1\input texinfo @c -*-texinfo-*- 2@setfilename gprof.info 3@c Copyright (C) 1988-2024 Free Software Foundation, Inc. 4@settitle GNU gprof 5@setchapternewpage odd 6 7@c man begin INCLUDE 8@include bfdver.texi 9@c man end 10 11@ifnottex 12@c This is a dir.info fragment to support semi-automated addition of 13@c manuals to an info tree. zoo@cygnus.com is developing this facility. 14@dircategory Software development 15@direntry 16* gprof: (gprof). Profiling your program's execution 17@end direntry 18@end ifnottex 19 20@copying 21This file documents the gprof profiler of the GNU system. 22 23@c man begin COPYRIGHT 24Copyright @copyright{} 1988-2024 Free Software Foundation, Inc. 25 26Permission is granted to copy, distribute and/or modify this document 27under the terms of the GNU Free Documentation License, Version 1.3 28or any later version published by the Free Software Foundation; 29with no Invariant Sections, with no Front-Cover Texts, and with no 30Back-Cover Texts. A copy of the license is included in the 31section entitled ``GNU Free Documentation License''. 32 33@c man end 34@end copying 35 36@finalout 37@smallbook 38 39@titlepage 40@title GNU gprof 41@subtitle The @sc{gnu} Profiler 42@ifset VERSION_PACKAGE 43@subtitle @value{VERSION_PACKAGE} 44@end ifset 45@subtitle Version @value{VERSION} 46@author Jay Fenlason and Richard Stallman 47 48@page 49 50This manual describes the @sc{gnu} profiler, @code{gprof}, and how you 51can use it to determine which parts of a program are taking most of the 52execution time. We assume that you know how to write, compile, and 53execute programs. @sc{gnu} @code{gprof} was written by Jay Fenlason. 54Eric S. Raymond made some minor corrections and additions in 2003. 55 56@vskip 0pt plus 1filll 57Copyright @copyright{} 1988-2024 Free Software Foundation, Inc. 58 59 Permission is granted to copy, distribute and/or modify this document 60 under the terms of the GNU Free Documentation License, Version 1.3 61 or any later version published by the Free Software Foundation; 62 with no Invariant Sections, with no Front-Cover Texts, and with no 63 Back-Cover Texts. A copy of the license is included in the 64 section entitled ``GNU Free Documentation License''. 65 66@end titlepage 67@contents 68 69@ifnottex 70@node Top 71@top Profiling a Program: Where Does It Spend Its Time? 72 73This manual describes the @sc{gnu} profiler, @code{gprof}, and how you 74can use it to determine which parts of a program are taking most of the 75execution time. We assume that you know how to write, compile, and 76execute programs. @sc{gnu} @code{gprof} was written by Jay Fenlason. 77 78This manual is for @code{gprof} 79@ifset VERSION_PACKAGE 80@value{VERSION_PACKAGE} 81@end ifset 82version @value{VERSION}. 83 84This document is distributed under the terms of the GNU Free 85Documentation License version 1.3. A copy of the license is included 86in the section entitled ``GNU Free Documentation License''. 87 88@menu 89* Introduction:: What profiling means, and why it is useful. 90 91* Compiling:: How to compile your program for profiling. 92* Executing:: Executing your program to generate profile data 93* Invoking:: How to run @code{gprof}, and its options 94 95* Output:: Interpreting @code{gprof}'s output 96 97* Inaccuracy:: Potential problems you should be aware of 98* How do I?:: Answers to common questions 99* Incompatibilities:: (between @sc{gnu} @code{gprof} and Unix @code{gprof}.) 100* Details:: Details of how profiling is done 101* GNU Free Documentation License:: GNU Free Documentation License 102@end menu 103@end ifnottex 104 105@node Introduction 106@chapter Introduction to Profiling 107 108@ifset man 109@c man title gprof display call graph profile data 110 111@smallexample 112@c man begin SYNOPSIS 113gprof [ -[abcDhilLrsTvwxyz] ] [ -[ABCeEfFJnNOpPqQRStZ][@var{name}] ] 114 [ -I @var{dirs} ] [ -d[@var{num}] ] [ -k @var{from/to} ] 115 [ -m @var{min-count} ] [ -R @var{map_file} ] [ -t @var{table-length} ] 116 [ --[no-]annotated-source[=@var{name}] ] 117 [ --[no-]exec-counts[=@var{name}] ] 118 [ --[no-]flat-profile[=@var{name}] ] [ --[no-]graph[=@var{name}] ] 119 [ --[no-]time=@var{name}] [ --all-lines ] [ --brief ] 120 [ --debug[=@var{level}] ] [ --function-ordering ] 121 [ --file-ordering @var{map_file} ] [ --directory-path=@var{dirs} ] 122 [ --display-unused-functions ] [ --file-format=@var{name} ] 123 [ --file-info ] [ --help ] [ --line ] [ --inline-file-names ] 124 [ --min-count=@var{n} ] [ --no-static ] [ --print-path ] 125 [ --separate-files ] [ --static-call-graph ] [ --sum ] 126 [ --table-length=@var{len} ] [ --traditional ] [ --version ] 127 [ --width=@var{n} ] [ --ignore-non-functions ] 128 [ --demangle[=@var{STYLE}] ] [ --no-demangle ] 129 [--external-symbol-table=name] 130 [ @var{image-file} ] [ @var{profile-file} @dots{} ] 131@c man end 132@end smallexample 133 134@c man begin DESCRIPTION 135@code{gprof} produces an execution profile of C, Pascal, or Fortran77 136programs. The effect of called routines is incorporated in the profile 137of each caller. The profile data is taken from the call graph profile file 138(@file{gmon.out} default) which is created by programs 139that are compiled with the @samp{-pg} option of 140@code{cc}, @code{pc}, and @code{f77}. 141The @samp{-pg} option also links in versions of the library routines 142that are compiled for profiling. @code{Gprof} reads the given object 143file (the default is @code{a.out}) and establishes the relation between 144its symbol table and the call graph profile from @file{gmon.out}. 145If more than one profile file is specified, the @code{gprof} 146output shows the sum of the profile information in the given profile files. 147 148@code{Gprof} calculates the amount of time spent in each routine. 149Next, these times are propagated along the edges of the call graph. 150Cycles are discovered, and calls into a cycle are made to share the time 151of the cycle. 152 153@c man end 154 155@c man begin BUGS 156The granularity of the sampling is shown, but remains 157statistical at best. 158We assume that the time for each execution of a function 159can be expressed by the total time for the function divided 160by the number of times the function is called. 161Thus the time propagated along the call graph arcs to the function's 162parents is directly proportional to the number of times that 163arc is traversed. 164 165Parents that are not themselves profiled will have the time of 166their profiled children propagated to them, but they will appear 167to be spontaneously invoked in the call graph listing, and will 168not have their time propagated further. 169Similarly, signal catchers, even though profiled, will appear 170to be spontaneous (although for more obscure reasons). 171Any profiled children of signal catchers should have their times 172propagated properly, unless the signal catcher was invoked during 173the execution of the profiling routine, in which case all is lost. 174 175The profiled program must call @code{exit}(2) 176or return normally for the profiling information to be saved 177in the @file{gmon.out} file. 178@c man end 179 180@c man begin FILES 181@table @code 182@item @file{a.out} 183the namelist and text space. 184@item @file{gmon.out} 185dynamic call graph and profile. 186@item @file{gmon.sum} 187summarized dynamic call graph and profile. 188@end table 189@c man end 190 191@c man begin SEEALSO 192monitor(3), profil(2), cc(1), prof(1), and the Info entry for @file{gprof}. 193 194``An Execution Profiler for Modular Programs'', 195by S. Graham, P. Kessler, M. McKusick; 196Software - Practice and Experience, 197Vol. 13, pp. 671-685, 1983. 198 199``gprof: A Call Graph Execution Profiler'', 200by S. Graham, P. Kessler, M. McKusick; 201Proceedings of the SIGPLAN '82 Symposium on Compiler Construction, 202SIGPLAN Notices, Vol. 17, No 6, pp. 120-126, June 1982. 203@c man end 204@end ifset 205 206Profiling allows you to learn where your program spent its time and which 207functions called which other functions while it was executing. This 208information can show you which pieces of your program are slower than you 209expected, and might be candidates for rewriting to make your program 210execute faster. It can also tell you which functions are being called more 211or less often than you expected. This may help you spot bugs that had 212otherwise been unnoticed. 213 214Since the profiler uses information collected during the actual execution 215of your program, it can be used on programs that are too large or too 216complex to analyze by reading the source. However, how your program is run 217will affect the information that shows up in the profile data. If you 218don't use some feature of your program while it is being profiled, no 219profile information will be generated for that feature. 220 221Profiling has several steps: 222 223@itemize @bullet 224@item 225You must compile and link your program with profiling enabled. 226@xref{Compiling, ,Compiling a Program for Profiling}. 227 228@item 229You must execute your program to generate a profile data file. 230@xref{Executing, ,Executing the Program}. 231 232@item 233You must run @code{gprof} to analyze the profile data. 234@xref{Invoking, ,@code{gprof} Command Summary}. 235@end itemize 236 237The next three chapters explain these steps in greater detail. 238 239@c man begin DESCRIPTION 240 241Several forms of output are available from the analysis. 242 243The @dfn{flat profile} shows how much time your program spent in each function, 244and how many times that function was called. If you simply want to know 245which functions burn most of the cycles, it is stated concisely here. 246@xref{Flat Profile, ,The Flat Profile}. 247 248The @dfn{call graph} shows, for each function, which functions called it, which 249other functions it called, and how many times. There is also an estimate 250of how much time was spent in the subroutines of each function. This can 251suggest places where you might try to eliminate function calls that use a 252lot of time. @xref{Call Graph, ,The Call Graph}. 253 254The @dfn{annotated source} listing is a copy of the program's 255source code, labeled with the number of times each line of the 256program was executed. @xref{Annotated Source, ,The Annotated Source 257Listing}. 258@c man end 259 260To better understand how profiling works, you may wish to read 261a description of its implementation. 262@xref{Implementation, ,Implementation of Profiling}. 263 264@node Compiling 265@chapter Compiling a Program for Profiling 266 267The first step in generating profile information for your program is 268to compile and link it with profiling enabled. 269 270To compile a source file for profiling, specify the @samp{-pg} option when 271you run the compiler. (This is in addition to the options you normally 272use.) 273 274To link the program for profiling, if you use a compiler such as @code{cc} 275to do the linking, simply specify @samp{-pg} in addition to your usual 276options. The same option, @samp{-pg}, alters either compilation or linking 277to do what is necessary for profiling. Here are examples: 278 279@example 280cc -g -c myprog.c utils.c -pg 281cc -o myprog myprog.o utils.o -pg 282@end example 283 284The @samp{-pg} option also works with a command that both compiles and links: 285 286@example 287cc -o myprog myprog.c utils.c -g -pg 288@end example 289 290Note: The @samp{-pg} option must be part of your compilation options 291as well as your link options. If it is not then no call-graph data 292will be gathered and when you run @code{gprof} you will get an error 293message like this: 294 295@example 296gprof: gmon.out file is missing call-graph data 297@end example 298 299If you add the @samp{-Q} switch to suppress the printing of the call 300graph data you will still be able to see the time samples: 301 302@example 303Flat profile: 304 305Each sample counts as 0.01 seconds. 