Commit | Line | Data |
---|---|---|
252b5132 RH |
1 | \input texinfo @c -*-texinfo-*- |
2 | @setfilename gprof.info | |
3 | @settitle GNU gprof | |
4 | @setchapternewpage odd | |
5 | ||
6 | @ifinfo | |
7 | @c This is a dir.info fragment to support semi-automated addition of | |
8 | @c manuals to an info tree. zoo@cygnus.com is developing this facility. | |
9 | @format | |
10 | START-INFO-DIR-ENTRY | |
11 | * gprof: (gprof). Profiling your program's execution | |
12 | END-INFO-DIR-ENTRY | |
13 | @end format | |
14 | @end ifinfo | |
15 | ||
16 | @ifinfo | |
17 | This file documents the gprof profiler of the GNU system. | |
18 | ||
5af11cab | 19 | Copyright (C) 1988, 92, 97, 98, 99, 2000 Free Software Foundation, Inc. |
252b5132 | 20 | |
cf055d54 NC |
21 | Permission is granted to copy, distribute and/or modify this document |
22 | under the terms of the GNU Free Documentation License, Version 1.1 | |
23 | or any later version published by the Free Software Foundation; | |
24 | with no Invariant Sections, with no Front-Cover Texts, and with no | |
25 | Back-Cover Texts. A copy of the license is included in the | |
26 | section entitled "GNU Free Documentation License". | |
252b5132 RH |
27 | |
28 | @ignore | |
29 | Permission is granted to process this file through Tex and print the | |
30 | results, provided the printed document carries copying permission | |
31 | notice identical to this one except for the removal of this paragraph | |
32 | (this paragraph not being relevant to the printed manual). | |
33 | ||
34 | @end ignore | |
252b5132 RH |
35 | @end ifinfo |
36 | ||
37 | @finalout | |
38 | @smallbook | |
39 | ||
40 | @titlepage | |
41 | @title GNU gprof | |
42 | @subtitle The @sc{gnu} Profiler | |
43 | @author Jay Fenlason and Richard Stallman | |
44 | ||
45 | @page | |
46 | ||
47 | This manual describes the @sc{gnu} profiler, @code{gprof}, and how you | |
48 | can use it to determine which parts of a program are taking most of the | |
49 | execution time. We assume that you know how to write, compile, and | |
50 | execute programs. @sc{gnu} @code{gprof} was written by Jay Fenlason. | |
51 | ||
252b5132 | 52 | @vskip 0pt plus 1filll |
5af11cab | 53 | Copyright @copyright{} 1988, 92, 97, 98, 99, 2000 Free Software Foundation, Inc. |
252b5132 | 54 | |
cf055d54 NC |
55 | Permission is granted to copy, distribute and/or modify this document |
56 | under the terms of the GNU Free Documentation License, Version 1.1 | |
57 | or any later version published by the Free Software Foundation; | |
58 | with no Invariant Sections, with no Front-Cover Texts, and with no | |
59 | Back-Cover Texts. A copy of the license is included in the | |
60 | section entitled "GNU Free Documentation License". | |
252b5132 RH |
61 | |
62 | @end titlepage | |
63 | ||
64 | @ifinfo | |
65 | @node Top | |
66 | @top Profiling a Program: Where Does It Spend Its Time? | |
67 | ||
68 | This manual describes the @sc{gnu} profiler, @code{gprof}, and how you | |
69 | can use it to determine which parts of a program are taking most of the | |
70 | execution time. We assume that you know how to write, compile, and | |
71 | execute programs. @sc{gnu} @code{gprof} was written by Jay Fenlason. | |
72 | ||
cf055d54 NC |
73 | This document is distributed under the terms of the GNU Free |
74 | Documentation License. A copy of the license is included in the | |
75 | section entitled "GNU Free Documentation License". | |
76 | ||
252b5132 RH |
77 | @menu |
78 | * Introduction:: What profiling means, and why it is useful. | |
79 | ||
80 | * Compiling:: How to compile your program for profiling. | |
81 | * Executing:: Executing your program to generate profile data | |
82 | * Invoking:: How to run @code{gprof}, and its options | |
83 | ||
84 | * Output:: Interpreting @code{gprof}'s output | |
85 | ||
86 | * Inaccuracy:: Potential problems you should be aware of | |
87 | * How do I?:: Answers to common questions | |
88 | * Incompatibilities:: (between @sc{gnu} @code{gprof} and Unix @code{gprof}.) | |
89 | * Details:: Details of how profiling is done | |
cf055d54 | 90 | * GNU Free Documentation License:: GNU Free Documentation License |
252b5132 RH |
91 | @end menu |
92 | @end ifinfo | |
93 | ||
94 | @node Introduction | |
95 | @chapter Introduction to Profiling | |
96 | ||
97 | Profiling allows you to learn where your program spent its time and which | |
98 | functions called which other functions while it was executing. This | |
99 | information can show you which pieces of your program are slower than you | |
100 | expected, and might be candidates for rewriting to make your program | |
101 | execute faster. It can also tell you which functions are being called more | |
102 | or less often than you expected. This may help you spot bugs that had | |
103 | otherwise been unnoticed. | |
104 | ||
105 | Since the profiler uses information collected during the actual execution | |
106 | of your program, it can be used on programs that are too large or too | |
107 | complex to analyze by reading the source. However, how your program is run | |
108 | will affect the information that shows up in the profile data. If you | |
109 | don't use some feature of your program while it is being profiled, no | |
110 | profile information will be generated for that feature. | |
111 | ||
112 | Profiling has several steps: | |
113 | ||
114 | @itemize @bullet | |
115 | @item | |
116 | You must compile and link your program with profiling enabled. | |
117 | @xref{Compiling}. | |
118 | ||
119 | @item | |
120 | You must execute your program to generate a profile data file. | |
121 | @xref{Executing}. | |
122 | ||
123 | @item | |
124 | You must run @code{gprof} to analyze the profile data. | |
125 | @xref{Invoking}. | |
126 | @end itemize | |
127 | ||
128 | The next three chapters explain these steps in greater detail. | |
129 | ||
130 | Several forms of output are available from the analysis. | |
131 | ||
132 | The @dfn{flat profile} shows how much time your program spent in each function, | |
133 | and how many times that function was called. If you simply want to know | |
134 | which functions burn most of the cycles, it is stated concisely here. | |
135 | @xref{Flat Profile}. | |
136 | ||
137 | The @dfn{call graph} shows, for each function, which functions called it, which | |
138 | other functions it called, and how many times. There is also an estimate | |
139 | of how much time was spent in the subroutines of each function. This can | |
140 | suggest places where you might try to eliminate function calls that use a | |
141 | lot of time. @xref{Call Graph}. | |
142 | ||
143 | The @dfn{annotated source} listing is a copy of the program's | |
144 | source code, labeled with the number of times each line of the | |
145 | program was executed. @xref{Annotated Source}. | |
146 | ||
147 | To better understand how profiling works, you may wish to read | |
148 | a description of its implementation. | |
149 | @xref{Implementation}. | |
150 | ||
151 | @node Compiling | |
152 | @chapter Compiling a Program for Profiling | |
153 | ||
154 | The first step in generating profile information for your program is | |
155 | to compile and link it with profiling enabled. | |
156 | ||
157 | To compile a source file for profiling, specify the @samp{-pg} option when | |
158 | you run the compiler. (This is in addition to the options you normally | |
159 | use.) | |
160 | ||
161 | To link the program for profiling, if you use a compiler such as @code{cc} | |
162 | to do the linking, simply specify @samp{-pg} in addition to your usual | |
163 | options. The same option, @samp{-pg}, alters either compilation or linking | |
164 | to do what is necessary for profiling. Here are examples: | |
165 | ||
166 | @example | |
167 | cc -g -c myprog.c utils.c -pg | |
168 | cc -o myprog myprog.o utils.o -pg | |
169 | @end example | |
170 | ||
171 | The @samp{-pg} option also works with a command that both compiles and links: | |
172 | ||
173 | @example | |
174 | cc -o myprog myprog.c utils.c -g -pg | |
175 | @end example | |
176 | ||
177 | If you run the linker @code{ld} directly instead of through a compiler | |
178 | such as @code{cc}, you may have to specify a profiling startup file | |
179 | @file{gcrt0.o} as the first input file instead of the usual startup | |
180 | file @file{crt0.o}. In addition, you would probably want to | |
181 | specify the profiling C library, @file{libc_p.a}, by writing | |
182 | @samp{-lc_p} instead of the usual @samp{-lc}. This is not absolutely | |
183 | necessary, but doing this gives you number-of-calls information for | |
184 | standard library functions such as @code{read} and @code{open}. For | |
185 | example: | |
186 | ||
187 | @example | |
188 | ld -o myprog /lib/gcrt0.o myprog.o utils.o -lc_p | |
189 | @end example | |
190 | ||
191 | If you compile only some of the modules of the program with @samp{-pg}, you | |
192 | can still profile the program, but you won't get complete information about | |
193 | the modules that were compiled without @samp{-pg}. The only information | |
194 | you get for the functions in those modules is the total time spent in them; | |
195 | there is no record of how many times they were called, or from where. This | |
196 | will not affect the flat profile (except that the @code{calls} field for | |
197 | the functions will be blank), but will greatly reduce the usefulness of the | |
198 | call graph. | |
199 | ||
200 | If you wish to perform line-by-line profiling, | |
201 | you will also need to specify the @samp{-g} option, | |
202 | instructing the compiler to insert debugging symbols into the program | |
203 | that match program addresses to source code lines. | |
204 | @xref{Line-by-line}. | |
205 | ||
206 | In addition to the @samp{-pg} and @samp{-g} options, | |
207 | you may also wish to specify the @samp{-a} option when compiling. | |
208 | This will instrument | |
209 | the program to perform basic-block counting. As the program runs, | |
210 | it will count how many times it executed each branch of each @samp{if} | |
211 | statement, each iteration of each @samp{do} loop, etc. This will | |
212 | enable @code{gprof} to construct an annotated source code | |
213 | listing showing how many times each line of code was executed. | |
214 | ||
215 | @node Executing | |
216 | @chapter Executing the Program | |
217 | ||
218 | Once the program is compiled for profiling, you must run it in order to | |
219 | generate the information that @code{gprof} needs. Simply run the program | |
220 | as usual, using the normal arguments, file names, etc. The program should | |
221 | run normally, producing the same output as usual. It will, however, run | |
222 | somewhat slower than normal because of the time spent collecting and the | |
223 | writing the profile data. | |
224 | ||
225 | The way you run the program---the arguments and input that you give | |
226 | it---may have a dramatic effect on what the profile information shows. The | |
227 | profile data will describe the parts of the program that were activated for | |
228 | the particular input you use. For example, if the first command you give | |
229 | to your program is to quit, the profile data will show the time used in | |
230 | initialization and in cleanup, but not much else. | |
231 | ||
232 | Your program will write the profile data into a file called @file{gmon.out} | |
233 | just before exiting. If there is already a file called @file{gmon.out}, | |
234 | its contents are overwritten. There is currently no way to tell the | |
235 | program to write the profile data under a different name, but you can rename | |
236 | the file afterward if you are concerned that it may be overwritten. | |
237 | ||
238 | In order to write the @file{gmon.out} file properly, your program must exit | |
239 | normally: by returning from @code{main} or by calling @code{exit}. Calling | |
240 | the low-level function @code{_exit} does not write the profile data, and | |
241 | neither does abnormal termination due to an unhandled signal. | |
242 | ||
243 | The @file{gmon.out} file is written in the program's @emph{current working | |
244 | directory} at the time it exits. This means that if your program calls | |
245 | @code{chdir}, the @file{gmon.out} file will be left in the last directory | |
246 | your program @code{chdir}'d to. If you don't have permission to write in | |
247 | this directory, the file is not written, and you will get an error message. | |
248 | ||
249 | Older versions of the @sc{gnu} profiling library may also write a file | |
250 | called @file{bb.out}. This file, if present, contains an human-readable | |
251 | listing of the basic-block execution counts. Unfortunately, the | |
252 | appearance of a human-readable @file{bb.out} means the basic-block | |
253 | counts didn't get written into @file{gmon.out}. | |
254 | The Perl script @code{bbconv.pl}, included with the @code{gprof} | |
255 | source distribution, will convert a @file{bb.out} file into | |
256 | a format readable by @code{gprof}. | |
257 | ||
258 | @node Invoking | |
259 | @chapter @code{gprof} Command Summary | |
260 | ||
261 | After you have a profile data file @file{gmon.out}, you can run @code{gprof} | |
262 | to interpret the information in it. The @code{gprof} program prints a | |
263 | flat profile and a call graph on standard output. Typically you would | |
264 | redirect the output of @code{gprof} into a file with @samp{>}. | |
265 | ||
266 | You run @code{gprof} like this: | |
267 | ||
268 | @smallexample | |
269 | gprof @var{options} [@var{executable-file} [@var{profile-data-files}@dots{}]] [> @var{outfile}] | |
270 | @end smallexample | |
271 | ||
272 | @noindent | |
273 | Here square-brackets indicate optional arguments. | |
274 | ||
275 | If you omit the executable file name, the file @file{a.out} is used. If | |
276 | you give no profile data file name, the file @file{gmon.out} is used. If | |
277 | any file is not in the proper format, or if the profile data file does not | |
278 | appear to belong to the executable file, an error message is printed. | |
279 | ||
280 | You can give more than one profile data file by entering all their names | |
281 | after the executable file name; then the statistics in all the data files | |
282 | are summed together. | |
283 | ||
284 | The order of these options does not matter. | |
285 | ||
286 | @menu | |
287 | * Output Options:: Controlling @code{gprof}'s output style | |
288 | * Analysis Options:: Controlling how @code{gprof} analyses its data | |
289 | * Miscellaneous Options:: | |
5af11cab | 290 | * Deprecated Options:: Options you no longer need to use, but which |
252b5132 RH |
291 | have been retained for compatibility |
292 | * Symspecs:: Specifying functions to include or exclude | |
293 | @end menu | |
294 | ||
295 | @node Output Options,Analysis Options,,Invoking | |
296 | @section Output Options | |
297 | ||
298 | These options specify which of several output formats | |
299 | @code{gprof} should produce. | |
300 | ||
301 | Many of these options take an optional @dfn{symspec} to specify | |
302 | functions to be included or excluded. These options can be | |
303 | specified multiple times, with different symspecs, to include | |
304 | or exclude sets of symbols. @xref{Symspecs}. | |
305 | ||
306 | Specifying any of these options overrides the default (@samp{-p -q}), | |
