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, 1992, 1997, 1998, 1999, 2000, 2001, 2003, 2007,
11 2008, 2009 Free Software Foundation, Inc.
13 Permission is granted to copy, distribute and/or modify this document
14 under the terms of the GNU Free Documentation License, Version 1.3 or
15 any later version published by the Free Software Foundation; with no
16 Invariant Sections, with no Front-Cover Texts, and with no Back-Cover
17 Texts. A copy of the license is included in the section entitled "GNU
18 Free Documentation License".
21 File: gprof.info, Node: Top, Next: Introduction, Up: (dir)
23 Profiling a Program: Where Does It Spend Its Time?
24 **************************************************
26 This manual describes the GNU profiler, `gprof', and how you can use it
27 to determine which parts of a program are taking most of the execution
28 time. We assume that you know how to write, compile, and execute
29 programs. GNU `gprof' was written by Jay Fenlason.
31 This manual is for `gprof' (GNU Binutils) version 2.21.90.
33 This document is distributed under the terms of the GNU Free
34 Documentation License version 1.3. A copy of the license is included
35 in the section entitled "GNU Free Documentation License".
39 * Introduction:: What profiling means, and why it is useful.
41 * Compiling:: How to compile your program for profiling.
42 * Executing:: Executing your program to generate profile data
43 * Invoking:: How to run `gprof', and its options
45 * Output:: Interpreting `gprof''s output
47 * Inaccuracy:: Potential problems you should be aware of
48 * How do I?:: Answers to common questions
49 * Incompatibilities:: (between GNU `gprof' and Unix `gprof'.)
50 * Details:: Details of how profiling is done
51 * GNU Free Documentation License:: GNU Free Documentation License
54 File: gprof.info, Node: Introduction, Next: Compiling, Prev: Top, Up: Top
56 1 Introduction to Profiling
57 ***************************
59 Profiling allows you to learn where your program spent its time and
60 which functions called which other functions while it was executing.
61 This information can show you which pieces of your program are slower
62 than you expected, and might be candidates for rewriting to make your
63 program execute faster. It can also tell you which functions are being
64 called more or less often than you expected. This may help you spot
65 bugs that had otherwise been unnoticed.
67 Since the profiler uses information collected during the actual
68 execution of your program, it can be used on programs that are too
69 large or too complex to analyze by reading the source. However, how
70 your program is run will affect the information that shows up in the
71 profile data. If you don't use some feature of your program while it
72 is being profiled, no profile information will be generated for that
75 Profiling has several steps:
77 * You must compile and link your program with profiling enabled.
78 *Note Compiling a Program for Profiling: Compiling.
80 * You must execute your program to generate a profile data file.
81 *Note Executing the Program: Executing.
83 * You must run `gprof' to analyze the profile data. *Note `gprof'
84 Command Summary: Invoking.
86 The next three chapters explain these steps in greater detail.
88 Several forms of output are available from the analysis.
90 The "flat profile" shows how much time your program spent in each
91 function, and how many times that function was called. If you simply
92 want to know which functions burn most of the cycles, it is stated
93 concisely here. *Note The Flat Profile: Flat Profile.
95 The "call graph" shows, for each function, which functions called
96 it, which other functions it called, and how many times. There is also
97 an estimate of how much time was spent in the subroutines of each
98 function. This can suggest places where you might try to eliminate
99 function calls that use a lot of time. *Note The Call Graph: Call
102 The "annotated source" listing is a copy of the program's source
103 code, labeled with the number of times each line of the program was
104 executed. *Note The Annotated Source Listing: Annotated Source.
106 To better understand how profiling works, you may wish to read a
107 description of its implementation. *Note Implementation of Profiling:
111 File: gprof.info, Node: Compiling, Next: Executing, Prev: Introduction, Up: Top
113 2 Compiling a Program for Profiling
114 ***********************************
116 The first step in generating profile information for your program is to
117 compile and link it with profiling enabled.
119 To compile a source file for profiling, specify the `-pg' option when
120 you run the compiler. (This is in addition to the options you normally
123 To link the program for profiling, if you use a compiler such as `cc'
124 to do the linking, simply specify `-pg' in addition to your usual
125 options. The same option, `-pg', alters either compilation or linking
126 to do what is necessary for profiling. Here are examples:
128 cc -g -c myprog.c utils.c -pg
129 cc -o myprog myprog.o utils.o -pg
131 The `-pg' option also works with a command that both compiles and
134 cc -o myprog myprog.c utils.c -g -pg
136 Note: The `-pg' option must be part of your compilation options as
137 well as your link options. If it is not then no call-graph data will
138 be gathered and when you run `gprof' you will get an error message like
141 gprof: gmon.out file is missing call-graph data
143 If you add the `-Q' switch to suppress the printing of the call
144 graph data you will still be able to see the time samples:
148 Each sample counts as 0.01 seconds.
149 % cumulative self self total
150 time seconds seconds calls Ts/call Ts/call name
151 44.12 0.07 0.07 zazLoop
153 20.59 0.17 0.04 bazMillion
155 If you run the linker `ld' directly instead of through a compiler
156 such as `cc', you may have to specify a profiling startup file
157 `gcrt0.o' as the first input file instead of the usual startup file
158 `crt0.o'. In addition, you would probably want to specify the
159 profiling C library, `libc_p.a', by writing `-lc_p' instead of the
160 usual `-lc'. This is not absolutely necessary, but doing this gives
161 you number-of-calls information for standard library functions such as
162 `read' and `open'. For example:
164 ld -o myprog /lib/gcrt0.o myprog.o utils.o -lc_p
166 If you are running the program on a system which supports shared
167 libraries you may run into problems with the profiling support code in
168 a shared library being called before that library has been fully
169 initialised. This is usually detected by the program encountering a
170 segmentation fault as soon as it is run. The solution is to link
171 against a static version of the library containing the profiling
172 support code, which for `gcc' users can be done via the `-static' or
173 `-static-libgcc' command line option. For example:
175 gcc -g -pg -static-libgcc myprog.c utils.c -o myprog
177 If you compile only some of the modules of the program with `-pg',
178 you can still profile the program, but you won't get complete
179 information about the modules that were compiled without `-pg'. The
180 only information you get for the functions in those modules is the
181 total time spent in them; there is no record of how many times they
182 were called, or from where. This will not affect the flat profile
183 (except that the `calls' field for the functions will be blank), but
184 will greatly reduce the usefulness of the call graph.
186 If you wish to perform line-by-line profiling you should use the
187 `gcov' tool instead of `gprof'. See that tool's manual or info pages
188 for more details of how to do this.
190 Note, older versions of `gcc' produce line-by-line profiling
191 information that works with `gprof' rather than `gcov' so there is
192 still support for displaying this kind of information in `gprof'. *Note
193 Line-by-line Profiling: Line-by-line.
195 It also worth noting that `gcc' implements a
196 `-finstrument-functions' command line option which will insert calls to
197 special user supplied instrumentation routines at the entry and exit of
198 every function in their program. This can be used to implement an
199 alternative profiling scheme.
202 File: gprof.info, Node: Executing, Next: Invoking, Prev: Compiling, Up: Top
204 3 Executing the Program
205 ***********************
207 Once the program is compiled for profiling, you must run it in order to
208 generate the information that `gprof' needs. Simply run the program as
209 usual, using the normal arguments, file names, etc. The program should
210 run normally, producing the same output as usual. It will, however, run
211 somewhat slower than normal because of the time spent collecting and
212 writing the profile data.
214 The way you run the program--the arguments and input that you give
215 it--may have a dramatic effect on what the profile information shows.
216 The profile data will describe the parts of the program that were
217 activated for the particular input you use. For example, if the first
218 command you give to your program is to quit, the profile data will show
219 the time used in initialization and in cleanup, but not much else.
221 Your program will write the profile data into a file called
222 `gmon.out' just before exiting. If there is already a file called
223 `gmon.out', its contents are overwritten. There is currently no way to
224 tell the program to write the profile data under a different name, but
225 you can rename the file afterwards if you are concerned that it may be
228 In order to write the `gmon.out' file properly, your program must
229 exit normally: by returning from `main' or by calling `exit'. Calling
230 the low-level function `_exit' does not write the profile data, and
231 neither does abnormal termination due to an unhandled signal.
233 The `gmon.out' file is written in the program's _current working
234 directory_ at the time it exits. This means that if your program calls
235 `chdir', the `gmon.out' file will be left in the last directory your
236 program `chdir''d to. If you don't have permission to write in this
237 directory, the file is not written, and you will get an error message.
239 Older versions of the GNU profiling library may also write a file
240 called `bb.out'. This file, if present, contains an human-readable
241 listing of the basic-block execution counts. Unfortunately, the
242 appearance of a human-readable `bb.out' means the basic-block counts
243 didn't get written into `gmon.out'. The Perl script `bbconv.pl',
244 included with the `gprof' source distribution, will convert a `bb.out'
245 file into a format readable by `gprof'. Invoke it like this:
247 bbconv.pl < bb.out > BH-DATA
249 This translates the information in `bb.out' into a form that `gprof'
250 can understand. But you still need to tell `gprof' about the existence
251 of this translated information. To do that, include BB-DATA on the
252 `gprof' command line, _along with `gmon.out'_, like this:
254 gprof OPTIONS EXECUTABLE-FILE gmon.out BB-DATA [YET-MORE-PROFILE-DATA-FILES...] [> OUTFILE]
257 File: gprof.info, Node: Invoking, Next: Output, Prev: Executing, Up: Top
259 4 `gprof' Command Summary
260 *************************
262 After you have a profile data file `gmon.out', you can run `gprof' to
263 interpret the information in it. The `gprof' program prints a flat
264 profile and a call graph on standard output. Typically you would
265 redirect the output of `gprof' into a file with `>'.
267 You run `gprof' like this:
269 gprof OPTIONS [EXECUTABLE-FILE [PROFILE-DATA-FILES...]] [> OUTFILE]
271 Here square-brackets indicate optional arguments.
273 If you omit the executable file name, the file `a.out' is used. If
274 you give no profile data file name, the file `gmon.out' is used. If
275 any file is not in the proper format, or if the profile data file does
276 not appear to belong to the executable file, an error message is
279 You can give more than one profile data file by entering all their
280 names after the executable file name; then the statistics in all the
281 data files are summed together.
283 The order of these options does not matter.
287 * Output Options:: Controlling `gprof''s output style
288 * Analysis Options:: Controlling how `gprof' analyzes its data
289 * Miscellaneous Options::
290 * Deprecated Options:: Options you no longer need to use, but which
291 have been retained for compatibility
292 * Symspecs:: Specifying functions to include or exclude
295 File: gprof.info, Node: Output Options, Next: Analysis Options, Up: Invoking
300 These options specify which of several output formats `gprof' should
303 Many of these options take an optional "symspec" to specify
304 functions to be included or excluded. These options can be specified
305 multiple times, with different symspecs, to include or exclude sets of
306 symbols. *Note Symspecs: Symspecs.
308 Specifying any of these options overrides the default (`-p -q'),
309 which prints a flat profile and call graph analysis for all functions.
312 `--annotated-source[=SYMSPEC]'
313 The `-A' option causes `gprof' to print annotated source code. If
314 SYMSPEC is specified, print output only for matching symbols.
315 *Note The Annotated Source Listing: Annotated Source.
319 If the `-b' option is given, `gprof' doesn't print the verbose
320 blurbs that try to explain the meaning of all of the fields in the
321 tables. This is useful if you intend to print out the output, or
322 are tired of seeing the blurbs.
325 `--exec-counts[=SYMSPEC]'
326 The `-C' option causes `gprof' to print a tally of functions and
327 the number of times each was called. If SYMSPEC is specified,
328 print tally only for matching symbols.
