Supernova `CompressedSNARK` #131

huitseeker · 2023-11-22T17:34:35Z

This is #108 re-opened from a feature branch.

src/traits/snark.rs

adr1anh · 2023-11-28T09:46:36Z

I think the order of merge should be #140 and then #136. The former has some factored out code snippets which are copied from batched_ppsnark, so I can update #136 to use these.

src/spartan/batched.rs

porcuquine · 2023-11-28T22:23:58Z

src/spartan/batched.rs

+    let (num_rounds_x, num_rounds_y): (Vec<_>, Vec<_>) = S
+      .iter()
+      .map(|s| (s.num_cons.log_2(), s.num_vars.log_2() + 1))
+      .unzip();
+    let num_rounds_x_max = *num_rounds_x.iter().max().unwrap();
+    let num_rounds_y_max = *num_rounds_y.iter().max().unwrap();


You could do this all in one pass with fold.

I'm not sure what you mean exactly. We need both the number of rounds and the max for x and y.

Sorry for late response, and I'm not particularly saying this is necessary. However, I just meant that you could still do that in one pass with fold. The accumulated value could be (Vec<usize>, Vec<usize>, usize, usize) and the function to fold would be something like (untested):

|acc, s| { let num_rounds_x = s.num_cons.log_2(); let num_rounds_y = s.num_vars.log_2(); (num_rounds_x, num_rounds_y, num_rounds_x.max(acc.2), num_rounds_y.max(acc.3)) }

src/spartan/batched.rs

porcuquine · 2023-11-28T23:41:48Z

src/spartan/batched.rs

+    )?;
+
+    // Collect evaluation point for each instance
+    let r_x = num_rounds_x


A little more documentation of exactly what's going on here would not hurt.

https://hackmd.io/@adr1anh/BJw1g0aBT#Non-uniform-instances

src/spartan/polys/multilinear.rs

src/spartan/batched_ppsnark.rs

src/spartan/polys/power.rs

- Irregular shape yields error - Remove (W,E) clones - Use Math module for log2 - Generate tau using only 1 challenge - Remove some `clone()` - Comments

This backports the following Arecibo PRs: - #2 - #3 - #10 - #16 - #23 - #30 - #28 - #41 - #45 - #50 - #56 - #51 - #72 - #92 - #95 - #97 - #101 - #110 - #106 - #112 - #114 - #119 - #120 - #127 - #123 - #131 - #174 - #175 - #182 Co-authored-by: WYATT <wyattbenno@gmail.com> Co-authored-by: Hanting Zhang <hantingz@usc.edu> Co-authored-by: Ming <hero78119@gmail.com> Co-authored-by: porcuquine <porcuquine@users.noreply.github.com> Co-authored-by: Samuel Burnham <45365069+samuelburnham@users.noreply.github.com> Co-authored-by: Matej Penciak <96667244+mpenciak@users.noreply.github.com> Co-authored-by: Adrian Hamelink <adrian.hamelink@gmail.com>

