Project

General

Profile

Task #3002

Task #2675: bonded CUDA offload task

consider splitting bonded work into local/nonlocal

Added by Szilárd Páll 4 months ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
mdrun
Target version:
-
Difficulty:
uncategorized
Close

Description

Splitting the bonded task into local and non-local would allow more uniform, simpler and easier to maintain code, in particular with GPU offload where the other short-range task (nonbondeds) is already split in this manner.
This will likely also allow some performance improvements to be had as a smaller nonlocal task will speed up the critical path allowing the halo-exchange to start earlier.

Also available in: Atom PDF