there is
191 still support for displaying this kind of information in `gprof'. *Note
192 Line-by-line Profiling: Line-by-line.
194 It also worth noting that `gcc' implements a
195 `-finstrument-functions' command line option which will insert calls to
196 special user supplied instrumentation routines at the entry and exit of
197 every function in their program. This can be used to implement an
198 alternative profiling scheme.
201 File: gprof.info, Node: Executing, Next: Invoking, Prev: Compiling, Up: Top
203 3 Executing the Program
204 ***********************
206 Once the program is compiled for profiling, you must run it in order to
207 generate the information that `gprof' needs. Simply run the program as
208 usual, using the normal arguments, file names, etc. The program should
209 run normally, producing the same output as usual. It will, however, run
210 somewhat slower than normal because of the time spent collecting and
211 writing the profile data.
213 The way you run the program--the arguments and input that you give
214 it--may have a dramatic effect on what the profile information shows.
215 The profile data will describe the parts of the program that were
216 activated for the particular input you use. For example, if the first
217 command you give to your program is to quit, the profile data will show
218 the time used in initialization and in cleanup, but not much else.
220 Your program will write the profile data into a file called
221 `gmon.out' just before exiting. If there is already a file called
222 `gmon.out', its contents are overwritten. There is currently no way to
223 tell the program to write the profile data under a different name, but
224 you can rename the file afterwards if you are concerned that it may be
227 In order to write the `gmon.out' file properly, your program must
228 exit normally: by returning from `main' or by calling `exit'. Calling
229 the low-level function `_exit' does not write the profile data, and
230 neither does abnormal termination due to an unhandled signal.
232 The `gmon.out' file is written in the program's _current working
233 directory_ at the time it exits. This means that if your program calls
234 `chdir', the `gmon.out' file will be left in the last directory your
235 program `chdir''d to. If you don't have permission to write in this
236 directory, the file is not written, and you will get an error message.
238 Older versions of the GNU profiling library may also write a file
239 called `bb.out'. This file, if present, contains an human-readable
240 listing of the basic-block execution counts. Unfortunately, the
241 appearance of a human-readable `bb.out' means the basic-block counts
242 didn't get written into `gmon.out'. The Perl script `bbconv.pl',
243 included with the `gprof' source distribution, will convert a `bb.out'
244 file into a format readable by `gprof'. Invoke it like this:
246 bbconv.pl < bb.out > BH-DATA
248 This translates the information in `bb.out' into a form that `gprof'
249 can understand. But you still need to tell `gprof' about the existence
250 of this translated information. To do that, include BB-DATA on the
251 `gprof' command line, _along with `gmon.out'_, like this:
253 gprof OPTIONS EXECUTABLE-FILE gmon.out BB-DATA [YET-MORE-PROFILE-DATA-FILES...] [> OUTFILE]
256 File: gprof.info, Node: Invoking, Next: Output, Prev: Executing, Up: Top
258 4 `gprof' Command Summary
259 *************************
261 After you have a profile data file `gmon.out', you can run `gprof' to
262 interpret the information in it. The `gprof' program prints a flat
263 profile and a call graph on standard output. Typically you would
264 redirect the output of `gprof' into a file with `>'.
266 You run `gprof' like this:
268 gprof OPTIONS [EXECUTABLE-FILE [PROFILE-DATA-FILES...]] [> OUTFILE]
270 Here square-brackets indicate optional arguments.
272 If you omit the executable file name, the file `a.out' is used. If
273 you give no profile data file name, the file `gmon.out' is used. If
274 any file is not in the proper format, or if the profile data file does
275 not appear to belong to the executable file, an error message is
278 You can give more than one profile data file by entering all their
279 names after the executable file name; then the statistics in all the
280 data files are summed together.
282 The order of these options does not matter.
286 * Output Options:: Controlling `gprof''s output style
287 * Analysis Options:: Controlling how `gprof' analyzes its data
288 * Miscellaneous Options::
289 * Deprecated Options:: Options you no longer need to use, but which
290 have been retained for compatibility
291 * Symspecs:: Specifying functions to include or exclude
294 File: gprof.info, Node: Output Options, Next: Analysis Options, Up: Invoking
299 These options specify which of several output formats `gprof' should
302 Many of these options take an optional "symspec" to specify
303 functions to be included or excluded. These options can be specified
304 multiple times, with different symspecs, to include or exclude sets of
305 symbols. *Note Symspecs: Symspecs.
307 Specifying any of these options overrides the default (`-p -q'),
308 which prints a flat profile and call graph analysis for all functions.
311 `--annotated-source[=SYMSPEC]'
312 The `-A' option causes `gprof' to print annotated source code. If
313 SYMSPEC is specified, print output only for matching symbols.
314 *Note The Annotated Source Listing: Annotated Source.
318 If the `-b' option is given, `gprof' doesn't print the verbose
319 blurbs that try to explain the meaning of all of the fields in the
320 tables. This is useful if you intend to print out the output, or
321 are tired of seeing the blurbs.
324 `--exec-counts[=SYMSPEC]'
325 The `-C' option causes `gprof' to print a tally of functions and
326 the number of times each was called. If SYMSPEC is specified,
327 print tally only for matching symbols.
329 If the profile data file contains basic-block count records,
330 specifying the `-l' option, along with `-C', will cause basic-block
331 execution counts to be tallied and displayed.
335 The `-i' option causes `gprof' to display summary information
336 about the profile data file(s) and then exit. The number of
337 histogram, call graph, and basic-block count records is displayed.
340 `--directory-path=DIRS'
341 The `-I' option specifies a list of search directories in which to
342 find source files. Environment variable GPROF_PATH can also be
343 used to convey this information. Used mostly for annotated source
347 `--no-annotated-source[=SYMSPEC]'
348 The `-J' option causes `gprof' not to print annotated source code.
349 If SYMSPEC is specified, `gprof' prints annotated source, but
350 excludes matching symbols.
354 Normally, source filenames are printed with the path component
355 suppressed. The `-L' option causes `gprof' to print the full
356 pathname of source filenames, which is determined from symbolic
357 debugging information in the image file and is relative to the
358 directory in which the compiler was invoked.
361 `--flat-profile[=SYMSPEC]'
362 The `-p' option causes `gprof' to print a flat profile. If
363 SYMSPEC is specified, print flat profile only for matching symbols.
364 *Note The Flat Profile: Flat Profile.
367 `--no-flat-profile[=SYMSPEC]'
368 The `-P' option causes `gprof' to suppress printing a flat profile.
369 If SYMSPEC is specified, `gprof' prints a flat profile, but
370 excludes matching symbols.
374 The `-q' option causes `gprof' to print the call graph analysis.
375 If SYMSPEC is specified, print call graph only for matching symbols
376 and their children. *Note The Call Graph: Call Graph.
379 `--no-graph[=SYMSPEC]'
380 The `-Q' option causes `gprof' to suppress printing the call graph.
381 If SYMSPEC is specified, `gprof' prints a call graph, but excludes
386 The `-t' option causes the NUM most active source lines in each
387 source file to be listed when source annotation is enabled. The
392 This option affects annotated source output only. Normally,
393 `gprof' prints annotated source files to standard-output. If this
394 option is specified, annotated source for a file named
395 `path/FILENAME' is generated in the file `FILENAME-ann'. If the
396 underlying file system would truncate `FILENAME-ann' so that it
397 overwrites the original `FILENAME', `gprof' generates annotated
398 source in the file `FILENAME.ann' instead (if the original file
399 name has an extension, that extension is _replaced_ with `.ann').
402 `--no-exec-counts[=SYMSPEC]'
403 The `-Z' option causes `gprof' not to print a tally of functions
404 and the number of times each was called. If SYMSPEC is specified,
405 print tally, but exclude matching symbols.
408 `--function-ordering'
409 The `--function-ordering' option causes `gprof' to print a
410 suggested function ordering for the program based on profiling
411 data. This option suggests an ordering which may improve paging,
412 tlb and cache behavior for the program on systems which support
413 arbitrary ordering of functions in an executable.
415 The exact details of how to force the linker to place functions in
416 a particular order is system dependent and out of the scope of this
420 `--file-ordering MAP_FILE'
421 The `--file-ordering' option causes `gprof' to print a suggested
422 .o link line ordering for the program based on profiling data.
423 This option suggests an ordering which may improve paging, tlb and
424 cache behavior for the program on systems which do not support
425 arbitrary ordering of functions in an executable.
427 Use of the `-a' argument is highly recommended with this option.
429 The MAP_FILE argument is a pathname to a file which provides
430 function name to object file mappings. The format of the file is
431 similar to the output of the program `nm'.
433 c-parse.o:00000000 T yyparse
434 c-parse.o:00000004 C yyerrflag
435 c-lang.o:00000000 T maybe_objc_method_name
436 c-lang.o:00000000 T print_lang_statistics
437 c-lang.o:00000000 T recognize_objc_keyword
438 c-decl.o:00000000 T print_lang_identifier
439 c-decl.o:00000000 T print_lang_type
442 To create a MAP_FILE with GNU `nm', type a command like `nm
443 --extern-only --defined-only -v --print-file-name program-name'.
447 The `-T' option causes `gprof' to print its output in
448 "traditional" BSD style.
452 Sets width of output lines to WIDTH. Currently only used when
453 printing the function index at the bottom of the call graph.
457 This option affects annotated source output only. By default,
458 only the lines at the beginning of a basic-block are annotated.
459 If this option is specified, every line in a basic-block is
460 annotated by repeating the annotation for the first line. This
461 behavior is similar to `tcov''s `-a'.
465 These options control whether C++ symbol names should be demangled
466 when printing output. The default is to demangle symbols. The
467 `--no-demangle' option may be used to turn off demangling.
468 Different compilers have different mangling styles. The optional
469 demangling style argument can be used to choose an appropriate
470 demangling style for your compiler.
473 File: gprof.info, Node: Analysis Options, Next: Miscellaneous Options, Prev: Output Options, Up: Invoking
480 The `-a' option causes `gprof' to suppress the printing of
481 statically declared (private) functions. (These are functions
482 whose names are not listed as global, and which are not visible
483 outside the file/function/block where they were defined.) Time
484 spent in these functions, calls to/from them, etc., will all be
485 attributed to the function that was loaded directly before it in
486 the executable file. This option affects both the flat profile
490 `--static-call-graph'
491 The `-c' option causes the call graph of the program to be
492 augmented by a heuristic which examines the text space of the
493 object file and identifies function calls in the binary machine
494 code. Since normal call graph records are only generated when
495 functions are entered, this option identifies children that could
496 have been called, but never were. Calls to functions that were
497 not compiled with profiling enabled are also identified, but only
498 if symbol table entries are present for them. Calls to dynamic
499 library routines are typically _not_ found by this option.
500 Parents or children identified via this heuristic are indicated in
501 the call graph with call counts of `0'.
