search

choderalab / alchemy

Alchemical tools for OpenMM
9 stars 3 forks source link

Look into performance regression introduced by exceptions overhaul in 1.2.1 #66

Open jchodera opened 7 years ago

jchodera commented 7 years ago

We should investigate whether reworking some of the alchemical states can speed up simulations.

cc: https://github.com/choderalab/yank/issues/577#issuecomment-277737517

andrrizzi commented 7 years ago

This is probably the case. The first scheme I implemented was optimized but a little ugly. @Lnaden suggested the current scheme which is way more readable since we were still debugging back then, but I think there's good room for performance improvement.

jchodera commented 7 years ago

Another possibility: Once we move to using a transparent System cache, we can add an option to have alchemy to create an unmodified System (no CustomForce objects) with scaled charges if appropriate. This will be just as fast as the standard unmodified System for many of the alchemical intermediates.

Lnaden commented 7 years ago

As in add all alchemical electrostatic interactions as exceptions through the standard NonbondedForce?

jchodera commented 7 years ago

This would be an alternative to the "alchemical state is controllable via context parameters" approach.