OT: A breakthrough astrophysics code rapidly models stellar collisions

[Gary Scott]

A breakthrough astrophysics code rapidly models stellar collisions:
https://phys.org/news/2021-04-breakthrough-astrophysics-code-rapidly-stellar.html

C of course. Imagine what they could achieve if they wrote it in a
proper language and used co-arrays...

[spectrum]

Thanks very much for sharing an interesting article :)
Because I had no idea what "HPX" means (which is used in the above project),
I've searched the net and it looks like some sort of framework (?) for parallel calculations
(based on "task-based" parallelism).

https://en.wikipedia.org/wiki/HPX
https://github.com/STEllAR-GROUP/hpx

Octo-Tiger: a new, 3D hydrodynamic code for stellar mergers that uses HPX parallelisation
https://academic.oup.com/mnras/advance-article/doi/10.1093/mnras/stab937/6219861

This article shows some comparison between HPX and MPI:

Performance Comparison of HPX Versus Traditional Parallelization
Strategies for the Discontinuous Galerkin Method
https://link.springer.com/article/10.1007/s10915-019-00960-z

> * Active Global Address Space
> While the threading subsystem operates within a single private address space, HPX extends its programming model to distributed runs via an active global address space (AGAS). Global address spaces attempt to emulate the ease of programming on a single node, while still maintaining tight control over data locality necessary for writing a performant distributed code. Two well-known global address space models are UPC [22] and Co-Array Fortran [48]. While global address space models like UPC’s partitioned global address space (PGAS) are more data-centric, e.g. by exposing pointers to memory addresses on different nodes, HPX approaches global address spaces in a more object-oriented manner.

> The key difference between AGAS and other global address space models is its native support for component migration. HPX’s AGAS layer allows the developer to relocate components to different localities during runtime. ...

I have no idea what is the advantage/disadvantage of this approach, but
they seem to be aware of Coarray Fortran also (I guess HPX is kind of an experimental
project for new parallel approaches).