307 | which prints a flat profile and call graph analysis | |
308 | for all functions. | |
309 | ||
310 | @table @code | |
311 | ||
312 | @item -A[@var{symspec}] | |
313 | @itemx --annotated-source[=@var{symspec}] | |
314 | The @samp{-A} option causes @code{gprof} to print annotated source code. | |
315 | If @var{symspec} is specified, print output only for matching symbols. | |
316 | @xref{Annotated Source}. | |
317 | ||
318 | @item -b | |
319 | @itemx --brief | |
320 | If the @samp{-b} option is given, @code{gprof} doesn't print the | |
321 | verbose blurbs that try to explain the meaning of all of the fields in | |
322 | the tables. This is useful if you intend to print out the output, or | |
323 | are tired of seeing the blurbs. | |
324 | ||
325 | @item -C[@var{symspec}] | |
326 | @itemx --exec-counts[=@var{symspec}] | |
327 | The @samp{-C} option causes @code{gprof} to | |
328 | print a tally of functions and the number of times each was called. | |
329 | If @var{symspec} is specified, print tally only for matching symbols. | |
330 | ||
5af11cab | 331 | If the profile data file contains basic-block count records, specifying |
252b5132 RH |
332 | the @samp{-l} option, along with @samp{-C}, will cause basic-block |
333 | execution counts to be tallied and displayed. | |
334 | ||
335 | @item -i | |
336 | @itemx --file-info | |
337 | The @samp{-i} option causes @code{gprof} to display summary information | |
338 | about the profile data file(s) and then exit. The number of histogram, | |
339 | call graph, and basic-block count records is displayed. | |
340 | ||
341 | @item -I @var{dirs} | |
342 | @itemx --directory-path=@var{dirs} | |
343 | The @samp{-I} option specifies a list of search directories in | |
344 | which to find source files. Environment variable @var{GPROF_PATH} | |
5af11cab | 345 | can also be used to convey this information. |
252b5132 RH |
346 | Used mostly for annotated source output. |
347 | ||
348 | @item -J[@var{symspec}] | |
349 | @itemx --no-annotated-source[=@var{symspec}] | |
350 | The @samp{-J} option causes @code{gprof} not to | |
351 | print annotated source code. | |
352 | If @var{symspec} is specified, @code{gprof} prints annotated source, | |
353 | but excludes matching symbols. | |
354 | ||
355 | @item -L | |
356 | @itemx --print-path | |
357 | Normally, source filenames are printed with the path | |
358 | component suppressed. The @samp{-L} option causes @code{gprof} | |
359 | to print the full pathname of | |
360 | source filenames, which is determined | |
361 | from symbolic debugging information in the image file | |
362 | and is relative to the directory in which the compiler | |
363 | was invoked. | |
364 | ||
365 | @item -p[@var{symspec}] | |
366 | @itemx --flat-profile[=@var{symspec}] | |
367 | The @samp{-p} option causes @code{gprof} to print a flat profile. | |
368 | If @var{symspec} is specified, print flat profile only for matching symbols. | |
369 | @xref{Flat Profile}. | |
370 | ||
371 | @item -P[@var{symspec}] | |
372 | @itemx --no-flat-profile[=@var{symspec}] | |
373 | The @samp{-P} option causes @code{gprof} to suppress printing a flat profile. | |
374 | If @var{symspec} is specified, @code{gprof} prints a flat profile, | |
375 | but excludes matching symbols. | |
376 | ||
377 | @item -q[@var{symspec}] | |
378 | @itemx --graph[=@var{symspec}] | |
379 | The @samp{-q} option causes @code{gprof} to print the call graph analysis. | |
380 | If @var{symspec} is specified, print call graph only for matching symbols | |
381 | and their children. | |
382 | @xref{Call Graph}. | |
383 | ||
384 | @item -Q[@var{symspec}] | |
385 | @itemx --no-graph[=@var{symspec}] | |
386 | The @samp{-Q} option causes @code{gprof} to suppress printing the | |
387 | call graph. | |
388 | If @var{symspec} is specified, @code{gprof} prints a call graph, | |
389 | but excludes matching symbols. | |
390 | ||
391 | @item -y | |
392 | @itemx --separate-files | |
393 | This option affects annotated source output only. | |
5af11cab | 394 | Normally, @code{gprof} prints annotated source files |
252b5132 | 395 | to standard-output. If this option is specified, |
5af11cab AM |
396 | annotated source for a file named @file{path/@var{filename}} |
397 | is generated in the file @file{@var{filename}-ann}. If the underlying | |
398 | filesystem would truncate @file{@var{filename}-ann} so that it | |
399 | overwrites the original @file{@var{filename}}, @code{gprof} generates | |
400 | annotated source in the file @file{@var{filename}.ann} instead (if the | |
401 | original file name has an extension, that extension is @emph{replaced} | |
402 | with @file{.ann}). | |
252b5132 RH |
403 | |
404 | @item -Z[@var{symspec}] | |
405 | @itemx --no-exec-counts[=@var{symspec}] | |
406 | The @samp{-Z} option causes @code{gprof} not to | |
407 | print a tally of functions and the number of times each was called. | |
408 | If @var{symspec} is specified, print tally, but exclude matching symbols. | |
409 | ||
410 | @item --function-ordering | |
411 | The @samp{--function-ordering} option causes @code{gprof} to print a | |
412 | suggested function ordering for the program based on profiling data. | |
413 | This option suggests an ordering which may improve paging, tlb and | |
414 | cache behavior for the program on systems which support arbitrary | |
415 | ordering of functions in an executable. | |
416 | ||
417 | The exact details of how to force the linker to place functions | |
418 | in a particular order is system dependent and out of the scope of this | |
419 | manual. | |
420 | ||
421 | @item --file-ordering @var{map_file} | |
422 | The @samp{--file-ordering} option causes @code{gprof} to print a | |
423 | suggested .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 arbitrary | |
426 | ordering of functions in an executable. | |
427 | ||
428 | Use of the @samp{-a} argument is highly recommended with this option. | |
429 | ||
430 | The @var{map_file} argument is a pathname to a file which provides | |
431 | function name to object file mappings. The format of the file is similar to | |
432 | the output of the program @code{nm}. | |
433 | ||
434 | @smallexample | |
435 | @group | |
436 | c-parse.o:00000000 T yyparse | |
437 | c-parse.o:00000004 C yyerrflag | |
438 | c-lang.o:00000000 T maybe_objc_method_name | |
439 | c-lang.o:00000000 T print_lang_statistics | |
440 | c-lang.o:00000000 T recognize_objc_keyword | |
441 | c-decl.o:00000000 T print_lang_identifier | |
442 | c-decl.o:00000000 T print_lang_type | |
443 | @dots{} | |
444 | ||
445 | @end group | |
446 | @end smallexample | |
447 | ||
5af11cab AM |
448 | To create a @var{map_file} with @sc{gnu} @code{nm}, type a command like |
449 | @kbd{nm --extern-only --defined-only -v --print-file-name program-name}. | |
252b5132 RH |
450 | |
451 | @item -T | |
452 | @itemx --traditional | |
453 | The @samp{-T} option causes @code{gprof} to print its output in | |
454 | ``traditional'' BSD style. | |
455 | ||
456 | @item -w @var{width} | |
457 | @itemx --width=@var{width} | |
458 | Sets width of output lines to @var{width}. | |
459 | Currently only used when printing the function index at the bottom | |
460 | of the call graph. | |
461 | ||
462 | @item -x | |
463 | @itemx --all-lines | |
464 | This option affects annotated source output only. | |
465 | By default, only the lines at the beginning of a basic-block | |
466 | are annotated. If this option is specified, every line in | |
467 | a basic-block is annotated by repeating the annotation for the | |
468 | first line. This behavior is similar to @code{tcov}'s @samp{-a}. | |
469 | ||
28c309a2 | 470 | @item --demangle[=@var{style}] |
252b5132 RH |
471 | @itemx --no-demangle |
472 | These options control whether C++ symbol names should be demangled when | |
473 | printing output. The default is to demangle symbols. The | |
28c309a2 NC |
474 | @code{--no-demangle} option may be used to turn off demangling. Different |
475 | compilers have different mangling styles. The optional demangling style | |
476 | argument can be used to choose an appropriate demangling style for your | |
477 | compiler. | |
252b5132 RH |
478 | @end table |
479 | ||
480 | @node Analysis Options,Miscellaneous Options,Output Options,Invoking | |
481 | @section Analysis Options | |
482 | ||
483 | @table @code | |
484 | ||
485 | @item -a | |
486 | @itemx --no-static | |
487 | The @samp{-a} option causes @code{gprof} to suppress the printing of | |
488 | statically declared (private) functions. (These are functions whose | |
489 | names are not listed as global, and which are not visible outside the | |
490 | file/function/block where they were defined.) Time spent in these | |
491 | functions, calls to/from them, etc, will all be attributed to the | |
492 | function that was loaded directly before it in the executable file. | |
493 | @c This is compatible with Unix @code{gprof}, but a bad idea. | |
494 | This option affects both the flat profile and the call graph. | |
495 | ||
496 | @item -c | |
497 | @itemx --static-call-graph | |
498 | The @samp{-c} option causes the call graph of the program to be | |
499 | augmented by a heuristic which examines the text space of the object | |
500 | file and identifies function calls in the binary machine code. | |
501 | Since normal call graph records are only generated when functions are | |
502 | entered, this option identifies children that could have been called, | |
503 | but never were. Calls to functions that were not compiled with | |
504 | profiling enabled are also identified, but only if symbol table | |
505 | entries are present for them. | |
506 | Calls to dynamic library routines are typically @emph{not} found | |
507 | by this option. | |
508 | Parents or children identified via this heuristic | |
509 | are indicated in the call graph with call counts of @samp{0}. | |
510 | ||
511 | @item -D | |
512 | @itemx --ignore-non-functions | |
513 | The @samp{-D} option causes @code{gprof} to ignore symbols which | |
514 | are not known to be functions. This option will give more accurate | |
515 | profile data on systems where it is supported (Solaris and HPUX for | |
516 | example). | |
517 | ||
518 | @item -k @var{from}/@var{to} | |
519 | The @samp{-k} option allows you to delete from the call graph any arcs from | |
520 | symbols matching symspec @var{from} to those matching symspec @var{to}. | |
521 | ||
522 | @item -l | |
523 | @itemx --line | |
524 | The @samp{-l} option enables line-by-line profiling, which causes | |
525 | histogram hits to be charged to individual source code lines, | |
526 | instead of functions. | |
527 | If the program was compiled with basic-block counting enabled, | |
528 | this option will also identify how many times each line of | |
529 | code was executed. | |
530 | While line-by-line profiling can help isolate where in a large function | |
531 | a program is spending its time, it also significantly increases | |
532 | the running time of @code{gprof}, and magnifies statistical | |
533 | inaccuracies. | |
534 | @xref{Sampling Error}. | |
535 | ||
536 | @item -m @var{num} | |
537 | @itemx --min-count=@var{num} | |
538 | This option affects execution count output only. | |
539 | Symbols that are executed less than @var{num} times are suppressed. | |
540 | ||
541 | @item -n[@var{symspec}] | |
542 | @itemx --time[=@var{symspec}] | |
543 | The @samp{-n} option causes @code{gprof}, in its call graph analysis, | |
544 | to only propagate times for symbols matching @var{symspec}. | |
545 | ||
546 | @item -N[@var{symspec}] | |
547 | @itemx --no-time[=@var{symspec}] | |
548 | The @samp{-n} option causes @code{gprof}, in its call graph analysis, | |
549 | not to propagate times for symbols matching @var{symspec}. | |
550 | ||
551 | @item -z | |
552 | @itemx --display-unused-functions | |
553 | If you give the @samp{-z} option, @code{gprof} will mention all | |
554 | functions in the flat profile, even those that were never called, and | |
555 | that had no time spent in them. This is useful in conjunction with the | |
556 | @samp{-c} option for discovering which routines were never called. | |
557 | ||
558 | @end table | |
559 | ||
5af11cab | 560 | @node Miscellaneous Options,Deprecated Options,Analysis Options,Invoking |
252b5132 RH |
561 | @section Miscellaneous Options |
562 | ||
563 | @table @code | |
564 | ||
565 | @item -d[@var{num}] | |
566 | @itemx --debug[=@var{num}] | |
567 | The @samp{-d @var{num}} option specifies debugging options. | |
568 | If @var{num} is not specified, enable all debugging. | |
569 | @xref{Debugging}. | |
570 | ||
571 | @item -O@var{name} | |
572 | @itemx --file-format=@var{name} | |
573 | Selects the format of the profile data files. Recognized formats are | |
574 | @samp{auto} (the default), @samp{bsd}, @samp{4.4bsd}, @samp{magic}, and | |
575 | @samp{prof} (not yet supported). | |
576 | ||
577 | @item -s | |
578 | @itemx --sum | |
579 | The @samp{-s} option causes @code{gprof} to summarize the information | |
580 | in the profile data files it read in, and write out a profile data | |
581 | file called @file{gmon.sum}, which contains all the information from | |
582 | the profile data files that @code{gprof} read in. The file @file{gmon.sum} | |
583 | may be one of the specified input files; the effect of this is to | |
584 | merge the data in the other input files into @file{gmon.sum}. | |
585 | ||
586 | Eventually you can run @code{gprof} again without @samp{-s} to analyze the | |
587 | cumulative data in the file @file{gmon.sum}. | |
588 | ||
589 | @item -v | |
590 | @itemx --version | |
591 | The @samp{-v} flag causes @code{gprof} to print the current version | |
592 | number, and then exit. | |
593 | ||
594 | @end table | |
595 | ||
5af11cab AM |
596 | @node Deprecated Options,Symspecs,Miscellaneous Options,Invoking |
597 | @section Deprecated Options | |
252b5132 RH |
598 | |
599 | @table @code | |
600 | ||
601 | These options have been replaced with newer versions that use symspecs. | |
602 | ||
603 | @item -e @var{function_name} | |
604 | The @samp{-e @var{function}} option tells @code{gprof} to not print | |
605 | information about the function @var{function_name} (and its | |
606 | children@dots{}) in the call graph. The function will still be listed | |
607 | as a child of any functions that call it, but its index number will be | |
608 | shown as @samp{[not printed]}. More than one @samp{-e} option may be | |
609 | given; only one @var{function_name} may be indicated with each @samp{-e} | |
610 | option. | |
611 | ||
612 | @item -E @var{function_name} | |
613 | The @code{-E @var{function}} option works like the @code{-e} option, but | |
614 | time spent in the function (and children who were not called from | |
615 | anywhere else), will not be used to compute the percentages-of-time for | |
616 | the call graph. More than one @samp{-E} option may be given; only one | |
617 | @var{function_name} may be indicated with each @samp{-E} option. | |
618 | ||
619 | @item -f @var{function_name} | |
620 | The @samp{-f @var{function}} option causes @code{gprof} to limit the | |
621 | call graph to the function @var{function_name} and its children (and | |
622 | their children@dots{}). More than one @samp{-f} option may be given; | |
623 | only one @var{function_name} may be indicated with each @samp{-f} | |
