TGB Linux gprof profile data
by Christopher \"Pacula\" Corkum · in Torque Game Builder · 10/18/2006 (7:30 am) · 2 replies
I don't know if this will be much - or ANY - use to anybody, but I'm only starting to figure out the results myself, I thought I'd post these logs here to see if anybody else would find them of use. These were generated with OS=LINUX (naturally), COMPILER=GCC4, and BUILD=DEBUG. I also added gprof's -pg flag to the CFLAGS.DEBUG and LFLAGS.DEBUG lines in conf.UNIX.mk. I did this both with the 1.1.1 code included in the 'Linux' download, as well as the 1.1.2 code from the OS X version.
Anyway, here are the resulting logs:
http://www.sweetleafstudios.com/tgb_debug_gprof-1.1.1.log.gz
http://www.sweetleafstudios.com/tgb_debug_gprof-1.1.2.log.gz
In each case, I ran the program, played the default 'falling planes' demo for several seconds, exited out to the GUI, and then quit the program.
One new mostly-unrelated thing came up of possible interest - 1.1.1 seemed to be notably faster than before. This possibly could have been because I built it with gcc4 rather than gcc3.3, but that's a guess at best. It's still not as fast as 1.1.2 is (though much closer), nor even remotely close to 'playable', but still seemed to be an improvement.
Note that I have no intention of leaving this 'as is' with hopes of somebody taking over from here - though I'm certainly open to any suggestions or help :). My current plan is to try another profiler that was just recommended to me by some associates (oprofile) and see what that tells me. After that, I'm going to delve into the TGB code and start playing around with the functions that the profilers have flagged as possibly problematic - possibly even stub out/comment out some of them completely - and see what happens.
Anyway, here are the resulting logs:
http://www.sweetleafstudios.com/tgb_debug_gprof-1.1.1.log.gz
http://www.sweetleafstudios.com/tgb_debug_gprof-1.1.2.log.gz
In each case, I ran the program, played the default 'falling planes' demo for several seconds, exited out to the GUI, and then quit the program.
One new mostly-unrelated thing came up of possible interest - 1.1.1 seemed to be notably faster than before. This possibly could have been because I built it with gcc4 rather than gcc3.3, but that's a guess at best. It's still not as fast as 1.1.2 is (though much closer), nor even remotely close to 'playable', but still seemed to be an improvement.
Note that I have no intention of leaving this 'as is' with hopes of somebody taking over from here - though I'm certainly open to any suggestions or help :). My current plan is to try another profiler that was just recommended to me by some associates (oprofile) and see what that tells me. After that, I'm going to delve into the TGB code and start playing around with the functions that the profilers have flagged as possibly problematic - possibly even stub out/comment out some of them completely - and see what happens.
#2
10/19/2006 (8:28 am)
TGE contains a profiler which you can use to narrow down problematic functions. I'm not sure if it was stripped during the creation of TGB or not, but assuming it hasn't have a read of tdn.garagegames.com/wiki/Torque/Profiler 
Torque 3D Owner Rodney Rindels - Torqued
Divide By Zero