306 % cumulative self self total 307 time seconds seconds calls Ts/call Ts/call name 308 44.12 0.07 0.07 zazLoop 309 35.29 0.14 0.06 main 310 20.59 0.17 0.04 bazMillion 311@end example 312 313If you run the linker @code{ld} directly instead of through a compiler 314such as @code{cc}, you may have to specify a profiling startup file 315@file{gcrt0.o} as the first input file instead of the usual startup 316file @file{crt0.o}. In addition, you would probably want to 317specify the profiling C library, @file{libc_p.a}, by writing 318@samp{-lc_p} instead of the usual @samp{-lc}. This is not absolutely 319necessary, but doing this gives you number-of-calls information for 320standard library functions such as @code{read} and @code{open}. For 321example: 322 323@example 324ld -o myprog /lib/gcrt0.o myprog.o utils.o -lc_p 325@end example 326 327If you are running the program on a system which supports shared 328libraries you may run into problems with the profiling support code in 329a shared library being called before that library has been fully 330initialised. This is usually detected by the program encountering a 331segmentation fault as soon as it is run. The solution is to link 332against a static version of the library containing the profiling 333support code, which for @code{gcc} users can be done via the 334@samp{-static} or @samp{-static-libgcc} command-line option. For 335example: 336 337@example 338gcc -g -pg -static-libgcc myprog.c utils.c -o myprog 339@end example 340 341If you compile only some of the modules of the program with @samp{-pg}, you 342can still profile the program, but you won't get complete information about 343the modules that were compiled without @samp{-pg}. The only information 344you get for the functions in those modules is the total time spent in them; 345there is no record of how many times they were called, or from where. This 346will not affect the flat profile (except that the @code{calls} field for 347the functions will be blank), but will greatly reduce the usefulness of the 348call graph. 349 350If you wish to perform line-by-line profiling you should use the 351@code{gcov} tool instead of @code{gprof}. See that tool's manual or 352info pages for more details of how to do this. 353 354Note, older versions of @code{gcc} produce line-by-line profiling 355information that works with @code{gprof} rather than @code{gcov} so 356there is still support for displaying this kind of information in 357@code{gprof}. @xref{Line-by-line, ,Line-by-line Profiling}. 358 359It also worth noting that @code{gcc} implements a 360@samp{-finstrument-functions} command-line option which will insert 361calls to special user supplied instrumentation routines at the entry 362and exit of every function in their program. This can be used to 363implement an alternative profiling scheme. 364 365@node Executing 366@chapter Executing the Program 367 368Once the program is compiled for profiling, you must run it in order to 369generate the information that @code{gprof} needs. Simply run the program 370as usual, using the normal arguments, file names, etc. The program should 371run normally, producing the same output as usual. It will, however, run 372somewhat slower than normal because of the time spent collecting and 373writing the profile data. 374 375The way you run the program---the arguments and input that you give 376it---may have a dramatic effect on what the profile information shows. The 377profile data will describe the parts of the program that were activated for 378the particular input you use. For example, if the first command you give 379to your program is to quit, the profile data will show the time used in 380initialization and in cleanup, but not much else. 381 382Your program will write the profile data into a file called @file{gmon.out} 383just before exiting. If there is already a file called @file{gmon.out}, 384its contents are overwritten. You can rename the file afterwards if you 385are concerned that it may be overwritten. If your system libc allows you 386may be able to write the profile data under a different name. Set the 387GMON_OUT_PREFIX environment variable; this name will be appended with 388the PID of the running program. 389 390In order to write the @file{gmon.out} file properly, your program must exit 391normally: by returning from @code{main} or by calling @code{exit}. Calling 392the low-level function @code{_exit} does not write the profile data, and 393neither does abnormal termination due to an unhandled signal. 394 395The @file{gmon.out} file is written in the program's @emph{current working 396directory} at the time it exits. This means that if your program calls 397@code{chdir}, the @file{gmon.out} file will be left in the last directory 398your program @code{chdir}'d to. If you don't have permission to write in 399this directory, the file is not written, and you will get an error message. 400 401Older versions of the @sc{gnu} profiling library may also write a file 402called @file{bb.out}. This file, if present, contains an human-readable 403listing of the basic-block execution counts. Unfortunately, the 404appearance of a human-readable @file{bb.out} means the basic-block 405counts didn't get written into @file{gmon.out}. 406The Perl script @code{bbconv.pl}, included with the @code{gprof} 407source distribution, will convert a @file{bb.out} file into 408a format readable by @code{gprof}. Invoke it like this: 409 410@smallexample 411bbconv.pl < bb.out > @var{bh-data} 412@end smallexample 413 414This translates the information in @file{bb.out} into a form that 415@code{gprof} can understand. But you still need to tell @code{gprof} 416about the existence of this translated information. To do that, include 417@var{bb-data} on the @code{gprof} command line, @emph{along with 418@file{gmon.out}}, like this: 419 420@smallexample 421gprof @var{options} @var{executable-file} gmon.out @var{bb-data} [@var{yet-more-profile-data-files}@dots{}] [> @var{outfile}] 422@end smallexample 423 424@node Invoking 425@chapter @code{gprof} Command Summary 426 427After you have a profile data file @file{gmon.out}, you can run @code{gprof} 428to interpret the information in it. The @code{gprof} program prints a 429flat profile and a call graph on standard output. Typically you would 430redirect the output of @code{gprof} into a file with @samp{>}. 431 432You run @code{gprof} like this: 433 434@smallexample 435gprof @var{options} [@var{executable-file} [@var{profile-data-files}@dots{}]] [> @var{outfile}] 436@end smallexample 437 438@noindent 439Here square-brackets indicate optional arguments. 440 441If you omit the executable file name, the file @file{a.out} is used. If 442you give no profile data file name, the file @file{gmon.out} is used. If 443any file is not in the proper format, or if the profile data file does not 444appear to belong to the executable file, an error message is printed. 445 446You can give more than one profile data file by entering all their names 447after the executable file name; then the statistics in all the data files 448are summed together. 449 450The order of these options does not matter. 451 452@menu 453* Output Options:: Controlling @code{gprof}'s output style 454* Analysis Options:: Controlling how @code{gprof} analyzes its data 455* Miscellaneous Options:: 456* Deprecated Options:: Options you no longer need to use, but which 457 have been retained for compatibility 458* Symspecs:: Specifying functions to include or exclude 459@end menu 460 461@node Output Options 462@section Output Options 463 464@c man begin OPTIONS 465These options specify which of several output formats 466@code{gprof} should produce. 467 468Many of these options take an optional @dfn{symspec} to specify 469functions to be included or excluded. These options can be 470specified multiple times, with different symspecs, to include 471or exclude sets of symbols. @xref{Symspecs, ,Symspecs}. 472 473Specifying any of these options overrides the default (@samp{-p -q}), 474which prints a flat profile and call graph analysis 475for all functions. 476 477@table @code 478 479@item -A[@var{symspec}] 480@itemx --annotated-source[=@var{symspec}] 481The @samp{-A} option causes @code{gprof} to print annotated source code. 482If @var{symspec} is specified, print output only for matching symbols. 483@xref{Annotated Source, ,The Annotated Source Listing}. 484 485@item -b 486@itemx --brief 487If the @samp{-b} option is given, @code{gprof} doesn't print the 488verbose blurbs that try to explain the meaning of all of the fields in 489the tables. This is useful if you intend to print out the output, or 490are tired of seeing the blurbs. 491 492@item -B 493The @samp{-B} option causes @code{gprof} to print the call graph analysis. 494 495@item -C[@var{symspec}] 496@itemx --exec-counts[=@var{symspec}] 497The @samp{-C} option causes @code{gprof} to 498print a tally of functions and the number of times each was called. 499If @var{symspec} is specified, print tally only for matching symbols. 500 501If the profile data file contains basic-block count records, specifying 502the @samp{-l} option, along with @samp{-C}, will cause basic-block 503execution counts to be tallied and displayed. 504 505@item -i 506@itemx --file-info 507The @samp{-i} option causes @code{gprof} to display summary information 508about the profile data file(s) and then exit. The number of histogram, 509call graph, and basic-block count records is displayed. 510 511@item -I @var{dirs} 512@itemx --directory-path=@var{dirs} 513The @samp{-I} option specifies a list of search directories in 514which to find source files. Environment variable @var{GPROF_PATH} 515can also be used to convey this information. 516Used mostly for annotated source output. 517 518@item -J[@var{symspec}] 519@itemx --no-annotated-source[=@var{symspec}] 520The @samp{-J} option causes @code{gprof} not to 521print annotated source code. 522If @var{symspec} is specified, @code{gprof} prints annotated source, 523but excludes matching symbols. 524 525@item -L 526@itemx --print-path 527Normally, source filenames are printed with the path 528component suppressed. The @samp{-L} option causes @code{gprof} 529to print the full pathname of 530source filenames, which is determined 531from symbolic debugging information in the image file 532and is relative to the directory in which the compiler 533was invoked. 534 535@item -p[@var{symspec}] 536@itemx --flat-profile[=@var{symspec}] 537The @samp{-p} option causes @code{gprof} to print a flat profile. 538If @var{symspec} is specified, print flat profile only for matching symbols. 539@xref{Flat Profile, ,The Flat Profile}. 540 541@item -P[@var{symspec}] 542@itemx --no-flat-profile[=@var{symspec}] 543The @samp{-P} option causes @code{gprof} to suppress printing a flat profile. 544If @var{symspec} is specified, @code{gprof} prints a flat profile, 545but excludes matching symbols. 546 547@item -q[@var{symspec}] 548@itemx --graph[=@var{symspec}] 549The @samp{-q} option causes @code{gprof} to print the call graph analysis. 550If @var{symspec} is specified, print call graph only for matching symbols 551and their children. 552@xref{Call Graph, ,The Call Graph}. 553 554@item -Q[@var{symspec}] 555@itemx --no-graph[=@var{symspec}] 556The @samp{-Q} option causes @code{gprof} to suppress printing the 557call graph. 558If @var{symspec} is specified, @code{gprof} prints a call graph, 559but excludes matching symbols. 560 561@item -t 562@itemx --table-length=@var{num} 563The @samp{-t} option causes the @var{num} most active source lines in 564each source file to be listed when source annotation is enabled. The 565default is 10. 566 567@item -y 568@itemx --separate-files 569This option affects annotated source output only. 570Normally, @code{gprof} prints annotated source files 571to standard-output. If this option is specified, 572annotated source for a file named @file{path/@var{filename}} 573is generated in the file @file{@var{filename}-ann}. If the underlying 574file system would truncate @file{@var{filename}-ann} so that it 575overwrites the original @file{@var{filename}}, @code{gprof} generates 576annotated source in the file @file{@var{filename}.ann} instead (if the 577original file name has an extension, that extension is @emph{replaced} 578with @file{.ann}). 