I am doing an initial investigation of the geometries that include a
number of 'MANY' volumes, instead of these volumes being defined 'ONLY'.
The geometry of the CC has 2 'MANY' volumes.
The geometry of the beamline (geom_foil) has 7 MANY volumes, and
The geometry of the MINI TORUS (geom_mg) has 17+ MANY volumes, and
The geometry of the ST has 8 MANY volumes.
The slowdown due to each MANY volume is combinatoric, so this is rather
bad news. If anybody who wrote some of this code is willing to help
clean it up and change the MANY into ONLY (making sure they don't conflict
with other volumes, and if they do, that they do so correctly.)
An example of possible speed gains:
Running the standard speed test on P90 laptop, 100 events with the FOIL
geometry all set to VACUUM material (to make a fair comparison) :
Code as is: 601.69 sec for 100 events.
MANY replaced with ONLY: 456.43 sec for 100 events.
The problem is that some of the MANY volumes changed to ONLY may be
interfering with other ONLY volumes causing unpredictable effects (until
we sort out all the detail.)
Anybody willing to volunteer some time to take a system and start cleaning
up those geometries ?
I'm starting with "FOIL" since it is unlikely Dale Conwell will touch it.
Best Regards,
Maurik
----------------------------------------------------------------------
Maurik Holtrop web: http://www.physics.unh.edu/Maurik
UNH Physics Department tel: (603) 862-2019 | fax: (603) 862-2998
DeMeritt Hall mailto:maurik.holtrop@unh.edu
Durham, NH 03824 Home: (603) 659-8765
----------------------------------------------------------------------