504 `--ignore-non-functions'
505 The `-D' option causes `gprof' to ignore symbols which are not
506 known to be functions. This option will give more accurate
507 profile data on systems where it is supported (Solaris and HPUX for
511 The `-k' option allows you to delete from the call graph any arcs
512 from symbols matching symspec FROM to those matching symspec TO.
516 The `-l' option enables line-by-line profiling, which causes
517 histogram hits to be charged to individual source code lines,
518 instead of functions. This feature only works with programs
519 compiled by older versions of the `gcc' compiler. Newer versions
520 of `gcc' are designed to work with the `gcov' tool instead.
522 If the program was compiled with basic-block counting enabled,
523 this option will also identify how many times each line of code
524 was executed. While line-by-line profiling can help isolate where
525 in a large function a program is spending its time, it also
526 significantly increases the running time of `gprof', and magnifies
527 statistical inaccuracies. *Note Statistical Sampling Error:
530 `--inline-file-names'
531 This option causes `gprof' to print the source file after each
532 symbol in both the flat profile and the call graph. The full path
533 to the file is printed if used with the `-L' option.
537 This option affects execution count output only. Symbols that are
538 executed less than NUM times are suppressed.
542 The `-n' option causes `gprof', in its call graph analysis, to
543 only propagate times for symbols matching SYMSPEC.
547 The `-n' option causes `gprof', in its call graph analysis, not to
548 propagate times for symbols matching SYMSPEC.
551 `--external-symbol-table=FILENAME'
552 The `-S' option causes `gprof' to read an external symbol table
553 file, such as `/proc/kallsyms', rather than read the symbol table
554 from the given object file (the default is `a.out'). This is useful
555 for profiling kernel modules.
558 `--display-unused-functions'
559 If you give the `-z' option, `gprof' will mention all functions in
560 the flat profile, even those that were never called, and that had
561 no time spent in them. This is useful in conjunction with the
562 `-c' option for discovering which routines were never called.
566 File: gprof.info, Node: Miscellaneous Options, Next: Deprecated Options, Prev: Analysis Options, Up: Invoking
568 4.3 Miscellaneous Options
569 =========================
573 The `-d NUM' option specifies debugging options. If NUM is not
574 specified, enable all debugging. *Note Debugging `gprof':
579 The `-h' option prints command line usage.
583 Selects the format of the profile data files. Recognized formats
584 are `auto' (the default), `bsd', `4.4bsd', `magic', and `prof'
589 The `-s' option causes `gprof' to summarize the information in the
590 profile data files it read in, and write out a profile data file
591 called `gmon.sum', which contains all the information from the
592 profile data files that `gprof' read in. The file `gmon.sum' may
593 be one of the specified input files; the effect of this is to
594 merge the data in the other input files into `gmon.sum'.
596 Eventually you can run `gprof' again without `-s' to analyze the
597 cumulative data in the file `gmon.sum'.
601 The `-v' flag causes `gprof' to print the current version number,
606 File: gprof.info, Node: Deprecated Options, Next: Symspecs, Prev: Miscellaneous Options, Up: Invoking
608 4.4 Deprecated Options
609 ======================
611 These options have been replaced with newer versions that use symspecs.
614 The `-e FUNCTION' option tells `gprof' to not print information
615 about the function FUNCTION_NAME (and its children...) in the call
616 graph. The function will still be listed as a child of any
617 functions that call it, but its index number will be shown as
618 `[not printed]'. More than one `-e' option may be given; only one
619 FUNCTION_NAME may be indicated with each `-e' option.
622 The `-E FUNCTION' option works like the `-e' option, but time
623 spent in the function (and children who were not called from
624 anywhere else), will not be used to compute the
625 percentages-of-time for the call graph. More than one `-E' option
626 may be given; only one FUNCTION_NAME may be indicated with each
630 The `-f FUNCTION' option causes `gprof' to limit the call graph to
631 the function FUNCTION_NAME and its children (and their
632 children...). More than one `-f' option may be given; only one
633 FUNCTION_NAME may be indicated with each `-f' option.
636 The `-F FUNCTION' option works like the `-f' option, but only time
637 spent in the function and its children (and their children...)
638 will be used to determine total-time and percentages-of-time for
639 the call graph. More than one `-F' option may be given; only one
640 FUNCTION_NAME may be indicated with each `-F' option. The `-F'
641 option overrides the `-E' option.
644 Note that only one function can be specified with each `-e', `-E',
645 `-f' or `-F' option. To specify more than one function, use multiple
646 options. For example, this command:
648 gprof -e boring -f foo -f bar myprogram > gprof.output
650 lists in the call graph all functions that were reached from either
651 `foo' or `bar' and were not reachable from `boring'.
654 File: gprof.info, Node: Symspecs, Prev: Deprecated Options, Up: Invoking
659 Many of the output options allow functions to be included or excluded
660 using "symspecs" (symbol specifications), which observe the following
663 filename_containing_a_dot
664 | funcname_not_containing_a_dot
666 | ( [ any_filename ] `:' ( any_funcname | linenumber ) )
668 Here are some sample symspecs:
671 Selects everything in file `main.c'--the dot in the string tells
672 `gprof' to interpret the string as a filename, rather than as a
673 function name. To select a file whose name does not contain a
674 dot, a trailing colon should be specified. For example, `odd:' is
675 interpreted as the file named `odd'.
678 Selects all functions named `main'.
680 Note that there may be multiple instances of the same function name
681 because some of the definitions may be local (i.e., static).
682 Unless a function name is unique in a program, you must use the
683 colon notation explained below to specify a function from a
684 specific source file.
686 Sometimes, function names contain dots. In such cases, it is
687 necessary to add a leading colon to the name. For example,
688 `:.mul' selects function `.mul'.
690 In some object file formats, symbols have a leading underscore.
691 `gprof' will normally not print these underscores. When you name a
692 symbol in a symspec, you should type it exactly as `gprof' prints
693 it in its output. For example, if the compiler produces a symbol
694 `_main' from your `main' function, `gprof' still prints it as
695 `main' in its output, so you should use `main' in symspecs.
698 Selects function `main' in file `main.c'.
701 Selects line 134 in file `main.c'.
704 File: gprof.info, Node: Output, Next: Inaccuracy, Prev: Invoking, Up: Top
706 5 Interpreting `gprof''s Output
707 *******************************
709 `gprof' can produce several different output styles, the most important
710 of which are described below. The simplest output styles (file
711 information, execution count, and function and file ordering) are not
712 described here, but are documented with the respective options that
713 trigger them. *Note Output Options: Output Options.
717 * Flat Profile:: The flat profile shows how much time was spent
718 executing directly in each function.
719 * Call Graph:: The call graph shows which functions called which
720 others, and how much time each function used
721 when its subroutine calls are included.
722 * Line-by-line:: `gprof' can analyze individual source code lines
723 * Annotated Source:: The annotated source listing displays source code
724 labeled with execution counts
727 File: gprof.info, Node: Flat Profile, Next: Call Graph, Up: Output
732 The "flat profile" shows the total amount of time your program spent
733 executing each function. Unless the `-z' option is given, functions
734 with no apparent time spent in them, and no apparent calls to them, are
735 not mentioned. Note that if a function was not compiled for profiling,
736 and didn't run long enough to show up on the program counter histogram,
737 it will be indistinguishable from a function that was never called.
739 This is part of a flat profile for a small program:
743 Each sample counts as 0.01 seconds.
744 % cumulative self self total
745 time seconds seconds calls ms/call ms/call name
746 33.34 0.02 0.02 7208 0.00 0.00 open
747 16.67 0.03 0.01 244 0.04 0.12 offtime
748 16.67 0.04 0.01 8 1.25 1.25 memccpy
749 16.67 0.05 0.01 7 1.43 1.43 write
750 16.67 0.06 0.01 mcount
751 0.00 0.06 0.00 236 0.00 0.00 tzset
752 0.00 0.06 0.00 192 0.00 0.00 tolower
753 0.00 0.06 0.00 47 0.00 0.00 strlen
754 0.00 0.06 0.00 45 0.00 0.00 strchr
755 0.00 0.06 0.00 1 0.00 50.00 main
756 0.00 0.06 0.00 1 0.00 0.00 memcpy
757 0.00 0.06 0.00 1 0.00 10.11 print
758 0.00 0.06 0.00 1 0.00 0.00 profil
759 0.00 0.06 0.00 1 0.00 50.00 report
762 The functions are sorted first by decreasing run-time spent in them,
763 then by decreasing number of calls, then alphabetically by name. The
764 functions `mcount' and `profil' are part of the profiling apparatus and
765 appear in every flat profile; their time gives a measure of the amount
766 of overhead due to profiling.
768 Just before the column headers, a statement appears indicating how
769 much time each sample counted as. This "sampling period" estimates the
770 margin of error in each of the time figures. A time figure that is not
771 much larger than this is not reliable. In this example, each sample
772 counted as 0.01 seconds, suggesting a 100 Hz sampling rate. The
773 program's total execution time was 0.06 seconds, as indicated by the
774 `cumulative seconds' field. Since each sample counted for 0.01
775 seconds, this means only six samples were taken during the run. Two of
776 the samples occurred while the program was in the `open' function, as
777 indicated by the `self seconds' field. Each of the other four samples
778 occurred one each in `offtime', `memccpy', `write', and `mcount'.
779 Since only six samples were taken, none of these values can be regarded
780 as particularly reliable. In another run, the `self seconds' field for
781 `mcount' might well be `0.00' or `0.02'. *Note Statistical Sampling
782 Error: Sampling Error, for a complete discussion.
784 The remaining functions in the listing (those whose `self seconds'
785 field is `0.00') didn't appear in the histogram samples at all.
786 However, the call graph indicated that they were called, so therefore
787 they are listed, sorted in decreasing order by the `calls' field.
788 Clearly some time was spent executing these functions, but the paucity
789 of histogram samples prevents any determination of how much time each
792 Here is what the fields in each line mean:
795 This is the percentage of the total execution time your program
796 spent in this function. These should all add up to 100%.
799 This is the cumulative total number of seconds the computer spent
800 executing this functions, plus the time spent in all the functions
801 above this one in this table.
804 This is the number of seconds accounted for by this function alone.
805 The flat profile listing is sorted first by this number.
808 This is the total number of times the function was called. If the
809 function was never called, or the number of times it was called
810 cannot be determined (probably because the function was not
811 compiled with profiling enabled), the "calls" field is blank.
814 This represents the average number of milliseconds spent in this
815 function per call, if this function is profiled. Otherwise, this
816 field is blank for this function.
819 This represents the average number of milliseconds spent in this
820 function and its descendants per call, if this function is
821 profiled. Otherwise, this field is blank for this function. This
822 is the only field in the flat profile that uses call graph
826 This is the name of the function. The flat profile is sorted by
827 this field alphabetically after the "self seconds" and "calls"
831 File: gprof.info, Node: Call Graph, Next: Line-by-line, Prev: Flat Profile, Up: Output
836 The "call graph" shows how much time was spent in each function and its
837 children. From this information, you can find functions that, while
838 they themselves may not have used much time, called other functions
839 that did use unusual amounts of time.
