| Re: cpu times [message #13555 is a reply to message #13554] |
Fri, 01 June 2012 11:15  |
Mohammad Al-Turany
Messages: 518
Registered: April 2004
Location: GSI, Germany
|
first-grade participant |
From: *gsi.de
|
|
Hi,
First of all I am not sure if we really need this discussion here, but any way I run the profiler on the svn rev 15651 (yesterday). The standard sim, digi and then I profile the reco with valgrind and callgrind, all with 10 events. The result is shown below:
I normalize the time to the exec time, if you look at the picture above you will see that we spend:
Kalman filter: 57 %
STT track finding 13 %
STT+MVD tracking 15 %
MVD Riemann 7 %
STTMVDGEM 7 %
now if you look at the "calls" you will notice that the STT and Kalman (first three in the list above) are taking about 85 % of the time, however going down in the picture you will see that 70 % of the 85% are spend in Geane and glpk code, to make it clear:
we spend 70 % of processing time in external packages
The rest of the code which is 30 % of time has definitely not that much time spent in STL but in other algorithms and IO etc. So assuming that C arrays are much faster (Which I do not agree on!) it make no sense to contaminate the code with non-readable stuff because of speed. One has also to think about debugging and tracing the code e.g: out of bound problems which we already have with the c arrays in the tracking code.
best regards,
Mohammad
-
Attachment: task.jpg
(Size: 255.22KB, Downloaded 436 times)
|
|
|
|
GSI Forum
Search
Help
Members
Login
Home