624 | option. | |
625 | ||
626 | @item -F @var{function_name} | |
627 | The @samp{-F @var{function}} option works like the @code{-f} option, but | |
628 | only time spent in the function and its children (and their | |
629 | children@dots{}) will be used to determine total-time and | |
630 | percentages-of-time for the call graph. More than one @samp{-F} option | |
631 | may be given; only one @var{function_name} may be indicated with each | |
632 | @samp{-F} option. The @samp{-F} option overrides the @samp{-E} option. | |
633 | ||
634 | @end table | |
635 | ||
636 | Note that only one function can be specified with each @code{-e}, | |
637 | @code{-E}, @code{-f} or @code{-F} option. To specify more than one | |
638 | function, use multiple options. For example, this command: | |
639 | ||
640 | @example | |
641 | gprof -e boring -f foo -f bar myprogram > gprof.output | |
642 | @end example | |
643 | ||
644 | @noindent | |
645 | lists in the call graph all functions that were reached from either | |
646 | @code{foo} or @code{bar} and were not reachable from @code{boring}. | |
647 | ||
5af11cab | 648 | @node Symspecs,,Deprecated Options,Invoking |
252b5132 RH |
649 | @section Symspecs |
650 | ||
651 | Many of the output options allow functions to be included or excluded | |
652 | using @dfn{symspecs} (symbol specifications), which observe the | |
653 | following syntax: | |
654 | ||
655 | @example | |
656 | filename_containing_a_dot | |
657 | | funcname_not_containing_a_dot | |
658 | | linenumber | |
659 | | ( [ any_filename ] `:' ( any_funcname | linenumber ) ) | |
660 | @end example | |
661 | ||
662 | Here are some sample symspecs: | |
663 | ||
664 | @table @samp | |
665 | @item main.c | |
666 | Selects everything in file @file{main.c}---the | |
5af11cab | 667 | dot in the string tells @code{gprof} to interpret |
252b5132 RH |
668 | the string as a filename, rather than as |
669 | a function name. To select a file whose | |
670 | name does not contain a dot, a trailing colon | |
671 | should be specified. For example, @samp{odd:} is | |
672 | interpreted as the file named @file{odd}. | |
673 | ||
674 | @item main | |
675 | Selects all functions named @samp{main}. | |
676 | ||
677 | Note that there may be multiple instances of the same function name | |
678 | because some of the definitions may be local (i.e., static). Unless a | |
679 | function name is unique in a program, you must use the colon notation | |
680 | explained below to specify a function from a specific source file. | |
681 | ||
a53f781e | 682 | Sometimes, function names contain dots. In such cases, it is necessary |
252b5132 RH |
683 | to add a leading colon to the name. For example, @samp{:.mul} selects |
684 | function @samp{.mul}. | |
685 | ||
5af11cab AM |
686 | In some object file formats, symbols have a leading underscore. |
687 | @code{gprof} will normally not print these underscores. When you name a | |
688 | symbol in a symspec, you should type it exactly as @code{gprof} prints | |
689 | it in its output. For example, if the compiler produces a symbol | |
690 | @samp{_main} from your @code{main} function, @code{gprof} still prints | |
691 | it as @samp{main} in its output, so you should use @samp{main} in | |
692 | symspecs. | |
252b5132 RH |
693 | |
694 | @item main.c:main | |
695 | Selects function @samp{main} in file @file{main.c}. | |
696 | ||
697 | @item main.c:134 | |
698 | Selects line 134 in file @file{main.c}. | |
699 | @end table | |
700 | ||
701 | @node Output | |
702 | @chapter Interpreting @code{gprof}'s Output | |
703 | ||
704 | @code{gprof} can produce several different output styles, the | |
705 | most important of which are described below. The simplest output | |
706 | styles (file information, execution count, and function and file ordering) | |
707 | are not described here, but are documented with the respective options | |
708 | that trigger them. | |
709 | @xref{Output Options}. | |
710 | ||
711 | @menu | |
712 | * Flat Profile:: The flat profile shows how much time was spent | |
713 | executing directly in each function. | |
714 | * Call Graph:: The call graph shows which functions called which | |
715 | others, and how much time each function used | |
716 | when its subroutine calls are included. | |
717 | * Line-by-line:: @code{gprof} can analyze individual source code lines | |
718 | * Annotated Source:: The annotated source listing displays source code | |
719 | labeled with execution counts | |
720 | @end menu | |
721 | ||
722 | ||
723 | @node Flat Profile,Call Graph,,Output | |
724 | @section The Flat Profile | |
725 | @cindex flat profile | |
726 | ||
727 | The @dfn{flat profile} shows the total amount of time your program | |
728 | spent executing each function. Unless the @samp{-z} option is given, | |
729 | functions with no apparent time spent in them, and no apparent calls | |
730 | to them, are not mentioned. Note that if a function was not compiled | |
731 | for profiling, and didn't run long enough to show up on the program | |
732 | counter histogram, it will be indistinguishable from a function that | |
733 | was never called. | |
734 | ||
735 | This is part of a flat profile for a small program: | |
736 | ||
737 | @smallexample | |
738 | @group | |
739 | Flat profile: | |
740 | ||
741 | Each sample counts as 0.01 seconds. | |
742 | % cumulative self self total | |
743 | time seconds seconds calls ms/call ms/call name | |
744 | 33.34 0.02 0.02 7208 0.00 0.00 open | |
745 | 16.67 0.03 0.01 244 0.04 0.12 offtime | |
746 | 16.67 0.04 0.01 8 1.25 1.25 memccpy | |
747 | 16.67 0.05 0.01 7 1.43 1.43 write | |
748 | 16.67 0.06 0.01 mcount | |
749 | 0.00 0.06 0.00 236 0.00 0.00 tzset | |
750 | 0.00 0.06 0.00 192 0.00 0.00 tolower | |
751 | 0.00 0.06 0.00 47 0.00 0.00 strlen | |
752 | 0.00 0.06 0.00 45 0.00 0.00 strchr | |
753 | 0.00 0.06 0.00 1 0.00 50.00 main | |
754 | 0.00 0.06 0.00 1 0.00 0.00 memcpy | |
755 | 0.00 0.06 0.00 1 0.00 10.11 print | |
756 | 0.00 0.06 0.00 1 0.00 0.00 profil | |
757 | 0.00 0.06 0.00 1 0.00 50.00 report | |
758 | @dots{} | |
759 | @end group | |
760 | @end smallexample | |
761 | ||
762 | @noindent | |
763 | The functions are sorted by first by decreasing run-time spent in them, | |
764 | then by decreasing number of calls, then alphabetically by name. The | |
765 | functions @samp{mcount} and @samp{profil} are part of the profiling | |
5af11cab | 766 | apparatus and appear in every flat profile; their time gives a measure of |
252b5132 RH |
767 | the amount of overhead due to profiling. |
768 | ||
769 | Just before the column headers, a statement appears indicating | |
770 | how much time each sample counted as. | |
771 | This @dfn{sampling period} estimates the margin of error in each of the time | |
772 | figures. A time figure that is not much larger than this is not | |
773 | reliable. In this example, each sample counted as 0.01 seconds, | |
774 | suggesting a 100 Hz sampling rate. | |
775 | The program's total execution time was 0.06 | |
776 | seconds, as indicated by the @samp{cumulative seconds} field. Since | |
777 | each sample counted for 0.01 seconds, this means only six samples | |
5af11cab | 778 | were taken during the run. Two of the samples occurred while the |
252b5132 RH |
779 | program was in the @samp{open} function, as indicated by the |
780 | @samp{self seconds} field. Each of the other four samples | |
5af11cab | 781 | occurred one each in @samp{offtime}, @samp{memccpy}, @samp{write}, |
252b5132 RH |
782 | and @samp{mcount}. |
783 | Since only six samples were taken, none of these values can | |
784 | be regarded as particularly reliable. | |
785 | In another run, | |
786 | the @samp{self seconds} field for | |
787 | @samp{mcount} might well be @samp{0.00} or @samp{0.02}. | |
788 | @xref{Sampling Error}, for a complete discussion. | |
789 | ||
790 | The remaining functions in the listing (those whose | |
791 | @samp{self seconds} field is @samp{0.00}) didn't appear | |
792 | in the histogram samples at all. However, the call graph | |
793 | indicated that they were called, so therefore they are listed, | |
794 | sorted in decreasing order by the @samp{calls} field. | |
795 | Clearly some time was spent executing these functions, | |
796 | but the paucity of histogram samples prevents any | |
797 | determination of how much time each took. | |
798 | ||
799 | Here is what the fields in each line mean: | |
800 | ||
801 | @table @code | |
802 | @item % time | |
803 | This is the percentage of the total execution time your program spent | |
804 | in this function. These should all add up to 100%. | |
805 | ||
806 | @item cumulative seconds | |
807 | This is the cumulative total number of seconds the computer spent | |
808 | executing this functions, plus the time spent in all the functions | |
809 | above this one in this table. | |
810 | ||
811 | @item self seconds | |
812 | This is the number of seconds accounted for by this function alone. | |
813 | The flat profile listing is sorted first by this number. | |
814 | ||
815 | @item calls | |
816 | This is the total number of times the function was called. If the | |
817 | function was never called, or the number of times it was called cannot | |
818 | be determined (probably because the function was not compiled with | |
819 | profiling enabled), the @dfn{calls} field is blank. | |
820 | ||
821 | @item self ms/call | |
822 | This represents the average number of milliseconds spent in this | |
823 | function per call, if this function is profiled. Otherwise, this field | |
824 | is blank for this function. | |
825 | ||
826 | @item total ms/call | |
827 | This represents the average number of milliseconds spent in this | |
828 | function and its descendants per call, if this function is profiled. | |
829 | Otherwise, this field is blank for this function. | |
830 | This is the only field in the flat profile that uses call graph analysis. | |
831 | ||
832 | @item name | |
833 | This is the name of the function. The flat profile is sorted by this | |
834 | field alphabetically after the @dfn{self seconds} and @dfn{calls} | |
835 | fields are sorted. | |
836 | @end table | |
837 | ||
838 | @node Call Graph,Line-by-line,Flat Profile,Output | |
839 | @section The Call Graph | |
840 | @cindex call graph | |
841 | ||
842 | The @dfn{call graph} shows how much time was spent in each function | |
843 | and its children. From this information, you can find functions that, | |
844 | while they themselves may not have used much time, called other | |
845 | functions that did use unusual amounts of time. | |
846 | ||
847 | Here is a sample call from a small program. This call came from the | |
848 | same @code{gprof} run as the flat profile example in the previous | |
849 | chapter. | |
850 | ||
851 | @smallexample | |
852 | @group | |
853 | granularity: each sample hit covers 2 byte(s) for 20.00% of 0.05 seconds | |
854 | ||
855 | index % time self children called name | |
856 | <spontaneous> | |
857 | [1] 100.0 0.00 0.05 start [1] | |
858 | 0.00 0.05 1/1 main [2] | |
859 | 0.00 0.00 1/2 on_exit [28] | |
860 | 0.00 0.00 1/1 exit [59] | |
861 | ----------------------------------------------- | |
862 | 0.00 0.05 1/1 start [1] | |
863 | [2] 100.0 0.00 0.05 1 main [2] | |
864 | 0.00 0.05 1/1 report [3] | |
865 | ----------------------------------------------- | |
866 | 0.00 0.05 1/1 main [2] | |
867 | [3] 100.0 0.00 0.05 1 report [3] | |
868 | 0.00 0.03 8/8 timelocal [6] | |
869 | 0.00 0.01 1/1 print [9] | |
870 | 0.00 0.01 9/9 fgets [12] | |
871 | 0.00 0.00 12/34 strncmp <cycle 1> [40] | |
872 | 0.00 0.00 8/8 lookup [20] | |
873 | 0.00 0.00 1/1 fopen [21] | |
874 | 0.00 0.00 8/8 chewtime [24] | |
875 | 0.00 0.00 8/16 skipspace [44] | |
876 | ----------------------------------------------- | |
877 | [4] 59.8 0.01 0.02 8+472 <cycle 2 as a whole> [4] | |
878 | 0.01 0.02 244+260 offtime <cycle 2> [7] | |
879 | 0.00 0.00 236+1 tzset <cycle 2> [26] | |
880 | ----------------------------------------------- | |
881 | @end group | |
882 | @end smallexample | |
883 | ||
884 | The lines full of dashes divide this table into @dfn{entries}, one for each | |
885 | function. Each entry has one or more lines. | |
886 | ||
887 | In each entry, the primary line is the one that starts with an index number | |
888 | in square brackets. The end of this line says which function the entry is | |
889 | for. The preceding lines in the entry describe the callers of this | |
890 | function and the following lines describe its subroutines (also called | |
891 | @dfn{children} when we speak of the call graph). | |
892 | ||
893 | The entries are sorted by time spent in the function and its subroutines. | |
894 | ||
895 | The internal profiling function @code{mcount} (@pxref{Flat Profile}) | |
896 | is never mentioned in the call graph. | |
897 | ||
898 | @menu | |
899 | * Primary:: Details of the primary line's contents. | |
900 | * Callers:: Details of caller-lines' contents. | |
901 | * Subroutines:: Details of subroutine-lines' contents. | |
902 | * Cycles:: When there are cycles of recursion, | |
903 | such as @code{a} calls @code{b} calls @code{a}@dots{} | |
904 | @end menu | |
905 | ||
906 | @node Primary | |
907 | @subsection The Primary Line | |
908 | ||
909 | The @dfn{primary line} in a call graph entry is the line that | |
910 | describes the function which the entry is about and gives the overall | |
911 | statistics for this function. | |
912 | ||
913 | For reference, we repeat the primary line from the entry for function | |
914 | @code{report} in our main example, together with the heading line that | |
915 | shows the names of the fields: | |
916 | ||
917 | @smallexample | |
918 | @group | |
919 | index % time self children called name | |
920 | @dots{} | |
921 | [3] 100.0 0.00 0.05 1 report [3] | |
922 | @end group | |
923 | @end smallexample | |
924 | ||
925 | Here is what the fields in the primary line mean: | |
926 | ||
927 | @table @code | |
928 | @item index | |
929 | Entries are numbered with consecutive integers. Each function | |
930 | therefore has an index number, which appears at the beginning of its | |
931 | primary line. | |
932 | ||
933 | Each cross-reference to a function, as a caller or subroutine of | |
934 | another, gives its index number as well as its name. The index number | |
935 | guides you if you wish to look for the entry for that function. | |
936 | ||
937 | @item % time | |
938 | This is the percentage of the total time that was spent in this | |
939 | function, including time spent in subroutines called from this | |
940 | function. | |
941 | ||
942 | The time spent in this function is counted again for the callers of | |
943 | this function. Therefore, adding up these percentages is meaningless. | |
944 | ||
945 | @item self | |
946 | This is the total amount of time spent in this function. This | |
947 | should be identical to the number printed in the @code{seconds} field | |
948 | for this function in the flat profile. | |
949 | ||
950 | @item children | |
951 | This is the total amount of time spent in the subroutine calls made by | |
952 | this function. This should be equal to the sum of all the @code{self} | |
953 | and @code{children} entries of the children listed directly below this | |
954 | function. | |
955 | ||
956 | @item called | |