841 Here is a sample call from a small program. This call came from the
842 same `gprof' run as the flat profile example in the previous section.
844 granularity: each sample hit covers 2 byte(s) for 20.00% of 0.05 seconds
846 index % time self children called name
848 [1] 100.0 0.00 0.05 start [1]
849 0.00 0.05 1/1 main [2]
850 0.00 0.00 1/2 on_exit [28]
851 0.00 0.00 1/1 exit [59]
852 -----------------------------------------------
853 0.00 0.05 1/1 start [1]
854 [2] 100.0 0.00 0.05 1 main [2]
855 0.00 0.05 1/1 report [3]
856 -----------------------------------------------
857 0.00 0.05 1/1 main [2]
858 [3] 100.0 0.00 0.05 1 report [3]
859 0.00 0.03 8/8 timelocal [6]
860 0.00 0.01 1/1 print [9]
861 0.00 0.01 9/9 fgets [12]
862 0.00 0.00 12/34 strncmp <cycle 1> [40]
863 0.00 0.00 8/8 lookup [20]
864 0.00 0.00 1/1 fopen [21]
865 0.00 0.00 8/8 chewtime [24]
866 0.00 0.00 8/16 skipspace [44]
867 -----------------------------------------------
868 [4] 59.8 0.01 0.02 8+472 <cycle 2 as a whole> [4]
869 0.01 0.02 244+260 offtime <cycle 2> [7]
870 0.00 0.00 236+1 tzset <cycle 2> [26]
871 -----------------------------------------------
873 The lines full of dashes divide this table into "entries", one for
874 each function. Each entry has one or more lines.
876 In each entry, the primary line is the one that starts with an index
877 number in square brackets. The end of this line says which function
878 the entry is for. The preceding lines in the entry describe the
879 callers of this function and the following lines describe its
880 subroutines (also called "children" when we speak of the call graph).
882 The entries are sorted by time spent in the function and its
885 The internal profiling function `mcount' (*note The Flat Profile:
886 Flat Profile.) is never mentioned in the call graph.
890 * Primary:: Details of the primary line's contents.
891 * Callers:: Details of caller-lines' contents.
892 * Subroutines:: Details of subroutine-lines' contents.
893 * Cycles:: When there are cycles of recursion,
894 such as `a' calls `b' calls `a'...
897 File: gprof.info, Node: Primary, Next: Callers, Up: Call Graph
899 5.2.1 The Primary Line
900 ----------------------
902 The "primary line" in a call graph entry is the line that describes the
903 function which the entry is about and gives the overall statistics for
906 For reference, we repeat the primary line from the entry for function
907 `report' in our main example, together with the heading line that shows
908 the names of the fields:
910 index % time self children called name
912 [3] 100.0 0.00 0.05 1 report [3]
914 Here is what the fields in the primary line mean:
917 Entries are numbered with consecutive integers. Each function
918 therefore has an index number, which appears at the beginning of
921 Each cross-reference to a function, as a caller or subroutine of
922 another, gives its index number as well as its name. The index
923 number guides you if you wish to look for the entry for that
927 This is the percentage of the total time that was spent in this
928 function, including time spent in subroutines called from this
931 The time spent in this function is counted again for the callers of
932 this function. Therefore, adding up these percentages is
936 This is the total amount of time spent in this function. This
937 should be identical to the number printed in the `seconds' field
938 for this function in the flat profile.
941 This is the total amount of time spent in the subroutine calls
942 made by this function. This should be equal to the sum of all the
943 `self' and `children' entries of the children listed directly
947 This is the number of times the function was called.
949 If the function called itself recursively, there are two numbers,
950 separated by a `+'. The first number counts non-recursive calls,
951 and the second counts recursive calls.
953 In the example above, the function `report' was called once from
957 This is the name of the current function. The index number is
960 If the function is part of a cycle of recursion, the cycle number
961 is printed between the function's name and the index number (*note
962 How Mutually Recursive Functions Are Described: Cycles.). For
963 example, if function `gnurr' is part of cycle number one, and has
964 index number twelve, its primary line would be end like this:
969 File: gprof.info, Node: Callers, Next: Subroutines, Prev: Primary, Up: Call Graph
971 5.2.2 Lines for a Function's Callers
972 ------------------------------------
974 A function's entry has a line for each function it was called by.
975 These lines' fields correspond to the fields of the primary line, but
976 their meanings are different because of the difference in context.
978 For reference, we repeat two lines from the entry for the function
979 `report', the primary line and one caller-line preceding it, together
980 with the heading line that shows the names of the fields:
982 index % time self children called name
984 0.00 0.05 1/1 main [2]
985 [3] 100.0 0.00 0.05 1 report [3]
987 Here are the meanings of the fields in the caller-line for `report'
991 An estimate of the amount of time spent in `report' itself when it
992 was called from `main'.
995 An estimate of the amount of time spent in subroutines of `report'
996 when `report' was called from `main'.
998 The sum of the `self' and `children' fields is an estimate of the
999 amount of time spent within calls to `report' from `main'.
1002 Two numbers: the number of times `report' was called from `main',
1003 followed by the total number of non-recursive calls to `report'
1004 from all its callers.
1006 `name and index number'
1007 The name of the caller of `report' to which this line applies,
1008 followed by the caller's index number.
1010 Not all functions have entries in the call graph; some options to
1011 `gprof' request the omission of certain functions. When a caller
1012 has no entry of its own, it still has caller-lines in the entries
1013 of the functions it calls.
1015 If the caller is part of a recursion cycle, the cycle number is
1016 printed between the name and the index number.
1018 If the identity of the callers of a function cannot be determined, a
1019 dummy caller-line is printed which has `<spontaneous>' as the "caller's
1020 name" and all other fields blank. This can happen for signal handlers.
1023 File: gprof.info, Node: Subroutines, Next: Cycles, Prev: Callers, Up: Call Graph
1025 5.2.3 Lines for a Function's Subroutines
1026 ----------------------------------------
1028 A function's entry has a line for each of its subroutines--in other
1029 words, a line for each other function that it called. These lines'
1030 fields correspond to the fields of the primary line, but their meanings
1031 are different because of the difference in context.
1033 For reference, we repeat two lines from the entry for the function
1034 `main', the primary line and a line for a subroutine, together with the
1035 heading line that shows the names of the fields:
1037 index % time self children called name
1039 [2] 100.0 0.00 0.05 1 main [2]
1040 0.00 0.05 1/1 report [3]
1042 Here are the meanings of the fields in the subroutine-line for `main'
1046 An estimate of the amount of time spent directly within `report'
1047 when `report' was called from `main'.
1050 An estimate of the amount of time spent in subroutines of `report'
1051 when `report' was called from `main'.
1053 The sum of the `self' and `children' fields is an estimate of the
1054 total time spent in calls to `report' from `main'.
1057 Two numbers, the number of calls to `report' from `main' followed
1058 by the total number of non-recursive calls to `report'. This
1059 ratio is used to determine how much of `report''s `self' and
1060 `children' time gets credited to `main'. *Note Estimating
1061 `children' Times: Assumptions.
1064 The name of the subroutine of `main' to which this line applies,
1065 followed by the subroutine's index number.
1067 If the caller is part of a recursion cycle, the cycle number is
1068 printed between the name and the index number.
1071 File: gprof.info, Node: Cycles, Prev: Subroutines, Up: Call Graph
1073 5.2.4 How Mutually Recursive Functions Are Described
1074 ----------------------------------------------------
1076 The graph may be complicated by the presence of "cycles of recursion"
1077 in the call graph. A cycle exists if a function calls another function
1078 that (directly or indirectly) calls (or appears to call) the original
1079 function. For example: if `a' calls `b', and `b' calls `a', then `a'
1080 and `b' form a cycle.
1082 Whenever there are call paths both ways between a pair of functions,
1083 they belong to the same cycle. If `a' and `b' call each other and `b'
1084 and `c' call each other, all three make one cycle. Note that even if
1085 `b' only calls `a' if it was not called from `a', `gprof' cannot
1086 determine this, so `a' and `b' are still considered a cycle.
1088 The cycles are numbered with consecutive integers. When a function
1089 belongs to a cycle, each time the function name appears in the call
1090 graph it is followed by `<cycle NUMBER>'.
1092 The reason cycles matter is that they make the time values in the
1093 call graph paradoxical. The "time spent in children" of `a' should
1094 include the time spent in its subroutine `b' and in `b''s
1095 subroutines--but one of `b''s subroutines is `a'! How much of `a''s
1096 time should be included in the children of `a', when `a' is indirectly
1099 The way `gprof' resolves this paradox is by creating a single entry
1100 for the cycle as a whole. The primary line of this entry describes the
1101 total time spent directly in the functions of the cycle. The
1102 "subroutines" of the cycle are the individual functions of the cycle,
1103 and all other functions that were called directly by them. The
1104 "callers" of the cycle are the functions, outside the cycle, that
1105 called functions in the cycle.
1107 Here is an example portion of a call graph which shows a cycle
1108 containing functions `a' and `b'. The cycle was entered by a call to
1109 `a' from `main'; both `a' and `b' called `c'.
1111 index % time self children called name
1112 ----------------------------------------
1114 [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3]
1115 1.02 0 3 b <cycle 1> [4]
1116 0.75 0 2 a <cycle 1> [5]
1117 ----------------------------------------
1119 [4] 52.85 1.02 0 0 b <cycle 1> [4]
1122 ----------------------------------------
1125 [5] 38.86 0.75 0 1 a <cycle 1> [5]
1128 ----------------------------------------
1130 (The entire call graph for this program contains in addition an entry
1131 for `main', which calls `a', and an entry for `c', with callers `a' and
1134 index % time self children called name
1136 [1] 100.00 0 1.93 0 start [1]
1137 0.16 1.77 1/1 main [2]
1138 ----------------------------------------
1139 0.16 1.77 1/1 start [1]
1140 [2] 100.00 0.16 1.77 1 main [2]
1141 1.77 0 1/1 a <cycle 1> [5]
1142 ----------------------------------------
1144 [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3]
1145 1.02 0 3 b <cycle 1> [4]
1146 0.75 0 2 a <cycle 1> [5]
1148 ----------------------------------------
1150 [4] 52.85 1.02 0 0 b <cycle 1> [4]
1153 ----------------------------------------
1156 [5] 38.86 0.75 0 1 a <cycle 1> [5]
1159 ----------------------------------------
1160 0 0 3/6 b <cycle 1> [4]
1161 0 0 3/6 a <cycle 1> [5]
1162 [6] 0.00 0 0 6 c [6]
1163 ----------------------------------------
1165 The `self' field of the cycle's primary line is the total time spent
1166 in all the functions of the cycle. It equals the sum of the `self'
1167 fields for the individual functions in the cycle, found in the entry in
1168 the subroutine lines for these functions.