330 If the profile data file contains basic-block count records,
331 specifying the `-l' option, along with `-C', will cause basic-block
332 execution counts to be tallied and displayed.
336 The `-i' option causes `gprof' to display summary information
337 about the profile data file(s) and then exit. The number of
338 histogram, call graph, and basic-block count records is displayed.
341 `--directory-path=DIRS'
342 The `-I' option specifies a list of search directories in which to
343 find source files. Environment variable GPROF_PATH can also be
344 used to convey this information. Used mostly for annotated source
348 `--no-annotated-source[=SYMSPEC]'
349 The `-J' option causes `gprof' not to print annotated source code.
350 If SYMSPEC is specified, `gprof' prints annotated source, but
351 excludes matching symbols.
355 Normally, source filenames are printed with the path component
356 suppressed. The `-L' option causes `gprof' to print the full
357 pathname of source filenames, which is determined from symbolic
358 debugging information in the image file and is relative to the
359 directory in which the compiler was invoked.
362 `--flat-profile[=SYMSPEC]'
363 The `-p' option causes `gprof' to print a flat profile. If
364 SYMSPEC is specified, print flat profile only for matching symbols.
365 *Note The Flat Profile: Flat Profile.
368 `--no-flat-profile[=SYMSPEC]'
369 The `-P' option causes `gprof' to suppress printing a flat profile.
370 If SYMSPEC is specified, `gprof' prints a flat profile, but
371 excludes matching symbols.
375 The `-q' option causes `gprof' to print the call graph analysis.
376 If SYMSPEC is specified, print call graph only for matching symbols
377 and their children. *Note The Call Graph: Call Graph.
380 `--no-graph[=SYMSPEC]'
381 The `-Q' option causes `gprof' to suppress printing the call graph.
382 If SYMSPEC is specified, `gprof' prints a call graph, but excludes
387 The `-t' option causes the NUM most active source lines in each
388 source file to be listed when source annotation is enabled. The
393 This option affects annotated source output only. Normally,
394 `gprof' prints annotated source files to standard-output. If this
395 option is specified, annotated source for a file named
396 `path/FILENAME' is generated in the file `FILENAME-ann'. If the
397 underlying file system would truncate `FILENAME-ann' so that it
398 overwrites the original `FILENAME', `gprof' generates annotated
399 source in the file `FILENAME.ann' instead (if the original file
400 name has an extension, that extension is _replaced_ with `.ann').
403 `--no-exec-counts[=SYMSPEC]'
404 The `-Z' option causes `gprof' not to print a tally of functions
405 and the number of times each was called. If SYMSPEC is specified,
406 print tally, but exclude matching symbols.
409 `--function-ordering'
410 The `--function-ordering' option causes `gprof' to print a
411 suggested function ordering for the program based on profiling
412 data. This option suggests an ordering which may improve paging,
413 tlb and cache behavior for the program on systems which support
414 arbitrary ordering of functions in an executable.
416 The exact details of how to force the linker to place functions in
417 a particular order is system dependent and out of the scope of this
421 `--file-ordering MAP_FILE'
422 The `--file-ordering' option causes `gprof' to print a suggested
423 .o link line ordering for the program based on profiling data.
424 This option suggests an ordering which may improve paging, tlb and
425 cache behavior for the program on systems which do not support
426 arbitrary ordering of functions in an executable.
428 Use of the `-a' argument is highly recommended with this option.
430 The MAP_FILE argument is a pathname to a file which provides
431 function name to object file mappings. The format of the file is
432 similar to the output of the program `nm'.
434 c-parse.o:00000000 T yyparse
435 c-parse.o:00000004 C yyerrflag
436 c-lang.o:00000000 T maybe_objc_method_name
437 c-lang.o:00000000 T print_lang_statistics
438 c-lang.o:00000000 T recognize_objc_keyword
439 c-decl.o:00000000 T print_lang_identifier
440 c-decl.o:00000000 T print_lang_type
443 To create a MAP_FILE with GNU `nm', type a command like `nm
444 --extern-only --defined-only -v --print-file-name program-name'.
448 The `-T' option causes `gprof' to print its output in
449 "traditional" BSD style.
453 Sets width of output lines to WIDTH. Currently only used when
454 printing the function index at the bottom of the call graph.
458 This option affects annotated source output only. By default,
459 only the lines at the beginning of a basic-block are annotated.
460 If this option is specified, every line in a basic-block is
461 annotated by repeating the annotation for the first line. This
462 behavior is similar to `tcov''s `-a'.
466 These options control whether C++ symbol names should be demangled
467 when printing output. The default is to demangle symbols. The
468 `--no-demangle' option may be used to turn off demangling.
469 Different compilers have different mangling styles. The optional
470 demangling style argument can be used to choose an appropriate
471 demangling style for your compiler.
474 File: gprof.info, Node: Analysis Options, Next: Miscellaneous Options, Prev: Output Options, Up: Invoking
481 The `-a' option causes `gprof' to suppress the printing of
482 statically declared (private) functions. (These are functions
483 whose names are not listed as global, and which are not visible
484 outside the file/function/block where they were defined.) Time
485 spent in these functions, calls to/from them, etc., will all be
486 attributed to the function that was loaded directly before it in
487 the executable file. This option affects both the flat profile
491 `--static-call-graph'
492 The `-c' option causes the call graph of the program to be
493 augmented by a heuristic which examines the text space of the
494 object file and identifies function calls in the binary machine
495 code. Since normal call graph records are only generated when
496 functions are entered, this option identifies children that could
497 have been called, but never were. Calls to functions that were
498 not compiled with profiling enabled are also identified, but only
499 if symbol table entries are present for them. Calls to dynamic
500 library routines are typically _not_ found by this option.
501 Parents or children identified via this heuristic are indicated in
502 the call graph with call counts of `0'.
505 `--ignore-non-functions'
506 The `-D' option causes `gprof' to ignore symbols which are not
507 known to be functions. This option will give more accurate
508 profile data on systems where it is supported (Solaris and HPUX for
512 The `-k' option allows you to delete from the call graph any arcs
513 from symbols matching symspec FROM to those matching symspec TO.
517 The `-l' option enables line-by-line profiling, which causes
518 histogram hits to be charged to individual source code lines,
519 instead of functions. This feature only works with programs
520 compiled by older versions of the `gcc' compiler. Newer versions
521 of `gcc' are designed to work with the `gcov' tool instead.
523 If the program was compiled with basic-block counting enabled,
524 this option will also identify how many times each line of code
525 was executed. While line-by-line profiling can help isolate where
526 in a large function a program is spending its time, it also
527 significantly increases the running time of `gprof', and magnifies
528 statistical inaccuracies. *Note Statistical Sampling Error:
533 This option affects execution count output only. Symbols that are
534 executed less than NUM times are suppressed.
538 The `-n' option causes `gprof', in its call graph analysis, to
539 only propagate times for symbols matching SYMSPEC.
543 The `-n' option causes `gprof', in its call graph analysis, not to
544 propagate times for symbols matching SYMSPEC.
547 `--external-symbol-table=FILENAME'
548 The `-S' option causes `gprof' to read an external symbol table
549 file, such as `/proc/kallsyms', rather than read the symbol table
550 from the given object file (the default is `a.out'). This is useful
551 for profiling kernel modules.
554 `--display-unused-functions'
555 If you give the `-z' option, `gprof' will mention all functions in
556 the flat profile, even those that were never called, and that had
557 no time spent in them. This is useful in conjunction with the
558 `-c' option for discovering which routines were never called.
562 File: gprof.info, Node: Miscellaneous Options, Next: Deprecated Options, Prev: Analysis Options, Up: Invoking
564 4.3 Miscellaneous Options
565 =========================
569 The `-d NUM' option specifies debugging options. If NUM is not
570 specified, enable all debugging. *Note Debugging `gprof':
575 The `-h' option prints command line usage.
579 Selects the format of the profile data files. Recognized formats
580 are `auto' (the default), `bsd', `4.4bsd', `magic', and `prof'
585 The `-s' option causes `gprof' to summarize the information in the
586 profile data files it read in, and write out a profile data file
587 called `gmon.sum', which contains all the information from the
588 profile data files that `gprof' read in. The file `gmon.sum' may
589 be one of the specified input files; the effect of this is to
590 merge the data in the other input files into `gmon.sum'.
592 Eventually you can run `gprof' again without `-s' to analyze the
593 cumulative data in the file `gmon.sum'.
597 The `-v' flag causes `gprof' to print the current version number,
602 File: gprof.info, Node: Deprecated Options, Next: Symspecs, Prev: Miscellaneous Options, Up: Invoking
604 4.4 Deprecated Options
605 ======================
607 These options have been replaced with newer versions that use symspecs.
610 The `-e FUNCTION' option tells `gprof' to not print information
611 about the function FUNCTION_NAME (and its children...) in the call
612 graph. The function will still be listed as a child of any
613 functions that call it, but its index number will be shown as
614 `[not printed]'. More than one `-e' option may be given; only one
615 FUNCTION_NAME may be indicated with each `-e' option.
618 The `-E FUNCTION' option works like the `-e' option, but time
619 spent in the function (and children who were not called from
620 anywhere else), will not be used to compute the
621 percentages-of-time for the call graph. More than one `-E' option
622 may be given; only one FUNCTION_NAME may be indicated with each
626 The `-f FUNCTION' option causes `gprof' to limit the call graph to
627 the function FUNCTION_NAME and its children (and their
628 children...). More than one `-f' option may be given; only one
629 FUNCTION_NAME may be indicated with each `-f' option.
632 The `-F FUNCTION' option works like the `-f' option, but only time
633 spent in the function and its children (and their children...)
634 will be used to determine total-time and percentages-of-time for
635 the call graph. More than one `-F' option may be given; only one
636 FUNCTION_NAME may be indicated with each `-F' option. The `-F'
637 option overrides the `-E' option.
640 Note that only one function can be specified with each `-e', `-E',
641 `-f' or `-F' option. To specify more than one function, use multiple
642 options. For example, this command:
644 gprof -e boring -f foo -f bar myprogram > gprof.output
646 lists in the call graph all functions that were reached from either
647 `foo' or `bar' and were not reachable from `boring'.
650 File: gprof.info, Node: Symspecs, Prev: Deprecated Options, Up: Invoking
655 Many of the output options allow functions to be included or excluded
656 using "symspecs" (symbol specifications), which observe the following
659 filename_containing_a_dot
660 | funcname_not_containing_a_dot
662 | ( [ any_filename ] `:' ( any_funcname | linenumber ) )
664 Here are some sample symspecs:
667 Selects everything in file `main.c'--the dot in the string tells
668 `gprof' to interpret the string as a filename, rather than as a
669 function name. To select a file whose name does not contain a
670 dot, a trailing colon should be specified. For example, `odd:' is
671 interpreted as the file named `odd'.
674 Selects all functions named `main'.
676 Note that there may be multiple instances of the same function name
677 because some of the definitions may be local (i.e., static).
678 Unless a function name is unique in a program, you must use the
679 colon notation explained below to specify a function from a
680 specific source file.
682 Sometimes, function names contain dots. In such cases, it is
683 necessary to add a leading colon to the name. For example,
684 `:.mul' selects function `.mul'.
686 In some object file formats, symbols have a leading underscore.
687 `gprof' will normally not print these underscores. When you name a
688 symbol in a symspec, you should type it exactly as `gprof' prints
689 it in its output. For example, if the compiler produces a symbol
690 `_main' from your `main' function, `gprof' still prints it as
691 `main' in its output, so you should use `main' in symspecs.
