-
Notifications
You must be signed in to change notification settings - Fork 590
Shriya/tp overlap patch #205
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.
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
Shriya/tp overlap patch #205
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If I understand correctly, the only functional change is to change some memory accesses to use atomicAdd_system. We also add some more information to an error message, which requires plumbing in layer_name throughout the whole PyTorch infrastructure. If this debugging info is not essential, I feel the extra messiness is not worth it. If it is absolutely needed, then at the very least we should change layer_name to something less awkward and confusing, maybe ub_layer_id or ub_layer_label.
b26cdba to
aaced36
Compare
|
Agreed, removed the logging commit from this PR |
timmoon10
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
|
@ShriyaPalsamudram please sign your commit. |
Signed-off-by: Sangkug Lym <[email protected]>
df43bf0 to
d178cea
Compare
userbuffer pushsend/recv fix with atomicAdd_system Signed-off-by: Sangkug Lym <[email protected]> Co-authored-by: Sangkug Lym <[email protected]>
userbuffer pushsend/recv fix with atomicAdd_system Signed-off-by: Sangkug Lym <[email protected]> Co-authored-by: Sangkug Lym <[email protected]>
Upstreaming @erhoo82's branch: tp_overlap_patch