Project

General

Profile

Task #3001

Task #2675: bonded CUDA offload task

explore simplifying virial and shift force reduction

Added by Szilárd Páll 2 months ago.

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

Description

Doing a single-stage instead of the two-stage local/global atomic reduction should both simplify the bonded kernels as well as likely improve performance (we do #SHIFTs*3 4-byte atomic on global and the central-conditional local reduction is unlikely to be more work).

Also available in: Atom PDF