This repository has been archived by the owner on Oct 28, 2023. It is now read-only.
Additional Performance gains #16
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
|
The main reason I made this a 2nd PR is the drawback to splitting better_match into 2 separate loops. It actually makes a difference due to cache hits, but it means that 'black on black' guide areas take a lot of loops to stack up cost. So, I put the color cost first, even if alpha is small. |
I ran cargo flamegraph, and it turns out a huge portion of the runtime was spent in find_match and find_better_match. It's a very, very hot loop. Almost all of the work done in the inner loop (find_better_match) is a function of two u8s... it can be precomputed! Also, the alpha masks can be rendered into these precomputed cost functions, avoiding the need to do any alpha computations in the loop. I ran all the examples and couldn't find any visible artifacts.
I found a few small bugs while looking at @zicklag's comments on EmbarkStudios#14 First: there is a numerical precision bug with the calculation of distance gaussians. The exp() function used to be f64::exp(), and I was using f32::exp(). Second: there were missing entries in the precomputed function table. Loop bounds are exclusive... but 256u8 is not a u8... so it needs 0..=255u8 which was made for this situation.
…ecialization for ExactSizeIterator Rust iterators have some great specialization features - and .zip() has fantastic performance when joining two `DoubleEndedIterator + ExactSizeIterator` - which std::Slice::Iter implements! The trick is that it can combine the bounds checks for the zipped-together items. And the benches look faster than they used to be - so llvm may be optimizing the bound checks away. These changes improve the example01 performance down to 13.06 real 39.56 user from: 15.50 real 50.72 sys
austinjones
force-pushed
the
pr.addl-performance
branch
from
c4aa777 to
55e4b7e
Compare
Merged
|
This PR was merged into #28, thanks again for the changes! |
|
No problem @Jake-Shadle. Thanks for merging! |
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 builds on PR1 to improve performance by another ~15%.
Rust iterators have some great specialization features - and .zip() has a neat specialization when joining two
DoubleEndedIterator + ExactSizeIterator- which std::Slice::Iter implements! See zip.rs.