1 .\" Copyright (c) 1983, 1990 The Regents of the University of California.
2 .\" All rights reserved.
4 .\" Redistribution and use in source and binary forms are permitted provided
5 .\" that: (1) source distributions retain this entire copyright notice and
6 .\" comment, and (2) distributions including binaries display the following
7 .\" acknowledgement: ``This product includes software developed by the
8 .\" University of California, Berkeley and its contributors'' in the
9 .\" documentation or other materials provided with the distribution and in
10 .\" all advertising materials mentioning features or use of this software.
11 .\" Neither the name of the University nor the names of its contributors may
12 .\" be used to endorse or promote products derived from this software without
13 .\" specific prior written permission.
14 .\" THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR IMPLIED
15 .\" WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF
16 .\" MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
18 .\" @(#)gprof.1 6.6 (Berkeley) 7/24/90
25 .Nd display call graph profile data
29 .Op Ar a.out Op Ar gmon.out ...
32 produces an execution profile of C, Pascal, or Fortran77 programs.
33 The effect of called routines is incorporated in the profile of each caller.
34 The profile data is taken from the call graph profile file
36 default) which is created by programs
37 that are compiled with the
46 option also links in versions of the library routines
47 that are compiled for profiling.
49 reads the given object file (the default is
51 and establishes the relation between it's symbol table
52 and the call graph profile from
54 If more than one profile file is specified,
57 output shows the sum of the profile information in the given profile files.
60 calculates the amount of time spent in each routine.
61 Next, these times are propagated along the edges of the call graph.
62 Cycles are discovered, and calls into a cycle are made to share the time
64 The first listing shows the functions
65 sorted according to the time they represent
66 including the time of their call graph descendents.
67 Below each function entry is shown its (direct) call graph children,
68 and how their times are propagated to this function.
69 A similar display above the function shows how this function's time and the
70 time of its descendents is propagated to its (direct) call graph parents.
72 Cycles are also shown, with an entry for the cycle as a whole and
73 a listing of the members of the cycle and their contributions to the
74 time and call counts of the cycle.
76 Second, a flat profile is given,
77 similar to that provided by
79 This listing gives the total execution times, the call counts,
80 the time in milleseconds the call spent in the routine itself, and
81 the time in milleseconds the call spent in the routine itself including
84 Finally, an index of the function names is provided.
86 The following options are available:
89 suppresses the printing of statically declared functions.
90 If this option is given, all relevant information about the static function
91 (e.g., time samples, calls to other functions, calls from other functions)
92 belongs to the function loaded just before the static function in the
96 suppresses the printing of a description of each field in the profile.
98 the static call graph of the program is discovered by a heuristic
99 that examines the text space of the object file.
100 Static-only parents or children are shown
101 with call counts of 0.
106 suppresses the printing of the graph profile entry for routine
108 and all its descendants
109 (unless they have other ancestors that aren't suppressed).
115 may be given with each
122 suppresses the printing of the graph profile entry for routine
124 (and its descendants) as
126 above, and also excludes the time spent in
128 (and its descendants) from the total and percentage time computations.
139 prints the graph profile entry of only the specified routine
147 may be given with each
154 prints the graph profile entry of only the routine
156 and its descendants (as
158 above) and also uses only the times of the printed routines
159 in total time and percentage computations.
165 may be given with each
181 will delete any arcs from routine
185 This can be used to break undesired cycles.
189 Only one pair of routine names may be given with each
195 is produced that represents
196 the sum of the profile information in all the specified profile files.
197 This summary profile file may be given to later
198 executions of gprof (probably also with a
200 to accumulate profile data across several runs of an
204 displays routines that have zero usage (as shown by call counts
205 and accumulated time).
206 This is useful with the
208 option for discovering which routines were never called.
214 the namelist and text space.
216 dynamic call graph and profile.
218 summarized dynamic call graph and profile.
226 .Em An Execution Profiler for Modular Programs ,
228 S. Graham, P. Kessler, M. McKusick;
229 Software - Practice and Experience,
230 Vol. 13, pp. 671-685, 1983.
232 .Em gprof: A Call Graph Execution Profiler ,
233 by S. Graham, P. Kessler, M. McKusick;
234 Proceedings of the SIGPLAN '82 Symposium on Compiler Construction,
235 SIGPLAN Notices, Vol. 17, No 6, pp. 120-126, June 1982.
240 The granularity of the sampling is shown, but remains
242 We assume that the time for each execution of a function
243 can be expressed by the total time for the function divided
244 by the number of times the function is called.
245 Thus the time propagated along the call graph arcs to the function's
246 parents is directly proportional to the number of times that
249 Parents that are not themselves profiled will have the time of
250 their profiled children propagated to them, but they will appear
251 to be spontaneously invoked in the call graph listing, and will
252 not have their time propagated further.
253 Similarly, signal catchers, even though profiled, will appear
254 to be spontaneous (although for more obscure reasons).
255 Any profiled children of signal catchers should have their times
256 propagated properly, unless the signal catcher was invoked during
257 the execution of the profiling routine, in which case all is lost.
259 The profiled program must call
261 or return normally for the profiling information to be saved