694 Selects function `main' in file `main.c'.
697 Selects line 134 in file `main.c'.
700 File: gprof.info, Node: Output, Next: Inaccuracy, Prev: Invoking, Up: Top
702 5 Interpreting `gprof''s Output
703 *******************************
705 `gprof' can produce several different output styles, the most important
706 of which are described below. The simplest output styles (file
707 information, execution count, and function and file ordering) are not
708 described here, but are documented with the respective options that
709 trigger them. *Note Output Options: Output Options.
713 * Flat Profile:: The flat profile shows how much time was spent
714 executing directly in each function.
715 * Call Graph:: The call graph shows which functions called which
716 others, and how much time each function used
717 when its subroutine calls are included.
718 * Line-by-line:: `gprof' can analyze individual source code lines
719 * Annotated Source:: The annotated source listing displays source code
720 labeled with execution counts
723 File: gprof.info, Node: Flat Profile, Next: Call Graph, Up: Output
728 The "flat profile" shows the total amount of time your program spent
729 executing each function. Unless the `-z' option is given, functions
730 with no apparent time spent in them, and no apparent calls to them, are
731 not mentioned. Note that if a function was not compiled for profiling,
732 and didn't run long enough to show up on the program counter histogram,
733 it will be indistinguishable from a function that was never called.
735 This is part of a flat profile for a small program:
739 Each sample counts as 0.01 seconds.
740 % cumulative self self total
741 time seconds seconds calls ms/call ms/call name
742 33.34 0.02 0.02 7208 0.00 0.00 open
743 16.67 0.03 0.01 244 0.04 0.12 offtime
744 16.67 0.04 0.01 8 1.25 1.25 memccpy
745 16.67 0.05 0.01 7 1.43 1.43 write
746 16.67 0.06 0.01 mcount
747 0.00 0.06 0.00 236 0.00 0.00 tzset
748 0.00 0.06 0.00 192 0.00 0.00 tolower
749 0.00 0.06 0.00 47 0.00 0.00 strlen
750 0.00 0.06 0.00 45 0.00 0.00 strchr
751 0.00 0.06 0.00 1 0.00 50.00 main
752 0.00 0.06 0.00 1 0.00 0.00 memcpy
753 0.00 0.06 0.00 1 0.00 10.11 print
754 0.00 0.06 0.00 1 0.00 0.00 profil
755 0.00 0.06 0.00 1 0.00 50.00 report
758 The functions are sorted first by decreasing run-time spent in them,
759 then by decreasing number of calls, then alphabetically by name. The
760 functions `mcount' and `profil' are part of the profiling apparatus and
761 appear in every flat profile; their time gives a measure of the amount
762 of overhead due to profiling.
764 Just before the column headers, a statement appears indicating how
765 much time each sample counted as. This "sampling period" estimates the
766 margin of error in each of the time figures. A time figure that is not
767 much larger than this is not reliable. In this example, each sample
768 counted as 0.01 seconds, suggesting a 100 Hz sampling rate. The
769 program's total execution time was 0.06 seconds, as indicated by the
770 `cumulative seconds' field. Since each sample counted for 0.01
771 seconds, this means only six samples were taken during the run. Two of
772 the samples occurred while the program was in the `open' function, as
773 indicated by the `self seconds' field. Each of the other four samples
774 occurred one each in `offtime', `memccpy', `write', and `mcount'.
775 Since only six samples were taken, none of these values can be regarded
776 as particularly reliable. In another run, the `self seconds' field for
777 `mcount' might well be `0.00' or `0.02'. *Note Statistical Sampling
778 Error: Sampling Error, for a complete discussion.
780 The remaining functions in the listing (those whose `self seconds'
781 field is `0.00') didn't appear in the histogram samples at all.
782 However, the call graph indicated that they were called, so therefore
783 they are listed, sorted in decreasing order by the `calls' field.
784 Clearly some time was spent executing these functions, but the paucity
785 of histogram samples prevents any determination of how much time each
788 Here is what the fields in each line mean:
791 This is the percentage of the total execution time your program
792 spent in this function. These should all add up to 100%.
795 This is the cumulative total number of seconds the computer spent
796 executing this functions, plus the time spent in all the functions
797 above this one in this table.
800 This is the number of seconds accounted for by this function alone.
801 The flat profile listing is sorted first by this number.
804 This is the total number of times the function was called. If the
805 function was never called, or the number of times it was called
806 cannot be determined (probably because the function was not
807 compiled with profiling enabled), the "calls" field is blank.
810 This represents the average number of milliseconds spent in this
811 function per call, if this function is profiled. Otherwise, this
812 field is blank for this function.
815 This represents the average number of milliseconds spent in this
816 function and its descendants per call, if this function is
817 profiled. Otherwise, this field is blank for this function. This
818 is the only field in the flat profile that uses call graph
822 This is the name of the function. The flat profile is sorted by
823 this field alphabetically after the "self seconds" and "calls"
827 File: gprof.info, Node: Call Graph, Next: Line-by-line, Prev: Flat Profile, Up: Output
832 The "call graph" shows how much time was spent in each function and its
833 children. From this information, you can find functions that, while
834 they themselves may not have used much time, called other functions
835 that did use unusual amounts of time.
837 Here is a sample call from a small program. This call came from the
838 same `gprof' run as the flat profile example in the previous section.
840 granularity: each sample hit covers 2 byte(s) for 20.00% of 0.05 seconds
842 index % time self children called name
844 [1] 100.0 0.00 0.05 start [1]
845 0.00 0.05 1/1 main [2]
846 0.00 0.00 1/2 on_exit [28]
847 0.00 0.00 1/1 exit [59]
848 -----------------------------------------------
849 0.00 0.05 1/1 start [1]
850 [2] 100.0 0.00 0.05 1 main [2]
851 0.00 0.05 1/1 report [3]
852 -----------------------------------------------
853 0.00 0.05 1/1 main [2]
854 [3] 100.0 0.00 0.05 1 report [3]
855 0.00 0.03 8/8 timelocal [6]
856 0.00 0.01 1/1 print [9]
857 0.00 0.01 9/9 fgets [12]
858 0.00 0.00 12/34 strncmp <cycle 1> [40]
859 0.00 0.00 8/8 lookup [20]
860 0.00 0.00 1/1 fopen [21]
861 0.00 0.00 8/8 chewtime [24]
862 0.00 0.00 8/16 skipspace [44]
863 -----------------------------------------------
864 [4] 59.8 0.01 0.02 8+472 <cycle 2 as a whole> [4]
865 0.01 0.02 244+260 offtime <cycle 2> [7]
866 0.00 0.00 236+1 tzset <cycle 2> [26]
867 -----------------------------------------------
869 The lines full of dashes divide this table into "entries", one for
870 each function. Each entry has one or more lines.
872 In each entry, the primary line is the one that starts with an index
873 number in square brackets. The end of this line says which function
874 the entry is for. The preceding lines in the entry describe the
875 callers of this function and the following lines describe its
876 subroutines (also called "children" when we speak of the call graph).
878 The entries are sorted by time spent in the function and its
881 The internal profiling function `mcount' (*note The Flat Profile:
882 Flat Profile.) is never mentioned in the call graph.
886 * Primary:: Details of the primary line's contents.
887 * Callers:: Details of caller-lines' contents.
888 * Subroutines:: Details of subroutine-lines' contents.
889 * Cycles:: When there are cycles of recursion,
890 such as `a' calls `b' calls `a'...
893 File: gprof.info, Node: Primary, Next: Callers, Up: Call Graph
895 5.2.1 The Primary Line
896 ----------------------
898 The "primary line" in a call graph entry is the line that describes the
899 function which the entry is about and gives the overall statistics for
902 For reference, we repeat the primary line from the entry for function
903 `report' in our main example, together with the heading line that shows
904 the names of the fields:
906 index % time self children called name
908 [3] 100.0 0.00 0.05 1 report [3]
910 Here is what the fields in the primary line mean:
913 Entries are numbered with consecutive integers. Each function
914 therefore has an index number, which appears at the beginning of
917 Each cross-reference to a function, as a caller or subroutine of
918 another, gives its index number as well as its name. The index
919 number guides you if you wish to look for the entry for that
923 This is the percentage of the total time that was spent in this
924 function, including time spent in subroutines called from this
927 The time spent in this function is counted again for the callers of
928 this function. Therefore, adding up these percentages is
932 This is the total amount of time spent in this function. This
933 should be identical to the number printed in the `seconds' field
934 for this function in the flat profile.
937 This is the total amount of time spent in the subroutine calls
938 made by this function. This should be equal to the sum of all the
939 `self' and `children' entries of the children listed directly
943 This is the number of times the function was called.
945 If the function called itself recursively, there are two numbers,
946 separated by a `+'. The first number counts non-recursive calls,
947 and the second counts recursive calls.
949 In the example above, the function `report' was called once from
953 This is the name of the current function. The index number is
956 If the function is part of a cycle of recursion, the cycle number
957 is printed between the function's name and the index number (*note
958 How Mutually Recursive Functions Are Described: Cycles.). For
959 example, if function `gnurr' is part of cycle number one, and has
960 index number twelve, its primary line would be end like this:
965 File: gprof.info, Node: Callers, Next: Subroutines, Prev: Primary, Up: Call Graph
967 5.2.2 Lines for a Function's Callers
968 ------------------------------------
970 A function's entry has a line for each function it was called by.
971 These lines' fields correspond to the fields of the primary line, but
972 their meanings are different because of the difference in context.
974 For reference, we repeat two lines from the entry for the function
975 `report', the primary line and one caller-line preceding it, together
976 with the heading line that shows the names of the fields:
978 index % time self children called name
980 0.00 0.05 1/1 main [2]
981 [3] 100.0 0.00 0.05 1 report [3]
983 Here are the meanings of the fields in the caller-line for `report'
987 An estimate of the amount of time spent in `report' itself when it
988 was called from `main'.
991 An estimate of the amount of time spent in subroutines of `report'
992 when `report' was called from `main'.
994 The sum of the `self' and `children' fields is an estimate of the
995 amount of time spent within calls to `report' from `main'.
998 Two numbers: the number of times `report' was called from `main',
999 followed by the total number of non-recursive calls to `report'
1000 from all its callers.
1002 `name and index number'
1003 The name of the caller of `report' to which this line applies,
1004 followed by the caller's index number.
1006 Not all functions have entries in the call graph; some options to
1007 `gprof' request the omission of certain functions. When a caller
1008 has no entry of its own, it still has caller-lines in the entries
1009 of the functions it calls.
1011 If the caller is part of a recursion cycle, the cycle number is
1012 printed between the name and the index number.
1014 If the identity of the callers of a function cannot be determined, a
1015 dummy caller-line is printed which has `<spontaneous>' as the "caller's
1016 name" and all other fields blank. This can happen for signal handlers.
1019 File: gprof.info, Node: Subroutines, Next: Cycles, Prev: Callers, Up: Call Graph
1021 5.2.3 Lines for a Function's Subroutines
1022 ----------------------------------------
1024 A function's entry has a line for each of its subroutines--in other
1025 words, a line for each other function that it called. These lines'
1026 fields correspond to the fields of the primary line, but their meanings
1027 are different because of the difference in context.
1029 For reference, we repeat two lines from the entry for the function
1030 `main', the primary line and a line for a subroutine, together with the
1031 heading line that shows the names of the fields:
1033 index % time self children called name
1035 [2] 100.0 0.00 0.05 1 main [2]
1036 0.00 0.05 1/1 report [3]
1038 Here are the meanings of the fields in the subroutine-line for `main'
1042 An estimate of the amount of time spent directly within `report'
1043 when `report' was called from `main'.
1046 An estimate of the amount of time spent in subroutines of `report'
1047 when `report' was called from `main'.
1049 The sum of the `self' and `children' fields is an estimate of the
1050 total time spent in calls to `report' from `main'.
1053 Two numbers, the number of calls to `report' from `main' followed
1054 by the total number of non-recursive calls to `report'. This
1055 ratio is used to determine how much of `report''s `self' and
1056 `children' time gets credited to `main'. *Note Estimating
1057 `children' Times: Assumptions.
1060 The name of the subroutine of `main' to which this line applies,
1061 followed by the subroutine's index number.
1063 If the caller is part of a recursion cycle, the cycle number is
1064 printed between the name and the index number.
1067 File: gprof.info, Node: Cycles, Prev: Subroutines, Up: Call Graph
1069 5.2.4 How Mutually Recursive Functions Are Described
1070 ----------------------------------------------------
1072 The graph may be complicated by the presence of "cycles of recursion"
1073 in the call graph. A cycle exists if a function calls another function
1074 that (directly or indirectly) calls (or appears to call) the original
1075 function. For example: if `a' calls `b', and `b' calls `a', then `a'
1076 and `b' form a cycle.
1078 Whenever there are call paths both ways between a pair of functions,
1079 they belong to the same cycle. If `a' and `b' call each other and `b'
1080 and `c' call each other, all three make one cycle. Note that even if
1081 `b' only calls `a' if it was not called from `a', `gprof' cannot
1082 determine this, so `a' and `b' are still considered a cycle.
1084 The cycles are numbered with consecutive integers. When a function
1085 belongs to a cycle, each time the function name appears in the call
1086 graph it is followed by `<cycle NUMBER>'.
1088 The reason cycles matter is that they make the time values in the
1089 call graph paradoxical. The "time spent in children" of `a' should
1090 include the time spent in its subroutine `b' and in `b''s
1091 subroutines--but one of `b''s subroutines is `a'! How much of `a''s
1092 time should be included in the children of `a', when `a' is indirectly
1095 The way `gprof' resolves this paradox is by creating a single entry
1096 for the cycle as a whole. The primary line of this entry describes the
1097 total time spent directly in the functions of the cycle. The
1098 "subroutines" of the cycle are the individual functions of the cycle,
1099 and all other functions that were called directly by them. The
1100 "callers" of the cycle are the functions, outside the cycle, that
1101 called functions in the cycle.
1103 Here is an example portion of a call graph which shows a cycle
1104 containing functions `a' and `b'. The cycle was entered by a call to
1105 `a' from `main'; both `a' and `b' called `c'.
1107 index % time self children called name
1108 ----------------------------------------
1110 [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3]
1111 1.02 0 3 b <cycle 1> [4]
1112 0.75 0 2 a <cycle 1> [5]
1113 ----------------------------------------
1115 [4] 52.85 1.02 0 0 b <cycle 1> [4]
1118 ----------------------------------------
1121 [5] 38.86 0.75 0 1 a <cycle 1> [5]
1124 ----------------------------------------
1126 (The entire call graph for this program contains in addition an entry
1127 for `main', which calls `a', and an entry for `c', with callers `a' and
1130 index % time self children called name
1132 [1] 100.00 0 1.93 0 start [1]
1133 0.16 1.77 1/1 main [2]
1134 ----------------------------------------
1135 0.16 1.77 1/1 start [1]
1136 [2] 100.00 0.16 1.77 1 main [2]
1137 1.77 0 1/1 a <cycle 1> [5]
1138 ----------------------------------------
1140 [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3]
1141 1.02 0 3 b <cycle 1> [4]
1142 0.75 0 2 a <cycle 1> [5]
1144 ----------------------------------------
1146 [4] 52.85 1.02 0 0 b <cycle 1> [4]
1149 ----------------------------------------
1152 [5] 38.86 0.75 0 1 a <cycle 1> [5]
1155 ----------------------------------------
1156 0 0 3/6 b <cycle 1> [4]
1157 0 0 3/6 a <cycle 1> [5]
1158 [6] 0.00 0 0 6 c [6]
1159 ----------------------------------------
1161 The `self' field of the cycle's primary line is the total time spent
1162 in all the functions of the cycle. It equals the sum of the `self'
1163 fields for the individual functions in the cycle, found in the entry in
1164 the subroutine lines for these functions.
1166 The `children' fields of the cycle's primary line and subroutine
1167 lines count only subroutines outside the cycle. Even though `a' calls
1168 `b', the time spent in those calls to `b' is not counted in `a''s
1169 `children' time. Thus, we do not encounter the problem of what to do
1170 when the time in those calls to `b' includes indirect recursive calls
1173 The `children' field of a caller-line in the cycle's entry estimates
1174 the amount of time spent _in the whole cycle_, and its other
1175 subroutines, on the times when that caller called a function in the
1178 The `called' field in the primary line for the cycle has two numbers:
1179 first, the number of times functions in the cycle were called by
1180 functions outside the cycle; second, the number of times they were
1181 called by functions in the cycle (including times when a function in
1182 the cycle calls itself). This is a generalization of the usual split
1183 into non-recursive and recursive calls.
1185 The `called' field of a subroutine-line for a cycle member in the
1186 cycle's entry says how many time that function was called from
1187 functions in the cycle. The total of all these is the second number in
1188 the primary line's `called' field.
1190 In the individual entry for a function in a cycle, the other
1191 functions in the same cycle can appear as subroutines and as callers.
1192 These lines show how many times each function in the cycle called or
1193 was called from each other function in the cycle. The `self' and
1194 `children' fields in these lines are blank because of the difficulty of
1195 defining meanings for them when recursion is going on.
1198 File: gprof.info, Node: Line-by-line, Next: Annotated Source, Prev: Call Graph, Up: Output
1200 5.3 Line-by-line Profiling
1201 ==========================
1203 `gprof''s `-l' option causes the program to perform "line-by-line"
1204 profiling. In this mode, histogram samples are assigned not to
1205 functions, but to individual lines of source code. This only works
1206 with programs compiled with older versions of the `gcc' compiler.
1207 Newer versions of `gcc' use a different program - `gcov' - to display
1208 line-by-line profiling information.
1210 With the older versions of `gcc' the program usually has to be
1211 compiled with a `-g' option, in addition to `-pg', in order to generate
1212 debugging symbols for tracking source code lines. Note, in much older
1213 versions of `gcc' the program had to be compiled with the `-a' command
1214 line option as well.
1216 The flat profile is the most useful output table in line-by-line
1217 mode. The call graph isn't as useful as normal, since the current
1218 version of `gprof' does not propagate call graph arcs from source code
1219 lines to the enclosing function. The call graph does, however, show
1220 each line of code that called each function, along with a count.
1222 Here is a section of `gprof''s output, without line-by-line
1223 profiling. Note that `ct_init' accounted for four histogram hits, and
1224 13327 calls to `init_block'.
1228 Each sample counts as 0.01 seconds.
1229 % cumulative self self total
1230 time seconds seconds calls us/call us/call name
1231 30.77 0.13 0.04 6335 6.31 6.31 ct_init
1234 Call graph (explanation follows)
1237 granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds
1239 index % time self children called name
1241 0.00 0.00 1/13496 name_too_long
1242 0.00 0.00 40/13496 deflate
1243 0.00 0.00 128/13496 deflate_fast
1244 0.00 0.00 13327/13496 ct_init
1245 [7] 0.0 0.00 0.00 13496 init_block
1247 Now let's look at some of `gprof''s output from the same program run,
1248 this time with line-by-line profiling enabled. Note that `ct_init''s
1249 four histogram hits are broken down into four lines of source code--one
1250 hit occurred on each of lines 349, 351, 382 and 385. In the call graph,
1251 note how `ct_init''s 13327 calls to `init_block' are broken down into
1252 one call from line 396, 3071 calls from line 384, 3730 calls from line
1253 385, and 6525 calls from 387.
1257 Each sample counts as 0.01 seconds.
1259 time seconds seconds calls name
1260 7.69 0.10 0.01 ct_init (trees.c:349)
1261 7.69 0.11 0.01 ct_init (trees.c:351)
1262 7.69 0.12 0.01 ct_init (trees.c:382)
1263 7.69 0.13 0.01 ct_init (trees.c:385)
1266 Call graph (explanation follows)
1269 granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds
1271 % time self children called name
1273 0.00 0.00 1/13496 name_too_long (gzip.c:1440)
1274 0.00 0.00 1/13496 deflate (deflate.c:763)
1275 0.00 0.00 1/13496 ct_init (trees.c:396)
1276 0.00 0.00 2/13496 deflate (deflate.c:727)
1277 0.00 0.00 4/13496 deflate (deflate.c:686)
1278 0.00 0.00 5/13496 deflate (deflate.c:675)
1279 0.00 0.00 12/13496 deflate (deflate.c:679)
1280 0.00 0.00 16/13496 deflate (deflate.c:730)
1281 0.00 0.00 128/13496 deflate_fast (deflate.c:654)
1282 0.00 0.00 3071/13496 ct_init (trees.c:384)
1283 0.00 0.00 3730/13496 ct_init (trees.c:385)
1284 0.00 0.00 6525/13496 ct_init (trees.c:387)
1285 [6] 0.0 0.00 0.00 13496 init_block (trees.c:408)
1288 File: gprof.info, Node: Annotated Source, Prev: Line-by-line, Up: Output
1290 5.4 The Annotated Source Listing
1291 ================================
1293 `gprof''s `-A' option triggers an annotated source listing, which lists
1294 the program's source code, each function labeled with the number of
1295 times it was called. You may also need to specify the `-I' option, if
1296 `gprof' can't find the source code files.
1298 With older versions of `gcc' compiling with `gcc ... -g -pg -a'
1299 augments your program with basic-block counting code, in addition to
1300 function counting code. This enables `gprof' to determine how many
1301 times each line of code was executed. With newer versions of `gcc'
1302 support for displaying basic-block counts is provided by the `gcov'
1305 For example, consider the following function, taken from gzip, with
1314 7 static ulg crc = (ulg)0xffffffffL;
1321 14 c = crc_32_tab[...];
1325 18 return c ^ 0xffffffffL;
1328 `updcrc' has at least five basic-blocks. One is the function
1329 itself. The `if' statement on line 9 generates two more basic-blocks,
1330 one for each branch of the `if'. A fourth basic-block results from the
1331 `if' on line 13, and the contents of the `do' loop form the fifth
1332 basic-block. The compiler may also generate additional basic-blocks to
1333 handle various special cases.
1335 A program augmented for basic-block counting can be analyzed with
1336 `gprof -l -A'. The `-x' option is also helpful, to ensure that each
1337 line of code is labeled at least once. Here is `updcrc''s annotated
1338 source listing for a sample `gzip' run:
1346 static ulg crc = (ulg)0xffffffffL;
1348 2 -> if (s == NULL) {
1349 1 -> c = 0xffffffffL;
1353 26312 -> c = crc_32_tab[...];
1354 26312,1,26311 -> } while (--n);
1357 2 -> return c ^ 0xffffffffL;
1360 In this example, the function was called twice, passing once through
1361 each branch of the `if' statement. The body of the `do' loop was
1362 executed a total of 26312 times. Note how the `while' statement is
1363 annotated. It began execution 26312 times, once for each iteration
1364 through the loop. One of those times (the last time) it exited, while
1365 it branched back to the beginning of the loop 26311 times.
1368 File: gprof.info, Node: Inaccuracy, Next: How do I?, Prev: Output, Up: Top
1370 6 Inaccuracy of `gprof' Output
1371 ******************************
1375 * Sampling Error:: Statistical margins of error
1376 * Assumptions:: Estimating children times
1379 File: gprof.info, Node: Sampling Error, Next: Assumptions, Up: Inaccuracy
1381 6.1 Statistical Sampling Error
1382 ==============================
1384 The run-time figures that `gprof' gives you are based on a sampling
1385 process, so they are subject to statistical inaccuracy. If a function
1386 runs only a small amount of time, so that on the average the sampling
1387 process ought to catch that function in the act only once, there is a
1388 pretty good chance it will actually find that function zero times, or
1391 By contrast, the number-of-calls and basic-block figures are derived
1392 by counting, not sampling. They are completely accurate and will not
1393 vary from run to run if your program is deterministic and single
1394 threaded. In multi-threaded applications, or single threaded
1395 applications that link with multi-threaded libraries, the counts are
1396 only deterministic if the counting function is thread-safe. (Note:
1397 beware that the mcount counting function in glibc is _not_
1398 thread-safe). *Note Implementation of Profiling: Implementation.
1400 The "sampling period" that is printed at the beginning of the flat
1401 profile says how often samples are taken. The rule of thumb is that a
1402 run-time figure is accurate if it is considerably bigger than the
1405 The actual amount of error can be predicted. For N samples, the
1406 _expected_ error is the square-root of N. For example, if the sampling
1407 period is 0.01 seconds and `foo''s run-time is 1 second, N is 100
1408 samples (1 second/0.01 seconds), sqrt(N) is 10 samples, so the expected
1409 error in `foo''s run-time is 0.1 seconds (10*0.01 seconds), or ten
1410 percent of the observed value. Again, if the sampling period is 0.01
1411 seconds and `bar''s run-time is 100 seconds, N is 10000 samples,
1412 sqrt(N) is 100 samples, so the expected error in `bar''s run-time is 1
1413 second, or one percent of the observed value. It is likely to vary
1414 this much _on the average_ from one profiling run to the next.
1415 (_Sometimes_ it will vary more.)
1417 This does not mean that a small run-time figure is devoid of
1418 information. If the program's _total_ run-time is large, a small
1419 run-time for one function does tell you that that function used an
1420 insignificant fraction of the whole program's time. Usually this means
1421 it is not worth optimizing.
1423 One way to get more accuracy is to give your program more (but
1424 similar) input data so it will take longer. Another way is to combine
1425 the data from several runs, using the `-s' option of `gprof'. Here is
1428 1. Run your program once.
1430 2. Issue the command `mv gmon.out gmon.sum'.
1432 3. Run your program again, the same as before.
1434 4. Merge the new data in `gmon.out' into `gmon.sum' with this command:
1436 gprof -s EXECUTABLE-FILE gmon.out gmon.sum
1438 5. Repeat the last two steps as often as you wish.
1440 6. Analyze the cumulative data using this command:
1442 gprof EXECUTABLE-FILE gmon.sum > OUTPUT-FILE
1445 File: gprof.info, Node: Assumptions, Prev: Sampling Error, Up: Inaccuracy
1447 6.2 Estimating `children' Times
1448 ===============================
1450 Some of the figures in the call graph are estimates--for example, the
1451 `children' time values and all the time figures in caller and
1454 There is no direct information about these measurements in the
1455 profile data itself. Instead, `gprof' estimates them by making an
1456 assumption about your program that might or might not be true.
1458 The assumption made is that the average time spent in each call to
1459 any function `foo' is not correlated with who called `foo'. If `foo'
1460 used 5 seconds in all, and 2/5 of the calls to `foo' came from `a',
1461 then `foo' contributes 2 seconds to `a''s `children' time, by
1464 This assumption is usually true enough, but for some programs it is
1465 far from true. Suppose that `foo' returns very quickly when its
1466 argument is zero; suppose that `a' always passes zero as an argument,
1467 while other callers of `foo' pass other arguments. In this program,
1468 all the time spent in `foo' is in the calls from callers other than `a'.
1469 But `gprof' has no way of knowing this; it will blindly and incorrectly
1470 charge 2 seconds of time in `foo' to the children of `a'.
1472 We hope some day to put more complete data into `gmon.out', so that
1473 this assumption is no longer needed, if we can figure out how. For the
1474 novice, the estimated figures are usually more useful than misleading.
1477 File: gprof.info, Node: How do I?, Next: Incompatibilities, Prev: Inaccuracy, Up: Top
1479 7 Answers to Common Questions
1480 *****************************
1482 How can I get more exact information about hot spots in my program?
1483 Looking at the per-line call counts only tells part of the story.
1484 Because `gprof' can only report call times and counts by function,
1485 the best way to get finer-grained information on where the program
1486 is spending its time is to re-factor large functions into sequences
1487 of calls to smaller ones. Beware however that this can introduce
1488 artificial hot spots since compiling with `-pg' adds a significant
1489 overhead to function calls. An alternative solution is to use a
1490 non-intrusive profiler, e.g. oprofile.
1492 How do I find which lines in my program were executed the most times?
1493 Use the `gcov' program.
1495 How do I find which lines in my program called a particular function?
1496 Use `gprof -l' and lookup the function in the call graph. The
1497 callers will be broken down by function and line number.
1499 How do I analyze a program that runs for less than a second?
1500 Try using a shell script like this one:
1502 for i in `seq 1 100`; do
1504 mv gmon.out gmon.out.$i
1507 gprof -s fastprog gmon.out.*
1509 gprof fastprog gmon.sum
1511 If your program is completely deterministic, all the call counts
1512 will be simple multiples of 100 (i.e., a function called once in
1513 each run will appear with a call count of 100).
1517 File: gprof.info, Node: Incompatibilities, Next: Details, Prev: How do I?, Up: Top
1519 8 Incompatibilities with Unix `gprof'
1520 *************************************
1522 GNU `gprof' and Berkeley Unix `gprof' use the same data file
1523 `gmon.out', and provide essentially the same information. But there
1524 are a few differences.
1526 * GNU `gprof' uses a new, generalized file format with support for
1527 basic-block execution counts and non-realtime histograms. A magic
1528 cookie and version number allows `gprof' to easily identify new
1529 style files. Old BSD-style files can still be read. *Note
1530 Profiling Data File Format: File Format.
1532 * For a recursive function, Unix `gprof' lists the function as a
1533 parent and as a child, with a `calls' field that lists the number
1534 of recursive calls. GNU `gprof' omits these lines and puts the
1535 number of recursive calls in the primary line.
1537 * When a function is suppressed from the call graph with `-e', GNU
1538 `gprof' still lists it as a subroutine of functions that call it.
1540 * GNU `gprof' accepts the `-k' with its argument in the form
1541 `from/to', instead of `from to'.
1543 * In the annotated source listing, if there are multiple basic
1544 blocks on the same line, GNU `gprof' prints all of their counts,
1545 separated by commas.
1547 * The blurbs, field widths, and output formats are different. GNU
1548 `gprof' prints blurbs after the tables, so that you can see the
1549 tables without skipping the blurbs.
1552 File: gprof.info, Node: Details, Next: GNU Free Documentation License, Prev: Incompatibilities, Up: Top
1554 9 Details of Profiling
1555 **********************
1559 * Implementation:: How a program collects profiling information
1560 * File Format:: Format of `gmon.out' files
1561 * Internals:: `gprof''s internal operation
1562 * Debugging:: Using `gprof''s `-d' option
1565 File: gprof.info, Node: Implementation, Next: File Format, Up: Details
1567 9.1 Implementation of Profiling
1568 ===============================
1570 Profiling works by changing how every function in your program is
1571 compiled so that when it is called, it will stash away some information
1572 about where it was called from. From this, the profiler can figure out
1573 what function called it, and can count how many times it was called.
1574 This change is made by the compiler when your program is compiled with
1575 the `-pg' option, which causes every function to call `mcount' (or
1576 `_mcount', or `__mcount', depending on the OS and compiler) as one of
1577 its first operations.
1579 The `mcount' routine, included in the profiling library, is
1580 responsible for recording in an in-memory call graph table both its
1581 parent routine (the child) and its parent's parent. This is typically
1582 done by examining the stack frame to find both the address of the
1583 child, and the return address in the original parent. Since this is a
1584 very machine-dependent operation, `mcount' itself is typically a short
1585 assembly-language stub routine that extracts the required information,
1586 and then calls `__mcount_internal' (a normal C function) with two
1587 arguments--`frompc' and `selfpc'. `__mcount_internal' is responsible
1588 for maintaining the in-memory call graph, which records `frompc',
1589 `selfpc', and the number of times each of these call arcs was traversed.
1591 GCC Version 2 provides a magical function
1592 (`__builtin_return_address'), which allows a generic `mcount' function
1593 to extract the required information from the stack frame. However, on
1594 some architectures, most notably the SPARC, using this builtin can be
1595 very computationally expensive, and an assembly language version of
1596 `mcount' is used for performance reasons.
1598 Number-of-calls information for library routines is collected by
1599 using a special version of the C library. The programs in it are the
1600 same as in the usual C library, but they were compiled with `-pg'. If
1601 you link your program with `gcc ... -pg', it automatically uses the
1602 profiling version of the library.
1604 Profiling also involves watching your program as it runs, and
1605 keeping a histogram of where the program counter happens to be every
1606 now and then. Typically the program counter is looked at around 100
1607 times per second of run time, but the exact frequency may vary from
1610 This is done is one of two ways. Most UNIX-like operating systems
1611 provide a `profil()' system call, which registers a memory array with
1612 the kernel, along with a scale factor that determines how the program's
1613 address space maps into the array. Typical scaling values cause every
1614 2 to 8 bytes of address space to map into a single array slot. On
1615 every tick of the system clock (assuming the profiled program is
1616 running), the value of the program counter is examined and the
1617 corresponding slot in the memory array is incremented. Since this is
1618 done in the kernel, which had to interrupt the process anyway to handle
1619 the clock interrupt, very little additional system overhead is required.
1621 However, some operating systems, most notably Linux 2.0 (and
1622 earlier), do not provide a `profil()' system call. On such a system,
1623 arrangements are made for the kernel to periodically deliver a signal
1624 to the process (typically via `setitimer()'), which then performs the
1625 same operation of examining the program counter and incrementing a slot
1626 in the memory array. Since this method requires a signal to be
1627 delivered to user space every time a sample is taken, it uses
1628 considerably more overhead than kernel-based profiling. Also, due to
1629 the added delay required to deliver the signal, this method is less
1632 A special startup routine allocates memory for the histogram and
1633 either calls `profil()' or sets up a clock signal handler. This
1634 routine (`monstartup') can be invoked in several ways. On Linux
1635 systems, a special profiling startup file `gcrt0.o', which invokes
1636 `monstartup' before `main', is used instead of the default `crt0.o'.
1637 Use of this special startup file is one of the effects of using `gcc
1638 ... -pg' to link. On SPARC systems, no special startup files are used.
1639 Rather, the `mcount' routine, when it is invoked for the first time
1640 (typically when `main' is called), calls `monstartup'.
1642 If the compiler's `-a' option was used, basic-block counting is also
1643 enabled. Each object file is then compiled with a static array of
1644 counts, initially zero. In the executable code, every time a new
1645 basic-block begins (i.e., when an `if' statement appears), an extra
1646 instruction is inserted to increment the corresponding count in the
1647 array. At compile time, a paired array was constructed that recorded
1648 the starting address of each basic-block. Taken together, the two
1649 arrays record the starting address of every basic-block, along with the
1650 number of times it was executed.
1652 The profiling library also includes a function (`mcleanup') which is
1653 typically registered using `atexit()' to be called as the program
1654 exits, and is responsible for writing the file `gmon.out'. Profiling
1655 is turned off, various headers are output, and the histogram is
1656 written, followed by the call-graph arcs and the basic-block counts.
1658 The output from `gprof' gives no indication of parts of your program
1659 that are limited by I/O or swapping bandwidth. This is because samples
1660 of the program counter are taken at fixed intervals of the program's
1661 run time. Therefore, the time measurements in `gprof' output say
1662 nothing about time that your program was not running. For example, a
1663 part of the program that creates so much data that it cannot all fit in
1664 physical memory at once may run very slowly due to thrashing, but
1665 `gprof' will say it uses little time. On the other hand, sampling by
1666 run time has the advantage that the amount of load due to other users
1667 won't directly affect the output you get.
1670 File: gprof.info, Node: File Format, Next: Internals, Prev: Implementation, Up: Details
1672 9.2 Profiling Data File Format
1673 ==============================
1675 The old BSD-derived file format used for profile data does not contain a
1676 magic cookie that allows to check whether a data file really is a
1677 `gprof' file. Furthermore, it does not provide a version number, thus
1678 rendering changes to the file format almost impossible. GNU `gprof'
1679 uses a new file format that provides these features. For backward
1680 compatibility, GNU `gprof' continues to support the old BSD-derived
1681 format, but not all features are supported with it. For example,
1682 basic-block execution counts cannot be accommodated by the old file
1685 The new file format is defined in header file `gmon_out.h'. It
1686 consists of a header containing the magic cookie and a version number,
1687 as well as some spare bytes available for future extensions. All data
1688 in a profile data file is in the native format of the target for which
1689 the profile was collected. GNU `gprof' adapts automatically to the
1692 In the new file format, the header is followed by a sequence of
1693 records. Currently, there are three different record types: histogram
1694 records, call-graph arc records, and basic-block execution count
1695 records. Each file can contain any number of each record type. When
1696 reading a file, GNU `gprof' will ensure records of the same type are
1697 compatible with each other and compute the union of all records. For
1698 example, for basic-block execution counts, the union is simply the sum
1699 of all execution counts for each basic-block.
1701 9.2.1 Histogram Records
1702 -----------------------
1704 Histogram records consist of a header that is followed by an array of
1705 bins. The header contains the text-segment range that the histogram
1706 spans, the size of the histogram in bytes (unlike in the old BSD
1707 format, this does not include the size of the header), the rate of the
1708 profiling clock, and the physical dimension that the bin counts
1709 represent after being scaled by the profiling clock rate. The physical
1710 dimension is specified in two parts: a long name of up to 15 characters
1711 and a single character abbreviation. For example, a histogram
1712 representing real-time would specify the long name as "seconds" and the
1713 abbreviation as "s". This feature is useful for architectures that
1714 support performance monitor hardware (which, fortunately, is becoming
1715 increasingly common). For example, under DEC OSF/1, the "uprofile"
1716 command can be used to produce a histogram of, say, instruction cache
1717 misses. In this case, the dimension in the histogram header could be
1718 set to "i-cache misses" and the abbreviation could be set to "1"
1719 (because it is simply a count, not a physical dimension). Also, the
1720 profiling rate would have to be set to 1 in this case.
1722 Histogram bins are 16-bit numbers and each bin represent an equal
1723 amount of text-space. For example, if the text-segment is one thousand
1724 bytes long and if there are ten bins in the histogram, each bin
1725 represents one hundred bytes.
1727 9.2.2 Call-Graph Records
1728 ------------------------
1730 Call-graph records have a format that is identical to the one used in
1731 the BSD-derived file format. It consists of an arc in the call graph
1732 and a count indicating the number of times the arc was traversed during
1733 program execution. Arcs are specified by a pair of addresses: the
1734 first must be within caller's function and the second must be within
1735 the callee's function. When performing profiling at the function
1736 level, these addresses can point anywhere within the respective
1737 function. However, when profiling at the line-level, it is better if
1738 the addresses are as close to the call-site/entry-point as possible.
1739 This will ensure that the line-level call-graph is able to identify
1740 exactly which line of source code performed calls to a function.
1742 9.2.3 Basic-Block Execution Count Records
1743 -----------------------------------------
1745 Basic-block execution count records consist of a header followed by a
1746 sequence of address/count pairs. The header simply specifies the
1747 length of the sequence. In an address/count pair, the address
1748 identifies a basic-block and the count specifies the number of times
1749 that basic-block was executed. Any address within the basic-address can
1753 File: gprof.info, Node: Internals, Next: Debugging, Prev: File Format, Up: Details
1755 9.3 `gprof''s Internal Operation
1756 ================================
1758 Like most programs, `gprof' begins by processing its options. During
1759 this stage, it may building its symspec list (`sym_ids.c:sym_id_add'),
1760 if options are specified which use symspecs. `gprof' maintains a
1761 single linked list of symspecs, which will eventually get turned into
1762 12 symbol tables, organized into six include/exclude pairs--one pair
1763 each for the flat profile (INCL_FLAT/EXCL_FLAT), the call graph arcs
1764 (INCL_ARCS/EXCL_ARCS), printing in the call graph
1765 (INCL_GRAPH/EXCL_GRAPH), timing propagation in the call graph
1766 (INCL_TIME/EXCL_TIME), the annotated source listing
1767 (INCL_ANNO/EXCL_ANNO), and the execution count listing
1768 (INCL_EXEC/EXCL_EXEC).
1770 After option processing, `gprof' finishes building the symspec list
1771 by adding all the symspecs in `default_excluded_list' to the exclude
1772 lists EXCL_TIME and EXCL_GRAPH, and if line-by-line profiling is
1773 specified, EXCL_FLAT as well. These default excludes are not added to
1774 EXCL_ANNO, EXCL_ARCS, and EXCL_EXEC.
1776 Next, the BFD library is called to open the object file, verify that
1777 it is an object file, and read its symbol table (`core.c:core_init'),
1778 using `bfd_canonicalize_symtab' after mallocing an appropriately sized
1779 array of symbols. At this point, function mappings are read (if the
1780 `--file-ordering' option has been specified), and the core text space
1781 is read into memory (if the `-c' option was given).
1783 `gprof''s own symbol table, an array of Sym structures, is now built.
1784 This is done in one of two ways, by one of two routines, depending on
1785 whether line-by-line profiling (`-l' option) has been enabled. For
1786 normal profiling, the BFD canonical symbol table is scanned. For
1787 line-by-line profiling, every text space address is examined, and a new
1788 symbol table entry gets created every time the line number changes. In
1789 either case, two passes are made through the symbol table--one to count
1790 the size of the symbol table required, and the other to actually read
1791 the symbols. In between the two passes, a single array of type `Sym'
1792 is created of the appropriate length. Finally,
1793 `symtab.c:symtab_finalize' is called to sort the symbol table and
1794 remove duplicate entries (entries with the same memory address).
1796 The symbol table must be a contiguous array for two reasons. First,
1797 the `qsort' library function (which sorts an array) will be used to
1798 sort the symbol table. Also, the symbol lookup routine
1799 (`symtab.c:sym_lookup'), which finds symbols based on memory address,
1800 uses a binary search algorithm which requires the symbol table to be a
1801 sorted array. Function symbols are indicated with an `is_func' flag.
1802 Line number symbols have no special flags set. Additionally, a symbol
1803 can have an `is_static' flag to indicate that it is a local symbol.
1805 With the symbol table read, the symspecs can now be translated into
1806 Syms (`sym_ids.c:sym_id_parse'). Remember that a single symspec can
1807 match multiple symbols. An array of symbol tables (`syms') is created,
1808 each entry of which is a symbol table of Syms to be included or
1809 excluded from a particular listing. The master symbol table and the
1810 symspecs are examined by nested loops, and every symbol that matches a
1811 symspec is inserted into the appropriate syms table. This is done
1812 twice, once to count the size of each required symbol table, and again
1813 to build the tables, which have been malloced between passes. From now
1814 on, to determine whether a symbol is on an include or exclude symspec
1815 list, `gprof' simply uses its standard symbol lookup routine on the
1816 appropriate table in the `syms' array.
1818 Now the profile data file(s) themselves are read
1819 (`gmon_io.c:gmon_out_read'), first by checking for a new-style
1820 `gmon.out' header, then assuming this is an old-style BSD `gmon.out' if
1821 the magic number test failed.
1823 New-style histogram records are read by `hist.c:hist_read_rec'. For
1824 the first histogram record, allocate a memory array to hold all the
1825 bins, and read them in. When multiple profile data files (or files
1826 with multiple histogram records) are read, the memory ranges of each
1827 pair of histogram records must be either equal, or non-overlapping.
1828 For each pair of histogram records, the resolution (memory region size
1829 divided by the number of bins) must be the same. The time unit must be
1830 the same for all histogram records. If the above containts are met, all
1831 histograms for the same memory range are merged.
1833 As each call graph record is read (`call_graph.c:cg_read_rec'), the
1834 parent and child addresses are matched to symbol table entries, and a
1835 call graph arc is created by `cg_arcs.c:arc_add', unless the arc fails
1836 a symspec check against INCL_ARCS/EXCL_ARCS. As each arc is added, a
1837 linked list is maintained of the parent's child arcs, and of the child's
1838 parent arcs. Both the child's call count and the arc's call count are
1839 incremented by the record's call count.
1841 Basic-block records are read (`basic_blocks.c:bb_read_rec'), but
1842 only if line-by-line profiling has been selected. Each basic-block
1843 address is matched to a corresponding line symbol in the symbol table,
1844 and an entry made in the symbol's bb_addr and bb_calls arrays. Again,
1845 if multiple basic-block records are present for the same address, the
1846 call counts are cumulative.
1848 A gmon.sum file is dumped, if requested (`gmon_io.c:gmon_out_write').
1850 If histograms were present in the data files, assign them to symbols
1851 (`hist.c:hist_assign_samples') by iterating over all the sample bins
1852 and assigning them to symbols. Since the symbol table is sorted in
1853 order of ascending memory addresses, we can simple follow along in the
1854 symbol table as we make our pass over the sample bins. This step
1855 includes a symspec check against INCL_FLAT/EXCL_FLAT. Depending on the
1856 histogram scale factor, a sample bin may span multiple symbols, in
1857 which case a fraction of the sample count is allocated to each symbol,
1858 proportional to the degree of overlap. This effect is rare for normal
1859 profiling, but overlaps are more common during line-by-line profiling,
1860 and can cause each of two adjacent lines to be credited with half a
1863 If call graph data is present, `cg_arcs.c:cg_assemble' is called.
1864 First, if `-c' was specified, a machine-dependent routine (`find_call')
1865 scans through each symbol's machine code, looking for subroutine call
1866 instructions, and adding them to the call graph with a zero call count.
1867 A topological sort is performed by depth-first numbering all the
1868 symbols (`cg_dfn.c:cg_dfn'), so that children are always numbered less
1869 than their parents, then making a array of pointers into the symbol
1870 table and sorting it into numerical order, which is reverse topological
1871 order (children appear before parents). Cycles are also detected at
1872 this point, all members of which are assigned the same topological
1873 number. Two passes are now made through this sorted array of symbol
1874 pointers. The first pass, from end to beginning (parents to children),
1875 computes the fraction of child time to propagate to each parent and a
1876 print flag. The print flag reflects symspec handling of
1877 INCL_GRAPH/EXCL_GRAPH, with a parent's include or exclude (print or no
1878 print) property being propagated to its children, unless they
1879 themselves explicitly appear in INCL_GRAPH or EXCL_GRAPH. A second
1880 pass, from beginning to end (children to parents) actually propagates
1881 the timings along the call graph, subject to a check against
1882 INCL_TIME/EXCL_TIME. With the print flag, fractions, and timings now
1883 stored in the symbol structures, the topological sort array is now
1884 discarded, and a new array of pointers is assembled, this time sorted
1887 Finally, print the various outputs the user requested, which is now
1888 fairly straightforward. The call graph (`cg_print.c:cg_print') and
1889 flat profile (`hist.c:hist_print') are regurgitations of values already
1890 computed. The annotated source listing
1891 (`basic_blocks.c:print_annotated_source') uses basic-block information,
1892 if present, to label each line of code with call counts, otherwise only
1893 the function call counts are presented.