1170 The `children' fields of the cycle's primary line and subroutine
1171 lines count only subroutines outside the cycle. Even though `a' calls
1172 `b', the time spent in those calls to `b' is not counted in `a''s
1173 `children' time. Thus, we do not encounter the problem of what to do
1174 when the time in those calls to `b' includes indirect recursive calls
1177 The `children' field of a caller-line in the cycle's entry estimates
1178 the amount of time spent _in the whole cycle_, and its other
1179 subroutines, on the times when that caller called a function in the
1182 The `called' field in the primary line for the cycle has two numbers:
1183 first, the number of times functions in the cycle were called by
1184 functions outside the cycle; second, the number of times they were
1185 called by functions in the cycle (including times when a function in
1186 the cycle calls itself). This is a generalization of the usual split
1187 into non-recursive and recursive calls.
1189 The `called' field of a subroutine-line for a cycle member in the
1190 cycle's entry says how many time that function was called from
1191 functions in the cycle. The total of all these is the second number in
1192 the primary line's `called' field.
1194 In the individual entry for a function in a cycle, the other
1195 functions in the same cycle can appear as subroutines and as callers.
1196 These lines show how many times each function in the cycle called or
1197 was called from each other function in the cycle. The `self' and
1198 `children' fields in these lines are blank because of the difficulty of
1199 defining meanings for them when recursion is going on.
1202 File: gprof.info, Node: Line-by-line, Next: Annotated Source, Prev: Call Graph, Up: Output
1204 5.3 Line-by-line Profiling
1205 ==========================
1207 `gprof''s `-l' option causes the program to perform "line-by-line"
1208 profiling. In this mode, histogram samples are assigned not to
1209 functions, but to individual lines of source code. This only works
1210 with programs compiled with older versions of the `gcc' compiler.
1211 Newer versions of `gcc' use a different program - `gcov' - to display
1212 line-by-line profiling information.
1214 With the older versions of `gcc' the program usually has to be
1215 compiled with a `-g' option, in addition to `-pg', in order to generate
1216 debugging symbols for tracking source code lines. Note, in much older
1217 versions of `gcc' the program had to be compiled with the `-a' command
1218 line option as well.
1220 The flat profile is the most useful output table in line-by-line
1221 mode. The call graph isn't as useful as normal, since the current
1222 version of `gprof' does not propagate call graph arcs from source code
1223 lines to the enclosing function. The call graph does, however, show
1224 each line of code that called each function, along with a count.
1226 Here is a section of `gprof''s output, without line-by-line
1227 profiling. Note that `ct_init' accounted for four histogram hits, and
1228 13327 calls to `init_block'.
1232 Each sample counts as 0.01 seconds.
1233 % cumulative self self total
1234 time seconds seconds calls us/call us/call name
1235 30.77 0.13 0.04 6335 6.31 6.31 ct_init
1238 Call graph (explanation follows)
1241 granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds
1243 index % time self children called name
1245 0.00 0.00 1/13496 name_too_long
1246 0.00 0.00 40/13496 deflate
1247 0.00 0.00 128/13496 deflate_fast
1248 0.00 0.00 13327/13496 ct_init
1249 [7] 0.0 0.00 0.00 13496 init_block
1251 Now let's look at some of `gprof''s output from the same program run,
1252 this time with line-by-line profiling enabled. Note that `ct_init''s
1253 four histogram hits are broken down into four lines of source code--one
1254 hit occurred on each of lines 349, 351, 382 and 385. In the call graph,
1255 note how `ct_init''s 13327 calls to `init_block' are broken down into
1256 one call from line 396, 3071 calls from line 384, 3730 calls from line
1257 385, and 6525 calls from 387.
1261 Each sample counts as 0.01 seconds.
1263 time seconds seconds calls name
1264 7.69 0.10 0.01 ct_init (trees.c:349)
1265 7.69 0.11 0.01 ct_init (trees.c:351)
1266 7.69 0.12 0.01 ct_init (trees.c:382)
1267 7.69 0.13 0.01 ct_init (trees.c:385)
1270 Call graph (explanation follows)
1273 granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds
1275 % time self children called name
1277 0.00 0.00 1/13496 name_too_long (gzip.c:1440)
1278 0.00 0.00 1/13496 deflate (deflate.c:763)
1279 0.00 0.00 1/13496 ct_init (trees.c:396)
1280 0.00 0.00 2/13496 deflate (deflate.c:727)
1281 0.00 0.00 4/13496 deflate (deflate.c:686)
1282 0.00 0.00 5/13496 deflate (deflate.c:675)
1283 0.00 0.00 12/13496 deflate (deflate.c:679)
1284 0.00 0.00 16/13496 deflate (deflate.c:730)
1285 0.00 0.00 128/13496 deflate_fast (deflate.c:654)
1286 0.00 0.00 3071/13496 ct_init (trees.c:384)
1287 0.00 0.00 3730/13496 ct_init (trees.c:385)
1288 0.00 0.00 6525/13496 ct_init (trees.c:387)
1289 [6] 0.0 0.00 0.00 13496 init_block (trees.c:408)
1292 File: gprof.info, Node: Annotated Source, Prev: Line-by-line, Up: Output
1294 5.4 The Annotated Source Listing
1295 ================================
1297 `gprof''s `-A' option triggers an annotated source listing, which lists
1298 the program's source code, each function labeled with the number of
1299 times it was called. You may also need to specify the `-I' option, if
1300 `gprof' can't find the source code files.
1302 With older versions of `gcc' compiling with `gcc ... -g -pg -a'
1303 augments your program with basic-block counting code, in addition to
1304 function counting code. This enables `gprof' to determine how many
1305 times each line of code was executed. With newer versions of `gcc'
1306 support for displaying basic-block counts is provided by the `gcov'
1309 For example, consider the following function, taken from gzip, with
1318 7 static ulg crc = (ulg)0xffffffffL;
1325 14 c = crc_32_tab[...];
1329 18 return c ^ 0xffffffffL;
1332 `updcrc' has at least five basic-blocks. One is the function
1333 itself. The `if' statement on line 9 generates two more basic-blocks,
1334 one for each branch of the `if'. A fourth basic-block results from the
1335 `if' on line 13, and the contents of the `do' loop form the fifth
1336 basic-block. The compiler may also generate additional basic-blocks to
1337 handle various special cases.
1339 A program augmented for basic-block counting can be analyzed with
1340 `gprof -l -A'. The `-x' option is also helpful, to ensure that each
1341 line of code is labeled at least once. Here is `updcrc''s annotated
1342 source listing for a sample `gzip' run:
1350 static ulg crc = (ulg)0xffffffffL;
1352 2 -> if (s == NULL) {
1353 1 -> c = 0xffffffffL;
1357 26312 -> c = crc_32_tab[...];
1358 26312,1,26311 -> } while (--n);
1361 2 -> return c ^ 0xffffffffL;
1364 In this example, the function was called twice, passing once through
1365 each branch of the `if' statement. The body of the `do' loop was
1366 executed a total of 26312 times. Note how the `while' statement is
1367 annotated. It began execution 26312 times, once for each iteration
1368 through the loop. One of those times (the last time) it exited, while
1369 it branched back to the beginning of the loop 26311 times.
1372 File: gprof.info, Node: Inaccuracy, Next: How do I?, Prev: Output, Up: Top
1374 6 Inaccuracy of `gprof' Output
1375 ******************************
1379 * Sampling Error:: Statistical margins of error
1380 * Assumptions:: Estimating children times
1383 File: gprof.info, Node: Sampling Error, Next: Assumptions, Up: Inaccuracy
1385 6.1 Statistical Sampling Error
1386 ==============================
1388 The run-time figures that `gprof' gives you are based on a sampling
1389 process, so they are subject to statistical inaccuracy. If a function
1390 runs only a small amount of time, so that on the average the sampling
1391 process ought to catch that function in the act only once, there is a
1392 pretty good chance it will actually find that function zero times, or
1395 By contrast, the number-of-calls and basic-block figures are derived
1396 by counting, not sampling. They are completely accurate and will not
1397 vary from run to run if your program is deterministic and single
1398 threaded. In multi-threaded applications, or single threaded
1399 applications that link with multi-threaded libraries, the counts are
1400 only deterministic if the counting function is thread-safe. (Note:
1401 beware that the mcount counting function in glibc is _not_
1402 thread-safe). *Note Implementation of Profiling: Implementation.
1404 The "sampling period" that is printed at the beginning of the flat
1405 profile says how often samples are taken. The rule of thumb is that a
1406 run-time figure is accurate if it is considerably bigger than the
1409 The actual amount of error can be predicted. For N samples, the
1410 _expected_ error is the square-root of N. For example, if the sampling
1411 period is 0.01 seconds and `foo''s run-time is 1 second, N is 100
1412 samples (1 second/0.01 seconds), sqrt(N) is 10 samples, so the expected
1413 error in `foo''s run-time is 0.1 seconds (10*0.01 seconds), or ten
1414 percent of the observed value. Again, if the sampling period is 0.01
1415 seconds and `bar''s run-time is 100 seconds, N is 10000 samples,
1416 sqrt(N) is 100 samples, so the expected error in `bar''s run-time is 1
1417 second, or one percent of the observed value. It is likely to vary
1418 this much _on the average_ from one profiling run to the next.
1419 (_Sometimes_ it will vary more.)
1421 This does not mean that a small run-time figure is devoid of
1422 information. If the program's _total_ run-time is large, a small
1423 run-time for one function does tell you that that function used an
1424 insignificant fraction of the whole program's time. Usually this means
1425 it is not worth optimizing.
1427 One way to get more accuracy is to give your program more (but
1428 similar) input data so it will take longer. Another way is to combine
1429 the data from several runs, using the `-s' option of `gprof'. Here is
1432 1. Run your program once.
1434 2. Issue the command `mv gmon.out gmon.sum'.
1436 3. Run your program again, the same as before.
1438 4. Merge the new data in `gmon.out' into `gmon.sum' with this command:
1440 gprof -s EXECUTABLE-FILE gmon.out gmon.sum
1442 5. Repeat the last two steps as often as you wish.
1444 6. Analyze the cumulative data using this command:
1446 gprof EXECUTABLE-FILE gmon.sum > OUTPUT-FILE
1449 File: gprof.info, Node: Assumptions, Prev: Sampling Error, Up: Inaccuracy
1451 6.2 Estimating `children' Times
1452 ===============================
1454 Some of the figures in the call graph are estimates--for example, the
1455 `children' time values and all the time figures in caller and
1458 There is no direct information about these measurements in the
1459 profile data itself. Instead, `gprof' estimates them by making an
1460 assumption about your program that might or might not be true.
