Profiling Boost Graph Library with gprof
Hello All, I have been trying to profile a few graph algorithms written using Boost Graph Library. I am using gprof for profiling. My graph has 100 vertices. It takes more than 10 minutes to compile, I suspect it is because of the size of the graph. The profile does not show any details though. It only shows the amount of time spent by the profiler. What am I doing wrong? Any help is greatly appreciated! Thanks. __________________________________________ Yahoo! DSL Something to write home about. Just $16.99/mo. or less. dsl.yahoo.com
On Jan 9, 2006, at 3:24 PM, Dave wrote:
I have been trying to profile a few graph algorithms written using Boost Graph Library. I am using gprof for profiling.
We've never had good luck using gprof for profiling BGL algorithms. There are far too many calls to small, inlined functions, which tend to make it very hard for profilers to give accurate results.
My graph has 100 vertices. It takes more than 10 minutes to compile, I suspect it is because of the size of the graph. The profile does not show any details though. It only shows the amount of time spent by the profiler. What am I doing wrong? Any help is greatly appreciated! Thanks.
We typically place a few timers in the code, manually, and use those for performance estimates. Doug
Doug Gregor wrote:
On Jan 9, 2006, at 3:24 PM, Dave wrote:
I have been trying to profile a few graph algorithms written using Boost Graph Library. I am using gprof for profiling.
We've never had good luck using gprof for profiling BGL algorithms. There are far too many calls to small, inlined functions, which tend to make it very hard for profilers to give accurate results.
My graph has 100 vertices. It takes more than 10 minutes to compile, I suspect it is because of the size of the graph. The profile does not show any details though. It only shows the amount of time spent by the profiler. What am I doing wrong? Any help is greatly appreciated! Thanks.
We typically place a few timers in the code, manually, and use those for performance estimates.
Or use callgrind + kcachegrind. This one emulates the program so works more reliably than gprof. - Volodya
participants (3)
-
Dave -
Doug Gregor -
Vladimir Prus