search

gandalfcode / gandalf

GANDALF (Graphical Astrophysics code for N-body Dynamics And Lagrangian Fluids)
GNU General Public License v2.0
44 stars 12 forks source link

Nbody fixes and paper figure scripts #137

Closed dhubber closed 7 years ago

dhubber commented 7 years ago

Small pull request with some small fixes and modifications to the N-body parts, including :

Also added files and scripts to generate star-only and hybrid plummer spheres, run their simulations and generate figures for the paper.

dhubber commented 7 years ago

Also I added conditional OpenMP parallelisation to only use OpenMP when a reasonably high number of particles per thread are being simulated (to prevent OpenMP being excessively deployed when only a small number of stars/sinks are present).

rbooth200 commented 7 years ago

Looks ok to me. I have one question though: does it not make more sense to set the maximum number of OpenMP threads used rather than switch it on or off. Given the work is N^2 you could do nMaxThread = 1 + Npart^2 / 16^2 or similar...

dhubber commented 7 years ago

Looks ok to me. I have one question though: does it not make more sense to set the maximum number of OpenMP threads used rather than switch it on or off. Given the work is N^2 you could do nMaxThread = 1 + Npart^2 / 16^2 or similar..

Hmmm, perhaps. I just went with the solution I was thinking of at the time but it's easy enough to change since it's only a few lines really.

giovanni-rosotti commented 7 years ago

Hi, this looks mostly fine to me. Isn't more natural to call maxNbodyOpenMp minNbodyOpenMp (it's the minimum number of particles needed to use openMP)?

David, are you going to do what Richard suggested, or do you think it's not worth the effort?

dhubber commented 7 years ago

David, are you going to do what Richard suggested, or do you think it's not worth the effort?

Oh well, it's an improvement but not essential (i.e. it's not a bug fix). But it also kind of applies to other OpenMP parts of the code (e.g. when there are only a few active particles/cells) so I'd be tempted to say make an Issue (for the whole code) and look at it later.