579 580@item -Z[@var{symspec}] 581@itemx --no-exec-counts[=@var{symspec}] 582The @samp{-Z} option causes @code{gprof} not to 583print a tally of functions and the number of times each was called. 584If @var{symspec} is specified, print tally, but exclude matching symbols. 585 586@item -r 587@itemx --function-ordering 588The @samp{--function-ordering} option causes @code{gprof} to print a 589suggested function ordering for the program based on profiling data. 590This option suggests an ordering which may improve paging, tlb and 591cache behavior for the program on systems which support arbitrary 592ordering of functions in an executable. 593 594The exact details of how to force the linker to place functions 595in a particular order is system dependent and out of the scope of this 596manual. 597 598@item -R @var{map_file} 599@itemx --file-ordering @var{map_file} 600The @samp{--file-ordering} option causes @code{gprof} to print a 601suggested .o link line ordering for the program based on profiling data. 602This option suggests an ordering which may improve paging, tlb and 603cache behavior for the program on systems which do not support arbitrary 604ordering of functions in an executable. 605 606Use of the @samp{-a} argument is highly recommended with this option. 607 608The @var{map_file} argument is a pathname to a file which provides 609function name to object file mappings. The format of the file is similar to 610the output of the program @code{nm}. 611 612@smallexample 613@group 614c-parse.o:00000000 T yyparse 615c-parse.o:00000004 C yyerrflag 616c-lang.o:00000000 T maybe_objc_method_name 617c-lang.o:00000000 T print_lang_statistics 618c-lang.o:00000000 T recognize_objc_keyword 619c-decl.o:00000000 T print_lang_identifier 620c-decl.o:00000000 T print_lang_type 621@dots{} 622 623@end group 624@end smallexample 625 626To create a @var{map_file} with @sc{gnu} @code{nm}, type a command like 627@kbd{nm --extern-only --defined-only -v --print-file-name program-name}. 628 629@item -T 630@itemx --traditional 631The @samp{-T} option causes @code{gprof} to print its output in 632``traditional'' BSD style. 633 634@item -w @var{width} 635@itemx --width=@var{width} 636Sets width of output lines to @var{width}. 637Currently only used when printing the function index at the bottom 638of the call graph. 639 640@item -x 641@itemx --all-lines 642This option affects annotated source output only. 643By default, only the lines at the beginning of a basic-block 644are annotated. If this option is specified, every line in 645a basic-block is annotated by repeating the annotation for the 646first line. This behavior is similar to @code{tcov}'s @samp{-a}. 647 648@item --demangle[=@var{style}] 649@itemx --no-demangle 650These options control whether C++ symbol names should be demangled when 651printing output. The default is to demangle symbols. The 652@code{--no-demangle} option may be used to turn off demangling. Different 653compilers have different mangling styles. The optional demangling style 654argument can be used to choose an appropriate demangling style for your 655compiler. 656@end table 657 658@node Analysis Options 659@section Analysis Options 660 661@table @code 662 663@item -a 664@itemx --no-static 665The @samp{-a} option causes @code{gprof} to suppress the printing of 666statically declared (private) functions. (These are functions whose 667names are not listed as global, and which are not visible outside the 668file/function/block where they were defined.) Time spent in these 669functions, calls to/from them, etc., will all be attributed to the 670function that was loaded directly before it in the executable file. 671@c This is compatible with Unix @code{gprof}, but a bad idea. 672This option affects both the flat profile and the call graph. 673 674@item -c 675@itemx --static-call-graph 676The @samp{-c} option causes the call graph of the program to be 677augmented by a heuristic which examines the text space of the object 678file and identifies function calls in the binary machine code. 679Since normal call graph records are only generated when functions are 680entered, this option identifies children that could have been called, 681but never were. Calls to functions that were not compiled with 682profiling enabled are also identified, but only if symbol table 683entries are present for them. 684Calls to dynamic library routines are typically @emph{not} found 685by this option. 686Parents or children identified via this heuristic 687are indicated in the call graph with call counts of @samp{0}. 688 689@item -D 690@itemx --ignore-non-functions 691The @samp{-D} option causes @code{gprof} to ignore symbols which 692are not known to be functions. This option will give more accurate 693profile data on systems where it is supported (Solaris and HPUX for 694example). 695 696@item -k @var{from}/@var{to} 697The @samp{-k} option allows you to delete from the call graph any arcs from 698symbols matching symspec @var{from} to those matching symspec @var{to}. 699 700@item -l 701@itemx --line 702The @samp{-l} option enables line-by-line profiling, which causes 703histogram hits to be charged to individual source code lines, 704instead of functions. This feature only works with programs compiled 705by older versions of the @code{gcc} compiler. Newer versions of 706@code{gcc} are designed to work with the @code{gcov} tool instead. 707 708If the program was compiled with basic-block counting enabled, 709this option will also identify how many times each line of 710code was executed. 711While line-by-line profiling can help isolate where in a large function 712a program is spending its time, it also significantly increases 713the running time of @code{gprof}, and magnifies statistical 714inaccuracies. 715@xref{Sampling Error, ,Statistical Sampling Error}. 716 717@item --inline-file-names 718This option causes @code{gprof} to print the source file after each 719symbol in both the flat profile and the call graph. The full path to the 720file is printed if used with the @samp{-L} option. 721 722@item -m @var{num} 723@itemx --min-count=@var{num} 724This option affects execution count output only. 725Symbols that are executed less than @var{num} times are suppressed. 726 727@item -n@var{symspec} 728@itemx --time=@var{symspec} 729The @samp{-n} option causes @code{gprof}, in its call graph analysis, 730to only propagate times for symbols matching @var{symspec}. 731 732@item -N@var{symspec} 733@itemx --no-time=@var{symspec} 734The @samp{-n} option causes @code{gprof}, in its call graph analysis, 735not to propagate times for symbols matching @var{symspec}. 736 737@item -S@var{filename} 738@itemx --external-symbol-table=@var{filename} 739The @samp{-S} option causes @code{gprof} to read an external symbol table 740file, such as @file{/proc/kallsyms}, rather than read the symbol table 741from the given object file (the default is @code{a.out}). This is useful 742for profiling kernel modules. 743 744@item -z 745@itemx --display-unused-functions 746If you give the @samp{-z} option, @code{gprof} will mention all 747functions in the flat profile, even those that were never called, and 748that had no time spent in them. This is useful in conjunction with the 749@samp{-c} option for discovering which routines were never called. 750 751@end table 752 753@node Miscellaneous Options 754@section Miscellaneous Options 755 756@table @code 757 758@item -d[@var{num}] 759@itemx --debug[=@var{num}] 760The @samp{-d @var{num}} option specifies debugging options. 761If @var{num} is not specified, enable all debugging. 762@xref{Debugging, ,Debugging @code{gprof}}. 763 764@item -h 765@itemx --help 766The @samp{-h} option prints command line usage. 767 768@item -O@var{name} 769@itemx --file-format=@var{name} 770Selects the format of the profile data files. Recognized formats are 771@samp{auto} (the default), @samp{bsd}, @samp{4.4bsd}, @samp{magic}, and 772@samp{prof} (not yet supported). 773 774@item -s 775@itemx --sum 776The @samp{-s} option causes @code{gprof} to summarize the information 777in the profile data files it read in, and write out a profile data 778file called @file{gmon.sum}, which contains all the information from 779the profile data files that @code{gprof} read in. The file @file{gmon.sum} 780may be one of the specified input files; the effect of this is to 781merge the data in the other input files into @file{gmon.sum}. 782 783Eventually you can run @code{gprof} again without @samp{-s} to analyze the 784cumulative data in the file @file{gmon.sum}. 785 786@item -v 787@itemx --version 788The @samp{-v} flag causes @code{gprof} to print the current version 789number, and then exit. 790 791@end table 792 793@node Deprecated Options 794@section Deprecated Options 795 796These options have been replaced with newer versions that use symspecs. 797 798@table @code 799 800@item -e @var{function_name} 801The @samp{-e @var{function}} option tells @code{gprof} to not print 802information about the function @var{function_name} (and its 803children@dots{}) in the call graph. The function will still be listed 804as a child of any functions that call it, but its index number will be 805shown as @samp{[not printed]}. More than one @samp{-e} option may be 806given; only one @var{function_name} may be indicated with each @samp{-e} 807option. 808 809@item -E @var{function_name} 810The @code{-E @var{function}} option works like the @code{-e} option, but 811time spent in the function (and children who were not called from 812anywhere else), will not be used to compute the percentages-of-time for 813the call graph. More than one @samp{-E} option may be given; only one 814@var{function_name} may be indicated with each @samp{-E} option. 815 816@item -f @var{function_name} 817The @samp{-f @var{function}} option causes @code{gprof} to limit the 818call graph to the function @var{function_name} and its children (and 819their children@dots{}). More than one @samp{-f} option may be given; 820only one @var{function_name} may be indicated with each @samp{-f} 821option. 822 823@item -F @var{function_name} 824The @samp{-F @var{function}} option works like the @code{-f} option, but 825only time spent in the function and its children (and their 826children@dots{}) will be used to determine total-time and 827percentages-of-time for the call graph. More than one @samp{-F} option 828may be given; only one @var{function_name} may be indicated with each 829@samp{-F} option. The @samp{-F} option overrides the @samp{-E} option. 830 831@end table 832 833@c man end 834 835Note that only one function can be specified with each @code{-e}, 836@code{-E}, @code{-f} or @code{-F} option. To specify more than one 837function, use multiple options. For example, this command: 838 839@example 840gprof -e boring -f foo -f bar myprogram > gprof.output 841@end example 842 843@noindent 844lists in the call graph all functions that were reached from either 845@code{foo} or @code{bar} and were not reachable from @code{boring}. 846 847@node Symspecs 848@section Symspecs 849 850Many of the output options allow functions to be included or excluded 851using @dfn{symspecs} (symbol specifications), which observe the 852following syntax: 853 854@example 855 filename_containing_a_dot 856| funcname_not_containing_a_dot 857| linenumber 858| ( [ any_filename ] `:' ( any_funcname | linenumber ) ) 859@end example 860 861Here are some sample symspecs: 862 863@table @samp 864@item main.c 865Selects everything in file @file{main.c}---the 866dot in the string tells @code{gprof} to interpret 867the string as a filename, rather than as 868a function name. To select a file whose 869name does not contain a dot, a trailing colon 870should be specified. For example, @samp{odd:} is 871interpreted as the file named @file{odd}. 872 873@item main 874Selects all functions named @samp{main}. 