search

plumed / plumed2

Development version of plumed 2
https://www.plumed.org
GNU Lesser General Public License v3.0
364 stars 289 forks source link

CONTACTMAP saturates all the available cores #1121

Open riccardocapelli opened 1 month ago

riccardocapelli commented 1 month ago

Hello,

I recently used CONTACTMAP (more in detail, I used the CMDIST), and it worked fine (I used the v2.9 branch). I moved some replicas on another machine with the same gromacs version (2024.3, but it returns the same problem with 2023.5) patched with the 2.11.dev version, and the same calculation saturated all the cores of the machine, generating more threads than the number of cores (ignoring the instructions given to gromacs).

For the (few) steps it did, the results were ok also with this thread issue (and maybe this is why the regtests did not find the problem).

I tried to look into the history of the contactmap.cpp file, and I think it broke in the hack-the-tree integration (I tried to follow it, but it's a bit confusing to me, sorry).

Thanks!

R

Iximiel commented 1 month ago

Can you reproduce the thread issue as a driver in input? In that way I may try to run it through perf without gromacs ( and why not with, just to check if we have problems in the inner code and not in the action) to see where we are using all that CPU