I have a fixed wall of particles set in place by the “fix setforce” command. There’s no need to calculate forces between these wall particles so I chose to use the neigh_modify command with the exclude option to save needless computation.
Here’s my question:
The code runs faster (by about 2 CPU seconds every 100 timesteps) without the exclude option in place. How could this be?
I have a fixed wall of particles set in place by the “fix setforce” command. There’s no need to calculate forces between these wall particles so I chose to use the neigh_modify command with the exclude option to save needless computation.
Here’s my question:
The code runs faster (by about 2 CPU seconds every 100 timesteps) without the exclude option in place. How could this be?
I have a fixed wall of particles set in place by the "fix setforce"
command. There's no need to calculate forces between these wall
particles so I chose to use the neigh_modify command with the exclude
option to save needless computation.Here's my question:
The code runs faster (by about 2 CPU seconds every 100 timesteps)
without the exclude option in place. How could this be?
brian,
have you looked at where the time is lost?
is it the pair part of the code or the neigborlist part?
can you give an example of system size and how many
processors you are using? these days optimizing memory
accesses is becoming increasingly important and having
more cache misses can be worse than having more compute
work to do.
cheers,
axel.
2% is pretty small. If you compare the stats printed from the
2 runs, you should see less neighbors when you exclude. But
the pair time may go down and the neigh time may go up since
it has to do more checks to do the exclusion. If you're not excluding
much, then it probably isn't worth it.
Steve