875 876Note that there may be multiple instances of the same function name 877because some of the definitions may be local (i.e., static). Unless a 878function name is unique in a program, you must use the colon notation 879explained below to specify a function from a specific source file. 880 881Sometimes, function names contain dots. In such cases, it is necessary 882to add a leading colon to the name. For example, @samp{:.mul} selects 883function @samp{.mul}. 884 885In some object file formats, symbols have a leading underscore. 886@code{gprof} will normally not print these underscores. When you name a 887symbol in a symspec, you should type it exactly as @code{gprof} prints 888it in its output. For example, if the compiler produces a symbol 889@samp{_main} from your @code{main} function, @code{gprof} still prints 890it as @samp{main} in its output, so you should use @samp{main} in 891symspecs. 892 893@item main.c:main 894Selects function @samp{main} in file @file{main.c}. 895 896@item main.c:134 897Selects line 134 in file @file{main.c}. 898@end table 899 900@node Output 901@chapter Interpreting @code{gprof}'s Output 902 903@code{gprof} can produce several different output styles, the 904most important of which are described below. The simplest output 905styles (file information, execution count, and function and file ordering) 906are not described here, but are documented with the respective options 907that trigger them. 908@xref{Output Options, ,Output Options}. 909 910@menu 911* Flat Profile:: The flat profile shows how much time was spent 912 executing directly in each function. 913* Call Graph:: The call graph shows which functions called which 914 others, and how much time each function used 915 when its subroutine calls are included. 916* Line-by-line:: @code{gprof} can analyze individual source code lines 917* Annotated Source:: The annotated source listing displays source code 918 labeled with execution counts 919@end menu 920 921 922@node Flat Profile 923@section The Flat Profile 924@cindex flat profile 925 926The @dfn{flat profile} shows the total amount of time your program 927spent executing each function. Unless the @samp{-z} option is given, 928functions with no apparent time spent in them, and no apparent calls 929to them, are not mentioned. Note that if a function was not compiled 930for profiling, and didn't run long enough to show up on the program 931counter histogram, it will be indistinguishable from a function that 932was never called. 933 934This is part of a flat profile for a small program: 935 936@smallexample 937@group 938Flat profile: 939 940Each sample counts as 0.01 seconds. 941 % cumulative self self total 942 time seconds seconds calls ms/call ms/call name 943 33.34 0.02 0.02 7208 0.00 0.00 open 944 16.67 0.03 0.01 244 0.04 0.12 offtime 945 16.67 0.04 0.01 8 1.25 1.25 memccpy 946 16.67 0.05 0.01 7 1.43 1.43 write 947 16.67 0.06 0.01 mcount 948 0.00 0.06 0.00 236 0.00 0.00 tzset 949 0.00 0.06 0.00 192 0.00 0.00 tolower 950 0.00 0.06 0.00 47 0.00 0.00 strlen 951 0.00 0.06 0.00 45 0.00 0.00 strchr 952 0.00 0.06 0.00 1 0.00 50.00 main 953 0.00 0.06 0.00 1 0.00 0.00 memcpy 954 0.00 0.06 0.00 1 0.00 10.11 print 955 0.00 0.06 0.00 1 0.00 0.00 profil 956 0.00 0.06 0.00 1 0.00 50.00 report 957@dots{} 958@end group 959@end smallexample 960 961@noindent 962The functions are sorted first by decreasing run-time spent in them, 963then by decreasing number of calls, then alphabetically by name. The 964functions @samp{mcount} and @samp{profil} are part of the profiling 965apparatus and appear in every flat profile; their time gives a measure of 966the amount of overhead due to profiling. 967 968Just before the column headers, a statement appears indicating 969how much time each sample counted as. 970This @dfn{sampling period} estimates the margin of error in each of the time 971figures. A time figure that is not much larger than this is not 972reliable. In this example, each sample counted as 0.01 seconds, 973suggesting a 100 Hz sampling rate. 974The program's total execution time was 0.06 975seconds, as indicated by the @samp{cumulative seconds} field. Since 976each sample counted for 0.01 seconds, this means only six samples 977were taken during the run. Two of the samples occurred while the 978program was in the @samp{open} function, as indicated by the 979@samp{self seconds} field. Each of the other four samples 980occurred one each in @samp{offtime}, @samp{memccpy}, @samp{write}, 981and @samp{mcount}. 982Since only six samples were taken, none of these values can 983be regarded as particularly reliable. 984In another run, 985the @samp{self seconds} field for 986@samp{mcount} might well be @samp{0.00} or @samp{0.02}. 987@xref{Sampling Error, ,Statistical Sampling Error}, 988for a complete discussion. 989 990The remaining functions in the listing (those whose 991@samp{self seconds} field is @samp{0.00}) didn't appear 992in the histogram samples at all. However, the call graph 993indicated that they were called, so therefore they are listed, 994sorted in decreasing order by the @samp{calls} field. 995Clearly some time was spent executing these functions, 996but the paucity of histogram samples prevents any 997determination of how much time each took. 998 999Here is what the fields in each line mean: 1000 1001@table @code 1002@item % time 1003This is the percentage of the total execution time your program spent 1004in this function. These should all add up to 100%. 1005 1006@item cumulative seconds 1007This is the cumulative total number of seconds the computer spent 1008executing this functions, plus the time spent in all the functions 1009above this one in this table. 1010 1011@item self seconds 1012This is the number of seconds accounted for by this function alone. 1013The flat profile listing is sorted first by this number. 1014 1015@item calls 1016This is the total number of times the function was called. If the 1017function was never called, or the number of times it was called cannot 1018be determined (probably because the function was not compiled with 1019profiling enabled), the @dfn{calls} field is blank. 1020 1021@item self ms/call 1022This represents the average number of milliseconds spent in this 1023function per call, if this function is profiled. Otherwise, this field 1024is blank for this function. 1025 1026@item total ms/call 1027This represents the average number of milliseconds spent in this 1028function and its descendants per call, if this function is profiled. 1029Otherwise, this field is blank for this function. 1030This is the only field in the flat profile that uses call graph analysis. 1031 1032@item name 1033This is the name of the function. The flat profile is sorted by this 1034field alphabetically after the @dfn{self seconds} and @dfn{calls} 1035fields are sorted. 1036@end table 1037 1038@node Call Graph 1039@section The Call Graph 1040@cindex call graph 1041 1042The @dfn{call graph} shows how much time was spent in each function 1043and its children. From this information, you can find functions that, 1044while they themselves may not have used much time, called other 1045functions that did use unusual amounts of time. 1046 1047Here is a sample call from a small program. This call came from the 1048same @code{gprof} run as the flat profile example in the previous 1049section. 1050 1051@smallexample 1052@group 1053granularity: each sample hit covers 2 byte(s) for 20.00% of 0.05 seconds 1054 1055index % time self children called name 1056 <spontaneous> 1057[1] 100.0 0.00 0.05 start [1] 1058 0.00 0.05 1/1 main [2] 1059 0.00 0.00 1/2 on_exit [28] 1060 0.00 0.00 1/1 exit [59] 1061----------------------------------------------- 1062 0.00 0.05 1/1 start [1] 1063[2] 100.0 0.00 0.05 1 main [2] 1064 0.00 0.05 1/1 report [3] 1065----------------------------------------------- 1066 0.00 0.05 1/1 main [2] 1067[3] 100.0 0.00 0.05 1 report [3] 1068 0.00 0.03 8/8 timelocal [6] 1069 0.00 0.01 1/1 print [9] 1070 0.00 0.01 9/9 fgets [12] 1071 0.00 0.00 12/34 strncmp <cycle 1> [40] 1072 0.00 0.00 8/8 lookup [20] 1073 0.00 0.00 1/1 fopen [21] 1074 0.00 0.00 8/8 chewtime [24] 1075 0.00 0.00 8/16 skipspace [44] 1076----------------------------------------------- 1077[4] 59.8 0.01 0.02 8+472 <cycle 2 as a whole> [4] 1078 0.01 0.02 244+260 offtime <cycle 2> [7] 1079 0.00 0.00 236+1 tzset <cycle 2> [26] 1080----------------------------------------------- 1081@end group 1082@end smallexample 1083 1084The lines full of dashes divide this table into @dfn{entries}, one for each 1085function. Each entry has one or more lines. 1086 1087In each entry, the primary line is the one that starts with an index number 1088in square brackets. The end of this line says which function the entry is 1089for. The preceding lines in the entry describe the callers of this 1090function and the following lines describe its subroutines (also called 1091@dfn{children} when we speak of the call graph). 1092 1093The entries are sorted by time spent in the function and its subroutines. 1094 1095The internal profiling function @code{mcount} (@pxref{Flat Profile, ,The 1096Flat Profile}) is never mentioned in the call graph. 1097 1098@menu 1099* Primary:: Details of the primary line's contents. 1100* Callers:: Details of caller-lines' contents. 1101* Subroutines:: Details of subroutine-lines' contents. 1102* Cycles:: When there are cycles of recursion, 1103 such as @code{a} calls @code{b} calls @code{a}@dots{} 1104@end menu 1105 1106@node Primary 1107@subsection The Primary Line 1108 1109The @dfn{primary line} in a call graph entry is the line that 1110describes the function which the entry is about and gives the overall 1111statistics for this function. 1112 1113For reference, we repeat the primary line from the entry for function 1114@code{report} in our main example, together with the heading line that 1115shows the names of the fields: 1116 1117@smallexample 1118@group 1119index % time self children called name 1120@dots{} 1121[3] 100.0 0.00 0.05 1 report [3] 1122@end group 1123@end smallexample 1124 1125Here is what the fields in the primary line mean: 1126 1127@table @code 1128@item index 1129Entries are numbered with consecutive integers. Each function 1130therefore has an index number, which appears at the beginning of its 1131primary line. 1132 1133Each cross-reference to a function, as a caller or subroutine of 1134another, gives its index number as well as its name. The index number 1135guides you if you wish to look for the entry for that function. 1136 1137@item % time 1138This is the percentage of the total time that was spent in this 1139function, including time spent in subroutines called from this 1140function. 1141 1142The time spent in this function is counted again for the callers of 1143this function. Therefore, adding up these percentages is meaningless. 1144 1145@item self 1146This is the total amount of time spent in this function. This 1147should be identical to the number printed in the @code{seconds} field 1148for this function in the flat profile. 1149 1150@item children 1151This is the total amount of time spent in the subroutine calls made by 1152this function. This should be equal to the sum of all the @code{self} 1153and @code{children} entries of the children listed directly below this 1154function. 1155 1156@item called 1157This is the number of times the function was called. 1158 1159If the function called itself recursively, there are two numbers, 1160separated by a @samp{+}. The first number counts non-recursive calls, 1161and the second counts recursive calls. 1162 1163In the example above, the function @code{report} was called once from 1164@code{main}. 1165 1166@item name 1167This is the name of the current function. The index number is 1168repeated after it. 1169 1170If the function is part of a cycle of recursion, the cycle number is 1171printed between the function's name and the index number 1172(@pxref{Cycles, ,How Mutually Recursive Functions Are Described}). 1173For example, if function @code{gnurr} is part of 1174cycle number one, and has index number twelve, its primary line would 1175be end like this: 1176 1177@example 1178gnurr <cycle 1> [12] 1179@end example 1180@end table 1181 1182@node Callers 1183@subsection Lines for a Function's Callers 1184 1185A function's entry has a line for each function it was called by. 1186These lines' fields correspond to the fields of the primary line, but 1187their meanings are different because of the difference in context. 1188 1189For reference, we repeat two lines from the entry for the function 1190@code{report}, the primary line and one caller-line preceding it, together 1191with the heading line that shows the names of the fields: 1192 1193@smallexample 1194index % time self children called name 1195@dots{} 1196 0.00 0.05 1/1 main [2] 1197[3] 100.0 0.00 0.05 1 report [3] 1198@end smallexample 1199 1200Here are the meanings of the fields in the caller-line for @code{report} 1201called from @code{main}: 1202 1203@table @code 1204@item self 1205An estimate of the amount of time spent in @code{report} itself when it was 1206called from @code{main}. 1207 1208@item children 1209An estimate of the amount of time spent in subroutines of @code{report} 1210when @code{report} was called from @code{main}. 1211 1212The sum of the @code{self} and @code{children} fields is an estimate 1213of the amount of time spent within calls to @code{report} from @code{main}. 1214 1215@item called 1216Two numbers: the number of times @code{report} was called from @code{main}, 1217followed by the total number of non-recursive calls to @code{report} from 1218all its callers. 1219 1220@item name and index number 1221The name of the caller of @code{report} to which this line applies, 1222followed by the caller's index number. 1223 1224Not all functions have entries in the call graph; some 1225options to @code{gprof} request the omission of certain functions. 1226When a caller has no entry of its own, it still has caller-lines 1227in the entries of the functions it calls. 1228 1229If the caller is part of a recursion cycle, the cycle number is 1230printed between the name and the index number. 1231@end table 1232 1233If the identity of the callers of a function cannot be determined, a 1234dummy caller-line is printed which has @samp{<spontaneous>} as the 1235``caller's name'' and all other fields blank. This can happen for 1236signal handlers. 1237@c What if some calls have determinable callers' names but not all? 1238@c FIXME - still relevant? 1239 1240@node Subroutines 1241@subsection Lines for a Function's Subroutines 1242 1243A function's entry has a line for each of its subroutines---in other 1244words, a line for each other function that it called. These lines' 1245fields correspond to the fields of the primary line, but their meanings 1246are different because of the difference in context. 1247 1248For reference, we repeat two lines from the entry for the function 1249@code{main}, the primary line and a line for a subroutine, together 1250with the heading line that shows the names of the fields: 1251 1252@smallexample 1253index % time self children called name 1254@dots{} 1255[2] 100.0 0.00 0.05 1 main [2] 1256 0.00 0.05 1/1 report [3] 1257@end smallexample 1258 1259Here are the meanings of the fields in the subroutine-line for @code{main} 1260calling @code{report}: 1261 1262@table @code 1263@item self 1264An estimate of the amount of time spent directly within @code{report} 1265when @code{report} was called from @code{main}. 1266 1267@item children 1268An estimate of the amount of time spent in subroutines of @code{report} 1269when @code{report} was called from @code{main}. 1270 1271The sum of the @code{self} and @code{children} fields is an estimate 1272of the total time spent in calls to @code{report} from @code{main}. 1273 1274@item called 1275Two numbers, the number of calls to @code{report} from @code{main} 1276followed by the total number of non-recursive calls to @code{report}. 1277This ratio is used to determine how much of @code{report}'s @code{self} 1278and @code{children} time gets credited to @code{main}. 1279@xref{Assumptions, ,Estimating @code{children} Times}. 1280 1281@item name 1282The name of the subroutine of @code{main} to which this line applies, 1283followed by the subroutine's index number. 1284 1285If the caller is part of a recursion cycle, the cycle number is 1286printed between the name and the index number. 1287@end table 1288 1289@node Cycles 1290@subsection How Mutually Recursive Functions Are Described 1291@cindex cycle 1292@cindex recursion cycle 1293 1294The graph may be complicated by the presence of @dfn{cycles of 1295recursion} in the call graph. A cycle exists if a function calls 1296another function that (directly or indirectly) calls (or appears to 1297call) the original function. For example: if @code{a} calls @code{b}, 1298and @code{b} calls @code{a}, then @code{a} and @code{b} form a cycle. 1299 1300Whenever there are call paths both ways between a pair of functions, they 1301belong to the same cycle. If @code{a} and @code{b} call each other and 1302@code{b} and @code{c} call each other, all three make one cycle. Note that 1303even if @code{b} only calls @code{a} if it was not called from @code{a}, 1304@code{gprof} cannot determine this, so @code{a} and @code{b} are still 1305considered a cycle. 1306 1307The cycles are numbered with consecutive integers. When a function 1308belongs to a cycle, each time the function name appears in the call graph 1309it is followed by @samp{<cycle @var{number}>}. 1310 1311The reason cycles matter is that they make the time values in the call 1312graph paradoxical. The ``time spent in children'' of @code{a} should 1313include the time spent in its subroutine @code{b} and in @code{b}'s 1314subroutines---but one of @code{b}'s subroutines is @code{a}! How much of 1315@code{a}'s time should be included in the children of @code{a}, when 1316@code{a} is indirectly recursive? 1317 1318The way @code{gprof} resolves this paradox is by creating a single entry 1319for the cycle as a whole. The primary line of this entry describes the 1320total time spent directly in the functions of the cycle. The 1321``subroutines'' of the cycle are the individual functions of the cycle, and 1322all other functions that were called directly by them. The ``callers'' of 1323the cycle are the functions, outside the cycle, that called functions in 1324the cycle. 1325 1326Here is an example portion of a call graph which shows a cycle containing 1327functions @code{a} and @code{b}. The cycle was entered by a call to 1328@code{a} from @code{main}; both @code{a} and @code{b} called @code{c}. 1329 1330@smallexample 1331index % time self children called name 1332---------------------------------------- 1333 1.77 0 1/1 main [2] 1334[3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3] 1335 1.02 0 3 b <cycle 1> [4] 1336 0.75 0 2 a <cycle 1> [5] 1337---------------------------------------- 1338 3 a <cycle 1> [5] 1339[4] 52.85 1.02 0 0 b <cycle 1> [4] 1340 2 a <cycle 1> [5] 1341 0 0 3/6 c [6] 1342---------------------------------------- 1343 1.77 0 1/1 main [2] 1344 2 b <cycle 1> [4] 1345[5] 38.86 0.75 0 1 a <cycle 1> [5] 1346 3 b <cycle 1> [4] 1347 0 0 3/6 c [6] 1348---------------------------------------- 1349@end smallexample 1350 1351@noindent 1352(The entire call graph for this program contains in addition an entry for 1353@code{main}, which calls @code{a}, and an entry for @code{c}, with callers 1354@code{a} and @code{b}.) 1355 1356@smallexample 1357index % time self children called name 1358 <spontaneous> 1359[1] 100.00 0 1.93 0 start [1] 1360 0.16 1.77 1/1 main [2] 1361---------------------------------------- 1362 0.16 1.77 1/1 start [1] 1363[2] 100.00 0.16 1.77 1 main [2] 1364 1.77 0 1/1 a <cycle 1> [5] 1365---------------------------------------- 1366 1.77 0 1/1 main [2] 1367[3] 91.71 1.77 0 1+5 <cycle 1 as a whole> [3] 1368 1.02 0 3 b <cycle 1> [4] 1369 0.75 0 2 a <cycle 1> [5] 1370 0 0 6/6 c [6] 1371---------------------------------------- 1372 3 a <cycle 1> [5] 1373[4] 52.85 1.02 0 0 b <cycle 1> [4] 1374 2 a <cycle 1> [5] 1375 0 0 3/6 c [6] 1376---------------------------------------- 1377 1.77 0 1/1 main [2] 1378 2 b <cycle 1> [4] 1379[5] 38.86 0.75 0 1 a <cycle 1> [5] 1380 3 b <cycle 1> [4] 1381 0 0 3/6 c [6] 1382---------------------------------------- 1383 0 0 3/6 b <cycle 1> [4] 1384 0 0 3/6 a <cycle 1> [5] 1385[6] 0.00 0 0 6 c [6] 1386---------------------------------------- 1387@end smallexample 1388 1389The @code{self} field of the cycle's primary line is the total time 1390spent in all the functions of the cycle. It equals the sum of the 1391@code{self} fields for the individual functions in the cycle, found 1392in the entry in the subroutine lines for these functions. 1393 1394The @code{children} fields of the cycle's primary line and subroutine lines 1395count only subroutines outside the cycle. Even though @code{a} calls 1396@code{b}, the time spent in those calls to @code{b} is not counted in 1397@code{a}'s @code{children} time. Thus, we do not encounter the problem of 1398what to do when the time in those calls to @code{b} includes indirect 1399recursive calls back to @code{a}. 1400 1401The @code{children} field of a caller-line in the cycle's entry estimates 1402the amount of time spent @emph{in the whole cycle}, and its other 1403subroutines, on the times when that caller called a function in the cycle. 1404 1405The @code{called} field in the primary line for the cycle has two numbers: 1406first, the number of times functions in the cycle were called by functions 1407outside the cycle; second, the number of times they were called by 1408functions in the cycle (including times when a function in the cycle calls 1409itself). This is a generalization of the usual split into non-recursive and 1410recursive calls. 1411 1412The @code{called} field of a subroutine-line for a cycle member in the 1413cycle's entry says how many time that function was called from functions in 1414the cycle. The total of all these is the second number in the primary line's 1415@code{called} field. 1416 1417In the individual entry for a function in a cycle, the other functions in 1418the same cycle can appear as subroutines and as callers. These lines show 1419how many times each function in the cycle called or was called from each other 1420function in the cycle. The @code{self} and @code{children} fields in these 1421lines are blank because of the difficulty of defining meanings for them 1422when recursion is going on. 1423 1424@node Line-by-line 1425@section Line-by-line Profiling 1426 1427@code{gprof}'s @samp{-l} option causes the program to perform 1428@dfn{line-by-line} profiling. In this mode, histogram 1429samples are assigned not to functions, but to individual 1430lines of source code. This only works with programs compiled with 1431older versions of the @code{gcc} compiler. Newer versions of @code{gcc} 1432use a different program - @code{gcov} - to display line-by-line 1433profiling information. 1434 1435With the older versions of @code{gcc} the program usually has to be 1436compiled with a @samp{-g} option, in addition to @samp{-pg}, in order 1437to generate debugging symbols for tracking source code lines. 1438Note, in much older versions of @code{gcc} the program had to be 1439compiled with the @samp{-a} command-line option as well. 1440 1441The flat profile is the most useful output table 1442in line-by-line mode. 1443The call graph isn't as useful as normal, since 1444the current version of @code{gprof} does not propagate 1445call graph arcs from source code lines to the enclosing function. 1446The call graph does, however, show each line of code 1447that called each function, along with a count. 1448 1449Here is a section of @code{gprof}'s output, without line-by-line profiling. 1450Note that @code{ct_init} accounted for four histogram hits, and 145113327 calls to @code{init_block}. 1452 1453@smallexample 1454Flat profile: 1455 1456Each sample counts as 0.01 seconds. 