You suggested one model for where the computation time is going in
GSIM in your email of 19 March:
"When combined with what I saw before (most of the calls to gtnext are
from the
DC volume), this implies that our biggest time sink is particles
showering
in the shielding foils, which we then track through the rest of the
detector."
My current mental picture of what takes a lot of cpu time is that all
the time is spent tracking secondaries which are primarily *within* the
inert structures, which is different from your model.
One way we could check this is to turn off secondaries in the inert
volumes, e.g. the minitorus, and check the number of calls in the DC
volumes. Your message of March 13 stated the following for how many
calls were made in each section:
"The main geometry sections:
Foil: 28%
EC1: 1%
Mini: 11%
LAS2: 60%
For foil, 70% of the time is spent in the SHD1 piece, which appears to
be a 25
cm long cone about the beamline.
For the LAS2, 65% is spent in one of the DC volumes (roughly equal
except DC1B
which is small). About 18% in the EC, 8% in the CC, 4% in the SC, and
about 6%
in the MG+MGSR pieces (ie, the magnet)."
If the 65% number drops dramatically when you turn off minitorus
secondaries, it would support your model. Did you already look at this?
- Will