957 | This is the number of times the function was called. | |
958 | ||
959 | If the function called itself recursively, there are two numbers, | |
960 | separated by a @samp{+}. The first number counts non-recursive calls, | |
961 | and the second counts recursive calls. | |
962 | ||
963 | In the example above, the function @code{report} was called once from | |
964 | @code{main}. | |
965 | ||
966 | @item name | |
967 | This is the name of the current function. The index number is | |
968 | repeated after it. | |
969 | ||
970 | If the function is part of a cycle of recursion, the cycle number is | |
971 | printed between the function's name and the index number | |
972 | (@pxref{Cycles}). For example, if function @code{gnurr} is part of | |
973 | cycle number one, and has index number twelve, its primary line would | |
974 | be end like this: | |
975 | ||
976 | @example | |
977 | gnurr <cycle 1> [12] | |
978 | @end example | |
979 | @end table | |
980 | ||
981 | @node Callers, Subroutines, Primary, Call Graph | |
982 | @subsection Lines for a Function's Callers | |
983 | ||
984 | A function's entry has a line for each function it was called by. | |
985 | These lines' fields correspond to the fields of the primary line, but | |
986 | their meanings are different because of the difference in context. | |
987 | ||
988 | For reference, we repeat two lines from the entry for the function | |
989 | @code{report}, the primary line and one caller-line preceding it, together | |
990 | with the heading line that shows the names of the fields: | |
991 | ||
992 | @smallexample | |
993 | index % time self children called name | |
994 | @dots{} | |
995 | 0.00 0.05 1/1 main [2] | |
996 | [3] 100.0 0.00 0.05 1 report [3] | |
997 | @end smallexample | |
998 | ||
999 | Here are the meanings of the fields in the caller-line for @code{report} | |
1000 | called from @code{main}: | |
1001 | ||
1002 | @table @code | |
1003 | @item self | |
1004 | An estimate of the amount of time spent in @code{report} itself when it was | |
1005 | called from @code{main}. | |
1006 | ||
1007 | @item children | |
1008 | An estimate of the amount of time spent in subroutines of @code{report} | |
1009 | when @code{report} was called from @code{main}. | |
1010 | ||
1011 | The sum of the @code{self} and @code{children} fields is an estimate | |
1012 | of the amount of time spent within calls to @code{report} from @code{main}. | |
1013 | ||
1014 | @item called | |
1015 | Two numbers: the number of times @code{report} was called from @code{main}, | |
5af11cab | 1016 | followed by the total number of non-recursive calls to @code{report} from |
252b5132 RH |
1017 | all its callers. |
1018 | ||
1019 | @item name and index number | |
1020 | The name of the caller of @code{report} to which this line applies, | |
1021 | followed by the caller's index number. | |
1022 | ||
1023 | Not all functions have entries in the call graph; some | |
1024 | options to @code{gprof} request the omission of certain functions. | |
1025 | When a caller has no entry of its own, it still has caller-lines | |
1026 | in the entries of the functions it calls. | |
1027 | ||
1028 | If the caller is part of a recursion cycle, the cycle number is | |
1029 | printed between the name and the index number. | |
1030 | @end table | |
1031 | ||
1032 | If the identity of the callers of a function cannot be determined, a | |
1033 | dummy caller-line is printed which has @samp{<spontaneous>} as the | |
1034 | ``caller's name'' and all other fields blank. This can happen for | |
1035 | signal handlers. | |
1036 | @c What if some calls have determinable callers' names but not all? | |
1037 | @c FIXME - still relevant? | |
1038 | ||
1039 | @node Subroutines, Cycles, Callers, Call Graph | |
1040 | @subsection Lines for a Function's Subroutines | |
1041 | ||
1042 | A function's entry has a line for each of its subroutines---in other | |
1043 | words, a line for each other function that it called. These lines' | |
1044 | fields correspond to the fields of the primary line, but their meanings | |
1045 | are different because of the difference in context. | |
1046 | ||
1047 | For reference, we repeat two lines from the entry for the function | |
1048 | @code{main}, the primary line and a line for a subroutine, together | |
1049 | with the heading line that shows the names of the fields: | |
1050 | ||
1051 | @smallexample | |
1052 | index % time self children called name | |
1053 | @dots{} | |
1054 | [2] 100.0 0.00 0.05 1 main [2] | |
1055 | 0.00 0.05 1/1 report [3] | |
1056 | @end smallexample | |
1057 | ||
1058 | Here are the meanings of the fields in the subroutine-line for @code{main} | |
1059 | calling @code{report}: | |
1060 | ||
1061 | @table @code | |
1062 | @item self | |
1063 | An estimate of the amount of time spent directly within @code{report} | |
1064 | when @code{report} was called from @code{main}. | |
1065 | ||
1066 | @item children | |
1067 | An estimate of the amount of time spent in subroutines of @code{report} | |
1068 | when @code{report} was called from @code{main}. | |
1069 | ||
1070 | The sum of the @code{self} and @code{children} fields is an estimate | |
1071 | of the total time spent in calls to @code{report} from @code{main}. | |
1072 | ||
1073 | @item called | |
1074 | Two numbers, the number of calls to @code{report} from @code{main} | |
5af11cab | 1075 | followed by the total number of non-recursive calls to @code{report}. |
252b5132 RH |
1076 | This ratio is used to determine how much of @code{report}'s @code{self} |
1077 | and @code{children} time gets credited to @code{main}. | |
1078 | @xref{Assumptions}. | |
1079 | ||
1080 | @item name | |
1081 | The name of the subroutine of @code{main} to which this line applies, | |
1082 | followed by the subroutine's index number. | |
1083 | ||
1084 | If the caller is part of a recursion cycle, the cycle number is | |
1085 | printed between the name and the index number. | |
1086 | @end table | |
1087 | ||
1088 | @node Cycles,, Subroutines, Call Graph | |
1089 | @subsection How Mutually Recursive Functions Are Described | |
1090 | @cindex cycle | |
1091 | @cindex recursion cycle | |
1092 | ||
1093 | The graph may be complicated by the presence of @dfn{cycles of | |
1094 | recursion} in the call graph. A cycle exists if a function calls | |
1095 | another function that (directly or indirectly) calls (or appears to | |
1096 | call) the original function. For example: if @code{a} calls @code{b}, | |
1097 | and @code{b} calls @code{a}, then @code{a} and @code{b} form a cycle. | |
1098 | ||
1099 | Whenever there are call paths both ways between a pair of functions, they | |
1100 | belong to the same cycle. If @code{a} and @code{b} call each other and | |
1101 | @code{b} and @code{c} call each other, all three make one cycle. Note that | |
1102 | even if @code{b} only calls @code{a} if it was not called from @code{a}, | |
1103 | @code{gprof} cannot determine this, so @code{a} and @code{b} are still | |
1104 | considered a cycle. | |
1105 | ||
1106 | The cycles are numbered with consecutive integers. When a function | |
1107 | belongs to a cycle, each time the function name appears in the call graph | |
1108 | it is followed by @samp{<cycle @var{number}>}. | |
1109 | ||
1110 | The reason cycles matter is that they make the time values in the call | |
1111 | graph paradoxical. The ``time spent in children'' of @code{a} should | |
1112 | include the time spent in its subroutine @code{b} and in @code{b}'s | |
1113 | subroutines---but one of @code{b}'s subroutines is @code{a}! How much of | |
1114 | @code{a}'s time should be included in the children of @code{a}, when | |
1115 | @code{a} is indirectly recursive? | |
1116 | ||
1117 | The way @code{gprof} resolves this paradox is by creating a single entry | |
1118 | for the cycle as a whole. The primary line of this entry describes the | |
1119 | total time spent directly in the functions of the cycle. The | |
1120 | ``subroutines'' of the cycle are the individual functions of the cycle, and | |
1121 | all other functions that were called directly by them. The ``callers'' of | |
1122 | the cycle are the functions, outside the cycle, that called functions in | |
1123 | the cycle. | |
1124 | ||
1125 | Here is an example portion of a call graph which shows a cycle containing | |
1126 | functions @code{a} and @code{b}. The cycle was entered by a call to | |
1127 | @code{a} from @code{main}; both @code{a} and @code{b} called @code{c}. | |
1128 | ||
1129 | @smallexample | |
1130 | index % time self children called name | |
1131 | ---------------------------------------- | |
1132 | 1.77 0 1/1 main [2] | |
1133 | [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3] | |
1134 | 1.02 0 3 b <cycle 1> [4] | |
1135 | 0.75 0 2 a <cycle 1> [5] | |
1136 | ---------------------------------------- | |
1137 | 3 a <cycle 1> [5] | |
1138 | [4] 52.85 1.02 0 0 b <cycle 1> [4] | |
1139 | 2 a <cycle 1> [5] | |
1140 | 0 0 3/6 c [6] | |
1141 | ---------------------------------------- | |
1142 | 1.77 0 1/1 main [2] | |
1143 | 2 b <cycle 1> [4] | |
1144 | [5] 38.86 0.75 0 1 a <cycle 1> [5] | |
1145 | 3 b <cycle 1> [4] | |
1146 | 0 0 3/6 c [6] | |
1147 | ---------------------------------------- | |
1148 | @end smallexample | |
1149 | ||
1150 | @noindent | |
1151 | (The entire call graph for this program contains in addition an entry for | |
1152 | @code{main}, which calls @code{a}, and an entry for @code{c}, with callers | |
1153 | @code{a} and @code{b}.) | |
1154 | ||
1155 | @smallexample | |
1156 | index % time self children called name | |
1157 | <spontaneous> | |
1158 | [1] 100.00 0 1.93 0 start [1] | |
1159 | 0.16 1.77 1/1 main [2] | |
1160 | ---------------------------------------- | |
1161 | 0.16 1.77 1/1 start [1] | |
1162 | [2] 100.00 0.16 1.77 1 main [2] | |
1163 | 1.77 0 1/1 a <cycle 1> [5] | |
1164 | ---------------------------------------- | |
1165 | 1.77 0 1/1 main [2] | |
1166 | [3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3] | |
1167 | 1.02 0 3 b <cycle 1> [4] | |
1168 | 0.75 0 2 a <cycle 1> [5] | |
1169 | 0 0 6/6 c [6] | |
1170 | ---------------------------------------- | |
1171 | 3 a <cycle 1> [5] | |
1172 | [4] 52.85 1.02 0 0 b <cycle 1> [4] | |
1173 | 2 a <cycle 1> [5] | |
1174 | 0 0 3/6 c [6] | |
1175 | ---------------------------------------- | |
1176 | 1.77 0 1/1 main [2] | |
1177 | 2 b <cycle 1> [4] | |
1178 | [5] 38.86 0.75 0 1 a <cycle 1> [5] | |
1179 | 3 b <cycle 1> [4] | |
1180 | 0 0 3/6 c [6] | |
1181 | ---------------------------------------- | |
1182 | 0 0 3/6 b <cycle 1> [4] | |
1183 | 0 0 3/6 a <cycle 1> [5] | |
1184 | [6] 0.00 0 0 6 c [6] | |
1185 | ---------------------------------------- | |
1186 | @end smallexample | |
1187 | ||
1188 | The @code{self} field of the cycle's primary line is the total time | |
1189 | spent in all the functions of the cycle. It equals the sum of the | |
1190 | @code{self} fields for the individual functions in the cycle, found | |
1191 | in the entry in the subroutine lines for these functions. | |
1192 | ||
1193 | The @code{children} fields of the cycle's primary line and subroutine lines | |
1194 | count only subroutines outside the cycle. Even though @code{a} calls | |
1195 | @code{b}, the time spent in those calls to @code{b} is not counted in | |
1196 | @code{a}'s @code{children} time. Thus, we do not encounter the problem of | |
1197 | what to do when the time in those calls to @code{b} includes indirect | |
1198 | recursive calls back to @code{a}. | |
1199 | ||
1200 | The @code{children} field of a caller-line in the cycle's entry estimates | |
1201 | the amount of time spent @emph{in the whole cycle}, and its other | |
1202 | subroutines, on the times when that caller called a function in the cycle. | |
1203 | ||
1204 | The @code{calls} field in the primary line for the cycle has two numbers: | |
1205 | first, the number of times functions in the cycle were called by functions | |
1206 | outside the cycle; second, the number of times they were called by | |
1207 | functions in the cycle (including times when a function in the cycle calls | |
5af11cab | 1208 | itself). This is a generalization of the usual split into non-recursive and |
252b5132 RH |
1209 | recursive calls. |
1210 | ||
1211 | The @code{calls} field of a subroutine-line for a cycle member in the | |
1212 | cycle's entry says how many time that function was called from functions in | |
1213 | the cycle. The total of all these is the second number in the primary line's | |
1214 | @code{calls} field. | |
1215 | ||
1216 | In the individual entry for a function in a cycle, the other functions in | |
1217 | the same cycle can appear as subroutines and as callers. These lines show | |
1218 | how many times each function in the cycle called or was called from each other | |
1219 | function in the cycle. The @code{self} and @code{children} fields in these | |
1220 | lines are blank because of the difficulty of defining meanings for them | |
1221 | when recursion is going on. | |
1222 | ||
1223 | @node Line-by-line,Annotated Source,Call Graph,Output | |
1224 | @section Line-by-line Profiling | |
1225 | ||
1226 | @code{gprof}'s @samp{-l} option causes the program to perform | |
1227 | @dfn{line-by-line} profiling. In this mode, histogram | |
1228 | samples are assigned not to functions, but to individual | |
1229 | lines of source code. The program usually must be compiled | |
1230 | with a @samp{-g} option, in addition to @samp{-pg}, in order | |
1231 | to generate debugging symbols for tracking source code lines. | |
1232 | ||
1233 | The flat profile is the most useful output table | |
1234 | in line-by-line mode. | |
1235 | The call graph isn't as useful as normal, since | |
1236 | the current version of @code{gprof} does not propagate | |
1237 | call graph arcs from source code lines to the enclosing function. | |
1238 | The call graph does, however, show each line of code | |
1239 | that called each function, along with a count. | |
1240 | ||
1241 | Here is a section of @code{gprof}'s output, without line-by-line profiling. | |
1242 | Note that @code{ct_init} accounted for four histogram hits, and | |
1243 | 13327 calls to @code{init_block}. | |
1244 | ||
1245 | @smallexample | |
1246 | Flat profile: | |
1247 | ||
1248 | Each sample counts as 0.01 seconds. | |
1249 | % cumulative self self total | |
1250 | time seconds seconds calls us/call us/call name | |
1251 | 30.77 0.13 0.04 6335 6.31 6.31 ct_init | |
1252 | ||
1253 | ||
1254 | Call graph (explanation follows) | |
1255 | ||
1256 | ||
1257 | granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds | |
1258 | ||
1259 | index % time self children called name | |
1260 | ||
1261 | 0.00 0.00 1/13496 name_too_long | |
1262 | 0.00 0.00 40/13496 deflate | |
1263 | 0.00 0.00 128/13496 deflate_fast | |
1264 | 0.00 0.00 13327/13496 ct_init | |
1265 | [7] 0.0 0.00 0.00 13496 init_block | |
1266 | ||
1267 | @end smallexample | |
1268 | ||
1269 | Now let's look at some of @code{gprof}'s output from the same program run, | |
1270 | this time with line-by-line profiling enabled. Note that @code{ct_init}'s | |
1271 | four histogram hits are broken down into four lines of source code - one hit | |
5af11cab | 1272 | occurred on each of lines 349, 351, 382 and 385. In the call graph, |
252b5132 RH |
1273 | note how |
1274 | @code{ct_init}'s 13327 calls to @code{init_block} are broken down | |
1275 | into one call from line 396, 3071 calls from line 384, 3730 calls | |
1276 | from line 385, and 6525 calls from 387. | |
1277 | ||
1278 | @smallexample | |
1279 | Flat profile: | |
1280 | ||
1281 | Each sample counts as 0.01 seconds. | |
1282 | % cumulative self | |
1283 | time seconds seconds calls name | |
1284 | 7.69 0.10 0.01 ct_init (trees.c:349) | |
1285 | 7.69 0.11 0.01 ct_init (trees.c:351) | |
1286 | 7.69 0.12 0.01 ct_init (trees.c:382) | |
1287 | 7.69 0.13 0.01 ct_init (trees.c:385) | |
1288 | ||
1289 | ||
1290 | Call graph (explanation follows) | |
1291 | ||
1292 | ||
1293 | granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds | |
1294 | ||
1295 | % time self children called name | |
1296 | ||
1297 | 0.00 0.00 1/13496 name_too_long (gzip.c:1440) | |
1298 | 0.00 0.00 1/13496 deflate (deflate.c:763) | |
1299 | 0.00 0.00 1/13496 ct_init (trees.c:396) | |
1300 | 0.00 0.00 2/13496 deflate (deflate.c:727) | |
1301 | 0.00 0.00 4/13496 deflate (deflate.c:686) | |
1302 | 0.00 0.00 5/13496 deflate (deflate.c:675) | |
1303 | 0.00 0.00 12/13496 deflate (deflate.c:679) | |
1304 | 0.00 0.00 16/13496 deflate (deflate.c:730) | |
1305 | 0.00 0.00 128/13496 deflate_fast (deflate.c:654) | |
1306 | 0.00 0.00 3071/13496 ct_init (trees.c:384) | |
1307 | 0.00 0.00 3730/13496 ct_init (trees.c:385) | |
1308 | 0.00 0.00 6525/13496 ct_init (trees.c:387) | |
1309 | [6] 0.0 0.00 0.00 13496 init_block (trees.c:408) | |
1310 | ||
1311 | @end smallexample | |
1312 | ||
1313 | ||
1314 | @node Annotated Source,,Line-by-line,Output | |
1315 | @section The Annotated Source Listing | |
1316 | ||
1317 | @code{gprof}'s @samp{-A} option triggers an annotated source listing, | |
1318 | which lists the program's source code, each function labeled with the | |
1319 | number of times it was called. You may also need to specify the | |
1320 | @samp{-I} option, if @code{gprof} can't find the source code files. | |
1321 | ||
1322 | Compiling with @samp{gcc @dots{} -g -pg -a} augments your program | |
1323 | with basic-block counting code, in addition to function counting code. | |
1324 | This enables @code{gprof} to determine how many times each line | |
5af11cab | 1325 | of code was executed. |
252b5132 RH |
1326 | For example, consider the following function, taken from gzip, |
1327 | with line numbers added: | |
1328 | ||
1329 | @smallexample | |
1330 | 1 ulg updcrc(s, n) | |
1331 | 2 uch *s; | |
1332 | 3 unsigned n; | |
1333 | 4 @{ | |
1334 | 5 register ulg c; | |
1335 | 6 | |
1336 | 7 static ulg crc = (ulg)0xffffffffL; | |
1337 | 8 | |
1338 | 9 if (s == NULL) @{ | |
1339 | 10 c = 0xffffffffL; | |
1340 | 11 @} else @{ | |
1341 | 12 c = crc; | |
1342 | 13 if (n) do @{ | |
1343 | 14 c = crc_32_tab[...]; | |
1344 | 15 @} while (--n); | |
1345 | 16 @} | |
1346 | 17 crc = c; | |
1347 | 18 return c ^ 0xffffffffL; | |
1348 | 19 @} | |
1349 | ||
1350 | @end smallexample | |
1351 | ||
1352 | @code{updcrc} has at least five basic-blocks. | |
1353 | One is the function itself. The | |
1354 | @code{if} statement on line 9 generates two more basic-blocks, one | |
1355 | for each branch of the @code{if}. A fourth basic-block results from | |
1356 | the @code{if} on line 13, and the contents of the @code{do} loop form | |
1357 | the fifth basic-block. The compiler may also generate additional | |
1358 | basic-blocks to handle various special cases. | |
1359 | ||
1360 | A program augmented for basic-block counting can be analyzed with | |
5af11cab | 1361 | @samp{gprof -l -A}. I also suggest use of the @samp{-x} option, |
252b5132 RH |
1362 | which ensures that each line of code is labeled at least once. |
1363 | Here is @code{updcrc}'s | |
1364 | annotated source listing for a sample @code{gzip} run: | |
1365 | ||
1366 | @smallexample | |
1367 | ulg updcrc(s, n) | |
1368 | uch *s; | |
1369 | unsigned n; | |
1370 | 2 ->@{ | |
1371 | register ulg c; | |
1372 | ||
1373 | static ulg crc = (ulg)0xffffffffL; | |
1374 | ||
1375 | 2 -> if (s == NULL) @{ | |
1376 | 1 -> c = 0xffffffffL; | |
1377 | 1 -> @} else @{ | |
1378 | 1 -> c = crc; | |
1379 | 1 -> if (n) do @{ | |
1380 | 26312 -> c = crc_32_tab[...]; | |
1381 | 26312,1,26311 -> @} while (--n); | |
1382 | @} | |
1383 | 2 -> crc = c; | |
1384 | 2 -> return c ^ 0xffffffffL; | |
1385 | 2 ->@} | |
1386 | @end smallexample | |
1387 | ||
1388 | In this example, the function was called twice, passing once through | |
1389 | each branch of the @code{if} statement. The body of the @code{do} | |
1390 | loop was executed a total of 26312 times. Note how the @code{while} | |
1391 | statement is annotated. It began execution 26312 times, once for | |
1392 | each iteration through the loop. One of those times (the last time) | |
1393 | it exited, while it branched back to the beginning of the loop 26311 times. | |
1394 | ||
1395 | @node Inaccuracy | |
1396 | @chapter Inaccuracy of @code{gprof} Output | |
1397 | ||
1398 | @menu | |
1399 | * Sampling Error:: Statistical margins of error | |
1400 | * Assumptions:: Estimating children times | |
1401 | @end menu | |
1402 | ||
1403 | @node Sampling Error,Assumptions,,Inaccuracy | |
1404 | @section Statistical Sampling Error | |
1405 | ||
1406 | The run-time figures that @code{gprof} gives you are based on a sampling | |
1407 | process, so they are subject to statistical inaccuracy. If a function runs | |
1408 | only a small amount of time, so that on the average the sampling process | |
1409 | ought to catch that function in the act only once, there is a pretty good | |
1410 | chance it will actually find that function zero times, or twice. | |
1411 | ||
1412 | By contrast, the number-of-calls and basic-block figures | |
1413 | are derived by counting, not | |
1414 | sampling. They are completely accurate and will not vary from run to run | |
1415 | if your program is deterministic. | |
1416 | ||
1417 | The @dfn{sampling period} that is printed at the beginning of the flat | |
1418 | profile says how often samples are taken. The rule of thumb is that a | |
1419 | run-time figure is accurate if it is considerably bigger than the sampling | |
1420 | period. | |
1421 | ||
1422 | The actual amount of error can be predicted. | |
1423 | For @var{n} samples, the @emph{expected} error | |
1424 | is the square-root of @var{n}. For example, | |
1425 | if the sampling period is 0.01 seconds and @code{foo}'s run-time is 1 second, | |
1426 | @var{n} is 100 samples (1 second/0.01 seconds), sqrt(@var{n}) is 10 samples, so | |
1427 | the expected error in @code{foo}'s run-time is 0.1 seconds (10*0.01 seconds), | |
1428 | or ten percent of the observed value. | |
1429 | Again, if the sampling period is 0.01 seconds and @code{bar}'s run-time is | |
1430 | 100 seconds, @var{n} is 10000 samples, sqrt(@var{n}) is 100 samples, so | |
1431 | the expected error in @code{bar}'s run-time is 1 second, | |
1432 | or one percent of the observed value. | |
1433 | It is likely to | |
1434 | vary this much @emph{on the average} from one profiling run to the next. | |
1435 | (@emph{Sometimes} it will vary more.) | |
1436 | ||
1437 | This does not mean that a small run-time figure is devoid of information. | |
1438 | If the program's @emph{total} run-time is large, a small run-time for one | |
1439 | function does tell you that that function used an insignificant fraction of | |
1440 | the whole program's time. Usually this means it is not worth optimizing. | |
1441 | ||
1442 | One way to get more accuracy is to give your program more (but similar) | |
1443 | input data so it will take longer. Another way is to combine the data from | |
1444 | several runs, using the @samp{-s} option of @code{gprof}. Here is how: | |
1445 | ||
1446 | @enumerate | |
1447 | @item | |
1448 | Run your program once. | |
1449 | ||
1450 | @item | |
1451 | Issue the command @samp{mv gmon.out gmon.sum}. | |
1452 | ||
1453 | @item | |
1454 | Run your program again, the same as before. | |
1455 | ||
1456 | @item | |
1457 | Merge the new data in @file{gmon.out} into @file{gmon.sum} with this command: | |
1458 | ||
1459 | @example | |
1460 | gprof -s @var{executable-file} gmon.out gmon.sum | |
1461 | @end example | |
1462 | ||
1463 | @item | |
1464 | Repeat the last two steps as often as you wish. | |
1465 | ||
1466 | @item | |
1467 | Analyze the cumulative data using this command: | |
1468 | ||
1469 | @example | |
1470 | gprof @var{executable-file} gmon.sum > @var{output-file} | |
1471 | @end example | |
1472 | @end enumerate | |
1473 | ||
1474 | @node Assumptions,,Sampling Error,Inaccuracy | |
1475 | @section Estimating @code{children} Times | |
1476 | ||
1477 | Some of the figures in the call graph are estimates---for example, the | |
1478 | @code{children} time values and all the the time figures in caller and | |
1479 | subroutine lines. | |
1480 | ||
1481 | There is no direct information about these measurements in the profile | |
1482 | data itself. Instead, @code{gprof} estimates them by making an assumption | |
1483 | about your program that might or might not be true. | |
1484 | ||
1485 | The assumption made is that the average time spent in each call to any | |
1486 | function @code{foo} is not correlated with who called @code{foo}. If | |
1487 | @code{foo} used 5 seconds in all, and 2/5 of the calls to @code{foo} came | |
1488 | from @code{a}, then @code{foo} contributes 2 seconds to @code{a}'s | |
1489 | @code{children} time, by assumption. | |
1490 | ||
1491 | This assumption is usually true enough, but for some programs it is far | |
1492 | from true. Suppose that @code{foo} returns very quickly when its argument | |
1493 | is zero; suppose that @code{a} always passes zero as an argument, while | |
1494 | other callers of @code{foo} pass other arguments. In this program, all the | |
1495 | time spent in @code{foo} is in the calls from callers other than @code{a}. | |
1496 | But @code{gprof} has no way of knowing this; it will blindly and | |
1497 | incorrectly charge 2 seconds of time in @code{foo} to the children of | |
1498 | @code{a}. | |
1499 | ||
1500 | @c FIXME - has this been fixed? | |
1501 | We hope some day to put more complete data into @file{gmon.out}, so that | |
1502 | this assumption is no longer needed, if we can figure out how. For the | |
1503 | nonce, the estimated figures are usually more useful than misleading. | |
1504 | ||
1505 | @node How do I? | |
1506 | @chapter Answers to Common Questions | |
1507 | ||
1508 | @table @asis | |
1509 | @item How do I find which lines in my program were executed the most times? | |
1510 | ||
1511 | Compile your program with basic-block counting enabled, run it, then | |
1512 | use the following pipeline: | |
1513 | ||
1514 | @example | |
1515 | gprof -l -C @var{objfile} | sort -k 3 -n -r | |
1516 | @end example | |
1517 | ||
1518 | This listing will show you the lines in your code executed most often, | |
1519 | but not necessarily those that consumed the most time. | |
1520 | ||
1521 | @item How do I find which lines in my program called a particular function? | |
1522 | ||
5af11cab | 1523 | Use @samp{gprof -l} and lookup the function in the call graph. |
252b5132 RH |
1524 | The callers will be broken down by function and line number. |
1525 | ||
1526 | @item How do I analyze a program that runs for less than a second? | |
1527 | ||
1528 | Try using a shell script like this one: | |
1529 | ||
1530 | @example | |
1531 | for i in `seq 1 100`; do | |
1532 | fastprog | |
1533 | mv gmon.out gmon.out.$i | |
1534 | done | |
1535 | ||
1536 | gprof -s fastprog gmon.out.* | |
1537 | ||
1538 | gprof fastprog gmon.sum | |
1539 | @end example | |
1540 | ||
1541 | If your program is completely deterministic, all the call counts | |
1542 | will be simple multiples of 100 (i.e. a function called once in | |
1543 | each run will appear with a call count of 100). | |
1544 | ||
1545 | @end table | |
1546 | ||
1547 | @node Incompatibilities | |
1548 | @chapter Incompatibilities with Unix @code{gprof} | |
1549 | ||
1550 | @sc{gnu} @code{gprof} and Berkeley Unix @code{gprof} use the same data | |
1551 | file @file{gmon.out}, and provide essentially the same information. But | |
1552 | there are a few differences. | |
1553 | ||
1554 | @itemize @bullet | |
1555 | @item | |
1556 | @sc{gnu} @code{gprof} uses a new, generalized file format with support | |
1557 | for basic-block execution counts and non-realtime histograms. A magic | |
1558 | cookie and version number allows @code{gprof} to easily identify | |
1559 | new style files. Old BSD-style files can still be read. | |
1560 | @xref{File Format}. | |
1561 | ||
1562 | @item | |
1563 | For a recursive function, Unix @code{gprof} lists the function as a | |
1564 | parent and as a child, with a @code{calls} field that lists the number | |
1565 | of recursive calls. @sc{gnu} @code{gprof} omits these lines and puts | |
1566 | the number of recursive calls in the primary line. | |
1567 | ||
1568 | @item | |
1569 | When a function is suppressed from the call graph with @samp{-e}, @sc{gnu} | |
1570 | @code{gprof} still lists it as a subroutine of functions that call it. | |
1571 | ||
1572 | @item | |
1573 | @sc{gnu} @code{gprof} accepts the @samp{-k} with its argument | |
1574 | in the form @samp{from/to}, instead of @samp{from to}. | |
1575 | ||
1576 | @item | |
1577 | In the annotated source listing, | |
1578 | if there are multiple basic blocks on the same line, | |
5af11cab | 1579 | @sc{gnu} @code{gprof} prints all of their counts, separated by commas. |
252b5132 RH |
1580 | |
1581 | @ignore - it does this now | |
1582 | @item | |
1583 | The function names printed in @sc{gnu} @code{gprof} output do not include | |
1584 | the leading underscores that are added internally to the front of all | |
1585 | C identifiers on many operating systems. | |
1586 | @end ignore | |
1587 | ||
1588 | @item | |
1589 | The blurbs, field widths, and output formats are different. @sc{gnu} | |
1590 | @code{gprof} prints blurbs after the tables, so that you can see the | |
1591 | tables without skipping the blurbs. | |
1592 | @end itemize | |
1593 | ||
1594 | @node Details | |
1595 | @chapter Details of Profiling | |
1596 | ||
1597 | @menu | |
5af11cab | 1598 | * Implementation:: How a program collects profiling information |
252b5132 RH |
1599 | * File Format:: Format of @samp{gmon.out} files |
1600 | * Internals:: @code{gprof}'s internal operation | |
1601 | * Debugging:: Using @code{gprof}'s @samp{-d} option | |
1602 | @end menu | |
1603 | ||
1604 | @node Implementation,File Format,,Details | |
1605 | @section Implementation of Profiling | |
1606 | ||
1607 | Profiling works by changing how every function in your program is compiled | |
1608 | so that when it is called, it will stash away some information about where | |
1609 | it was called from. From this, the profiler can figure out what function | |
1610 | called it, and can count how many times it was called. This change is made | |
1611 | by the compiler when your program is compiled with the @samp{-pg} option, | |
1612 | which causes every function to call @code{mcount} | |
1613 | (or @code{_mcount}, or @code{__mcount}, depending on the OS and compiler) | |
1614 | as one of its first operations. | |
1615 | ||
1616 | The @code{mcount} routine, included in the profiling library, | |
1617 | is responsible for recording in an in-memory call graph table | |
1618 | both its parent routine (the child) and its parent's parent. This is | |
1619 | typically done by examining the stack frame to find both | |
1620 | the address of the child, and the return address in the original parent. | |
5af11cab | 1621 | Since this is a very machine-dependent operation, @code{mcount} |
252b5132 RH |
1622 | itself is typically a short assembly-language stub routine |
1623 | that extracts the required | |
1624 | information, and then calls @code{__mcount_internal} | |
1625 | (a normal C function) with two arguments - @code{frompc} and @code{selfpc}. | |
1626 | @code{__mcount_internal} is responsible for maintaining | |
1627 | the in-memory call graph, which records @code{frompc}, @code{selfpc}, | |
5af11cab | 1628 | and the number of times each of these call arcs was traversed. |
252b5132 RH |
1629 | |
1630 | GCC Version 2 provides a magical function (@code{__builtin_return_address}), | |
1631 | which allows a generic @code{mcount} function to extract the | |
1632 | required information from the stack frame. However, on some | |
1633 | architectures, most notably the SPARC, using this builtin can be | |
1634 | very computationally expensive, and an assembly language version | |
1635 | of @code{mcount} is used for performance reasons. | |
1636 | ||
1637 | Number-of-calls information for library routines is collected by using a | |
1638 | special version of the C library. The programs in it are the same as in | |
1639 | the usual C library, but they were compiled with @samp{-pg}. If you | |
1640 | link your program with @samp{gcc @dots{} -pg}, it automatically uses the | |
1641 | profiling version of the library. | |
1642 | ||
1643 | Profiling also involves watching your program as it runs, and keeping a | |
1644 | histogram of where the program counter happens to be every now and then. | |
1645 | Typically the program counter is looked at around 100 times per second of | |
1646 | run time, but the exact frequency may vary from system to system. | |
1647 | ||
1648 | This is done is one of two ways. Most UNIX-like operating systems | |
1649 | provide a @code{profil()} system call, which registers a memory | |
1650 | array with the kernel, along with a scale | |
1651 | factor that determines how the program's address space maps | |
1652 | into the array. | |
1653 | Typical scaling values cause every 2 to 8 bytes of address space | |
1654 | to map into a single array slot. | |
1655 | On every tick of the system clock | |
1656 | (assuming the profiled program is running), the value of the | |
1657 | program counter is examined and the corresponding slot in | |
1658 | the memory array is incremented. Since this is done in the kernel, | |
1659 | which had to interrupt the process anyway to handle the clock | |
1660 | interrupt, very little additional system overhead is required. | |
1661 | ||
1662 | However, some operating systems, most notably Linux 2.0 (and earlier), | |
1663 | do not provide a @code{profil()} system call. On such a system, | |
1664 | arrangements are made for the kernel to periodically deliver | |
1665 | a signal to the process (typically via @code{setitimer()}), | |
1666 | which then performs the same operation of examining the | |
1667 | program counter and incrementing a slot in the memory array. | |
1668 | Since this method requires a signal to be delivered to | |
1669 | user space every time a sample is taken, it uses considerably | |
1670 | more overhead than kernel-based profiling. Also, due to the | |
1671 | added delay required to deliver the signal, this method is | |
1672 | less accurate as well. | |
1673 | ||
1674 | A special startup routine allocates memory for the histogram and | |
1675 | either calls @code{profil()} or sets up | |
1676 | a clock signal handler. | |
1677 | This routine (@code{monstartup}) can be invoked in several ways. | |
1678 | On Linux systems, a special profiling startup file @code{gcrt0.o}, | |
1679 | which invokes @code{monstartup} before @code{main}, | |
1680 | is used instead of the default @code{crt0.o}. | |
1681 | Use of this special startup file is one of the effects | |
1682 | of using @samp{gcc @dots{} -pg} to link. | |
1683 | On SPARC systems, no special startup files are used. | |
1684 | Rather, the @code{mcount} routine, when it is invoked for | |
1685 | the first time (typically when @code{main} is called), | |
1686 | calls @code{monstartup}. | |
1687 | ||
1688 | If the compiler's @samp{-a} option was used, basic-block counting | |
1689 | is also enabled. Each object file is then compiled with a static array | |
1690 | of counts, initially zero. | |
1691 | In the executable code, every time a new basic-block begins | |
1692 | (i.e. when an @code{if} statement appears), an extra instruction | |
1693 | is inserted to increment the corresponding count in the array. | |
1694 | At compile time, a paired array was constructed that recorded | |
1695 | the starting address of each basic-block. Taken together, | |
1696 | the two arrays record the starting address of every basic-block, | |
1697 | along with the number of times it was executed. | |
1698 | ||
1699 | The profiling library also includes a function (@code{mcleanup}) which is | |
1700 | typically registered using @code{atexit()} to be called as the | |
1701 | program exits, and is responsible for writing the file @file{gmon.out}. | |
1702 | Profiling is turned off, various headers are output, and the histogram | |
1703 | is written, followed by the call-graph arcs and the basic-block counts. | |
1704 | ||
1705 | The output from @code{gprof} gives no indication of parts of your program that | |
1706 | are limited by I/O or swapping bandwidth. This is because samples of the | |
1707 | program counter are taken at fixed intervals of the program's run time. | |
1708 | Therefore, the | |
1709 | time measurements in @code{gprof} output say nothing about time that your | |
1710 | program was not running. For example, a part of the program that creates | |
1711 | so much data that it cannot all fit in physical memory at once may run very | |
1712 | slowly due to thrashing, but @code{gprof} will say it uses little time. On | |
1713 | the other hand, sampling by run time has the advantage that the amount of | |
1714 | load due to other users won't directly affect the output you get. | |
1715 | ||
1716 | @node File Format,Internals,Implementation,Details | |
1717 | @section Profiling Data File Format | |
1718 | ||
1719 | The old BSD-derived file format used for profile data does not contain a | |
1720 | magic cookie that allows to check whether a data file really is a | |
5af11cab | 1721 | @code{gprof} file. Furthermore, it does not provide a version number, thus |
252b5132 RH |
1722 | rendering changes to the file format almost impossible. @sc{gnu} @code{gprof} |
1723 | uses a new file format that provides these features. For backward | |
1724 | compatibility, @sc{gnu} @code{gprof} continues to support the old BSD-derived | |
1725 | format, but not all features are supported with it. For example, | |
1726 | basic-block execution counts cannot be accommodated by the old file | |
1727 | format. | |
1728 | ||
1729 | The new file format is defined in header file @file{gmon_out.h}. It | |
1730 | consists of a header containing the magic cookie and a version number, | |
1731 | as well as some spare bytes available for future extensions. All data | |
1732 | in a profile data file is in the native format of the host on which | |
1733 | the profile was collected. @sc{gnu} @code{gprof} adapts automatically to the | |
1734 | byte-order in use. | |
1735 | ||
1736 | In the new file format, the header is followed by a sequence of | |
1737 | records. Currently, there are three different record types: histogram | |
1738 | records, call-graph arc records, and basic-block execution count | |
1739 | records. Each file can contain any number of each record type. When | |
1740 | reading a file, @sc{gnu} @code{gprof} will ensure records of the same type are | |
1741 | compatible with each other and compute the union of all records. For | |
1742 | example, for basic-block execution counts, the union is simply the sum | |
1743 | of all execution counts for each basic-block. | |
1744 | ||
1745 | @subsection Histogram Records | |
1746 | ||
1747 | Histogram records consist of a header that is followed by an array of | |
1748 | bins. The header contains the text-segment range that the histogram | |
1749 | spans, the size of the histogram in bytes (unlike in the old BSD | |
1750 | format, this does not include the size of the header), the rate of the | |
1751 | profiling clock, and the physical dimension that the bin counts | |
1752 | represent after being scaled by the profiling clock rate. The | |
1753 | physical dimension is specified in two parts: a long name of up to 15 | |
1754 | characters and a single character abbreviation. For example, a | |
1755 | histogram representing real-time would specify the long name as | |
1756 | "seconds" and the abbreviation as "s". This feature is useful for | |
1757 | architectures that support performance monitor hardware (which, | |
1758 | fortunately, is becoming increasingly common). For example, under DEC | |
1759 | OSF/1, the "uprofile" command can be used to produce a histogram of, | |
1760 | say, instruction cache misses. In this case, the dimension in the | |
1761 | histogram header could be set to "i-cache misses" and the abbreviation | |
1762 | could be set to "1" (because it is simply a count, not a physical | |
1763 | dimension). Also, the profiling rate would have to be set to 1 in | |
1764 | this case. | |
1765 | ||
1766 | Histogram bins are 16-bit numbers and each bin represent an equal | |
1767 | amount of text-space. For example, if the text-segment is one | |
1768 | thousand bytes long and if there are ten bins in the histogram, each | |
1769 | bin represents one hundred bytes. | |
1770 | ||
1771 | ||
1772 | @subsection Call-Graph Records | |
1773 | ||
1774 | Call-graph records have a format that is identical to the one used in | |
1775 | the BSD-derived file format. It consists of an arc in the call graph | |
1776 | and a count indicating the number of times the arc was traversed | |
1777 | during program execution. Arcs are specified by a pair of addresses: | |
1778 | the first must be within caller's function and the second must be | |
1779 | within the callee's function. When performing profiling at the | |
1780 | function level, these addresses can point anywhere within the | |
1781 | respective function. However, when profiling at the line-level, it is | |
1782 | better if the addresses are as close to the call-site/entry-point as | |
1783 | possible. This will ensure that the line-level call-graph is able to | |
1784 | identify exactly which line of source code performed calls to a | |
1785 | function. | |
1786 | ||
1787 | @subsection Basic-Block Execution Count Records | |
1788 | ||
1789 | Basic-block execution count records consist of a header followed by a | |
1790 | sequence of address/count pairs. The header simply specifies the | |
1791 | length of the sequence. In an address/count pair, the address | |
1792 | identifies a basic-block and the count specifies the number of times | |
1793 | that basic-block was executed. Any address within the basic-address can | |
1794 | be used. | |
1795 | ||
1796 | @node Internals,Debugging,File Format,Details | |
1797 | @section @code{gprof}'s Internal Operation | |
1798 | ||
1799 | Like most programs, @code{gprof} begins by processing its options. | |
1800 | During this stage, it may building its symspec list | |
1801 | (@code{sym_ids.c:sym_id_add}), if | |
1802 | options are specified which use symspecs. | |
1803 | @code{gprof} maintains a single linked list of symspecs, | |
1804 | which will eventually get turned into 12 symbol tables, | |
1805 | organized into six include/exclude pairs - one | |
1806 | pair each for the flat profile (INCL_FLAT/EXCL_FLAT), | |
1807 | the call graph arcs (INCL_ARCS/EXCL_ARCS), | |
1808 | printing in the call graph (INCL_GRAPH/EXCL_GRAPH), | |
1809 | timing propagation in the call graph (INCL_TIME/EXCL_TIME), | |
1810 | the annotated source listing (INCL_ANNO/EXCL_ANNO), | |
1811 | and the execution count listing (INCL_EXEC/EXCL_EXEC). | |
1812 | ||
1813 | After option processing, @code{gprof} finishes | |
1814 | building the symspec list by adding all the symspecs in | |
1815 | @code{default_excluded_list} to the exclude lists | |
1816 | EXCL_TIME and EXCL_GRAPH, and if line-by-line profiling is specified, | |
1817 | EXCL_FLAT as well. | |
1818 | These default excludes are not added to EXCL_ANNO, EXCL_ARCS, and EXCL_EXEC. | |
1819 | ||
1820 | Next, the BFD library is called to open the object file, | |
1821 | verify that it is an object file, | |
1822 | and read its symbol table (@code{core.c:core_init}), | |
1823 | using @code{bfd_canonicalize_symtab} after mallocing | |
5af11cab | 1824 | an appropriately sized array of symbols. At this point, |
252b5132 RH |
1825 | function mappings are read (if the @samp{--file-ordering} option |
1826 | has been specified), and the core text space is read into | |
1827 | memory (if the @samp{-c} option was given). | |
1828 | ||
1829 | @code{gprof}'s own symbol table, an array of Sym structures, | |
1830 | is now built. | |
1831 | This is done in one of two ways, by one of two routines, depending | |
1832 | on whether line-by-line profiling (@samp{-l} option) has been | |
1833 | enabled. | |
1834 | For normal profiling, the BFD canonical symbol table is scanned. | |
1835 | For line-by-line profiling, every | |
1836 | text space address is examined, and a new symbol table entry | |
1837 | gets created every time the line number changes. | |
1838 | In either case, two passes are made through the symbol | |
1839 | table - one to count the size of the symbol table required, | |
1840 | and the other to actually read the symbols. In between the | |
1841 | two passes, a single array of type @code{Sym} is created of | |
5af11cab | 1842 | the appropriate length. |
252b5132 RH |
1843 | Finally, @code{symtab.c:symtab_finalize} |
1844 | is called to sort the symbol table and remove duplicate entries | |
1845 | (entries with the same memory address). | |
1846 | ||
1847 | The symbol table must be a contiguous array for two reasons. | |
1848 | First, the @code{qsort} library function (which sorts an array) | |
1849 | will be used to sort the symbol table. | |
1850 | Also, the symbol lookup routine (@code{symtab.c:sym_lookup}), | |
1851 | which finds symbols | |
1852 | based on memory address, uses a binary search algorithm | |
1853 | which requires the symbol table to be a sorted array. | |
1854 | Function symbols are indicated with an @code{is_func} flag. | |
1855 | Line number symbols have no special flags set. | |
1856 | Additionally, a symbol can have an @code{is_static} flag | |
1857 | to indicate that it is a local symbol. | |
1858 | ||
1859 | With the symbol table read, the symspecs can now be translated | |
1860 | into Syms (@code{sym_ids.c:sym_id_parse}). Remember that a single | |
1861 | symspec can match multiple symbols. | |
1862 | An array of symbol tables | |
1863 | (@code{syms}) is created, each entry of which is a symbol table | |
1864 | of Syms to be included or excluded from a particular listing. | |
1865 | The master symbol table and the symspecs are examined by nested | |
1866 | loops, and every symbol that matches a symspec is inserted | |
1867 | into the appropriate syms table. This is done twice, once to | |
1868 | count the size of each required symbol table, and again to build | |
1869 | the tables, which have been malloced between passes. | |
1870 | From now on, to determine whether a symbol is on an include | |
1871 | or exclude symspec list, @code{gprof} simply uses its | |
1872 | standard symbol lookup routine on the appropriate table | |
1873 | in the @code{syms} array. | |
1874 | ||
1875 | Now the profile data file(s) themselves are read | |
1876 | (@code{gmon_io.c:gmon_out_read}), | |
1877 | first by checking for a new-style @samp{gmon.out} header, | |
1878 | then assuming this is an old-style BSD @samp{gmon.out} | |
1879 | if the magic number test failed. | |
1880 | ||
1881 | New-style histogram records are read by @code{hist.c:hist_read_rec}. | |
1882 | For the first histogram record, allocate a memory array to hold | |
1883 | all the bins, and read them in. | |
1884 | When multiple profile data files (or files with multiple histogram | |
1885 | records) are read, the starting address, ending address, number | |
1886 | of bins and sampling rate must match between the various histograms, | |
1887 | or a fatal error will result. | |
1888 | If everything matches, just sum the additional histograms into | |
1889 | the existing in-memory array. | |
1890 | ||
1891 | As each call graph record is read (@code{call_graph.c:cg_read_rec}), | |
1892 | the parent and child addresses | |
1893 | are matched to symbol table entries, and a call graph arc is | |
1894 | created by @code{cg_arcs.c:arc_add}, unless the arc fails a symspec | |
1895 | check against INCL_ARCS/EXCL_ARCS. As each arc is added, | |
1896 | a linked list is maintained of the parent's child arcs, and of the child's | |
1897 | parent arcs. | |
1898 | Both the child's call count and the arc's call count are | |
1899 | incremented by the record's call count. | |
1900 | ||
1901 | Basic-block records are read (@code{basic_blocks.c:bb_read_rec}), | |
1902 | but only if line-by-line profiling has been selected. | |
1903 | Each basic-block address is matched to a corresponding line | |
1904 | symbol in the symbol table, and an entry made in the symbol's | |
1905 | bb_addr and bb_calls arrays. Again, if multiple basic-block | |
1906 | records are present for the same address, the call counts | |
1907 | are cumulative. | |
1908 | ||
1909 | A gmon.sum file is dumped, if requested (@code{gmon_io.c:gmon_out_write}). | |
1910 | ||
1911 | If histograms were present in the data files, assign them to symbols | |
1912 | (@code{hist.c:hist_assign_samples}) by iterating over all the sample | |
1913 | bins and assigning them to symbols. Since the symbol table | |
1914 | is sorted in order of ascending memory addresses, we can | |
1915 | simple follow along in the symbol table as we make our pass | |
1916 | over the sample bins. | |
1917 | This step includes a symspec check against INCL_FLAT/EXCL_FLAT. | |
1918 | Depending on the histogram | |
1919 | scale factor, a sample bin may span multiple symbols, | |
1920 | in which case a fraction of the sample count is allocated | |
1921 | to each symbol, proportional to the degree of overlap. | |
1922 | This effect is rare for normal profiling, but overlaps | |
1923 | are more common during line-by-line profiling, and can | |
1924 | cause each of two adjacent lines to be credited with half | |
1925 | a hit, for example. | |
1926 | ||
1927 | If call graph data is present, @code{cg_arcs.c:cg_assemble} is called. | |
5af11cab | 1928 | First, if @samp{-c} was specified, a machine-dependent |
252b5132 RH |
1929 | routine (@code{find_call}) scans through each symbol's machine code, |
1930 | looking for subroutine call instructions, and adding them | |
1931 | to the call graph with a zero call count. | |
1932 | A topological sort is performed by depth-first numbering | |
1933 | all the symbols (@code{cg_dfn.c:cg_dfn}), so that | |
1934 | children are always numbered less than their parents, | |
1935 | then making a array of pointers into the symbol table and sorting it into | |
1936 | numerical order, which is reverse topological | |
1937 | order (children appear before parents). | |
1938 | Cycles are also detected at this point, all members | |
1939 | of which are assigned the same topological number. | |
1940 | Two passes are now made through this sorted array of symbol pointers. | |
1941 | The first pass, from end to beginning (parents to children), | |
5af11cab | 1942 | computes the fraction of child time to propagate to each parent |
252b5132 RH |
1943 | and a print flag. |
1944 | The print flag reflects symspec handling of INCL_GRAPH/EXCL_GRAPH, | |
1945 | with a parent's include or exclude (print or no print) property | |
1946 | being propagated to its children, unless they themselves explicitly appear | |
1947 | in INCL_GRAPH or EXCL_GRAPH. | |
1948 | A second pass, from beginning to end (children to parents) actually | |
5af11cab | 1949 | propagates the timings along the call graph, subject |
252b5132 RH |
1950 | to a check against INCL_TIME/EXCL_TIME. |
1951 | With the print flag, fractions, and timings now stored in the symbol | |
1952 | structures, the topological sort array is now discarded, and a | |
1953 | new array of pointers is assembled, this time sorted by propagated time. | |
1954 | ||
1955 | Finally, print the various outputs the user requested, which is now fairly | |
1956 | straightforward. The call graph (@code{cg_print.c:cg_print}) and | |
1957 | flat profile (@code{hist.c:hist_print}) are regurgitations of values | |
1958 | already computed. The annotated source listing | |
1959 | (@code{basic_blocks.c:print_annotated_source}) uses basic-block | |
1960 | information, if present, to label each line of code with call counts, | |
1961 | otherwise only the function call counts are presented. | |
1962 | ||
1963 | The function ordering code is marginally well documented | |
1964 | in the source code itself (@code{cg_print.c}). Basically, | |
1965 | the functions with the most use and the most parents are | |
1966 | placed first, followed by other functions with the most use, | |
1967 | followed by lower use functions, followed by unused functions | |
1968 | at the end. | |
1969 | ||
1970 | @node Debugging,,Internals,Details | |
1971 | @subsection Debugging @code{gprof} | |
1972 | ||
1973 | If @code{gprof} was compiled with debugging enabled, | |
1974 | the @samp{-d} option triggers debugging output | |
1975 | (to stdout) which can be helpful in understanding its operation. | |
1976 | The debugging number specified is interpreted as a sum of the following | |
1977 | options: | |
1978 | ||
1979 | @table @asis | |
1980 | @item 2 - Topological sort | |
1981 | Monitor depth-first numbering of symbols during call graph analysis | |
1982 | @item 4 - Cycles | |
1983 | Shows symbols as they are identified as cycle heads | |
1984 | @item 16 - Tallying | |
1985 | As the call graph arcs are read, show each arc and how | |
1986 | the total calls to each function are tallied | |
1987 | @item 32 - Call graph arc sorting | |
1988 | Details sorting individual parents/children within each call graph entry | |
1989 | @item 64 - Reading histogram and call graph records | |
1990 | Shows address ranges of histograms as they are read, and each | |
1991 | call graph arc | |
1992 | @item 128 - Symbol table | |
1993 | Reading, classifying, and sorting the symbol table from the object file. | |
1994 | For line-by-line profiling (@samp{-l} option), also shows line numbers | |
1995 | being assigned to memory addresses. | |
1996 | @item 256 - Static call graph | |
1997 | Trace operation of @samp{-c} option | |
1998 | @item 512 - Symbol table and arc table lookups | |
1999 | Detail operation of lookup routines | |
2000 | @item 1024 - Call graph propagation | |
2001 | Shows how function times are propagated along the call graph | |
2002 | @item 2048 - Basic-blocks | |
2003 | Shows basic-block records as they are read from profile data | |
2004 | (only meaningful with @samp{-l} option) | |
2005 | @item 4096 - Symspecs | |
2006 | Shows symspec-to-symbol pattern matching operation | |
2007 | @item 8192 - Annotate source | |
2008 | Tracks operation of @samp{-A} option | |
2009 | @end table | |
2010 | ||
cf055d54 NC |
2011 | @node GNU Free Documentation License |
2012 | @chapter GNU Free Documentation License | |
2013 | ||
2014 | GNU Free Documentation License | |
2015 | ||
2016 | Version 1.1, March 2000 | |
2017 | ||
2018 | Copyright (C) 2000 Free Software Foundation, Inc. | |
2019 | 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
2020 | ||
2021 | Everyone is permitted to copy and distribute verbatim copies | |
2022 | of this license document, but changing it is not allowed. | |
2023 | ||
2024 | ||
2025 | 0. PREAMBLE | |
2026 | ||
2027 | The purpose of this License is to make a manual, textbook, or other | |
2028 | written document "free" in the sense of freedom: to assure everyone | |
2029 | the effective freedom to copy and redistribute it, with or without | |
2030 | modifying it, either commercially or noncommercially. Secondarily, | |
2031 | this License preserves for the author and publisher a way to get | |
2032 | credit for their work, while not being considered responsible for | |
2033 | modifications made by others. | |
2034 | ||
2035 | This License is a kind of "copyleft", which means that derivative | |
2036 | works of the document must themselves be free in the same sense. It | |
2037 | complements the GNU General Public License, which is a copyleft | |
2038 | license designed for free software. | |
2039 | ||
2040 | We have designed this License in order to use it for manuals for free | |
2041 | software, because free software needs free documentation: a free | |
2042 | program should come with manuals providing the same freedoms that the | |
2043 | software does. But this License is not limited to software manuals; | |
2044 | it can be used for any textual work, regardless of subject matter or | |
2045 | whether it is published as a printed book. We recommend this License | |
2046 | principally for works whose purpose is instruction or reference. | |
2047 | ||
2048 | ||
2049 | 1. APPLICABILITY AND DEFINITIONS | |
2050 | ||
2051 | This License applies to any manual or other work that contains a | |
2052 | notice placed by the copyright holder saying it can be distributed | |
2053 | under the terms of this License. The "Document", below, refers to any | |
2054 | such manual or work. Any member of the public is a licensee, and is | |
2055 | addressed as "you". | |
2056 | ||
2057 | A "Modified Version" of the Document means any work containing the | |
2058 | Document or a portion of it, either copied verbatim, or with | |
2059 | modifications and/or translated into another language. | |
2060 | ||
2061 | A "Secondary Section" is a named appendix or a front-matter section of | |
2062 | the Document that deals exclusively with the relationship of the | |
2063 | publishers or authors of the Document to the Document's overall subject | |
2064 | (or to related matters) and contains nothing that could fall directly | |
2065 | within that overall subject. (For example, if the Document is in part a | |
2066 | textbook of mathematics, a Secondary Section may not explain any | |
2067 | mathematics.) The relationship could be a matter of historical | |
2068 | connection with the subject or with related matters, or of legal, | |
2069 | commercial, philosophical, ethical or political position regarding | |
2070 | them. | |
2071 | ||
2072 | The "Invariant Sections" are certain Secondary Sections whose titles | |
2073 | are designated, as being those of Invariant Sections, in the notice | |
2074 | that says that the Document is released under this License. | |
2075 | ||
2076 | The "Cover Texts" are certain short passages of text that are listed, | |
2077 | as Front-Cover Texts or Back-Cover Texts, in the notice that says that | |
2078 | the Document is released under this License. | |
2079 | ||
2080 | A "Transparent" copy of the Document means a machine-readable copy, | |
2081 | represented in a format whose specification is available to the | |
2082 | general public, whose contents can be viewed and edited directly and | |
2083 | straightforwardly with generic text editors or (for images composed of | |
2084 | pixels) generic paint programs or (for drawings) some widely available | |
2085 | drawing editor, and that is suitable for input to text formatters or | |
2086 | for automatic translation to a variety of formats suitable for input | |
2087 | to text formatters. A copy made in an otherwise Transparent file | |
2088 | format whose markup has been designed to thwart or discourage | |
2089 | subsequent modification by readers is not Transparent. A copy that is | |
2090 | not "Transparent" is called "Opaque". | |
2091 | ||
2092 | Examples of suitable formats for Transparent copies include plain | |
2093 | ASCII without markup, Texinfo input format, LaTeX input format, SGML | |
2094 | or XML using a publicly available DTD, and standard-conforming simple | |
2095 | HTML designed for human modification. Opaque formats include | |
2096 | PostScript, PDF, proprietary formats that can be read and edited only | |
2097 | by proprietary word processors, SGML or XML for which the DTD and/or | |
2098 | processing tools are not generally available, and the | |
2099 | machine-generated HTML produced by some word processors for output | |
2100 | purposes only. | |
2101 | ||
2102 | The "Title Page" means, for a printed book, the title page itself, | |
2103 | plus such following pages as are needed to hold, legibly, the material | |
2104 | this License requires to appear in the title page. For works in | |
2105 | formats which do not have any title page as such, "Title Page" means | |
2106 | the text near the most prominent appearance of the work's title, | |
2107 | preceding the beginning of the body of the text. | |
2108 | ||
2109 | ||
2110 | 2. VERBATIM COPYING | |
2111 | ||
2112 | You may copy and distribute the Document in any medium, either | |
2113 | commercially or noncommercially, provided that this License, the | |
2114 | copyright notices, and the license notice saying this License applies | |
2115 | to the Document are reproduced in all copies, and that you add no other | |
2116 | conditions whatsoever to those of this License. You may not use | |
2117 | technical measures to obstruct or control the reading or further | |
2118 | copying of the copies you make or distribute. However, you may accept | |
2119 | compensation in exchange for copies. If you distribute a large enough | |
2120 | number of copies you must also follow the conditions in section 3. | |
2121 | ||
2122 | You may also lend copies, under the same conditions stated above, and | |
2123 | you may publicly display copies. | |
2124 | ||
2125 | ||
2126 | 3. COPYING IN QUANTITY | |
2127 | ||
2128 | If you publish printed copies of the Document numbering more than 100, | |
2129 | and the Document's license notice requires Cover Texts, you must enclose | |
2130 | the copies in covers that carry, clearly and legibly, all these Cover | |
