Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

fix caffe stream #2984

Merged
merged 3 commits into from
Aug 11, 2016
Merged

fix caffe stream #2984

merged 3 commits into from
Aug 11, 2016

Conversation

piiswrong
Copy link
Contributor

@piiswrong piiswrong commented Aug 10, 2016

@tqchen @HrWangChengdu

@Taco-W
Copy link
Contributor

Taco-W commented Aug 10, 2016

@piiswrong I still cannot pass example after adding cudaStreamSyn on my machine.

@piiswrong
Copy link
Contributor Author

@HrWangChengdu "device" works for me but 'local_allreduce_device' doesn't. It's so strange. They should be the same thing

@Taco-W
Copy link
Contributor

Taco-W commented Aug 10, 2016

Me too. 'Device' could work with or without streamSync. 'local_allreduce_device' cannot work in either setting.

@piiswrong
Copy link
Contributor Author

@HrWangChengdu Looks like caffeop is fine. It's local_allreduce_device that's broken @tqchen

@piiswrong
Copy link
Contributor Author

@HrWangChengdu http://docs.nvidia.com/cuda/cuda-driver-api/stream-sync-behavior.html#stream-sync-behavior
looks like this fix is not necessary. Caffe uses the default stream. It blocks all other streams

@tqchen
Copy link
Member

tqchen commented Aug 10, 2016

should be fixed by now, please check

@Taco-W
Copy link
Contributor

Taco-W commented Aug 10, 2016

@tqchen It works now!
@piiswrong Yes. Example code could run without this fix. I'll double check this later this week. Thank you.

@piiswrong piiswrong merged commit d98e823 into apache:master Aug 11, 2016
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants