Gather-scatter is missing Min/Max operations for device aware MPI backend

ExtremeFLOW / neko

/ᐠ. ｡.ᐟ\ᵐᵉᵒʷˎˊ˗

https://neko.cfd/

Other

159 stars 27 forks source link

Gather-scatter is missing Min/Max operations for device aware MPI backend #1333

Open njansson opened 5 days ago

njansson commented 5 days ago

These has not been implemented due to the lack of floating point support in atomicMin and atomicMax routines in CUDA/HIP. A proposed solution is to implemented these using CAS.

njansson commented 5 days ago

This is related to the issues observed in #1328