1895 The function ordering code is marginally well documented in the
1896 source code itself (`cg_print.c'). Basically, the functions with the
1897 most use and the most parents are placed first, followed by other
1898 functions with the most use, followed by lower use functions, followed
1899 by unused functions at the end.
1902 File: gprof.info, Node: Debugging, Prev: Internals, Up: Details
1904 9.4 Debugging `gprof'
1905 =====================
1907 If `gprof' was compiled with debugging enabled, the `-d' option
1908 triggers debugging output (to stdout) which can be helpful in
1909 understanding its operation. The debugging number specified is
1910 interpreted as a sum of the following options:
1912 2 - Topological sort
1913 Monitor depth-first numbering of symbols during call graph analysis
1916 Shows symbols as they are identified as cycle heads
1919 As the call graph arcs are read, show each arc and how the total
1920 calls to each function are tallied
1922 32 - Call graph arc sorting
1923 Details sorting individual parents/children within each call graph
1926 64 - Reading histogram and call graph records
1927 Shows address ranges of histograms as they are read, and each call
1931 Reading, classifying, and sorting the symbol table from the object
1932 file. For line-by-line profiling (`-l' option), also shows line
1933 numbers being assigned to memory addresses.
1935 256 - Static call graph
1936 Trace operation of `-c' option
1938 512 - Symbol table and arc table lookups
1939 Detail operation of lookup routines
1941 1024 - Call graph propagation
1942 Shows how function times are propagated along the call graph
1945 Shows basic-block records as they are read from profile data (only
1946 meaningful with `-l' option)
1949 Shows symspec-to-symbol pattern matching operation
1951 8192 - Annotate source
1952 Tracks operation of `-A' option
1955 File: gprof.info, Node: GNU Free Documentation License, Prev: Details, Up: Top
1957 Appendix A GNU Free Documentation License
1958 *****************************************
1960 Version 1.3, 3 November 2008
1962 Copyright (C) 2000, 2001, 2002, 2007, 2008 Free Software Foundation, Inc.
1965 Everyone is permitted to copy and distribute verbatim copies
1966 of this license document, but changing it is not allowed.
1970 The purpose of this License is to make a manual, textbook, or other
1971 functional and useful document "free" in the sense of freedom: to
1972 assure everyone the effective freedom to copy and redistribute it,
1973 with or without modifying it, either commercially or
1974 noncommercially. Secondarily, this License preserves for the
1975 author and publisher a way to get credit for their work, while not
1976 being considered responsible for modifications made by others.
1978 This License is a kind of "copyleft", which means that derivative
1979 works of the document must themselves be free in the same sense.
1980 It complements the GNU General Public License, which is a copyleft
1981 license designed for free software.
1983 We have designed this License in order to use it for manuals for
1984 free software, because free software needs free documentation: a
1985 free program should come with manuals providing the same freedoms
1986 that the software does. But this License is not limited to
1987 software manuals; it can be used for any textual work, regardless
1988 of subject matter or whether it is published as a printed book.
1989 We recommend this License principally for works whose purpose is
1990 instruction or reference.
1992 1. APPLICABILITY AND DEFINITIONS
1994 This License applies to any manual or other work, in any medium,
1995 that contains a notice placed by the copyright holder saying it
1996 can be distributed under the terms of this License. Such a notice
1997 grants a world-wide, royalty-free license, unlimited in duration,
1998 to use that work under the conditions stated herein. The
1999 "Document", below, refers to any such manual or work. Any member
2000 of the public is a licensee, and is addressed as "you". You
2001 accept the license if you copy, modify or distribute the work in a
2002 way requiring permission under copyright law.
2004 A "Modified Version" of the Document means any work containing the
2005 Document or a portion of it, either copied verbatim, or with
2006 modifications and/or translated into another language.
2008 A "Secondary Section" is a named appendix or a front-matter section
2009 of the Document that deals exclusively with the relationship of the
2010 publishers or authors of the Document to the Document's overall
2011 subject (or to related matters) and contains nothing that could
2012 fall directly within that overall subject. (Thus, if the Document
2013 is in part a textbook of mathematics, a Secondary Section may not
2014 explain any mathematics.) The relationship could be a matter of
2015 historical connection with the subject or with related matters, or
2016 of legal, commercial, philosophical, ethical or political position
2019 The "Invariant Sections" are certain Secondary Sections whose
2020 titles are designated, as being those of Invariant Sections, in
2021 the notice that says that the Document is released under this
2022 License. If a section does not fit the above definition of
2023 Secondary then it is not allowed to be designated as Invariant.
2024 The Document may contain zero Invariant Sections. If the Document
2025 does not identify any Invariant Sections then there are none.
2027 The "Cover Texts" are certain short passages of text that are
2028 listed, as Front-Cover Texts or Back-Cover Texts, in the notice
2029 that says that the Document is released under this License. A
2030 Front-Cover Text may be at most 5 words, and a Back-Cover Text may
2031 be at most 25 words.
2033 A "Transparent" copy of the Document means a machine-readable copy,
2034 represented in a format whose specification is available to the
2035 general public, that is suitable for revising the document
2036 straightforwardly with generic text editors or (for images
2037 composed of pixels) generic paint programs or (for drawings) some
2038 widely available drawing editor, and that is suitable for input to
2039 text formatters or for automatic translation to a variety of
2040 formats suitable for input to text formatters. A copy made in an
2041 otherwise Transparent file format whose markup, or absence of
2042 markup, has been arranged to thwart or discourage subsequent
2043 modification by readers is not Transparent. An image format is
2044 not Transparent if used for any substantial amount of text. A
2045 copy that is not "Transparent" is called "Opaque".
2047 Examples of suitable formats for Transparent copies include plain
2048 ASCII without markup, Texinfo input format, LaTeX input format,
2049 SGML or XML using a publicly available DTD, and
2050 standard-conforming simple HTML, PostScript or PDF designed for
2051 human modification. Examples of transparent image formats include
2052 PNG, XCF and JPG. Opaque formats include proprietary formats that
2053 can be read and edited only by proprietary word processors, SGML or
2054 XML for which the DTD and/or processing tools are not generally
2055 available, and the machine-generated HTML, PostScript or PDF
2056 produced by some word processors for output purposes only.
2058 The "Title Page" means, for a printed book, the title page itself,
2059 plus such following pages as are needed to hold, legibly, the
2060 material this License requires to appear in the title page. For
2061 works in formats which do not have any title page as such, "Title
2062 Page" means the text near the most prominent appearance of the
2063 work's title, preceding the beginning of the body of the text.
2065 The "publisher" means any person or entity that distributes copies
2066 of the Document to the public.
2068 A section "Entitled XYZ" means a named subunit of the Document
2069 whose title either is precisely XYZ or contains XYZ in parentheses
2070 following text that translates XYZ in another language. (Here XYZ
2071 stands for a specific section name mentioned below, such as
2072 "Acknowledgements", "Dedications", "Endorsements", or "History".)
2073 To "Preserve the Title" of such a section when you modify the
2074 Document means that it remains a section "Entitled XYZ" according
2077 The Document may include Warranty Disclaimers next to the notice
2078 which states that this License applies to the Document. These
2079 Warranty Disclaimers are considered to be included by reference in
2080 this License, but only as regards disclaiming warranties: any other
2081 implication that these Warranty Disclaimers may have is void and
2082 has no effect on the meaning of this License.
2086 You may copy and distribute the Document in any medium, either
2087 commercially or noncommercially, provided that this License, the
2088 copyright notices, and the license notice saying this License
2089 applies to the Document are reproduced in all copies, and that you
2090 add no other conditions whatsoever to those of this License. You
2091 may not use technical measures to obstruct or control the reading
2092 or further copying of the copies you make or distribute. However,
2093 you may accept compensation in exchange for copies. If you
2094 distribute a large enough number of copies you must also follow
2095 the conditions in section 3.
2097 You may also lend copies, under the same conditions stated above,
2098 and you may publicly display copies.
2100 3. COPYING IN QUANTITY
2102 If you publish printed copies (or copies in media that commonly
2103 have printed covers) of the Document, numbering more than 100, and
2104 the Document's license notice requires Cover Texts, you must
2105 enclose the copies in covers that carry, clearly and legibly, all
2106 these Cover Texts: Front-Cover Texts on the front cover, and
2107 Back-Cover Texts on the back cover. Both covers must also clearly
2108 and legibly identify you as the publisher of these copies. The
2109 front cover must present the full title with all words of the
2110 title equally prominent and visible. You may add other material
2111 on the covers in addition. Copying with changes limited to the
2112 covers, as long as they preserve the title of the Document and
2113 satisfy these conditions, can be treated as verbatim copying in
2116 If the required texts for either cover are too voluminous to fit
2117 legibly, you should put the first ones listed (as many as fit
2118 reasonably) on the actual cover, and continue the rest onto
2121 If you publish or distribute Opaque copies of the Document
2122 numbering more than 100, you must either include a
2123 machine-readable Transparent copy along with each Opaque copy, or
2124 state in or with each Opaque copy a computer-network location from
2125 which the general network-using public has access to download
2126 using public-standard network protocols a complete Transparent
2127 copy of the Document, free of added material. If you use the
2128 latter option, you must take reasonably prudent steps, when you
2129 begin distribution of Opaque copies in quantity, to ensure that
2130 this Transparent copy will remain thus accessible at the stated
2131 location until at least one year after the last time you
2132 distribute an Opaque copy (directly or through your agents or
2133 retailers) of that edition to the public.
2135 It is requested, but not required, that you contact the authors of
2136 the Document well before redistributing any large number of
2137 copies, to give them a chance to provide you with an updated
2138 version of the Document.
2142 You may copy and distribute a Modified Version of the Document
2143 under the conditions of sections 2 and 3 above, provided that you
2144 release the Modified Version under precisely this License, with
2145 the Modified Version filling the role of the Document, thus
2146 licensing distribution and modification of the Modified Version to
2147 whoever possesses a copy of it. In addition, you must do these
2148 things in the Modified Version:
2150 A. Use in the Title Page (and on the covers, if any) a title
2151 distinct from that of the Document, and from those of
2152 previous versions (which should, if there were any, be listed
2153 in the History section of the Document). You may use the
2154 same title as a previous version if the original publisher of
2155 that version gives permission.
2157 B. List on the Title Page, as authors, one or more persons or
2158 entities responsible for authorship of the modifications in
2159 the Modified Version, together with at least five of the
2160 principal authors of the Document (all of its principal
2161 authors, if it has fewer than five), unless they release you
2162 from this requirement.
2164 C. State on the Title page the name of the publisher of the
2165 Modified Version, as the publisher.
2167 D. Preserve all the copyright notices of the Document.
2169 E. Add an appropriate copyright notice for your modifications
2170 adjacent to the other copyright notices.
