You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The serial sim currently uses a watered down version of the POETS algorithm, which sees it split the volume into small cells, but iterates through them serially like any (3D) array and doesn’t operate using events. This is fine for x86, it benefits from it.
Having this as an option is good for some debug, but it would be quite useful to have a serial simulator that uses the full POETS version as seen in inc/gals.h and inc/sync.h. It could even receive and sends messages in the form of packets that are placed in a queue or set of some form and can be received in order or randomly. Debugging the code when you can output whenever you want would be very useful. It would also help to demonstrate the algorithm to people who don’t understand POETS in detail, or for people to run who don’t have access to POETS boxes.
Long term (not a goal for this issue) would be to make it parallel in some way, using queues or MPI, but more focussed on small cells and communication, following the POETS design flow, as opposed to osprey DPD which has large cells and packs message full of beads. It would make for a very nice comparison.
The text was updated successfully, but these errors were encountered:
The serial sim currently uses a watered down version of the POETS algorithm, which sees it split the volume into small cells, but iterates through them serially like any (3D) array and doesn’t operate using events. This is fine for x86, it benefits from it.
Having this as an option is good for some debug, but it would be quite useful to have a serial simulator that uses the full POETS version as seen in inc/gals.h and inc/sync.h. It could even receive and sends messages in the form of packets that are placed in a queue or set of some form and can be received in order or randomly. Debugging the code when you can output whenever you want would be very useful. It would also help to demonstrate the algorithm to people who don’t understand POETS in detail, or for people to run who don’t have access to POETS boxes.
Long term (not a goal for this issue) would be to make it parallel in some way, using queues or MPI, but more focussed on small cells and communication, following the POETS design flow, as opposed to osprey DPD which has large cells and packs message full of beads. It would make for a very nice comparison.
The text was updated successfully, but these errors were encountered: