-
Notifications
You must be signed in to change notification settings - Fork 195
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
Improve performance of recursive #163
Improve performance of recursive #163
Conversation
This comment was marked as off-topic.
This comment was marked as off-topic.
@microsoft-github-policy-service agree company="OROCHI NETWORK PTE. LTD." |
@chiro-hiro Is this still a draft? Could you please rebase with the main branch and let me know when this is ready? |
Thank you, I will rebase my code and try to complete this in the weekend. |
4438e1d
to
ced1675
Compare
ced1675
to
179f716
Compare
Here is the original code, there are multiple instance of // produce a recursive SNARK
let mut recursive_snark: Option<
RecursiveSNARK<
G1,
G2,
TrivialTestCircuit<<G1 as Group>::Scalar>,
CubicCircuit<<G2 as Group>::Scalar>,
>,
> = None;
for _i in 0..num_steps {
let res = RecursiveSNARK::prove_step(
&pp,
recursive_snark,
circuit_primary.clone(),
circuit_secondary.clone(),
vec![<G1 as Group>::Scalar::ONE],
vec![<G2 as Group>::Scalar::ZERO],
);
assert!(res.is_ok());
recursive_snark = Some(res.unwrap());
}
assert!(recursive_snark.is_some());
let recursive_snark = recursive_snark.unwrap();
// verify the recursive SNARK
let res = recursive_snark.verify(
&pp,
num_steps,
vec![<G1 as Group>::Scalar::ONE],
vec![<G2 as Group>::Scalar::ZERO],
);
assert!(res.is_ok()); Instead of create new instance of // produce a recursive SNARK
let mut recursive_snark = RecursiveSNARK::<
G1,
G2,
TrivialTestCircuit<<G1 as Group>::Scalar>,
CubicCircuit<<G2 as Group>::Scalar>,
>::new(
&pp,
&circuit_primary,
&circuit_secondary,
vec![<G1 as Group>::Scalar::one()],
vec![<G2 as Group>::Scalar::zero()],
);
for _i in 0..num_steps {
let res = recursive_snark.prove_step(
&pp,
&circuit_primary,
&circuit_secondary,
vec![<G1 as Group>::Scalar::one()],
vec![<G2 as Group>::Scalar::zero()],
);
assert!(res.is_ok());
}
// verify the recursive SNARK
let res = recursive_snark.verify(
&pp,
num_steps,
vec![<G1 as Group>::Scalar::ONE],
vec![<G2 as Group>::Scalar::ZERO],
);
assert!(res.is_ok()); In my observations, the circuit and z0 were passed as immutable parameters. I'm also change the code to borrow these values instead of |
Fantastic! Thanks @chiro-hiro! I had a quick comment about avoiding a small piece of code in prove_step. Please check. |
My pleasure, Indeed, that part of code looks ugly but I can't find the better way to keep the backwards compatibility. |
Hi @srinathsetty, there are pretty much changes were made in |
Hi @chiro-hiro! I'll see if I can find a fix for the issue noted above. We'll prioritize this PR. |
Thank you, I will focus on fixing the conflict with the main branch. |
91c143d
to
8253bb7
Compare
@chiro-hiro this is neat! I've opened orochi-network#1 to refine it further - you might find this or a similar way of getting the same results interesting. |
@huitseeker Thank you, everything looks great. I do check and rerun the test/benchmark as well. All set, just wait for @srinathsetty to review this PR. |
@chiro-hiro clippy warnings should disappear as soon as you rebase on the latest master. |
Hi, @chiro-hiro Could we please update this branch soon (with a rebase + a change to address the discussion thread above) so we can merge this ahead of other pending PRs? (cc: @huitseeker) |
Hi @srinathsetty, sorry for the delay. I have rebasing my fork with the upstream's main branch. |
- Update function arguments to use borrowing instead of passing ownership
Updates to version 0.21 See - microsoft/Nova#163
Updates to version 0.22 See - microsoft/Nova#163
Updates to version 0.22 See - microsoft/Nova#163
Updates to version 0.22 See - microsoft/Nova#163
Updates to version 0.22 See - microsoft/Nova#163
* Include number of instances in transcript * Clarify Sumcheck and verifier * Fix clippy
* skeleton of what is to be done * wip inner and outer sumchecks done * fix borrow checker issues * batched spartan prove done * Small improvements - Irregular shape yields error - Remove (W,E) clones - Use Math module for log2 - Generate tau using only 1 challenge - Remove some `clone()` - Comments * Fix evaluation points * Implement verifier * `CompressedSNARK` implementation * Tests fixed, failing * fix sumcheck, test passes * bound -> bind * remove todo * add test * Document BatchedRelaxedR1CSSNARKTrait and SuperNova CompressedSNARK. * working * - Remove debug code - Add Sumcheck::verify_batch - Uniformize batching of evals - Comments * Adress review comments - add itertools for multiunzip - avoid clones - avoid recomputing sizes for asserts * Fix asserts * batched ppsnark * Fix scaling * - Fix padding of poly_W - Optimized evaluation - use batch_diff_size - use multiunzip * rebase fixes * Only create taus once. * Avoid clone by not allocating PowPolynomial. * Riff on review. * refactor: Implement `.zip_eq()` for equal length iterator safety - Replaced `.zip()` with `.zip_eq()` across various files including `sumcheck.rs`, `mod.rs`, `batched_ppsnark.rs`, `batched.rs` and `snark.rs`, ensuring equal length enforcement in iterations and reducing out-of-bound errors. - Updated a multiunzip resulting in a double iteration in batched_ppsnark::prove. * add benchmark * fix witnes * fix: clean up iterators * feat: Introduce new utility macros in spartan module - Introduced a macro `zip_with` in `src/spartan/mod.rs` to simplify the process of zipping multiple iterators and applying a single parallel function. - Developed additional macros `nested_tuple` and `nested_idents` for detailed nested tuple patterns and identifiers respectively. - Implemented `zip_all` macro that acts as a fold-right zipping operator for all given expressions. * refactor: Refactor batched.rs to enhance code readability - Refactored `src/spartan/batched.rs` to use `zip_with()` macro improving the readability, simplification, and consistent usage in the code. - Inserted comments highlighting sections marked for review or possible future code simplification. * refactor: use zip_with! in batch_ppsnark - Updated `batched_ppsnark.rs` in the `src/spartan` directory with uses of zip_with macro * ignore local macro doctest * Use zip_with. * Implement and use zip_with_for_each. * Dubious double zip_with. * More zip_with. * Small fixes - `zip_with` for `batched_ppsnark` - Complete comment of `BatchedRelaxedR1CSSNARK` - Restore Abomonation bounds - Fix nits in `multilinear.rs` * refactor: more zip_with instances (microsoft#153) * refactor: more instances of zip_with * Supernova: Some rewrites (microsoft#154) * refactor: more zip_with * feat: Implement `FromIterator` for polynomial types - Enhanced `EqPolynomial`, `MultilinearPolynomial`, and `SparsePolynomial` with `FromIterator` implementation. * refactor: use instaces of Polynomial Fromiterator where possible * refactor: remove unused impls of FromIterator * refactor: address review comments * refactor: enhance scope visibility scope in Spartan and Supernova modules (microsoft#156) Absolute visibility modifiers tend to not inadvertently make things public when you move code, contrarily to (super). - Changed the visibility scope of several fields and methods within multiple substrates in the `ppsnark.rs` file to be within the `crate::spartan`. - Altered the module import path in the `circuit.rs` file from `super` to `crate::supernova`. - Modified the visibility of the `WitnessBoundSumcheck` in `batched_ppsnark.rs`, the visibility scope is now set to `crate::spartan`. - Scoped `batch_eval_prove` and `batch_eval_verify` functions within `crate::spartan` in `snark.rs`. * Avoid EqPolynomial when only for evals. * refactor: a few more instances of EqPolynomial::evals_from_points - in `multilinear.rs` and `ipa_pc.rs`, replaces the previous method `EqPolynomial::new(r.to_vec()).evals()` with a more efficient `EqPolynomial::evals_from_points(r)`. * doc: fix a few nits (microsoft#160) * refactor: rename powers -> squares - Renamed the `powers` method to `squares` in `PowPolynomial` struct within `src/spartan/polys/power.rs` and changed its visibility level. * fix: comment * fix: comment of batch * derive fixes * fix: compressed snark benchmark - Introduced `compressed-snark-supernova` within the `Cargo.toml` file. - fixed Rust * ci: Add trigger for PRs to batched_spartan (microsoft#165) * More zip with (microsoft#158) * Expand zip_with macros to accommodate more use. * Clippy. * Move Spartan macros to own module. * Only use zip_with_fn to implement convenience macros. * Use convenience zip_with macros more. * Remove for_each variants of zip_with macros. * Remove flat_map variants of zip_with. --------- Co-authored-by: porcuquine <[email protected]> * Introduce proper `MaskedEq` poly (microsoft#162) * Introduce proper `MaskedEq` poly * refactor: Refactor MaskedEq poly for improved testing and performance - Introduce lifetime parameters to `MaskedEqPolynomial` struct to improve memory management. - Modify the `MaskedEqPolynomial` creation process so it now requires a reference to an existing `EqPolynomial`. - Adapt corresponding tests to reflect the above changes. * Fix fmt & derive debug --------- Co-authored-by: François Garillot <[email protected]> * Restore zip_with_for_each. (microsoft#166) Co-authored-by: porcuquine <[email protected]> * fix benchmark (microsoft#168) * Sumcheck update (microsoft#163) * Include number of instances in transcript * Clarify Sumcheck and verifier * Fix clippy * Streamline macro syntax (microsoft#169) * refactor: Refactor Spartan code for streamlined use of `zip_with_fn` Details: - Simplified the implementation of `BatchedRelaxedR1CSSNARKTrait` by replacing usage of `zip_with_iter` and related macros with `zip_with_fn`. - Eliminated `zip_with_iter`, `zip_with_par_iter`, `zip_with_into_iter`, and `zip_with_into_par_iter` macros from the codebase, reducing repetition and enhancing code organization. * refactor: Refactor zip functions - Replaced the `zip_with_fn!` function with the `zip_with!` function in various files, - Refactored `zip_with` and `zip_with_for_each` macros in `src/spartan/macros.rs` for improved handling of iterator projection specification. - Consolidated `zip_all_with_fn` macro's functionality into `zip_all` and eliminated unused macros. * fix: remove brackets * Small fixes (microsoft#170) * refactor: removed needless copies of `R1CSShape` - Modified the `primary_r1cs_shapes` function in `supernova/mod.rs` to return a vector of references, improving memory efficiency. - Use those `primary_r1cs_shapes` in the `setup` and `prove` methods of `CompressedSNARK`. - Altered the `BatchedRelaxedR1CSSNARKTrait` trait functions in `traits/snark.rs` to accept vector of references instead of reference, * refactor: Refactor polynomial size assertion checks in sumcheck.rs - Refactors the length assertion checks in the `prove_quad_batch` and `prove_cubic_with_additive_term_batch` methods in `src/spartan/sumcheck.rs`. - Simplified the code by replacing multiple `assert_eq` calls with a loop. * refactor: Remove "nested_tuple" macro from spartan macros - Removed the "nested_tuple" macro from the "src/spartan/macros.rs" file. * refactor: pedantic calls in tests --------- Co-authored-by: Matej Penciak <[email protected]> Co-authored-by: Adrian Hamelink <[email protected]> Co-authored-by: porcuquine <[email protected]> Co-authored-by: Hanting Zhang <[email protected]> Co-authored-by: Samuel Burnham <[email protected]> Co-authored-by: porcuquine <[email protected]> Co-authored-by: Adrian Hamelink <[email protected]> Co-authored-by: Matej Penciak <[email protected]>
Im try to minimize the cost to
.copy()
,.clone()
multiple structures around. You might expect ~10% in total performance and ~5% in memory saving.2 tests are broken, so I'm try to fix it. If this worth, please let me know.