round reduction size up to nearest power of two to avoid overloading cache #65
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Outdated suggestions cannot be applied.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.
Hi, in
reduction.py
,ReductionKernel._get_basic_kernel
is cached according to its argumentsmaxls
andnd
, wheremaxls
is the smaller of the reduction size andself.init_local_size
(= 1024 on my machine). If there are a lot of small reductions, there will be many different versions of the basic kernel in the cache, too many to fit in the cache. However, one of the first things that the basic kernel does is to roundmaxls
up to the nearest power of two. This patch does the rounding up before caching so that there are at most lg(self.init_local_size) different versions of the basic kernel for each value ofnd
. (It also does the rounding again afterwards, to avoid breaking anything.) I think this fixes the problem with small reductions without costing anything for larger reductions.