-
Notifications
You must be signed in to change notification settings - Fork 164
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
perf: cache-friendly MemoryCell
#1669
Closed
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
Oppen
force-pushed
the
perf/compact_memory_cell
branch
from
March 15, 2024 19:16
3e461b0
to
cfcab8c
Compare
Oppen
requested review from
igaray,
fmoletta,
entropidelic,
juanbono and
pefontana
as code owners
March 15, 2024 19:16
Store data and metadata inline as a single `[u64; 4]` with `32` bytes alignment, fitting a whole number of cells per cache line to reduce evictions and double sharing and to avoid split line access. Besides performance, an observable change is that now `Memory::get_value` always returns a `Cow::Owned` variant because the decoding process always creates a new value.
Oppen
force-pushed
the
perf/compact_memory_cell
branch
from
March 15, 2024 19:24
cfcab8c
to
b328c10
Compare
Benchmark Results for unmodified programs 🚀
|
This reverts commit 9bd3905.
Already merged #1672 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Refactors
MemoryCell
to store a simple[u64; 4]
to use the cache more appropriately.This structure is now a fixed 32 bytes with 32 bytes alignment, meaning any given cell
in the VM memory will never be split across two different cache lines, and eviction will
be minimized in virtue of being the optimal size for a
Felt
of 252 bits.The cost of this change is no longer being able to extract the
MemoryCell
value as asimple reference, but requiring a conversion for the "normal"
MaybeRelocatable
, thatis, it implies extra computation and extra copying to extract values. This behaves well
for Lambdaworks where the data is inline in the array, but quite a hit for the
BigUint
implementation, which now allocates and frees memory much more often.
Upgrading is recommended only when using Lambdaworks. The tradeoff is deemed acceptable
since only the Lambdaworks (actually
types-rs
) implementation is preserved for the1.x branch.
Checklist