This implements [Supernova](https://eprint.iacr.org/2022/1758), allowing: - a 'pay-as-you-go' cost structure for folding operations, through the SuperNova folding scheme, - a final SNARK that efficiently compresses an instance of this folded proof, through batching techniques. References: - the [blog post](https://blog.lurk-lang.org/posts/arecibo-supernova/#technical-release-note-supernova-protocol-integration-into-nova) goes into our construction, and links to two more specialized notes on [the `CompressedSNARK` for Supernova](https://hackmd.io/@adr1anh/BJw1g0aBT) along with our variant of the [public input padding issue](https://hackmd.io/@adr1anh/Sy08YaVBa). - the Readme at `src/supernova/Readme.md` This backports the following Arecibo PRs: - lurk-lang/arecibo#2 - lurk-lang/arecibo#3 - lurk-lang/arecibo#10 - lurk-lang/arecibo#16 - lurk-lang/arecibo#23 - lurk-lang/arecibo#30 - lurk-lang/arecibo#28 - lurk-lang/arecibo#41 - lurk-lang/arecibo#45 - lurk-lang/arecibo#50 - lurk-lang/arecibo#56 - lurk-lang/arecibo#51 - lurk-lang/arecibo#72 - lurk-lang/arecibo#92 - lurk-lang/arecibo#95 - lurk-lang/arecibo#97 - lurk-lang/arecibo#101 - lurk-lang/arecibo#110 - lurk-lang/arecibo#106 - lurk-lang/arecibo#112 - lurk-lang/arecibo#114 - lurk-lang/arecibo#119 - lurk-lang/arecibo#120 - lurk-lang/arecibo#127 - lurk-lang/arecibo#123 - lurk-lang/arecibo#131 - lurk-lang/arecibo#174 - lurk-lang/arecibo#175 - lurk-lang/arecibo#182 Co-authored-by: WYATT <wyattbenno@gmail.com> Co-authored-by: Hanting Zhang <hantingz@usc.edu> Co-authored-by: Ming <hero78119@gmail.com> Co-authored-by: porcuquine <porcuquine@users.noreply.github.com> Co-authored-by: Samuel Burnham <45365069+samuelburnham@users.noreply.github.com> Co-authored-by: Matej Penciak <96667244+mpenciak@users.noreply.github.com> Co-authored-by: Adrian Hamelink <adrian.hamelink@gmail.com>

This gathers the changes to the pre-processing SNARK excerpted from the Supernova implementation in PR microsoft#283. The main changes extracted here are the introduction of the masked eq polynomial (and its use fixing [this issue](https://hackmd.io/@adr1anh/Sy08YaVBa)), additional sum-check tooling, removal of two calls to evaluation argument. Main reference Arecibo PRs: - lurk-lang/arecibo#106 - lurk-lang/arecibo#131 - lurk-lang/arecibo#174 - lurk-lang/arecibo#182 - Enhancement of polynomial related code in Spartan protocol including new polynomial types, modified existing types, better evaluation methods, and improved polynomial operations. - Introduction of `squares` function and change in the generation of `t_pow` in Spartan. - Addition of a new polynomial type through `MaskedEqPolynomial` with methods for its creation and evaluation. - Enhancements in `UniPoly` struct by addition of `PartialEq`, and `Eq` traits. - Improvements in `snark.rs` for proving and verifying batch evaluation claims, leveraging `gamma` and `rho` for random linear combinations, and optimizing various variable computations. - Updates in `multilinear.rs` with refactoring, optimization, error handling, and supplementing unit tests. - Refactor in `spartan/mod.rs` with import updates, function overhauls, struct visibility changes, and asynchronous operations for efficient calculations. - Additions and amendments in `sumcheck.rs` for batch verification, handling of vectors of claims, handling of cubic bounds with additive terms, visibility adjustments, and typo fixes. - Modifications in `eq.rs` including a debug derive for `EqPolynomial`, enhanced visibility of `r` vector, provision of `evals_from_points` for enhanced evaluation, and addition of `FromIterator` implementation. Co-authored-by: porcuquine <porcuquine@users.noreply.github.com> Co-authored-by: Matej Penciak <96667244+mpenciak@users.noreply.github.com> Co-authored-by: Adrian Hamelink <adrian.hamelink@gmail.com>

