Optimise SSZ encoding and decoding #55
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #55 +/- ##
==========================================
- Coverage 58.62% 58.33% -0.29%
==========================================
Files 8 8
Lines 290 324 +34
==========================================
+ Hits 170 189 +19
- Misses 120 135 +15 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
michaelsproul
changed the title
| len: 0, | ||
| expected: 1, | ||
| }) | ||
| } else if mem::size_of::<T>() == 1 && mem::align_of::<T>() == 1 { |
There was a problem hiding this comment.
Using size_of and align_of removes the need to add any bounds on T. Other crates like bytemuck would have required us to constrain T quite a lot, which sort of defeats the point of a generic impl.
Even using TypeId would have required us to add 'static.
The other advantage of this is that it works for types that encode as u8, like the ParticipationFlags in the BeaconState.
There was a problem hiding this comment.
Yeah I guess it could be, but we never use bools in any consensus data structures, do we?
Would be a good test case
| // Safety: We've verified T is u8, so Vec<T> is Vec<u8> | ||
| // and bytes.to_vec() produces Vec<u8> | ||
| let vec_u8 = bytes.to_vec(); | ||
| let vec_t = unsafe { std::mem::transmute::<Vec<u8>, Vec<T>>(vec_u8) }; |
There was a problem hiding this comment.
Should add a test for a type that is not u8
Summary of changes
unsafefast path for decoding lists and vectors ofu8chunks_exactand an explicit loop instead oftry_fold.Benchmarks
Additional Info
Depends on a new release of
ethereum_sszfor:Encoding is not meaningfully faster, but would need to be optimised in
sszitself (we need to optimise theVecimpl).