Project

General

Profile

Task #2695

Task #2675: bonded CUDA offload task

bonded GPU module timing

Added by Szilárd Páll about 1 year ago. Updated 10 months ago.

Status:
New
Priority:
Low
Assignee:
-
Category:
mdrun
Target version:
Difficulty:
simple
Close

Description

The timing implementation is straightforward, but not critical given that, due to the buggy cudaEven-timing facilities, it is not possible to time kernels when there is concurrent work launched (i.e. multiple streams).

Host-side (launch) timing should also be added to avoid leaking time into "Rest".

History

#1 Updated by Szilárd Páll about 1 year ago

  • Priority changed from Normal to Low

Note: not a blocker for the beta (nor the release IMO).

#2 Updated by Szilárd Páll 12 months ago

  • Description updated (diff)

#3 Updated by Gerrit Code Review Bot 12 months ago

Gerrit received a related patchset '2' for Issue #2695.
Uploader: Szilárd Páll ()
Change-Id: gromacs~release-2019~Ib3f18b8285b979b818ab79713253bc7f7bb89e2a
Gerrit URL: https://gerrit.gromacs.org/8784

#4 Updated by Szilárd Páll 11 months ago

Host side cycle counting has been fixed. Device-side timing is TODO

#5 Updated by Paul Bauer 11 months ago

  • Target version changed from 2019 to 2019.1

moved to 2019.1

#6 Updated by Mark Abraham 10 months ago

  • Target version changed from 2019.1 to 2020

Too useless for 2019 branch

Also available in: Atom PDF