This repository has been archived by the owner on May 3, 2024. It is now read-only.
enabled miopen pooling #6
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@dagamayank - Yes, I noticed that cuDNN disables MAX pool, and after digging in a bit, this is what I found: Looks like it was added as a workaround for indexing for in place computation in a layer after max pooling. I am still not sure why this is an issue or would affect miopen pool. We need some testing, especially training runs on the scenerio mentioned in the original issue. I am not sure if cuDNN fixed that problem, the date of the work-around to turn off max pooling is from Jan 2016. |
|
Thank you for the PR, Mayank. We'll take a look at training accuracy with these mods. |
|
A slightly modified version of this PR has been integrated into the 'tuning' branch. Closing. Thanks! |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This fix enables caffe to use MIOpen for MAX_POOLING. This was disabled for some reason which cuDNN does not / did not support. Looking at NVIDIA's github itself makes me believe that they have fixed this limitation.
Highly recommend to run training with this fix to ensure this does not manifest in accuracy issues. I am suspecting all this while we were not using MIOpen for pooling in our training runs.
cc\ @ashishfarmer This might also be the reason for the discrepancy you saw between ocl and hip caffe for pooling.