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Outdated suggestions cannot be applied.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.
Added a kernel with atomics for a single col reduce. It gets back the performance that we lost from the refactor but I don't like that I don't understand why.
In our reduction benchmark this is significantly slower than the looped version. Moreover the 8-bit and 16-bit versions are painfully slow ~5x in comparison to 32 bits and 64 bits doesn't work at all so I am routing only the 32 bit types there.
I am not proposing we merge this but if anybody feels like playing with it and figuring out how it can be slower to use a 2x faster kernel according to the micro-benchmark I 'd be very interested to learn why.