1462 The assumption made is that the average time spent in each call to
1463 any function `foo' is not correlated with who called `foo'. If `foo'
1464 used 5 seconds in all, and 2/5 of the calls to `foo' came from `a',
1465 then `foo' contributes 2 seconds to `a''s `children' time, by
1468 This assumption is usually true enough, but for some programs it is
1469 far from true. Suppose that `foo' returns very quickly when its
1470 argument is zero; suppose that `a' always passes zero as an argument,
1471 while other callers of `foo' pass other arguments. In this program,
1472 all the time spent in `foo' is in the calls from callers other than `a'.
1473 But `gprof' has no way of knowing this; it will blindly and incorrectly
1474 charge 2 seconds of time in `foo' to the children of `a'.
1476 We hope some day to put more complete data into `gmon.out', so that
1477 this assumption is no longer needed, if we can figure out how. For the
1478 novice, the estimated figures are usually more useful than misleading.
1481 File: gprof.info, Node: How do I?, Next: Incompatibilities, Prev: Inaccuracy, Up: Top
1483 7 Answers to Common Questions
1484 *****************************
1486 How can I get more exact information about hot spots in my program?
1487 Looking at the per-line call counts only tells part of the story.
1488 Because `gprof' can only report call times and counts by function,
1489 the best way to get finer-grained information on where the program
1490 is spending its time is to re-factor large functions into sequences
1491 of calls to smaller ones. Beware however that this can introduce
1492 artificial hot spots since compiling with `-pg' adds a significant
1493 overhead to function calls. An alternative solution is to use a
1494 non-intrusive profiler, e.g. oprofile.
1496 How do I find which lines in my program were executed the most times?
1497 Use the `gcov' program.
1499 How do I find which lines in my program called a particular function?
1500 Use `gprof -l' and lookup the function in the call graph. The
1501 callers will be broken down by function and line number.
1503 How do I analyze a program that runs for less than a second?
1504 Try using a shell script like this one:
1506 for i in `seq 1 100`; do
1508 mv gmon.out gmon.out.$i
1511 gprof -s fastprog gmon.out.*
1513 gprof fastprog gmon.sum
1515 If your program is completely deterministic, all the call counts
1516 will be simple multiples of 100 (i.e., a function called once in
1517 each run will appear with a call count of 100).
1521 File: gprof.info, Node: Incompatibilities, Next: Details, Prev: How do I?, Up: Top
1523 8 Incompatibilities with Unix `gprof'
1524 *************************************
1526 GNU `gprof' and Berkeley Unix `gprof' use the same data file
1527 `gmon.out', and provide essentially the same information. But there
1528 are a few differences.
1530 * GNU `gprof' uses a new, generalized file format with support for
1531 basic-block execution counts and non-realtime histograms. A magic
1532 cookie and version number allows `gprof' to easily identify new
1533 style files. Old BSD-style files can still be read. *Note
1534 Profiling Data File Format: File Format.
1536 * For a recursive function, Unix `gprof' lists the function as a
1537 parent and as a child, with a `calls' field that lists the number
1538 of recursive calls. GNU `gprof' omits these lines and puts the
1539 number of recursive calls in the primary line.
1541 * When a function is suppressed from the call graph with `-e', GNU
1542 `gprof' still lists it as a subroutine of functions that call it.
1544 * GNU `gprof' accepts the `-k' with its argument in the form
1545 `from/to', instead of `from to'.
1547 * In the annotated source listing, if there are multiple basic
1548 blocks on the same line, GNU `gprof' prints all of their counts,
1549 separated by commas.
1551 * The blurbs, field widths, and output formats are different. GNU
1552 `gprof' prints blurbs after the tables, so that you can see the
1553 tables without skipping the blurbs.
1556 File: gprof.info, Node: Details, Next: GNU Free Documentation License, Prev: Incompatibilities, Up: Top
1558 9 Details of Profiling
1559 **********************
1563 * Implementation:: How a program collects profiling information
1564 * File Format:: Format of `gmon.out' files
1565 * Internals:: `gprof''s internal operation
1566 * Debugging:: Using `gprof''s `-d' option
1569 File: gprof.info, Node: Implementation, Next: File Format, Up: Details
1571 9.1 Implementation of Profiling
1572 ===============================
1574 Profiling works by changing how every function in your program is
1575 compiled so that when it is called, it will stash away some information
1576 about where it was called from. From this, the profiler can figure out
1577 what function called it, and can count how many times it was called.
1578 This change is made by the compiler when your program is compiled with
1579 the `-pg' option, which causes every function to call `mcount' (or
1580 `_mcount', or `__mcount', depending on the OS and compiler) as one of
1581 its first operations.
1583 The `mcount' routine, included in the profiling library, is
1584 responsible for recording in an in-memory call graph table both its
1585 parent routine (the child) and its parent's parent. This is typically
1586 done by examining the stack frame to find both the address of the
1587 child, and the return address in the original parent. Since this is a
1588 very machine-dependent operation, `mcount' itself is typically a short
1589 assembly-language stub routine that extracts the required information,
1590 and then calls `__mcount_internal' (a normal C function) with two
1591 arguments--`frompc' and `selfpc'. `__mcount_internal' is responsible
1592 for maintaining the in-memory call graph, which records `frompc',
1593 `selfpc', and the number of times each of these call arcs was traversed.
1595 GCC Version 2 provides a magical function
1596 (`__builtin_return_address'), which allows a generic `mcount' function
1597 to extract the required information from the stack frame. However, on
1598 some architectures, most notably the SPARC, using this builtin can be
1599 very computationally expensive, and an assembly language version of
1600 `mcount' is used for performance reasons.
1602 Number-of-calls information for library routines is collected by
1603 using a special version of the C library. The programs in it are the
1604 same as in the usual C library, but they were compiled with `-pg'. If
1605 you link your program with `gcc ... -pg', it automatically uses the
1606 profiling version of the library.
1608 Profiling also involves watching your program as it runs, and
1609 keeping a histogram of where the program counter happens to be every
1610 now and then. Typically the program counter is looked at around 100
1611 times per second of run time, but the exact frequency may vary from
1614 This is done is one of two ways. Most UNIX-like operating systems
1615 provide a `profil()' system call, which registers a memory array with
1616 the kernel, along with a scale factor that determines how the program's
1617 address space maps into the array. Typical scaling values cause every
1618 2 to 8 bytes of address space to map into a single array slot. On
1619 every tick of the system clock (assuming the profiled program is
1620 running), the value of the program counter is examined and the
1621 corresponding slot in the memory array is incremented. Since this is
1622 done in the kernel, which had to interrupt the process anyway to handle
1623 the clock interrupt, very little additional system overhead is required.
1625 However, some operating systems, most notably Linux 2.0 (and
1626 earlier), do not provide a `profil()' system call. On such a system,
1627 arrangements are made for the kernel to periodically deliver a signal
1628 to the process (typically via `setitimer()'), which then performs the
1629 same operation of examining the program counter and incrementing a slot
1630 in the memory array. Since this method requires a signal to be
1631 delivered to user space every time a sample is taken, it uses
1632 considerably more overhead than kernel-based profiling. Also, due to
1633 the added delay required to deliver the signal, this method is less
1636 A special startup routine allocates memory for the histogram and
1637 either calls `profil()' or sets up a clock signal handler. This
1638 routine (`monstartup') can be invoked in several ways. On Linux
1639 systems, a special profiling startup file `gcrt0.o', which invokes
1640 `monstartup' before `main', is used instead of the default `crt0.o'.
1641 Use of this special startup file is one of the effects of using `gcc
1642 ... -pg' to link. On SPARC systems, no special startup files are used.
1643 Rather, the `mcount' routine, when it is invoked for the first time
1644 (typically when `main' is called), calls `monstartup'.
1646 If the compiler's `-a' option was used, basic-block counting is also
1647 enabled. Each object file is then compiled with a static array of
1648 counts, initially zero. In the executable code, every time a new
1649 basic-block begins (i.e., when an `if' statement appears), an extra
1650 instruction is inserted to increment the corresponding count in the
1651 array. At compile time, a paired array was constructed that recorded
1652 the starting address of each basic-block. Taken together, the two
1653 arrays record the starting address of every basic-block, along with the
1654 number of times it was executed.
1656 The profiling library also includes a function (`mcleanup') which is
1657 typically registered using `atexit()' to be called as the program
1658 exits, and is responsible for writing the file `gmon.out'. Profiling
1659 is turned off, various headers are output, and the histogram is
1660 written, followed by the call-graph arcs and the basic-block counts.
1662 The output from `gprof' gives no indication of parts of your program
1663 that are limited by I/O or swapping bandwidth. This is because samples
1664 of the program counter are taken at fixed intervals of the program's
1665 run time. Therefore, the time measurements in `gprof' output say
1666 nothing about time that your program was not running. For example, a
1667 part of the program that creates so much data that it cannot all fit in
1668 physical memory at once may run very slowly due to thrashing, but
1669 `gprof' will say it uses little time. On the other hand, sampling by
1670 run time has the advantage that the amount of load due to other users
1671 won't directly affect the output you get.
1674 File: gprof.info, Node: File Format, Next: Internals, Prev: Implementation, Up: Details
1676 9.2 Profiling Data File Format
1677 ==============================
1679 The old BSD-derived file format used for profile data does not contain a
1680 magic cookie that allows to check whether a data file really is a
1681 `gprof' file. Furthermore, it does not provide a version number, thus
1682 rendering changes to the file format almost impossible. GNU `gprof'
1683 uses a new file format that provides these features. For backward
1684 compatibility, GNU `gprof' continues to support the old BSD-derived
1685 format, but not all features are supported with it. For example,
1686 basic-block execution counts cannot be accommodated by the old file
1689 The new file format is defined in header file `gmon_out.h'. It
1690 consists of a header containing the magic cookie and a version number,
1691 as well as some spare bytes available for future extensions. All data
1692 in a profile data file is in the native format of the target for which
1693 the profile was collected. GNU `gprof' adapts automatically to the
1696 In the new file format, the header is followed by a sequence of
1697 records. Currently, there are three different record types: histogram
1698 records, call-graph arc records, and basic-block execution count
1699 records. Each file can contain any number of each record type. When
1700 reading a file, GNU `gprof' will ensure records of the same type are
1701 compatible with each other and compute the union of all records. For
1702 example, for basic-block execution counts, the union is simply the sum
1703 of all execution counts for each basic-block.