…ns (#293) This gathers the changes to the pre-processing SNARK excerpted from the Supernova implementation in PR #283. The main changes extracted here are the introduction of the masked eq polynomial (and its use fixing [this issue](https://hackmd.io/@adr1anh/Sy08YaVBa)), additional sum-check tooling, removal of two calls to evaluation argument. Main reference Arecibo PRs: - lurk-lang/arecibo#106 - lurk-lang/arecibo#131 - lurk-lang/arecibo#174 - lurk-lang/arecibo#182 - Enhancement of polynomial related code in Spartan protocol including new polynomial types, modified existing types, better evaluation methods, and improved polynomial operations. - Introduction of `squares` function and change in the generation of `t_pow` in Spartan. - Addition of a new polynomial type through `MaskedEqPolynomial` with methods for its creation and evaluation. - Enhancements in `UniPoly` struct by addition of `PartialEq`, and `Eq` traits. - Improvements in `snark.rs` for proving and verifying batch evaluation claims, leveraging `gamma` and `rho` for random linear combinations, and optimizing various variable computations. - Updates in `multilinear.rs` with refactoring, optimization, error handling, and supplementing unit tests. - Refactor in `spartan/mod.rs` with import updates, function overhauls, struct visibility changes, and asynchronous operations for efficient calculations. - Additions and amendments in `sumcheck.rs` for batch verification, handling of vectors of claims, handling of cubic bounds with additive terms, visibility adjustments, and typo fixes. - Modifications in `eq.rs` including a debug derive for `EqPolynomial`, enhanced visibility of `r` vector, provision of `evals_from_points` for enhanced evaluation, and addition of `FromIterator` implementation. Co-authored-by: porcuquine <porcuquine@users.noreply.github.com> Co-authored-by: Matej Penciak <96667244+mpenciak@users.noreply.github.com> Co-authored-by: Adrian Hamelink <adrian.hamelink@gmail.com>

This implements [Supernova](https://eprint.iacr.org/2022/1758), allowing: - a 'pay-as-you-go' cost structure for folding operations, through the SuperNova folding scheme, - a final SNARK that efficiently compresses an instance of this folded proof, through batching techniques. References: - the [blog post](https://blog.lurk-lang.org/posts/arecibo-supernova/#technical-release-note-supernova-protocol-integration-into-nova) goes into our construction, and links to two more specialized notes on [the `CompressedSNARK` for Supernova](https://hackmd.io/@adr1anh/BJw1g0aBT) along with our variant of the [public input padding issue](https://hackmd.io/@adr1anh/Sy08YaVBa). - the Readme at `src/supernova/Readme.md` This backports the following Arecibo PRs: - lurk-lang/arecibo#2 - lurk-lang/arecibo#3 - lurk-lang/arecibo#10 - lurk-lang/arecibo#16 - lurk-lang/arecibo#23 - lurk-lang/arecibo#30 - lurk-lang/arecibo#28 - lurk-lang/arecibo#41 - lurk-lang/arecibo#45 - lurk-lang/arecibo#50 - lurk-lang/arecibo#56 - lurk-lang/arecibo#51 - lurk-lang/arecibo#72 - lurk-lang/arecibo#92 - lurk-lang/arecibo#95 - lurk-lang/arecibo#97 - lurk-lang/arecibo#101 - lurk-lang/arecibo#110 - lurk-lang/arecibo#106 - lurk-lang/arecibo#112 - lurk-lang/arecibo#114 - lurk-lang/arecibo#119 - lurk-lang/arecibo#120 - lurk-lang/arecibo#127 - lurk-lang/arecibo#123 - lurk-lang/arecibo#131 - lurk-lang/arecibo#174 - lurk-lang/arecibo#175 - lurk-lang/arecibo#182 Co-authored-by: WYATT <wyattbenno@gmail.com> Co-authored-by: Hanting Zhang <hantingz@usc.edu> Co-authored-by: Ming <hero78119@gmail.com> Co-authored-by: porcuquine <porcuquine@users.noreply.github.com> Co-authored-by: Samuel Burnham <45365069+samuelburnham@users.noreply.github.com> Co-authored-by: Matej Penciak <96667244+mpenciak@users.noreply.github.com> Co-authored-by: Adrian Hamelink <adrian.hamelink@gmail.com>

huitseeker requested review from adr1anh, porcuquine and mpenciak November 22, 2023 17:34

huitseeker mentioned this pull request Nov 22, 2023

Supernova CompressedSNARK #108

Closed

adr1anh reviewed Nov 23, 2023

View reviewed changes

src/traits/snark.rs Outdated Show resolved Hide resolved

src/traits/snark.rs Outdated Show resolved Hide resolved

adr1anh force-pushed the batched_spartan branch from d15b333 to 66507af Compare November 23, 2023 16:02