2131 | Texts: Front-Cover Texts on the front cover, and Back-Cover Texts on | |
2132 | the back cover. Both covers must also clearly and legibly identify | |
2133 | you as the publisher of these copies. The front cover must present | |
2134 | the full title with all words of the title equally prominent and | |
2135 | visible. You may add other material on the covers in addition. | |
2136 | Copying with changes limited to the covers, as long as they preserve | |
2137 | the title of the Document and satisfy these conditions, can be treated | |
2138 | as verbatim copying in other respects. | |
2139 | ||
2140 | If the required texts for either cover are too voluminous to fit | |
2141 | legibly, you should put the first ones listed (as many as fit | |
2142 | reasonably) on the actual cover, and continue the rest onto adjacent | |
2143 | pages. | |
2144 | ||
2145 | If you publish or distribute Opaque copies of the Document numbering | |
2146 | more than 100, you must either include a machine-readable Transparent | |
2147 | copy along with each Opaque copy, or state in or with each Opaque copy | |
2148 | a publicly-accessible computer-network location containing a complete | |
2149 | Transparent copy of the Document, free of added material, which the | |
2150 | general network-using public has access to download anonymously at no | |
2151 | charge using public-standard network protocols. If you use the latter | |
2152 | option, you must take reasonably prudent steps, when you begin | |
2153 | distribution of Opaque copies in quantity, to ensure that this | |
2154 | Transparent copy will remain thus accessible at the stated location | |
2155 | until at least one year after the last time you distribute an Opaque | |
2156 | copy (directly or through your agents or retailers) of that edition to | |
2157 | the public. | |
2158 | ||
2159 | It is requested, but not required, that you contact the authors of the | |
2160 | Document well before redistributing any large number of copies, to give | |
2161 | them a chance to provide you with an updated version of the Document. | |
2162 | ||
2163 | ||
2164 | 4. MODIFICATIONS | |
2165 | ||
2166 | You may copy and distribute a Modified Version of the Document under | |
2167 | the conditions of sections 2 and 3 above, provided that you release | |
2168 | the Modified Version under precisely this License, with the Modified | |
2169 | Version filling the role of the Document, thus licensing distribution | |
2170 | and modification of the Modified Version to whoever possesses a copy | |
2171 | of it. In addition, you must do these things in the Modified Version: | |
2172 | ||
2173 | A. Use in the Title Page (and on the covers, if any) a title distinct | |
2174 | from that of the Document, and from those of previous versions | |
2175 | (which should, if there were any, be listed in the History section | |
2176 | of the Document). You may use the same title as a previous version | |
2177 | if the original publisher of that version gives permission. | |
2178 | B. List on the Title Page, as authors, one or more persons or entities | |
2179 | responsible for authorship of the modifications in the Modified | |
2180 | Version, together with at least five of the principal authors of the | |
2181 | Document (all of its principal authors, if it has less than five). | |
2182 | C. State on the Title page the name of the publisher of the | |
2183 | Modified Version, as the publisher. | |
2184 | D. Preserve all the copyright notices of the Document. | |
2185 | E. Add an appropriate copyright notice for your modifications | |
2186 | adjacent to the other copyright notices. | |
2187 | F. Include, immediately after the copyright notices, a license notice | |
2188 | giving the public permission to use the Modified Version under the | |
2189 | terms of this License, in the form shown in the Addendum below. | |
2190 | G. Preserve in that license notice the full lists of Invariant Sections | |
2191 | and required Cover Texts given in the Document's license notice. | |
2192 | H. Include an unaltered copy of this License. | |
2193 | I. Preserve the section entitled "History", and its title, and add to | |
2194 | it an item stating at least the title, year, new authors, and | |
2195 | publisher of the Modified Version as given on the Title Page. If | |
2196 | there is no section entitled "History" in the Document, create one | |
2197 | stating the title, year, authors, and publisher of the Document as | |
2198 | given on its Title Page, then add an item describing the Modified | |
2199 | Version as stated in the previous sentence. | |
2200 | J. Preserve the network location, if any, given in the Document for | |
2201 | public access to a Transparent copy of the Document, and likewise | |
2202 | the network locations given in the Document for previous versions | |
2203 | it was based on. These may be placed in the "History" section. | |
2204 | You may omit a network location for a work that was published at | |
2205 | least four years before the Document itself, or if the original | |
2206 | publisher of the version it refers to gives permission. | |
2207 | K. In any section entitled "Acknowledgements" or "Dedications", | |
2208 | preserve the section's title, and preserve in the section all the | |
2209 | substance and tone of each of the contributor acknowledgements | |
2210 | and/or dedications given therein. | |
2211 | L. Preserve all the Invariant Sections of the Document, | |
2212 | unaltered in their text and in their titles. Section numbers | |
2213 | or the equivalent are not considered part of the section titles. | |
2214 | M. Delete any section entitled "Endorsements". Such a section | |
2215 | may not be included in the Modified Version. | |
2216 | N. Do not retitle any existing section as "Endorsements" | |
2217 | or to conflict in title with any Invariant Section. | |
2218 | ||
2219 | If the Modified Version includes new front-matter sections or | |
2220 | appendices that qualify as Secondary Sections and contain no material | |
2221 | copied from the Document, you may at your option designate some or all | |
2222 | of these sections as invariant. To do this, add their titles to the | |
2223 | list of Invariant Sections in the Modified Version's license notice. | |
2224 | These titles must be distinct from any other section titles. | |
2225 | ||
2226 | You may add a section entitled "Endorsements", provided it contains | |
2227 | nothing but endorsements of your Modified Version by various | |
2228 | parties--for example, statements of peer review or that the text has | |
2229 | been approved by an organization as the authoritative definition of a | |
2230 | standard. | |
2231 | ||
2232 | You may add a passage of up to five words as a Front-Cover Text, and a | |
2233 | passage of up to 25 words as a Back-Cover Text, to the end of the list | |
2234 | of Cover Texts in the Modified Version. Only one passage of | |
2235 | Front-Cover Text and one of Back-Cover Text may be added by (or | |
2236 | through arrangements made by) any one entity. If the Document already | |
2237 | includes a cover text for the same cover, previously added by you or | |
2238 | by arrangement made by the same entity you are acting on behalf of, | |
2239 | you may not add another; but you may replace the old one, on explicit | |
2240 | permission from the previous publisher that added the old one. | |
2241 | ||
2242 | The author(s) and publisher(s) of the Document do not by this License | |
2243 | give permission to use their names for publicity for or to assert or | |
2244 | imply endorsement of any Modified Version. | |
2245 | ||
2246 | ||
2247 | 5. COMBINING DOCUMENTS | |
2248 | ||
2249 | You may combine the Document with other documents released under this | |
2250 | License, under the terms defined in section 4 above for modified | |
2251 | versions, provided that you include in the combination all of the | |
2252 | Invariant Sections of all of the original documents, unmodified, and | |
2253 | list them all as Invariant Sections of your combined work in its | |
2254 | license notice. | |
2255 | ||
2256 | The combined work need only contain one copy of this License, and | |
2257 | multiple identical Invariant Sections may be replaced with a single | |
2258 | copy. If there are multiple Invariant Sections with the same name but | |
2259 | different contents, make the title of each such section unique by | |
2260 | adding at the end of it, in parentheses, the name of the original | |
2261 | author or publisher of that section if known, or else a unique number. | |
2262 | Make the same adjustment to the section titles in the list of | |
2263 | Invariant Sections in the license notice of the combined work. | |
2264 | ||
2265 | In the combination, you must combine any sections entitled "History" | |
2266 | in the various original documents, forming one section entitled | |
2267 | "History"; likewise combine any sections entitled "Acknowledgements", | |
2268 | and any sections entitled "Dedications". You must delete all sections | |
2269 | entitled "Endorsements." | |
2270 | ||
2271 | ||
2272 | 6. COLLECTIONS OF DOCUMENTS | |
2273 | ||
2274 | You may make a collection consisting of the Document and other documents | |
2275 | released under this License, and replace the individual copies of this | |
2276 | License in the various documents with a single copy that is included in | |
2277 | the collection, provided that you follow the rules of this License for | |
2278 | verbatim copying of each of the documents in all other respects. | |
2279 | ||
2280 | You may extract a single document from such a collection, and distribute | |
2281 | it individually under this License, provided you insert a copy of this | |
2282 | License into the extracted document, and follow this License in all | |
2283 | other respects regarding verbatim copying of that document. | |
2284 | ||
2285 | ||
2286 | 7. AGGREGATION WITH INDEPENDENT WORKS | |
2287 | ||
2288 | A compilation of the Document or its derivatives with other separate | |
2289 | and independent documents or works, in or on a volume of a storage or | |
2290 | distribution medium, does not as a whole count as a Modified Version | |
2291 | of the Document, provided no compilation copyright is claimed for the | |
2292 | compilation. Such a compilation is called an "aggregate", and this | |
2293 | License does not apply to the other self-contained works thus compiled | |
2294 | with the Document, on account of their being thus compiled, if they | |
2295 | are not themselves derivative works of the Document. | |
2296 | ||
2297 | If the Cover Text requirement of section 3 is applicable to these | |
2298 | copies of the Document, then if the Document is less than one quarter | |
2299 | of the entire aggregate, the Document's Cover Texts may be placed on | |
2300 | covers that surround only the Document within the aggregate. | |
2301 | Otherwise they must appear on covers around the whole aggregate. | |
2302 | ||
2303 | ||
2304 | 8. TRANSLATION | |
2305 | ||
2306 | Translation is considered a kind of modification, so you may | |
2307 | distribute translations of the Document under the terms of section 4. | |
2308 | Replacing Invariant Sections with translations requires special | |
2309 | permission from their copyright holders, but you may include | |
2310 | translations of some or all Invariant Sections in addition to the | |
2311 | original versions of these Invariant Sections. You may include a | |
2312 | translation of this License provided that you also include the | |
2313 | original English version of this License. In case of a disagreement | |
2314 | between the translation and the original English version of this | |
2315 | License, the original English version will prevail. | |
2316 | ||
2317 | ||
2318 | 9. TERMINATION | |
2319 | ||
2320 | You may not copy, modify, sublicense, or distribute the Document except | |
2321 | as expressly provided for under this License. Any other attempt to | |
2322 | copy, modify, sublicense or distribute the Document is void, and will | |
2323 | automatically terminate your rights under this License. However, | |
2324 | parties who have received copies, or rights, from you under this | |
2325 | License will not have their licenses terminated so long as such | |
2326 | parties remain in full compliance. | |
2327 | ||
2328 | ||
2329 | 10. FUTURE REVISIONS OF THIS LICENSE | |
2330 | ||
2331 | The Free Software Foundation may publish new, revised versions | |
2332 | of the GNU Free Documentation License from time to time. Such new | |
2333 | versions will be similar in spirit to the present version, but may | |
2334 | differ in detail to address new problems or concerns. See | |
2335 | http://www.gnu.org/copyleft/. | |
2336 | ||
2337 | Each version of the License is given a distinguishing version number. | |
2338 | If the Document specifies that a particular numbered version of this | |
2339 | License "or any later version" applies to it, you have the option of | |
2340 | following the terms and conditions either of that specified version or | |
2341 | of any later version that has been published (not as a draft) by the | |
2342 | Free Software Foundation. If the Document does not specify a version | |
2343 | number of this License, you may choose any version ever published (not | |
2344 | as a draft) by the Free Software Foundation. | |
2345 | ||
2346 | ||
2347 | ADDENDUM: How to use this License for your documents | |
2348 | ||
2349 | To use this License in a document you have written, include a copy of | |
2350 | the License in the document and put the following copyright and | |
2351 | license notices just after the title page: | |
2352 | ||
2353 | @smallexample | |
2354 | Copyright (c) YEAR YOUR NAME. | |
2355 | Permission is granted to copy, distribute and/or modify this document | |
2356 | under the terms of the GNU Free Documentation License, Version 1.1 | |
2357 | or any later version published by the Free Software Foundation; | |
2358 | with the Invariant Sections being LIST THEIR TITLES, with the | |
2359 | Front-Cover Texts being LIST, and with the Back-Cover Texts being LIST. | |
2360 | A copy of the license is included in the section entitled "GNU | |
2361 | Free Documentation License". | |
2362 | @end smallexample | |
2363 | ||
2364 | If you have no Invariant Sections, write "with no Invariant Sections" | |
2365 | instead of saying which ones are invariant. If you have no | |
2366 | Front-Cover Texts, write "no Front-Cover Texts" instead of | |
2367 | "Front-Cover Texts being LIST"; likewise for Back-Cover Texts. | |
2368 | ||
2369 | If your document contains nontrivial examples of program code, we | |
2370 | recommend releasing these examples in parallel under your choice of | |
2371 | free software license, such as the GNU General Public License, | |
2372 | to permit their use in free software. | |
2373 | ||
252b5132 RH |
2374 | @contents |
2375 | @bye | |
2376 | ||
2377 | NEEDS AN INDEX | |
2378 | ||
2379 | -T - "traditional BSD style": How is it different? Should the | |
2380 | differences be documented? | |
2381 | ||
2382 | example flat file adds up to 100.01%... | |
2383 | ||
2384 | note: time estimates now only go out to one decimal place (0.0), where | |
2385 | they used to extend two (78.67). |