1705 9.2.1 Histogram Records
1706 -----------------------
1708 Histogram records consist of a header that is followed by an array of
1709 bins. The header contains the text-segment range that the histogram
1710 spans, the size of the histogram in bytes (unlike in the old BSD
1711 format, this does not include the size of the header), the rate of the
1712 profiling clock, and the physical dimension that the bin counts
1713 represent after being scaled by the profiling clock rate. The physical
1714 dimension is specified in two parts: a long name of up to 15 characters
1715 and a single character abbreviation. For example, a histogram
1716 representing real-time would specify the long name as "seconds" and the
1717 abbreviation as "s". This feature is useful for architectures that
1718 support performance monitor hardware (which, fortunately, is becoming
1719 increasingly common). For example, under DEC OSF/1, the "uprofile"
1720 command can be used to produce a histogram of, say, instruction cache
1721 misses. In this case, the dimension in the histogram header could be
1722 set to "i-cache misses" and the abbreviation could be set to "1"
1723 (because it is simply a count, not a physical dimension). Also, the
1724 profiling rate would have to be set to 1 in this case.
1726 Histogram bins are 16-bit numbers and each bin represent an equal
1727 amount of text-space. For example, if the text-segment is one thousand
1728 bytes long and if there are ten bins in the histogram, each bin
1729 represents one hundred bytes.
1731 9.2.2 Call-Graph Records
1732 ------------------------
1734 Call-graph records have a format that is identical to the one used in
1735 the BSD-derived file format. It consists of an arc in the call graph
1736 and a count indicating the number of times the arc was traversed during
1737 program execution. Arcs are specified by a pair of addresses: the
1738 first must be within caller's function and the second must be within
1739 the callee's function. When performing profiling at the function
1740 level, these addresses can point anywhere within the respective
1741 function. However, when profiling at the line-level, it is better if
1742 the addresses are as close to the call-site/entry-point as possible.
1743 This will ensure that the line-level call-graph is able to identify
1744 exactly which line of source code performed calls to a function.
1746 9.2.3 Basic-Block Execution Count Records
1747 -----------------------------------------
1749 Basic-block execution count records consist of a header followed by a
1750 sequence of address/count pairs. The header simply specifies the
1751 length of the sequence. In an address/count pair, the address
1752 identifies a basic-block and the count specifies the number of times
1753 that basic-block was executed. Any address within the basic-address can
1757 File: gprof.info, Node: Internals, Next: Debugging, Prev: File Format, Up: Details
1759 9.3 `gprof''s Internal Operation
1760 ================================
1762 Like most programs, `gprof' begins by processing its options. During
1763 this stage, it may building its symspec list (`sym_ids.c:sym_id_add'),
1764 if options are specified which use symspecs. `gprof' maintains a
1765 single linked list of symspecs, which will eventually get turned into
1766 12 symbol tables, organized into six include/exclude pairs--one pair
1767 each for the flat profile (INCL_FLAT/EXCL_FLAT), the call graph arcs
1768 (INCL_ARCS/EXCL_ARCS), printing in the call graph
1769 (INCL_GRAPH/EXCL_GRAPH), timing propagation in the call graph
1770 (INCL_TIME/EXCL_TIME), the annotated source listing
1771 (INCL_ANNO/EXCL_ANNO), and the execution count listing
1772 (INCL_EXEC/EXCL_EXEC).
1774 After option processing, `gprof' finishes building the symspec list
1775 by adding all the symspecs in `default_excluded_list' to the exclude
1776 lists EXCL_TIME and EXCL_GRAPH, and if line-by-line profiling is
1777 specified, EXCL_FLAT as well. These default excludes are not added to
1778 EXCL_ANNO, EXCL_ARCS, and EXCL_EXEC.
1780 Next, the BFD library is called to open the object file, verify that
1781 it is an object file, and read its symbol table (`core.c:core_init'),
1782 using `bfd_canonicalize_symtab' after mallocing an appropriately sized
1783 array of symbols. At this point, function mappings are read (if the
1784 `--file-ordering' option has been specified), and the core text space
1785 is read into memory (if the `-c' option was given).
1787 `gprof''s own symbol table, an array of Sym structures, is now built.
1788 This is done in one of two ways, by one of two routines, depending on
1789 whether line-by-line profiling (`-l' option) has been enabled. For
1790 normal profiling, the BFD canonical symbol table is scanned. For
1791 line-by-line profiling, every text space address is examined, and a new
1792 symbol table entry gets created every time the line number changes. In
1793 either case, two passes are made through the symbol table--one to count
1794 the size of the symbol table required, and the other to actually read
1795 the symbols. In between the two passes, a single array of type `Sym'
1796 is created of the appropriate length. Finally,
1797 `symtab.c:symtab_finalize' is called to sort the symbol table and
1798 remove duplicate entries (entries with the same memory address).
1800 The symbol table must be a contiguous array for two reasons. First,
1801 the `qsort' library function (which sorts an array) will be used to
1802 sort the symbol table. Also, the symbol lookup routine
1803 (`symtab.c:sym_lookup'), which finds symbols based on memory address,
1804 uses a binary search algorithm which requires the symbol table to be a
1805 sorted array. Function symbols are indicated with an `is_func' flag.
1806 Line number symbols have no special flags set. Additionally, a symbol
1807 can have an `is_static' flag to indicate that it is a local symbol.
1809 With the symbol table read, the symspecs can now be translated into
1810 Syms (`sym_ids.c:sym_id_parse'). Remember that a single symspec can
1811 match multiple symbols. An array of symbol tables (`syms') is created,
1812 each entry of which is a symbol table of Syms to be included or
1813 excluded from a particular listing. The master symbol table and the
1814 symspecs are examined by nested loops, and every symbol that matches a
1815 symspec is inserted into the appropriate syms table. This is done
1816 twice, once to count the size of each required symbol table, and again
1817 to build the tables, which have been malloced between passes. From now
1818 on, to determine whether a symbol is on an include or exclude symspec
1819 list, `gprof' simply uses its standard symbol lookup routine on the
1820 appropriate table in the `syms' array.
1822 Now the profile data file(s) themselves are read
1823 (`gmon_io.c:gmon_out_read'), first by checking for a new-style
1824 `gmon.out' header, then assuming this is an old-style BSD `gmon.out' if
1825 the magic number test failed.
1827 New-style histogram records are read by `hist.c:hist_read_rec'. For
1828 the first histogram record, allocate a memory array to hold all the
1829 bins, and read them in. When multiple profile data files (or files
1830 with multiple histogram records) are read, the memory ranges of each
1831 pair of histogram records must be either equal, or non-overlapping.
1832 For each pair of histogram records, the resolution (memory region size
1833 divided by the number of bins) must be the same. The time unit must be
1834 the same for all histogram records. If the above containts are met, all
1835 histograms for the same memory range are merged.
1837 As each call graph record is read (`call_graph.c:cg_read_rec'), the
1838 parent and child addresses are matched to symbol table entries, and a
1839 call graph arc is created by `cg_arcs.c:arc_add', unless the arc fails
1840 a symspec check against INCL_ARCS/EXCL_ARCS. As each arc is added, a
1841 linked list is maintained of the parent's child arcs, and of the child's
1842 parent arcs. Both the child's call count and the arc's call count are
1843 incremented by the record's call count.
1845 Basic-block records are read (`basic_blocks.c:bb_read_rec'), but
1846 only if line-by-line profiling has been selected. Each basic-block
1847 address is matched to a corresponding line symbol in the symbol table,
1848 and an entry made in the symbol's bb_addr and bb_calls arrays. Again,
1849 if multiple basic-block records are present for the same address, the
1850 call counts are cumulative.
1852 A gmon.sum file is dumped, if requested (`gmon_io.c:gmon_out_write').
1854 If histograms were present in the data files, assign them to symbols
1855 (`hist.c:hist_assign_samples') by iterating over all the sample bins
1856 and assigning them to symbols. Since the symbol table is sorted in
1857 order of ascending memory addresses, we can simple follow along in the
1858 symbol table as we make our pass over the sample bins. This step
1859 includes a symspec check against INCL_FLAT/EXCL_FLAT. Depending on the
1860 histogram scale factor, a sample bin may span multiple symbols, in
1861 which case a fraction of the sample count is allocated to each symbol,
1862 proportional to the degree of overlap. This effect is rare for normal
1863 profiling, but overlaps are more common during line-by-line profiling,
1864 and can cause each of two adjacent lines to be credited with half a
1867 If call graph data is present, `cg_arcs.c:cg_assemble' is called.
1868 First, if `-c' was specified, a machine-dependent routine (`find_call')
1869 scans through each symbol's machine code, looking for subroutine call
1870 instructions, and adding them to the call graph with a zero call count.
1871 A topological sort is performed by depth-first numbering all the
1872 symbols (`cg_dfn.c:cg_dfn'), so that children are always numbered less
1873 than their parents, then making a array of pointers into the symbol
1874 table and sorting it into numerical order, which is reverse topological
1875 order (children appear before parents). Cycles are also detected at
1876 this point, all members of which are assigned the same topological
1877 number. Two passes are now made through this sorted array of symbol
1878 pointers. The first pass, from end to beginning (parents to children),
1879 computes the fraction of child time to propagate to each parent and a
1880 print flag. The print flag reflects symspec handling of
1881 INCL_GRAPH/EXCL_GRAPH, with a parent's include or exclude (print or no
1882 print) property being propagated to its children, unless they
1883 themselves explicitly appear in INCL_GRAPH or EXCL_GRAPH. A second
1884 pass, from beginning to end (children to parents) actually propagates
1885 the timings along the call graph, subject to a check against
1886 INCL_TIME/EXCL_TIME. With the print flag, fractions, and timings now
1887 stored in the symbol structures, the topological sort array is now
1888 discarded, and a new array of pointers is assembled, this time sorted
1891 Finally, print the various outputs the user requested, which is now
1892 fairly straightforward. The call graph (`cg_print.c:cg_print') and
1893 flat profile (`hist.c:hist_print') are regurgitations of values already
1894 computed. The annotated source listing
1895 (`basic_blocks.c:print_annotated_source') uses basic-block information,
1896 if present, to label each line of code with call counts, otherwise only
1897 the function call counts are presented.
1899 The function ordering code is marginally well documented in the
1900 source code itself (`cg_print.c'). Basically, the functions with the
1901 most use and the most parents are placed first, followed by other
1902 functions with the most use, followed by lower use functions, followed
1903 by unused functions at the end.
1906 File: gprof.info, Node: Debugging, Prev: Internals, Up: Details
1908 9.4 Debugging `gprof'
1909 =====================
1911 If `gprof' was compiled with debugging enabled, the `-d' option
1912 triggers debugging output (to stdout) which can be helpful in
1913 understanding its operation. The debugging number specified is
1914 interpreted as a sum of the following options:
1916 2 - Topological sort
1917 Monitor depth-first numbering of symbols during call graph analysis
1920 Shows symbols as they are identified as cycle heads
1923 As the call graph arcs are read, show each arc and how the total
1924 calls to each function are tallied
1926 32 - Call graph arc sorting
1927 Details sorting individual parents/children within each call graph
1930 64 - Reading histogram and call graph records
1931 Shows address ranges of histograms as they are read, and each call
1935 Reading, classifying, and sorting the symbol table from the object
1936 file. For line-by-line profiling (`-l' option), also shows line
1937 numbers being assigned to memory addresses.