2172 F. Include, immediately after the copyright notices, a license
2173 notice giving the public permission to use the Modified
2174 Version under the terms of this License, in the form shown in
2177 G. Preserve in that license notice the full lists of Invariant
2178 Sections and required Cover Texts given in the Document's
2181 H. Include an unaltered copy of this License.
2183 I. Preserve the section Entitled "History", Preserve its Title,
2184 and add to it an item stating at least the title, year, new
2185 authors, and publisher of the Modified Version as given on
2186 the Title Page. If there is no section Entitled "History" in
2187 the Document, create one stating the title, year, authors,
2188 and publisher of the Document as given on its Title Page,
2189 then add an item describing the Modified Version as stated in
2190 the previous sentence.
2192 J. Preserve the network location, if any, given in the Document
2193 for public access to a Transparent copy of the Document, and
2194 likewise the network locations given in the Document for
2195 previous versions it was based on. These may be placed in
2196 the "History" section. You may omit a network location for a
2197 work that was published at least four years before the
2198 Document itself, or if the original publisher of the version
2199 it refers to gives permission.
2201 K. For any section Entitled "Acknowledgements" or "Dedications",
2202 Preserve the Title of the section, and preserve in the
2203 section all the substance and tone of each of the contributor
2204 acknowledgements and/or dedications given therein.
2206 L. Preserve all the Invariant Sections of the Document,
2207 unaltered in their text and in their titles. Section numbers
2208 or the equivalent are not considered part of the section
2211 M. Delete any section Entitled "Endorsements". Such a section
2212 may not be included in the Modified Version.
2214 N. Do not retitle any existing section to be Entitled
2215 "Endorsements" or to conflict in title with any Invariant
2218 O. Preserve any Warranty Disclaimers.
2220 If the Modified Version includes new front-matter sections or
2221 appendices that qualify as Secondary Sections and contain no
2222 material copied from the Document, you may at your option
2223 designate some or all of these sections as invariant. To do this,
2224 add their titles to the list of Invariant Sections in the Modified
2225 Version's license notice. These titles must be distinct from any
2226 other section titles.
2228 You may add a section Entitled "Endorsements", provided it contains
2229 nothing but endorsements of your Modified Version by various
2230 parties--for example, statements of peer review or that the text
2231 has been approved by an organization as the authoritative
2232 definition of a standard.
2234 You may add a passage of up to five words as a Front-Cover Text,
2235 and a passage of up to 25 words as a Back-Cover Text, to the end
2236 of the list of Cover Texts in the Modified Version. Only one
2237 passage of Front-Cover Text and one of Back-Cover Text may be
2238 added by (or through arrangements made by) any one entity. If the
2239 Document already includes a cover text for the same cover,
2240 previously added by you or by arrangement made by the same entity
2241 you are acting on behalf of, you may not add another; but you may
2242 replace the old one, on explicit permission from the previous
2243 publisher that added the old one.
2245 The author(s) and publisher(s) of the Document do not by this
2246 License give permission to use their names for publicity for or to
2247 assert or imply endorsement of any Modified Version.
2249 5. COMBINING DOCUMENTS
2251 You may combine the Document with other documents released under
2252 this License, under the terms defined in section 4 above for
2253 modified versions, provided that you include in the combination
2254 all of the Invariant Sections of all of the original documents,
2255 unmodified, and list them all as Invariant Sections of your
2256 combined work in its license notice, and that you preserve all
2257 their Warranty Disclaimers.
2259 The combined work need only contain one copy of this License, and
2260 multiple identical Invariant Sections may be replaced with a single
2261 copy. If there are multiple Invariant Sections with the same name
2262 but different contents, make the title of each such section unique
2263 by adding at the end of it, in parentheses, the name of the
2264 original author or publisher of that section if known, or else a
2265 unique number. Make the same adjustment to the section titles in
2266 the list of Invariant Sections in the license notice of the
2269 In the combination, you must combine any sections Entitled
2270 "History" in the various original documents, forming one section
2271 Entitled "History"; likewise combine any sections Entitled
2272 "Acknowledgements", and any sections Entitled "Dedications". You
2273 must delete all sections Entitled "Endorsements."
2275 6. COLLECTIONS OF DOCUMENTS
2277 You may make a collection consisting of the Document and other
2278 documents released under this License, and replace the individual
2279 copies of this License in the various documents with a single copy
2280 that is included in the collection, provided that you follow the
2281 rules of this License for verbatim copying of each of the
2282 documents in all other respects.
2284 You may extract a single document from such a collection, and
2285 distribute it individually under this License, provided you insert
2286 a copy of this License into the extracted document, and follow
2287 this License in all other respects regarding verbatim copying of
2290 7. AGGREGATION WITH INDEPENDENT WORKS
2292 A compilation of the Document or its derivatives with other
2293 separate and independent documents or works, in or on a volume of
2294 a storage or distribution medium, is called an "aggregate" if the
2295 copyright resulting from the compilation is not used to limit the
2296 legal rights of the compilation's users beyond what the individual
2297 works permit. When the Document is included in an aggregate, this
2298 License does not apply to the other works in the aggregate which
2299 are not themselves derivative works of the Document.
2301 If the Cover Text requirement of section 3 is applicable to these
2302 copies of the Document, then if the Document is less than one half
2303 of the entire aggregate, the Document's Cover Texts may be placed
2304 on covers that bracket the Document within the aggregate, or the
2305 electronic equivalent of covers if the Document is in electronic
2306 form. Otherwise they must appear on printed covers that bracket
2307 the whole aggregate.
2311 Translation is considered a kind of modification, so you may
2312 distribute translations of the Document under the terms of section
2313 4. Replacing Invariant Sections with translations requires special
2314 permission from their copyright holders, but you may include
2315 translations of some or all Invariant Sections in addition to the
2316 original versions of these Invariant Sections. You may include a
2317 translation of this License, and all the license notices in the
2318 Document, and any Warranty Disclaimers, provided that you also
2319 include the original English version of this License and the
2320 original versions of those notices and disclaimers. In case of a
2321 disagreement between the translation and the original version of
2322 this License or a notice or disclaimer, the original version will
2325 If a section in the Document is Entitled "Acknowledgements",
2326 "Dedications", or "History", the requirement (section 4) to
2327 Preserve its Title (section 1) will typically require changing the
2332 You may not copy, modify, sublicense, or distribute the Document
2333 except as expressly provided under this License. Any attempt
2334 otherwise to copy, modify, sublicense, or distribute it is void,
2335 and will automatically terminate your rights under this License.
2337 However, if you cease all violation of this License, then your
2338 license from a particular copyright holder is reinstated (a)
2339 provisionally, unless and until the copyright holder explicitly
2340 and finally terminates your license, and (b) permanently, if the
2341 copyright holder fails to notify you of the violation by some
2342 reasonable means prior to 60 days after the cessation.
2344 Moreover, your license from a particular copyright holder is
2345 reinstated permanently if the copyright holder notifies you of the
2346 violation by some reasonable means, this is the first time you have
2347 received notice of violation of this License (for any work) from
2348 that copyright holder, and you cure the violation prior to 30 days
2349 after your receipt of the notice.
2351 Termination of your rights under this section does not terminate
2352 the licenses of parties who have received copies or rights from
2353 you under this License. If your rights have been terminated and
2354 not permanently reinstated, receipt of a copy of some or all of
2355 the same material does not give you any rights to use it.
2357 10. FUTURE REVISIONS OF THIS LICENSE
2359 The Free Software Foundation may publish new, revised versions of
2360 the GNU Free Documentation License from time to time. Such new
2361 versions will be similar in spirit to the present version, but may
2362 differ in detail to address new problems or concerns. See
2363 `http://www.gnu.org/copyleft/'.
2365 Each version of the License is given a distinguishing version
2366 number. If the Document specifies that a particular numbered
2367 version of this License "or any later version" applies to it, you
2368 have the option of following the terms and conditions either of
2369 that specified version or of any later version that has been
2370 published (not as a draft) by the Free Software Foundation. If
2371 the Document does not specify a version number of this License,
2372 you may choose any version ever published (not as a draft) by the
2373 Free Software Foundation. If the Document specifies that a proxy
2374 can decide which future versions of this License can be used, that
2375 proxy's public statement of acceptance of a version permanently
2376 authorizes you to choose that version for the Document.
2380 "Massive Multiauthor Collaboration Site" (or "MMC Site") means any
2381 World Wide Web server that publishes copyrightable works and also
2382 provides prominent facilities for anybody to edit those works. A
2383 public wiki that anybody can edit is an example of such a server.
2384 A "Massive Multiauthor Collaboration" (or "MMC") contained in the
2385 site means any set of copyrightable works thus published on the MMC
2388 "CC-BY-SA" means the Creative Commons Attribution-Share Alike 3.0
2389 license published by Creative Commons Corporation, a not-for-profit
2390 corporation with a principal place of business in San Francisco,
2391 California, as well as future copyleft versions of that license
2392 published by that same organization.
2394 "Incorporate" means to publish or republish a Document, in whole or
2395 in part, as part of another Document.
2397 An MMC is "eligible for relicensing" if it is licensed under this
2398 License, and if all works that were first published under this
2399 License somewhere other than this MMC, and subsequently
2400 incorporated in whole or in part into the MMC, (1) had no cover
2401 texts or invariant sections, and (2) were thus incorporated prior
2402 to November 1, 2008.
2404 The operator of an MMC Site may republish an MMC contained in the
2405 site under CC-BY-SA on the same site at any time before August 1,
2406 2009, provided the MMC is eligible for relicensing.
2409 ADDENDUM: How to use this License for your documents
2410 ====================================================
2412 To use this License in a document you have written, include a copy of
2413 the License in the document and put the following copyright and license
2414 notices just after the title page:
2416 Copyright (C) YEAR YOUR NAME.
2417 Permission is granted to copy, distribute and/or modify this document
2418 under the terms of the GNU Free Documentation License, Version 1.3
2419 or any later version published by the Free Software Foundation;
2420 with no Invariant Sections, no Front-Cover Texts, and no Back-Cover
2421 Texts. A copy of the license is included in the section entitled ``GNU
2422 Free Documentation License''.
2424 If you have Invariant Sections, Front-Cover Texts and Back-Cover
2425 Texts, replace the "with...Texts." line with this:
2427 with the Invariant Sections being LIST THEIR TITLES, with
2428 the Front-Cover Texts being LIST, and with the Back-Cover Texts
2431 If you have Invariant Sections without Cover Texts, or some other
2432 combination of the three, merge those two alternatives to suit the
2435 If your document contains nontrivial examples of program code, we
2436 recommend releasing these examples in parallel under your choice of
2437 free software license, such as the GNU General Public License, to
2438 permit their use in free software.
2444 Node: Introduction
\7f2103
2445 Node: Compiling
\7f4595
2446 Node: Executing
\7f8651
2447 Node: Invoking
\7f11439
2448 Node: Output Options
\7f12854
2449 Node: Analysis Options
\7f19943
2450 Node: Miscellaneous Options
\7f23641
2451 Node: Deprecated Options
\7f24896
2452 Node: Symspecs
\7f26965
2453 Node: Output
\7f28791
2454 Node: Flat Profile
\7f29831
2455 Node: Call Graph
\7f34784
2456 Node: Primary
\7f38016
2457 Node: Callers
\7f40604
2458 Node: Subroutines
\7f42721
2459 Node: Cycles
\7f44562
2460 Node: Line-by-line
\7f51339
2461 Node: Annotated Source
\7f55412
2462 Node: Inaccuracy
\7f58411
2463 Node: Sampling Error
\7f58669
2464 Node: Assumptions
\7f61573
2465 Node: How do I?
\7f63043
2466 Node: Incompatibilities
\7f64597
2467 Node: Details
\7f66091
2468 Node: Implementation
\7f66484
2469 Node: File Format
\7f72381
2470 Node: Internals
\7f76671
2471 Node: Debugging
\7f85166
2472 Node: GNU Free Documentation License
\7f86767