1457 % cumulative self self total 1458 time seconds seconds calls us/call us/call name 1459 30.77 0.13 0.04 6335 6.31 6.31 ct_init 1460 1461 1462 Call graph (explanation follows) 1463 1464 1465granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds 1466 1467index % time self children called name 1468 1469 0.00 0.00 1/13496 name_too_long 1470 0.00 0.00 40/13496 deflate 1471 0.00 0.00 128/13496 deflate_fast 1472 0.00 0.00 13327/13496 ct_init 1473[7] 0.0 0.00 0.00 13496 init_block 1474 1475@end smallexample 1476 1477Now let's look at some of @code{gprof}'s output from the same program run, 1478this time with line-by-line profiling enabled. Note that @code{ct_init}'s 1479four histogram hits are broken down into four lines of source code---one hit 1480occurred on each of lines 349, 351, 382 and 385. In the call graph, 1481note how 1482@code{ct_init}'s 13327 calls to @code{init_block} are broken down 1483into one call from line 396, 3071 calls from line 384, 3730 calls 1484from line 385, and 6525 calls from 387. 1485 1486@smallexample 1487Flat profile: 1488 1489Each sample counts as 0.01 seconds. 1490 % cumulative self 1491 time seconds seconds calls name 1492 7.69 0.10 0.01 ct_init (trees.c:349) 1493 7.69 0.11 0.01 ct_init (trees.c:351) 1494 7.69 0.12 0.01 ct_init (trees.c:382) 1495 7.69 0.13 0.01 ct_init (trees.c:385) 1496 1497 1498 Call graph (explanation follows) 1499 1500 1501granularity: each sample hit covers 4 byte(s) for 7.69% of 0.13 seconds 1502 1503 % time self children called name 1504 1505 0.00 0.00 1/13496 name_too_long (gzip.c:1440) 1506 0.00 0.00 1/13496 deflate (deflate.c:763) 1507 0.00 0.00 1/13496 ct_init (trees.c:396) 1508 0.00 0.00 2/13496 deflate (deflate.c:727) 1509 0.00 0.00 4/13496 deflate (deflate.c:686) 1510 0.00 0.00 5/13496 deflate (deflate.c:675) 1511 0.00 0.00 12/13496 deflate (deflate.c:679) 1512 0.00 0.00 16/13496 deflate (deflate.c:730) 1513 0.00 0.00 128/13496 deflate_fast (deflate.c:654) 1514 0.00 0.00 3071/13496 ct_init (trees.c:384) 1515 0.00 0.00 3730/13496 ct_init (trees.c:385) 1516 0.00 0.00 6525/13496 ct_init (trees.c:387) 1517[6] 0.0 0.00 0.00 13496 init_block (trees.c:408) 1518 1519@end smallexample 1520 1521 1522@node Annotated Source 1523@section The Annotated Source Listing 1524 1525@code{gprof}'s @samp{-A} option triggers an annotated source listing, 1526which lists the program's source code, each function labeled with the 1527number of times it was called. You may also need to specify the 1528@samp{-I} option, if @code{gprof} can't find the source code files. 1529 1530With older versions of @code{gcc} compiling with @samp{gcc @dots{} -g 1531-pg -a} augments your program with basic-block counting code, in 1532addition to function counting code. This enables @code{gprof} to 1533determine how many times each line of code was executed. With newer 1534versions of @code{gcc} support for displaying basic-block counts is 1535provided by the @code{gcov} program. 1536 1537For example, consider the following function, taken from gzip, 1538with line numbers added: 1539 1540@smallexample 1541 1 ulg updcrc(s, n) 1542 2 uch *s; 1543 3 unsigned n; 1544 4 @{ 1545 5 register ulg c; 1546 6 1547 7 static ulg crc = (ulg)0xffffffffL; 1548 8 1549 9 if (s == NULL) @{ 155010 c = 0xffffffffL; 155111 @} else @{ 155212 c = crc; 155313 if (n) do @{ 155414 c = crc_32_tab[...]; 155515 @} while (--n); 155616 @} 155717 crc = c; 155818 return c ^ 0xffffffffL; 155919 @} 1560 1561@end smallexample 1562 1563@code{updcrc} has at least five basic-blocks. 1564One is the function itself. The 1565@code{if} statement on line 9 generates two more basic-blocks, one 1566for each branch of the @code{if}. A fourth basic-block results from 1567the @code{if} on line 13, and the contents of the @code{do} loop form 1568the fifth basic-block. The compiler may also generate additional 1569basic-blocks to handle various special cases. 1570 1571A program augmented for basic-block counting can be analyzed with 1572@samp{gprof -l -A}. 1573The @samp{-x} option is also helpful, 1574to ensure that each line of code is labeled at least once. 1575Here is @code{updcrc}'s 1576annotated source listing for a sample @code{gzip} run: 1577 1578@smallexample 1579 ulg updcrc(s, n) 1580 uch *s; 1581 unsigned n; 1582 2 ->@{ 1583 register ulg c; 1584 1585 static ulg crc = (ulg)0xffffffffL; 1586 1587 2 -> if (s == NULL) @{ 1588 1 -> c = 0xffffffffL; 1589 1 -> @} else @{ 1590 1 -> c = crc; 1591 1 -> if (n) do @{ 1592 26312 -> c = crc_32_tab[...]; 159326312,1,26311 -> @} while (--n); 1594 @} 1595 2 -> crc = c; 1596 2 -> return c ^ 0xffffffffL; 1597 2 ->@} 1598@end smallexample 1599 1600In this example, the function was called twice, passing once through 1601each branch of the @code{if} statement. The body of the @code{do} 1602loop was executed a total of 26312 times. Note how the @code{while} 1603statement is annotated. It began execution 26312 times, once for 1604each iteration through the loop. One of those times (the last time) 1605it exited, while it branched back to the beginning of the loop 26311 times. 1606 1607@node Inaccuracy 1608@chapter Inaccuracy of @code{gprof} Output 1609 1610@menu 1611* Sampling Error:: Statistical margins of error 1612* Assumptions:: Estimating children times 1613@end menu 1614 1615@node Sampling Error 1616@section Statistical Sampling Error 1617 1618The run-time figures that @code{gprof} gives you are based on a sampling 1619process, so they are subject to statistical inaccuracy. If a function runs 1620only a small amount of time, so that on the average the sampling process 1621ought to catch that function in the act only once, there is a pretty good 1622chance it will actually find that function zero times, or twice. 1623 1624By contrast, the number-of-calls and basic-block figures are derived 1625by counting, not sampling. They are completely accurate and will not 1626vary from run to run if your program is deterministic and single 1627threaded. In multi-threaded applications, or single threaded 1628applications that link with multi-threaded libraries, the counts are 1629only deterministic if the counting function is thread-safe. (Note: 1630beware that the mcount counting function in glibc is @emph{not} 1631thread-safe). @xref{Implementation, ,Implementation of Profiling}. 1632 1633The @dfn{sampling period} that is printed at the beginning of the flat 1634profile says how often samples are taken. The rule of thumb is that a 1635run-time figure is accurate if it is considerably bigger than the sampling 1636period. 1637 1638The actual amount of error can be predicted. 1639For @var{n} samples, the @emph{expected} error 1640is the square-root of @var{n}. For example, 1641if the sampling period is 0.01 seconds and @code{foo}'s run-time is 1 second, 1642@var{n} is 100 samples (1 second/0.01 seconds), sqrt(@var{n}) is 10 samples, so 1643the expected error in @code{foo}'s run-time is 0.1 seconds (10*0.01 seconds), 1644or ten percent of the observed value. 1645Again, if the sampling period is 0.01 seconds and @code{bar}'s run-time is 1646100 seconds, @var{n} is 10000 samples, sqrt(@var{n}) is 100 samples, so 1647the expected error in @code{bar}'s run-time is 1 second, 1648or one percent of the observed value. 1649It is likely to 1650vary this much @emph{on the average} from one profiling run to the next. 1651(@emph{Sometimes} it will vary more.) 1652 1653This does not mean that a small run-time figure is devoid of information. 1654If the program's @emph{total} run-time is large, a small run-time for one 1655function does tell you that that function used an insignificant fraction of 1656the whole program's time. Usually this means it is not worth optimizing. 1657 1658One way to get more accuracy is to give your program more (but similar) 1659input data so it will take longer. Another way is to combine the data from 1660several runs, using the @samp{-s} option of @code{gprof}. Here is how: 1661 1662@enumerate 1663@item 1664Run your program once. 1665 1666@item 1667Issue the command @samp{mv gmon.out gmon.sum}. 1668 1669@item 1670Run your program again, the same as before. 1671 1672@item 1673Merge the new data in @file{gmon.out} into @file{gmon.sum} with this command: 1674 1675@example 1676gprof -s @var{executable-file} gmon.out gmon.sum 1677@end example 1678 1679@item 1680Repeat the last two steps as often as you wish. 1681 1682@item 1683Analyze the cumulative data using this command: 1684 1685@example 1686gprof @var{executable-file} gmon.sum > @var{output-file} 1687@end example 1688@end enumerate 1689 1690@node Assumptions 1691@section Estimating @code{children} Times 1692 1693Some of the figures in the call graph are estimates---for example, the 1694@code{children} time values and all the time figures in caller and 1695subroutine lines. 1696 1697There is no direct information about these measurements in the profile 1698data itself. Instead, @code{gprof} estimates them by making an assumption 1699about your program that might or might not be true. 1700 1701The assumption made is that the average time spent in each call to any 1702function @code{foo} is not correlated with who called @code{foo}. If 1703@code{foo} used 5 seconds in all, and 2/5 of the calls to @code{foo} came 1704from @code{a}, then @code{foo} contributes 2 seconds to @code{a}'s 1705@code{children} time, by assumption. 1706 1707This assumption is usually true enough, but for some programs it is far 1708from true. Suppose that @code{foo} returns very quickly when its argument 1709is zero; suppose that @code{a} always passes zero as an argument, while 1710other callers of @code{foo} pass other arguments. In this program, all the 1711time spent in @code{foo} is in the calls from callers other than @code{a}. 1712But @code{gprof} has no way of knowing this; it will blindly and 1713incorrectly charge 2 seconds of time in @code{foo} to the children of 1714@code{a}. 1715 1716@c FIXME - has this been fixed? 1717We hope some day to put more complete data into @file{gmon.out}, so that 1718this assumption is no longer needed, if we can figure out how. For the 1719novice, the estimated figures are usually more useful than misleading. 1720 1721@node How do I? 1722@chapter Answers to Common Questions 1723 1724@table @asis 1725@item How can I get more exact information about hot spots in my program? 1726 1727Looking at the per-line call counts only tells part of the story. 1728Because @code{gprof} can only report call times and counts by function, 1729the best way to get finer-grained information on where the program 1730is spending its time is to re-factor large functions into sequences 1731of calls to smaller ones. Beware however that this can introduce 1732artificial hot spots since compiling with @samp{-pg} adds a significant 1733overhead to function calls. An alternative solution is to use a 1734non-intrusive profiler, e.g.@: oprofile. 1735 1736@item How do I find which lines in my program were executed the most times? 1737 1738Use the @code{gcov} program. 1739 1740@item How do I find which lines in my program called a particular function? 1741 1742Use @samp{gprof -l} and lookup the function in the call graph. 1743The callers will be broken down by function and line number. 1744 1745@item How do I analyze a program that runs for less than a second? 1746 1747Try using a shell script like this one: 1748 1749@example 1750for i in `seq 1 100`; do 1751 fastprog 1752 mv gmon.out gmon.out.$i 1753done 1754 1755gprof -s fastprog gmon.out.* 1756 1757gprof fastprog gmon.sum 1758@end example 1759 1760If your program is completely deterministic, all the call counts 1761will be simple multiples of 100 (i.e., a function called once in 1762each run will appear with a call count of 100). 1763 1764@end table 1765 1766@node Incompatibilities 1767@chapter Incompatibilities with Unix @code{gprof} 1768 1769@sc{gnu} @code{gprof} and Berkeley Unix @code{gprof} use the same data 1770file @file{gmon.out}, and provide essentially the same information. But 1771there are a few differences. 