porcuquine reviewed Nov 28, 2023

View reviewed changes

src/spartan/batched.rs Outdated Show resolved Hide resolved

porcuquine reviewed Nov 28, 2023

View reviewed changes

src/spartan/batched.rs Outdated Show resolved Hide resolved

porcuquine reviewed Nov 28, 2023

View reviewed changes

huitseeker added the supernova label Nov 29, 2023

huitseeker force-pushed the batched_spartan branch 3 times, most recently from 77eb249 to 6190738 Compare November 30, 2023 15:43

huitseeker commented Dec 1, 2023

View reviewed changes

src/spartan/polys/multilinear.rs Outdated Show resolved Hide resolved

huitseeker commented Dec 1, 2023

View reviewed changes

src/spartan/batched_ppsnark.rs Outdated Show resolved Hide resolved

adr1anh reviewed Dec 1, 2023

View reviewed changes

src/spartan/polys/power.rs Outdated Show resolved Hide resolved

porcuquine mentioned this pull request Dec 5, 2023

Support SuperNova CompressedSNARK. lurk-lab/lurk-beta#936

Merged

winston-h-zhang force-pushed the batched_spartan branch from 7ba38c3 to 355ee3d Compare December 5, 2023 16:14

porcuquine mentioned this pull request Dec 7, 2023

SuperNova CompressedSNARK benchmark #167

Closed

mpenciak and others added 9 commits December 8, 2023 10:09

Fix evaluation points

Verified

This commit was signed with the committer’s verified signature.

huitseeker François Garillot

GPG key ID: E7645C6B883A1E9A

Verified
Learn about vigilant mode

6599de3

Implement verifier

Verified

This commit was signed with the committer’s verified signature.

huitseeker François Garillot

GPG key ID: E7645C6B883A1E9A

Verified
Learn about vigilant mode

d672123

adr1anh self-requested a review December 11, 2023 16:40

adr1anh approved these changes Dec 11, 2023

View reviewed changes

mpenciak approved these changes Dec 11, 2023

View reviewed changes

huitseeker added this pull request to the merge queue Dec 11, 2023

Merged via the queue into dev with commit 8e411d0 Dec 11, 2023

huitseeker mentioned this pull request Dec 12, 2023

ci: various clean-ups after big merges #173

Merged

huitseeker deleted the batched_spartan branch December 13, 2023 18:34

huitseeker mentioned this pull request Dec 18, 2023

feat: implement SuperNova microsoft/Nova#283

Open

huitseeker mentioned this pull request Jan 17, 2024

refactor: Improve Spartan SNARK polynomial computations and evaluations microsoft/Nova#293

Merged

huitseeker mentioned this pull request Feb 9, 2024

🚨 Performance regression in #279 #281

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Supernova `CompressedSNARK` #131

Supernova `CompressedSNARK` #131

huitseeker commented Nov 22, 2023 •

edited

Loading

adr1anh commented Nov 28, 2023

porcuquine Nov 28, 2023

adr1anh Dec 1, 2023

porcuquine Dec 5, 2023

porcuquine Nov 28, 2023

huitseeker Dec 8, 2023

Supernova CompressedSNARK #131

Supernova CompressedSNARK #131

Conversation

huitseeker commented Nov 22, 2023 • edited Loading

adr1anh commented Nov 28, 2023

porcuquine Nov 28, 2023

Choose a reason for hiding this comment

adr1anh Dec 1, 2023

Choose a reason for hiding this comment

porcuquine Dec 5, 2023

Choose a reason for hiding this comment

porcuquine Nov 28, 2023

Choose a reason for hiding this comment

huitseeker Dec 8, 2023

Choose a reason for hiding this comment

Supernova `CompressedSNARK` #131

Supernova `CompressedSNARK` #131

huitseeker commented Nov 22, 2023 •

edited

Loading