1939 256 - Static call graph
1940 Trace operation of `-c' option
1942 512 - Symbol table and arc table lookups
1943 Detail operation of lookup routines
1945 1024 - Call graph propagation
1946 Shows how function times are propagated along the call graph
1949 Shows basic-block records as they are read from profile data (only
1950 meaningful with `-l' option)
1953 Shows symspec-to-symbol pattern matching operation
1955 8192 - Annotate source
1956 Tracks operation of `-A' option
1959 File: gprof.info, Node: GNU Free Documentation License, Prev: Details, Up: Top
1961 Appendix A GNU Free Documentation License
1962 *****************************************
1964 Version 1.3, 3 November 2008
1966 Copyright (C) 2000, 2001, 2002, 2007, 2008 Free Software Foundation, Inc.
1969 Everyone is permitted to copy and distribute verbatim copies
1970 of this license document, but changing it is not allowed.
1974 The purpose of this License is to make a manual, textbook, or other
1975 functional and useful document "free" in the sense of freedom: to
1976 assure everyone the effective freedom to copy and redistribute it,
1977 with or without modifying it, either commercially or
1978 noncommercially. Secondarily, this License preserves for the
1979 author and publisher a way to get credit for their work, while not
1980 being considered responsible for modifications made by others.
1982 This License is a kind of "copyleft", which means that derivative
1983 works of the document must themselves be free in the same sense.
1984 It complements the GNU General Public License, which is a copyleft
1985 license designed for free software.
1987 We have designed this License in order to use it for manuals for
1988 free software, because free software needs free documentation: a
1989 free program should come with manuals providing the same freedoms
1990 that the software does. But this License is not limited to
1991 software manuals; it can be used for any textual work, regardless
1992 of subject matter or whether it is published as a printed book.
1993 We recommend this License principally for works whose purpose is
1994 instruction or reference.
1996 1. APPLICABILITY AND DEFINITIONS
1998 This License applies to any manual or other work, in any medium,
1999 that contains a notice placed by the copyright holder saying it
2000 can be distributed under the terms of this License. Such a notice
2001 grants a world-wide, royalty-free license, unlimited in duration,
2002 to use that work under the conditions stated herein. The
2003 "Document", below, refers to any such manual or work. Any member
2004 of the public is a licensee, and is addressed as "you". You
2005 accept the license if you copy, modify or distribute the work in a
2006 way requiring permission under copyright law.
2008 A "Modified Version" of the Document means any work containing the
2009 Document or a portion of it, either copied verbatim, or with
2010 modifications and/or translated into another language.
2012 A "Secondary Section" is a named appendix or a front-matter section
2013 of the Document that deals exclusively with the relationship of the
2014 publishers or authors of the Document to the Document's overall
2015 subject (or to related matters) and contains nothing that could
2016 fall directly within that overall subject. (Thus, if the Document
2017 is in part a textbook of mathematics, a Secondary Section may not
2018 explain any mathematics.) The relationship could be a matter of
2019 historical connection with the subject or with related matters, or
2020 of legal, commercial, philosophical, ethical or political position
2023 The "Invariant Sections" are certain Secondary Sections whose
2024 titles are designated, as being those of Invariant Sections, in
2025 the notice that says that the Document is released under this
2026 License. If a section does not fit the above definition of
2027 Secondary then it is not allowed to be designated as Invariant.
2028 The Document may contain zero Invariant Sections. If the Document
2029 does not identify any Invariant Sections then there are none.
2031 The "Cover Texts" are certain short passages of text that are
2032 listed, as Front-Cover Texts or Back-Cover Texts, in the notice
2033 that says that the Document is released under this License. A
2034 Front-Cover Text may be at most 5 words, and a Back-Cover Text may
2035 be at most 25 words.
2037 A "Transparent" copy of the Document means a machine-readable copy,
2038 represented in a format whose specification is available to the
2039 general public, that is suitable for revising the document
2040 straightforwardly with generic text editors or (for images
2041 composed of pixels) generic paint programs or (for drawings) some
2042 widely available drawing editor, and that is suitable for input to
2043 text formatters or for automatic translation to a variety of
2044 formats suitable for input to text formatters. A copy made in an
2045 otherwise Transparent file format whose markup, or absence of
2046 markup, has been arranged to thwart or discourage subsequent
2047 modification by readers is not Transparent. An image format is
2048 not Transparent if used for any substantial amount of text. A
2049 copy that is not "Transparent" is called "Opaque".
2051 Examples of suitable formats for Transparent copies include plain
2052 ASCII without markup, Texinfo input format, LaTeX input format,
2053 SGML or XML using a publicly available DTD, and
2054 standard-conforming simple HTML, PostScript or PDF designed for
2055 human modification. Examples of transparent image formats include
2056 PNG, XCF and JPG. Opaque formats include proprietary formats that
2057 can be read and edited only by proprietary word processors, SGML or
2058 XML for which the DTD and/or processing tools are not generally
2059 available, and the machine-generated HTML, PostScript or PDF
2060 produced by some word processors for output purposes only.
2062 The "Title Page" means, for a printed book, the title page itself,
2063 plus such following pages as are needed to hold, legibly, the
2064 material this License requires to appear in the title page. For
2065 works in formats which do not have any title page as such, "Title
2066 Page" means the text near the most prominent appearance of the
2067 work's title, preceding the beginning of the body of the text.
2069 The "publisher" means any person or entity that distributes copies
2070 of the Document to the public.
2072 A section "Entitled XYZ" means a named subunit of the Document
2073 whose title either is precisely XYZ or contains XYZ in parentheses
2074 following text that translates XYZ in another language. (Here XYZ
2075 stands for a specific section name mentioned below, such as
2076 "Acknowledgements", "Dedications", "Endorsements", or "History".)
2077 To "Preserve the Title" of such a section when you modify the
2078 Document means that it remains a section "Entitled XYZ" according
2081 The Document may include Warranty Disclaimers next to the notice
2082 which states that this License applies to the Document. These
2083 Warranty Disclaimers are considered to be included by reference in
2084 this License, but only as regards disclaiming warranties: any other
2085 implication that these Warranty Disclaimers may have is void and
2086 has no effect on the meaning of this License.
2090 You may copy and distribute the Document in any medium, either
2091 commercially or noncommercially, provided that this License, the
2092 copyright notices, and the license notice saying this License
2093 applies to the Document are reproduced in all copies, and that you
2094 add no other conditions whatsoever to those of this License. You
2095 may not use technical measures to obstruct or control the reading
2096 or further copying of the copies you make or distribute. However,
2097 you may accept compensation in exchange for copies. If you
2098 distribute a large enough number of copies you must also follow
2099 the conditions in section 3.
2101 You may also lend copies, under the same conditions stated above,
2102 and you may publicly display copies.
2104 3. COPYING IN QUANTITY
2106 If you publish printed copies (or copies in media that commonly
2107 have printed covers) of the Document, numbering more than 100, and
2108 the Document's license notice requires Cover Texts, you must
2109 enclose the copies in covers that carry, clearly and legibly, all
2110 these Cover Texts: Front-Cover Texts on the front cover, and
2111 Back-Cover Texts on the back cover. Both covers must also clearly
2112 and legibly identify you as the publisher of these copies. The
2113 front cover must present the full title with all words of the
2114 title equally prominent and visible. You may add other material
2115 on the covers in addition. Copying with changes limited to the
2116 covers, as long as they preserve the title of the Document and
2117 satisfy these conditions, can be treated as verbatim copying in
2120 If the required texts for either cover are too voluminous to fit
2121 legibly, you should put the first ones listed (as many as fit
2122 reasonably) on the actual cover, and continue the rest onto
2125 If you publish or distribute Opaque copies of the Document
2126 numbering more than 100, you must either include a
2127 machine-readable Transparent copy along with each Opaque copy, or
2128 state in or with each Opaque copy a computer-network location from
2129 which the general network-using public has access to download
2130 using public-standard network protocols a complete Transparent
2131 copy of the Document, free of added material. If you use the
2132 latter option, you must take reasonably prudent steps, when you
2133 begin distribution of Opaque copies in quantity, to ensure that
2134 this Transparent copy will remain thus accessible at the stated
2135 location until at least one year after the last time you
2136 distribute an Opaque copy (directly or through your agents or
2137 retailers) of that edition to the public.
2139 It is requested, but not required, that you contact the authors of
2140 the Document well before redistributing any large number of
2141 copies, to give them a chance to provide you with an updated
2142 version of the Document.
2146 You may copy and distribute a Modified Version of the Document
2147 under the conditions of sections 2 and 3 above, provided that you
2148 release the Modified Version under precisely this License, with
2149 the Modified Version filling the role of the Document, thus
2150 licensing distribution and modification of the Modified Version to
2151 whoever possesses a copy of it. In addition, you must do these
2152 things in the Modified Version:
2154 A. Use in the Title Page (and on the covers, if any) a title
2155 distinct from that of the Document, and from those of
2156 previous versions (which should, if there were any, be listed
2157 in the History section of the Document). You may use the
2158 same title as a previous version if the original publisher of
2159 that version gives permission.
2161 B. List on the Title Page, as authors, one or more persons or
2162 entities responsible for authorship of the modifications in
2163 the Modified Version, together with at least five of the
2164 principal authors of the Document (all of its principal
2165 authors, if it has fewer than five), unless they release you
2166 from this requirement.
2168 C. State on the Title page the name of the publisher of the
2169 Modified Version, as the publisher.
2171 D. Preserve all the copyright notices of the Document.
2173 E. Add an appropriate copyright notice for your modifications
2174 adjacent to the other copyright notices.
2176 F. Include, immediately after the copyright notices, a license
2177 notice giving the public permission to use the Modified
2178 Version under the terms of this License, in the form shown in
2181 G. Preserve in that license notice the full lists of Invariant
2182 Sections and required Cover Texts given in the Document's
2185 H. Include an unaltered copy of this License.
2187 I. Preserve the section Entitled "History", Preserve its Title,
2188 and add to it an item stating at least the title, year, new
2189 authors, and publisher of the Modified Version as given on
2190 the Title Page. If there is no section Entitled "History" in
2191 the Document, create one stating the title, year, authors,
2192 and publisher of the Document as given on its Title Page,
2193 then add an item describing the Modified Version as stated in
2194 the previous sentence.
2196 J. Preserve the network location, if any, given in the Document
2197 for public access to a Transparent copy of the Document, and
2198 likewise the network locations given in the Document for
2199 previous versions it was based on. These may be placed in
2200 the "History" section. You may omit a network location for a
2201 work that was published at least four years before the
2202 Document itself, or if the original publisher of the version
2203 it refers to gives permission.
