Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add a function to pre-set infos for all streams/devices/... #665

Draft
wants to merge 2 commits into
base: master
Choose a base branch
from
Draft

Add a function to pre-set infos for all streams/devices/... #665

wants to merge 2 commits into from

Commits on Aug 7, 2024

  1. Add a function to pre-set infos for all streams/devices/... 

    Lazy/dynamic allocation of infos can be convenient for the users,
but also detrimental to predictive behaviors. Examples of drawbacks
are when testing memory constrained setups (e.g., when the test uses
zone_malloc to dynamically allocate objects in the info object
constructor, that may fail undeterministically because it depends
how many streams call the constructor), or small scale performance
runs where the number of cublas objects to initialize vary depending
on the run. Calling the new function incurs determinisitic overheads.
    @therault
    therault committed Aug 7, 2024
    Configuration menu
    Copy the full SHA
    956b2ab View commit details
    Browse the repository at this point in the history

  2. In multi-GPU runs, we need to set the target GPU before calling the u… 

    …ser's object info constructor (and potentially destructor). Allow the device to define a function to do so in order to keep the user code unchanged. This adds an additional function pointer call, but this call only happens once per info and per stream, ideally during warmup/initialization time
    @therault
    therault committed Aug 7, 2024
    Configuration menu
    Copy the full SHA
    8279fbd View commit details
    Browse the repository at this point in the history