Date: Wed, 1 Feb 1995 09:44:47 -0500 From: L Jonas Olsson <ljo@amcell2.caisr.cwru.edu> To: freebsd-questions@freefall.cdrom.com Subject: Bogus profiling info Message-ID: <199502011444.JAA09582@amcell2.caisr.cwru.edu>
next in thread | raw e-mail | index | archive | help
I'm profiling some code to find what parts to optimize. The problem is that most of the time is spent in some mystery place. This is on FreeBSD 2.0R, using -pg option to gcc, and gprof to get the results. The program uses a device driver that re-programs the timer0 to 400 Hz during some parts of the program. It also forks another program (not compiled with -pg) that it sends data to using pipes. All linked libraries are compiled for profiling. The profile looks something like: (only the flat profile part, the mystery functions are called spontaneously, so the graph gives no more info. % cumulative self self total time seconds seconds calls ms/call ms/call name 14.8 39.98 39.98 _Mystery1 [11] 11.9 72.31 32.32 346 93.42 303.23 _func1 [10] 7.4 92.48 20.17 346 58.29 58.29 _func2 [12] 6.4 109.73 17.25 _Mystery2 [13] 6.1 126.32 16.60 346 47.96 406.16 _func3 [9] 5.6 141.53 15.20 _Mystery3 [15] 4.8 154.45 12.92 _Mystery4 [16] 4.7 167.14 12.69 346 36.68 36.68 _func4 [18] 4.5 179.28 12.14 323966 0.04 0.05 _func5 [14] The func* functions have statistics that look fine. The Mystery* functions are functions in my program, but I know that they are never called (Besides if they were called there would be complete info about number of calls etc for these functions). I can guess that some of the time is spent in system calls, but which and called from what functions. What else can generate this bogus info? Is there any other way to improve the profiling information? Jonas
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199502011444.JAA09582>