Project

General

Profile

Task #3143

move the non-force accumulation/accounting tasks out of the GPU wait functions

Added by Szilárd Páll about 1 month ago.

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

Description

Current code works, but it is too error-prone due to the complex buffer ops/reduction code-path. We would benefit from having a clear separation between the force-only wait / reduction and energy or timing accounting.

To improve/split up:
  • gpu_try_finish_task()
  • pme_gpu_wait_and_reduce()

Also available in: Atom PDF