1772 1773@itemize @bullet 1774@item 1775@sc{gnu} @code{gprof} uses a new, generalized file format with support 1776for basic-block execution counts and non-realtime histograms. A magic 1777cookie and version number allows @code{gprof} to easily identify 1778new style files. Old BSD-style files can still be read. 1779@xref{File Format, ,Profiling Data File Format}. 1780 1781@item 1782For a recursive function, Unix @code{gprof} lists the function as a 1783parent and as a child, with a @code{calls} field that lists the number 1784of recursive calls. @sc{gnu} @code{gprof} omits these lines and puts 1785the number of recursive calls in the primary line. 1786 1787@item 1788When a function is suppressed from the call graph with @samp{-e}, @sc{gnu} 1789@code{gprof} still lists it as a subroutine of functions that call it. 1790 1791@item 1792@sc{gnu} @code{gprof} accepts the @samp{-k} with its argument 1793in the form @samp{from/to}, instead of @samp{from to}. 1794 1795@item 1796In the annotated source listing, 1797if there are multiple basic blocks on the same line, 1798@sc{gnu} @code{gprof} prints all of their counts, separated by commas. 1799 1800@ignore - it does this now 1801@item 1802The function names printed in @sc{gnu} @code{gprof} output do not include 1803the leading underscores that are added internally to the front of all 1804C identifiers on many operating systems. 1805@end ignore 1806 1807@item 1808The blurbs, field widths, and output formats are different. @sc{gnu} 1809@code{gprof} prints blurbs after the tables, so that you can see the 1810tables without skipping the blurbs. 1811@end itemize 1812 1813@node Details 1814@chapter Details of Profiling 1815 1816@menu 1817* Implementation:: How a program collects profiling information 1818* File Format:: Format of @samp{gmon.out} files 1819* Internals:: @code{gprof}'s internal operation 1820* Debugging:: Using @code{gprof}'s @samp{-d} option 1821@end menu 1822 1823@node Implementation 1824@section Implementation of Profiling 1825 1826Profiling works by changing how every function in your program is compiled 1827so that when it is called, it will stash away some information about where 1828it was called from. From this, the profiler can figure out what function 1829called it, and can count how many times it was called. This change is made 1830by the compiler when your program is compiled with the @samp{-pg} option, 1831which causes every function to call @code{mcount} 1832(or @code{_mcount}, or @code{__mcount}, depending on the OS and compiler) 1833as one of its first operations. 1834 1835The @code{mcount} routine, included in the profiling library, 1836is responsible for recording in an in-memory call graph table 1837both its parent routine (the child) and its parent's parent. This is 1838typically done by examining the stack frame to find both 1839the address of the child, and the return address in the original parent. 1840Since this is a very machine-dependent operation, @code{mcount} 1841itself is typically a short assembly-language stub routine 1842that extracts the required 1843information, and then calls @code{__mcount_internal} 1844(a normal C function) with two arguments---@code{frompc} and @code{selfpc}. 1845@code{__mcount_internal} is responsible for maintaining 1846the in-memory call graph, which records @code{frompc}, @code{selfpc}, 1847and the number of times each of these call arcs was traversed. 1848 1849GCC Version 2 provides a magical function (@code{__builtin_return_address}), 1850which allows a generic @code{mcount} function to extract the 1851required information from the stack frame. However, on some 1852architectures, most notably the SPARC, using this builtin can be 1853very computationally expensive, and an assembly language version 1854of @code{mcount} is used for performance reasons. 1855 1856Number-of-calls information for library routines is collected by using a 1857special version of the C library. The programs in it are the same as in 1858the usual C library, but they were compiled with @samp{-pg}. If you 1859link your program with @samp{gcc @dots{} -pg}, it automatically uses the 1860profiling version of the library. 1861 1862Profiling also involves watching your program as it runs, and keeping a 1863histogram of where the program counter happens to be every now and then. 1864Typically the program counter is looked at around 100 times per second of 1865run time, but the exact frequency may vary from system to system. 1866 1867This is done is one of two ways. Most UNIX-like operating systems 1868provide a @code{profil()} system call, which registers a memory 1869array with the kernel, along with a scale 1870factor that determines how the program's address space maps 1871into the array. 1872Typical scaling values cause every 2 to 8 bytes of address space 1873to map into a single array slot. 1874On every tick of the system clock 1875(assuming the profiled program is running), the value of the 1876program counter is examined and the corresponding slot in 1877the memory array is incremented. Since this is done in the kernel, 1878which had to interrupt the process anyway to handle the clock 1879interrupt, very little additional system overhead is required. 1880 1881However, some operating systems, most notably Linux 2.0 (and earlier), 1882do not provide a @code{profil()} system call. On such a system, 1883arrangements are made for the kernel to periodically deliver 1884a signal to the process (typically via @code{setitimer()}), 1885which then performs the same operation of examining the 1886program counter and incrementing a slot in the memory array. 1887Since this method requires a signal to be delivered to 1888user space every time a sample is taken, it uses considerably 1889more overhead than kernel-based profiling. Also, due to the 1890added delay required to deliver the signal, this method is 1891less accurate as well. 1892 1893A special startup routine allocates memory for the histogram and 1894either calls @code{profil()} or sets up 1895a clock signal handler. 1896This routine (@code{monstartup}) can be invoked in several ways. 1897On Linux systems, a special profiling startup file @code{gcrt0.o}, 1898which invokes @code{monstartup} before @code{main}, 1899is used instead of the default @code{crt0.o}. 1900Use of this special startup file is one of the effects 1901of using @samp{gcc @dots{} -pg} to link. 1902On SPARC systems, no special startup files are used. 1903Rather, the @code{mcount} routine, when it is invoked for 1904the first time (typically when @code{main} is called), 1905calls @code{monstartup}. 1906 1907If the compiler's @samp{-a} option was used, basic-block counting 1908is also enabled. Each object file is then compiled with a static array 1909of counts, initially zero. 1910In the executable code, every time a new basic-block begins 1911(i.e., when an @code{if} statement appears), an extra instruction 1912is inserted to increment the corresponding count in the array. 1913At compile time, a paired array was constructed that recorded 1914the starting address of each basic-block. Taken together, 1915the two arrays record the starting address of every basic-block, 1916along with the number of times it was executed. 1917 1918The profiling library also includes a function (@code{mcleanup}) which is 1919typically registered using @code{atexit()} to be called as the 1920program exits, and is responsible for writing the file @file{gmon.out}. 1921Profiling is turned off, various headers are output, and the histogram 1922is written, followed by the call-graph arcs and the basic-block counts. 1923 1924The output from @code{gprof} gives no indication of parts of your program that 1925are limited by I/O or swapping bandwidth. This is because samples of the 1926program counter are taken at fixed intervals of the program's run time. 1927Therefore, the 1928time measurements in @code{gprof} output say nothing about time that your 1929program was not running. For example, a part of the program that creates 1930so much data that it cannot all fit in physical memory at once may run very 1931slowly due to thrashing, but @code{gprof} will say it uses little time. On 1932the other hand, sampling by run time has the advantage that the amount of 1933load due to other users won't directly affect the output you get. 1934 1935@node File Format 1936@section Profiling Data File Format 1937 1938The old BSD-derived file format used for profile data does not contain a 1939magic cookie that allows one to check whether a data file really is a 1940@code{gprof} file. Furthermore, it does not provide a version number, thus 1941rendering changes to the file format almost impossible. @sc{gnu} @code{gprof} 1942uses a new file format that provides these features. For backward 1943compatibility, @sc{gnu} @code{gprof} continues to support the old BSD-derived 1944format, but not all features are supported with it. For example, 1945basic-block execution counts cannot be accommodated by the old file 1946format. 1947 1948The new file format is defined in header file @file{gmon_out.h}. It 1949consists of a header containing the magic cookie and a version number, 1950as well as some spare bytes available for future extensions. All data 1951in a profile data file is in the native format of the target for which 1952the profile was collected. @sc{gnu} @code{gprof} adapts automatically 1953to the byte-order in use. 1954 1955In the new file format, the header is followed by a sequence of 1956records. Currently, there are three different record types: histogram 1957records, call-graph arc records, and basic-block execution count 1958records. Each file can contain any number of each record type. When 1959reading a file, @sc{gnu} @code{gprof} will ensure records of the same type are 1960compatible with each other and compute the union of all records. For 1961example, for basic-block execution counts, the union is simply the sum 1962of all execution counts for each basic-block. 1963 1964@subsection Histogram Records 1965 1966Histogram records consist of a header that is followed by an array of 1967bins. The header contains the text-segment range that the histogram 1968spans, the size of the histogram in bytes (unlike in the old BSD 1969format, this does not include the size of the header), the rate of the 1970profiling clock, and the physical dimension that the bin counts 1971represent after being scaled by the profiling clock rate. The 1972physical dimension is specified in two parts: a long name of up to 15 1973characters and a single character abbreviation. For example, a 1974histogram representing real-time would specify the long name as 1975``seconds'' and the abbreviation as ``s''. This feature is useful for 1976architectures that support performance monitor hardware (which, 1977fortunately, is becoming increasingly common). For example, under DEC 1978OSF/1, the ``uprofile'' command can be used to produce a histogram of, 1979say, instruction cache misses. In this case, the dimension in the 1980histogram header could be set to ``i-cache misses'' and the abbreviation 1981could be set to ``1'' (because it is simply a count, not a physical 1982dimension). Also, the profiling rate would have to be set to 1 in 1983this case. 1984 1985Histogram bins are 16-bit numbers and each bin represent an equal 1986amount of text-space. For example, if the text-segment is one 1987thousand bytes long and if there are ten bins in the histogram, each 1988bin represents one hundred bytes. 1989 1990 1991@subsection Call-Graph Records 1992 1993Call-graph records have a format that is identical to the one used in 1994the BSD-derived file format. It consists of an arc in the call graph 1995and a count indicating the number of times the arc was traversed 1996during program execution. Arcs are specified by a pair of addresses: 1997the first must be within caller's function and the second must be 1998within the callee's function. When performing profiling at the 1999function level, these addresses can point anywhere within the 2000respective function. However, when profiling at the line-level, it is 2001better if the addresses are as close to the call-site/entry-point as 2002possible. This will ensure that the line-level call-graph is able to 2003identify exactly which line of source code performed calls to a 2004function. 