2205 K. For any section Entitled "Acknowledgements" or "Dedications",
2206 Preserve the Title of the section, and preserve in the
2207 section all the substance and tone of each of the contributor
2208 acknowledgements and/or dedications given therein.
2210 L. Preserve all the Invariant Sections of the Document,
2211 unaltered in their text and in their titles. Section numbers
2212 or the equivalent are not considered part of the section
2215 M. Delete any section Entitled "Endorsements". Such a section
2216 may not be included in the Modified Version.
2218 N. Do not retitle any existing section to be Entitled
2219 "Endorsements" or to conflict in title with any Invariant
2222 O. Preserve any Warranty Disclaimers.
2224 If the Modified Version includes new front-matter sections or
2225 appendices that qualify as Secondary Sections and contain no
2226 material copied from the Document, you may at your option
2227 designate some or all of these sections as invariant. To do this,
2228 add their titles to the list of Invariant Sections in the Modified
2229 Version's license notice. These titles must be distinct from any
2230 other section titles.
2232 You may add a section Entitled "Endorsements", provided it contains
2233 nothing but endorsements of your Modified Version by various
2234 parties--for example, statements of peer review or that the text
2235 has been approved by an organization as the authoritative
2236 definition of a standard.
2238 You may add a passage of up to five words as a Front-Cover Text,
2239 and a passage of up to 25 words as a Back-Cover Text, to the end
2240 of the list of Cover Texts in the Modified Version. Only one
2241 passage of Front-Cover Text and one of Back-Cover Text may be
2242 added by (or through arrangements made by) any one entity. If the
2243 Document already includes a cover text for the same cover,
2244 previously added by you or by arrangement made by the same entity
2245 you are acting on behalf of, you may not add another; but you may
2246 replace the old one, on explicit permission from the previous
2247 publisher that added the old one.
2249 The author(s) and publisher(s) of the Document do not by this
2250 License give permission to use their names for publicity for or to
2251 assert or imply endorsement of any Modified Version.
2253 5. COMBINING DOCUMENTS
2255 You may combine the Document with other documents released under
2256 this License, under the terms defined in section 4 above for
2257 modified versions, provided that you include in the combination
2258 all of the Invariant Sections of all of the original documents,
2259 unmodified, and list them all as Invariant Sections of your
2260 combined work in its license notice, and that you preserve all
2261 their Warranty Disclaimers.
2263 The combined work need only contain one copy of this License, and
2264 multiple identical Invariant Sections may be replaced with a single
2265 copy. If there are multiple Invariant Sections with the same name
2266 but different contents, make the title of each such section unique
2267 by adding at the end of it, in parentheses, the name of the
2268 original author or publisher of that section if known, or else a
2269 unique number. Make the same adjustment to the section titles in
2270 the list of Invariant Sections in the license notice of the
2273 In the combination, you must combine any sections Entitled
2274 "History" in the various original documents, forming one section
2275 Entitled "History"; likewise combine any sections Entitled
2276 "Acknowledgements", and any sections Entitled "Dedications". You
2277 must delete all sections Entitled "Endorsements."
2279 6. COLLECTIONS OF DOCUMENTS
2281 You may make a collection consisting of the Document and other
2282 documents released under this License, and replace the individual
2283 copies of this License in the various documents with a single copy
2284 that is included in the collection, provided that you follow the
2285 rules of this License for verbatim copying of each of the
2286 documents in all other respects.
2288 You may extract a single document from such a collection, and
2289 distribute it individually under this License, provided you insert
2290 a copy of this License into the extracted document, and follow
2291 this License in all other respects regarding verbatim copying of
2294 7. AGGREGATION WITH INDEPENDENT WORKS
2296 A compilation of the Document or its derivatives with other
2297 separate and independent documents or works, in or on a volume of
2298 a storage or distribution medium, is called an "aggregate" if the
2299 copyright resulting from the compilation is not used to limit the
2300 legal rights of the compilation's users beyond what the individual
2301 works permit. When the Document is included in an aggregate, this
2302 License does not apply to the other works in the aggregate which
2303 are not themselves derivative works of the Document.
2305 If the Cover Text requirement of section 3 is applicable to these
2306 copies of the Document, then if the Document is less than one half
2307 of the entire aggregate, the Document's Cover Texts may be placed
2308 on covers that bracket the Document within the aggregate, or the
2309 electronic equivalent of covers if the Document is in electronic
2310 form. Otherwise they must appear on printed covers that bracket
2311 the whole aggregate.
2315 Translation is considered a kind of modification, so you may
2316 distribute translations of the Document under the terms of section
2317 4. Replacing Invariant Sections with translations requires special
2318 permission from their copyright holders, but you may include
2319 translations of some or all Invariant Sections in addition to the
2320 original versions of these Invariant Sections. You may include a
2321 translation of this License, and all the license notices in the
2322 Document, and any Warranty Disclaimers, provided that you also
2323 include the original English version of this License and the
2324 original versions of those notices and disclaimers. In case of a
2325 disagreement between the translation and the original version of
2326 this License or a notice or disclaimer, the original version will
2329 If a section in the Document is Entitled "Acknowledgements",
2330 "Dedications", or "History", the requirement (section 4) to
2331 Preserve its Title (section 1) will typically require changing the
2336 You may not copy, modify, sublicense, or distribute the Document
2337 except as expressly provided under this License. Any attempt
2338 otherwise to copy, modify, sublicense, or distribute it is void,
2339 and will automatically terminate your rights under this License.
2341 However, if you cease all violation of this License, then your
2342 license from a particular copyright holder is reinstated (a)
2343 provisionally, unless and until the copyright holder explicitly
2344 and finally terminates your license, and (b) permanently, if the
2345 copyright holder fails to notify you of the violation by some
2346 reasonable means prior to 60 days after the cessation.
2348 Moreover, your license from a particular copyright holder is
2349 reinstated permanently if the copyright holder notifies you of the
2350 violation by some reasonable means, this is the first time you have
2351 received notice of violation of this License (for any work) from
2352 that copyright holder, and you cure the violation prior to 30 days
2353 after your receipt of the notice.
2355 Termination of your rights under this section does not terminate
2356 the licenses of parties who have received copies or rights from
2357 you under this License. If your rights have been terminated and
2358 not permanently reinstated, receipt of a copy of some or all of
2359 the same material does not give you any rights to use it.
2361 10. FUTURE REVISIONS OF THIS LICENSE
2363 The Free Software Foundation may publish new, revised versions of
2364 the GNU Free Documentation License from time to time. Such new
2365 versions will be similar in spirit to the present version, but may
2366 differ in detail to address new problems or concerns. See
2367 `http://www.gnu.org/copyleft/'.
2369 Each version of the License is given a distinguishing version
2370 number. If the Document specifies that a particular numbered
2371 version of this License "or any later version" applies to it, you
2372 have the option of following the terms and conditions either of
2373 that specified version or of any later version that has been
2374 published (not as a draft) by the Free Software Foundation. If
2375 the Document does not specify a version number of this License,
2376 you may choose any version ever published (not as a draft) by the
2377 Free Software Foundation. If the Document specifies that a proxy
2378 can decide which future versions of this License can be used, that
2379 proxy's public statement of acceptance of a version permanently
2380 authorizes you to choose that version for the Document.
2384 "Massive Multiauthor Collaboration Site" (or "MMC Site") means any
2385 World Wide Web server that publishes copyrightable works and also
2386 provides prominent facilities for anybody to edit those works. A
2387 public wiki that anybody can edit is an example of such a server.
2388 A "Massive Multiauthor Collaboration" (or "MMC") contained in the
2389 site means any set of copyrightable works thus published on the MMC
2392 "CC-BY-SA" means the Creative Commons Attribution-Share Alike 3.0
2393 license published by Creative Commons Corporation, a not-for-profit
2394 corporation with a principal place of business in San Francisco,
2395 California, as well as future copyleft versions of that license
2396 published by that same organization.
2398 "Incorporate" means to publish or republish a Document, in whole or
2399 in part, as part of another Document.
2401 An MMC is "eligible for relicensing" if it is licensed under this
2402 License, and if all works that were first published under this
2403 License somewhere other than this MMC, and subsequently
2404 incorporated in whole or in part into the MMC, (1) had no cover
2405 texts or invariant sections, and (2) were thus incorporated prior
2406 to November 1, 2008.
2408 The operator of an MMC Site may republish an MMC contained in the
2409 site under CC-BY-SA on the same site at any time before August 1,
2410 2009, provided the MMC is eligible for relicensing.
2413 ADDENDUM: How to use this License for your documents
2414 ====================================================
2416 To use this License in a document you have written, include a copy of
2417 the License in the document and put the following copyright and license
2418 notices just after the title page:
2420 Copyright (C) YEAR YOUR NAME.
2421 Permission is granted to copy, distribute and/or modify this document
2422 under the terms of the GNU Free Documentation License, Version 1.3
2423 or any later version published by the Free Software Foundation;
2424 with no Invariant Sections, no Front-Cover Texts, and no Back-Cover
2425 Texts. A copy of the license is included in the section entitled ``GNU
2426 Free Documentation License''.
2428 If you have Invariant Sections, Front-Cover Texts and Back-Cover
2429 Texts, replace the "with...Texts." line with this:
2431 with the Invariant Sections being LIST THEIR TITLES, with
2432 the Front-Cover Texts being LIST, and with the Back-Cover Texts
2435 If you have Invariant Sections without Cover Texts, or some other
2436 combination of the three, merge those two alternatives to suit the
2439 If your document contains nontrivial examples of program code, we
2440 recommend releasing these examples in parallel under your choice of
2441 free software license, such as the GNU General Public License, to
2442 permit their use in free software.
2448 Node: Introduction
\7f2045
2449 Node: Compiling
\7f4537
2450 Node: Executing
\7f8593
2451 Node: Invoking
\7f11381
2452 Node: Output Options
\7f12796
2453 Node: Analysis Options
\7f19885
2454 Node: Miscellaneous Options
\7f23803
2455 Node: Deprecated Options
\7f25058
2456 Node: Symspecs
\7f27127
2457 Node: Output
\7f28953
2458 Node: Flat Profile
\7f29993
2459 Node: Call Graph
\7f34946
2460 Node: Primary
\7f38178
2461 Node: Callers
\7f40766
2462 Node: Subroutines
\7f42883
2463 Node: Cycles
\7f44724
2464 Node: Line-by-line
\7f51501
2465 Node: Annotated Source
\7f55574
2466 Node: Inaccuracy
\7f58573
2467 Node: Sampling Error
\7f58831
2468 Node: Assumptions
\7f61735
2469 Node: How do I?
\7f63205
2470 Node: Incompatibilities
\7f64759
2471 Node: Details
\7f66253
2472 Node: Implementation
\7f66646
2473 Node: File Format
\7f72543
2474 Node: Internals
\7f76833
2475 Node: Debugging
\7f85328
2476 Node: GNU Free Documentation License
\7f86929
projects / dragonfly.git / blob