2005 2006@subsection Basic-Block Execution Count Records 2007 2008Basic-block execution count records consist of a header followed by a 2009sequence of address/count pairs. The header simply specifies the 2010length of the sequence. In an address/count pair, the address 2011identifies a basic-block and the count specifies the number of times 2012that basic-block was executed. Any address within the basic-address can 2013be used. 2014 2015@node Internals 2016@section @code{gprof}'s Internal Operation 2017 2018Like most programs, @code{gprof} begins by processing its options. 2019During this stage, it may building its symspec list 2020(@code{sym_ids.c:@-sym_id_add}), if 2021options are specified which use symspecs. 2022@code{gprof} maintains a single linked list of symspecs, 2023which will eventually get turned into 12 symbol tables, 2024organized into six include/exclude pairs---one 2025pair each for the flat profile (INCL_FLAT/EXCL_FLAT), 2026the call graph arcs (INCL_ARCS/EXCL_ARCS), 2027printing in the call graph (INCL_GRAPH/EXCL_GRAPH), 2028timing propagation in the call graph (INCL_TIME/EXCL_TIME), 2029the annotated source listing (INCL_ANNO/EXCL_ANNO), 2030and the execution count listing (INCL_EXEC/EXCL_EXEC). 2031 2032After option processing, @code{gprof} finishes 2033building the symspec list by adding all the symspecs in 2034@code{default_excluded_list} to the exclude lists 2035EXCL_TIME and EXCL_GRAPH, and if line-by-line profiling is specified, 2036EXCL_FLAT as well. 2037These default excludes are not added to EXCL_ANNO, EXCL_ARCS, and EXCL_EXEC. 2038 2039Next, the BFD library is called to open the object file, 2040verify that it is an object file, 2041and read its symbol table (@code{core.c:@-core_init}), 2042using @code{bfd_canonicalize_symtab} after mallocing 2043an appropriately sized array of symbols. At this point, 2044function mappings are read (if the @samp{--file-ordering} option 2045has been specified), and the core text space is read into 2046memory (if the @samp{-c} option was given). 2047 2048@code{gprof}'s own symbol table, an array of Sym structures, 2049is now built. 2050This is done in one of two ways, by one of two routines, depending 2051on whether line-by-line profiling (@samp{-l} option) has been 2052enabled. 2053For normal profiling, the BFD canonical symbol table is scanned. 2054For line-by-line profiling, every 2055text space address is examined, and a new symbol table entry 2056gets created every time the line number changes. 2057In either case, two passes are made through the symbol 2058table---one to count the size of the symbol table required, 2059and the other to actually read the symbols. In between the 2060two passes, a single array of type @code{Sym} is created of 2061the appropriate length. 2062Finally, @code{symtab.c:@-symtab_finalize} 2063is called to sort the symbol table and remove duplicate entries 2064(entries with the same memory address). 2065 2066The symbol table must be a contiguous array for two reasons. 2067First, the @code{qsort} library function (which sorts an array) 2068will be used to sort the symbol table. 2069Also, the symbol lookup routine (@code{symtab.c:@-sym_lookup}), 2070which finds symbols 2071based on memory address, uses a binary search algorithm 2072which requires the symbol table to be a sorted array. 2073Function symbols are indicated with an @code{is_func} flag. 2074Line number symbols have no special flags set. 2075Additionally, a symbol can have an @code{is_static} flag 2076to indicate that it is a local symbol. 2077 2078With the symbol table read, the symspecs can now be translated 2079into Syms (@code{sym_ids.c:@-sym_id_parse}). Remember that a single 2080symspec can match multiple symbols. 2081An array of symbol tables 2082(@code{syms}) is created, each entry of which is a symbol table 2083of Syms to be included or excluded from a particular listing. 2084The master symbol table and the symspecs are examined by nested 2085loops, and every symbol that matches a symspec is inserted 2086into the appropriate syms table. This is done twice, once to 2087count the size of each required symbol table, and again to build 2088the tables, which have been malloced between passes. 2089From now on, to determine whether a symbol is on an include 2090or exclude symspec list, @code{gprof} simply uses its 2091standard symbol lookup routine on the appropriate table 2092in the @code{syms} array. 2093 2094Now the profile data file(s) themselves are read 2095(@code{gmon_io.c:@-gmon_out_read}), 2096first by checking for a new-style @samp{gmon.out} header, 2097then assuming this is an old-style BSD @samp{gmon.out} 2098if the magic number test failed. 2099 2100New-style histogram records are read by @code{hist.c:@-hist_read_rec}. 2101For the first histogram record, allocate a memory array to hold 2102all the bins, and read them in. 2103When multiple profile data files (or files with multiple histogram 2104records) are read, the memory ranges of each pair of histogram records 2105must be either equal, or non-overlapping. For each pair of histogram 2106records, the resolution (memory region size divided by the number of 2107bins) must be the same. The time unit must be the same for all 2108histogram records. If the above containts are met, all histograms 2109for the same memory range are merged. 2110 2111As each call graph record is read (@code{call_graph.c:@-cg_read_rec}), 2112the parent and child addresses 2113are matched to symbol table entries, and a call graph arc is 2114created by @code{cg_arcs.c:@-arc_add}, unless the arc fails a symspec 2115check against INCL_ARCS/EXCL_ARCS. As each arc is added, 2116a linked list is maintained of the parent's child arcs, and of the child's 2117parent arcs. 2118Both the child's call count and the arc's call count are 2119incremented by the record's call count. 2120 2121Basic-block records are read (@code{basic_blocks.c:@-bb_read_rec}), 2122but only if line-by-line profiling has been selected. 2123Each basic-block address is matched to a corresponding line 2124symbol in the symbol table, and an entry made in the symbol's 2125bb_addr and bb_calls arrays. Again, if multiple basic-block 2126records are present for the same address, the call counts 2127are cumulative. 2128 2129A gmon.sum file is dumped, if requested (@code{gmon_io.c:@-gmon_out_write}). 2130 2131If histograms were present in the data files, assign them to symbols 2132(@code{hist.c:@-hist_assign_samples}) by iterating over all the sample 2133bins and assigning them to symbols. Since the symbol table 2134is sorted in order of ascending memory addresses, we can 2135simple follow along in the symbol table as we make our pass 2136over the sample bins. 2137This step includes a symspec check against INCL_FLAT/EXCL_FLAT. 2138Depending on the histogram 2139scale factor, a sample bin may span multiple symbols, 2140in which case a fraction of the sample count is allocated 2141to each symbol, proportional to the degree of overlap. 2142This effect is rare for normal profiling, but overlaps 2143are more common during line-by-line profiling, and can 2144cause each of two adjacent lines to be credited with half 2145a hit, for example. 2146 2147If call graph data is present, @code{cg_arcs.c:@-cg_assemble} is called. 2148First, if @samp{-c} was specified, a machine-dependent 2149routine (@code{find_call}) scans through each symbol's machine code, 2150looking for subroutine call instructions, and adding them 2151to the call graph with a zero call count. 2152A topological sort is performed by depth-first numbering 2153all the symbols (@code{cg_dfn.c:@-cg_dfn}), so that 2154children are always numbered less than their parents, 2155then making a array of pointers into the symbol table and sorting it into 2156numerical order, which is reverse topological 2157order (children appear before parents). 2158Cycles are also detected at this point, all members 2159of which are assigned the same topological number. 2160Two passes are now made through this sorted array of symbol pointers. 2161The first pass, from end to beginning (parents to children), 2162computes the fraction of child time to propagate to each parent 2163and a print flag. 2164The print flag reflects symspec handling of INCL_GRAPH/EXCL_GRAPH, 2165with a parent's include or exclude (print or no print) property 2166being propagated to its children, unless they themselves explicitly appear 2167in INCL_GRAPH or EXCL_GRAPH. 2168A second pass, from beginning to end (children to parents) actually 2169propagates the timings along the call graph, subject 2170to a check against INCL_TIME/EXCL_TIME. 2171With the print flag, fractions, and timings now stored in the symbol 2172structures, the topological sort array is now discarded, and a 2173new array of pointers is assembled, this time sorted by propagated time. 2174 2175Finally, print the various outputs the user requested, which is now fairly 2176straightforward. The call graph (@code{cg_print.c:@-cg_print}) and 2177flat profile (@code{hist.c:@-hist_print}) are regurgitations of values 2178already computed. The annotated source listing 2179(@code{basic_blocks.c:@-print_annotated_source}) uses basic-block 2180information, if present, to label each line of code with call counts, 2181otherwise only the function call counts are presented. 2182 2183The function ordering code is marginally well documented 2184in the source code itself (@code{cg_print.c}). Basically, 2185the functions with the most use and the most parents are 2186placed first, followed by other functions with the most use, 2187followed by lower use functions, followed by unused functions 2188at the end. 2189 2190@node Debugging 2191@section Debugging @code{gprof} 2192 2193If @code{gprof} was compiled with debugging enabled, 2194the @samp{-d} option triggers debugging output 2195(to stdout) which can be helpful in understanding its operation. 2196The debugging number specified is interpreted as a sum of the following 2197options: 2198 2199@table @asis 2200@item 2 - Topological sort 2201Monitor depth-first numbering of symbols during call graph analysis 2202@item 4 - Cycles 2203Shows symbols as they are identified as cycle heads 2204@item 16 - Tallying 2205As the call graph arcs are read, show each arc and how 2206the total calls to each function are tallied 2207@item 32 - Call graph arc sorting 2208Details sorting individual parents/children within each call graph entry 2209@item 64 - Reading histogram and call graph records 2210Shows address ranges of histograms as they are read, and each 2211call graph arc 2212@item 128 - Symbol table 2213Reading, classifying, and sorting the symbol table from the object file. 2214For line-by-line profiling (@samp{-l} option), also shows line numbers 2215being assigned to memory addresses. 2216@item 256 - Static call graph 2217Trace operation of @samp{-c} option 2218@item 512 - Symbol table and arc table lookups 2219Detail operation of lookup routines 2220@item 1024 - Call graph propagation 2221Shows how function times are propagated along the call graph 2222@item 2048 - Basic-blocks 2223Shows basic-block records as they are read from profile data 2224(only meaningful with @samp{-l} option) 2225@item 4096 - Symspecs 2226Shows symspec-to-symbol pattern matching operation 2227@item 8192 - Annotate source 2228Tracks operation of @samp{-A} option 2229@end table 2230 2231@node GNU Free Documentation License 2232@appendix GNU Free Documentation License 2233@include fdl.texi 2234 2235@bye 2236 2237NEEDS AN INDEX 2238 2239-T - "traditional BSD style": How is it different? Should the 2240differences be documented? 2241 2242example flat file adds up to 100.01%... 2243 2244note: time estimates now only go out to one decimal place (0.0), where 2245they used to extend two (78.67). 2246