Skip to content

Commit

Permalink
feat: add Dynamic Dory helper methods (#68)
Browse files Browse the repository at this point in the history
# Rationale for this change

We need to add the Dynamic Dory commitment scheme. This is the first PR.

# What changes are included in this PR?

* Added the `compute_dynamic_standard_basis_vecs` method, a helper
method used to compute the vectors in the Vector-Matrix-Vector product
in the dynamic dory scheme.
* Added `row_and_column_from_index` and `row_start_index` methods. These
methods define the structure of the new dynamic dory commitmemt scheme.

# Are these changes tested?

Yes
  • Loading branch information
JayWhite2357 authored Jul 31, 2024
2 parents 7128683 + cc57de7 commit 301674c
Show file tree
Hide file tree
Showing 3 changed files with 413 additions and 0 deletions.
Original file line number Diff line number Diff line change
@@ -0,0 +1,281 @@
//! This module provides the `build_standard_basis_vecs` method, which is used in converting a point to a
//! vector used in a Vector-Matrix-Vector product in the dynamic dory scheme.
use super::F;
use ark_ff::Field;

#[allow(dead_code)]
/// This method produces evaluation vectors from a point. This is a helper method for generating a Vector-Matrix-Vector product in the dynamic dory scheme.
///
/// The ith element of the lo_vec is essentially the ith monomial basis element (lexicographically).
/// The ith element of the hi_vec is essentially the jth monomial basis element where j = row_start_index(i).
///
/// NOTE: the lo_vec and hi_vec are scaled by lo_vec[0] and hi_vec[0] respectively.
/// NOTE: lo_vec and hi_vec should otherwise consist entirely of zeros in order to ensure correct output.
pub(super) fn compute_dynamic_standard_basis_vecs(point: &[F], lo_vec: &mut [F], hi_vec: &mut [F]) {
let nu = point.len() / 2 + 1;
debug_assert_eq!(lo_vec.len(), 1 << nu);
debug_assert_eq!(hi_vec.len(), 1 << nu);
for i in 1..nu {
build_partial_second_half_standard_basis_vecs(
&point[..2 * i - 1],
&mut lo_vec[..1 << i],
&mut hi_vec[..1 << i],
true,
);
}
// Note: if we don't have the "full" point, we shouldn't fill up the last quarter because it should be all zeros.
build_partial_second_half_standard_basis_vecs(point, lo_vec, hi_vec, point.len() % 2 == 1);
// Add the most significant variable, which was not included before in order to allow simple copying to work.
point.iter().skip(1).enumerate().for_each(|(i, v)| {
let p = i / 2;
let o = 2 + i % 2;
(o << p..(o + 1) << p).for_each(|k| hi_vec[k] *= v)
});
}

fn build_partial_second_half_standard_basis_vecs(
point: &[F],
lo_vec: &mut [F],
hi_vec: &mut [F],
add_last_quarter: bool,
) {
let nu = point.len() / 2 + 1;
debug_assert_eq!(lo_vec.len(), 1 << nu);
debug_assert_eq!(hi_vec.len(), 1 << nu);
if nu == 1 {
lo_vec[1] = if point.is_empty() {
F::ZERO
} else {
lo_vec[0] * point[0]
};
hi_vec[1] = hi_vec[0];
} else {
let (lo_half0, lo_half1) = lo_vec.split_at_mut(1 << (nu - 1));
lo_half0
.iter()
.zip(lo_half1)
.for_each(|(l, h)| *h = *l * point[nu - 1]);
if nu == 2 {
hi_vec[2] = hi_vec[0];
if add_last_quarter {
hi_vec[3] = hi_vec[1];
}
} else {
let (hi_half0, hi_half1) = hi_vec.split_at_mut(1 << (nu - 1));
let (_, hi_quarter1) = hi_half0.split_at(1 << (nu - 2));
let (hi_quarter2, hi_quarter3) = hi_half1.split_at_mut(1 << (nu - 2));
let (_, hi_eighth3) = hi_quarter1.split_at(1 << (nu - 3));
let (hi_eighth4, hi_eighth5) = hi_quarter2.split_at_mut(1 << (nu - 3));
let (hi_eighth6, hi_eighth7) = hi_quarter3.split_at_mut(1 << (nu - 3));
// Fill up quarter #2 (from 2/4..3/4).
hi_eighth3
.iter()
.zip(hi_eighth4.iter_mut().zip(hi_eighth5))
.for_each(|(&source, (target_lo, target_hi))| {
// Copy eighth #3 (from 3/8..4/8) to eighth #4 (4/8..5/8).
*target_lo = source;
// Copy eighth #3 (from 3/8..4/8) to eighth #5 (5/8..6/8)
// and multiply by the third from the last element in point.
*target_hi = source * point[2 * nu - 4];
});
if add_last_quarter {
// Fill up quarter #4 (from 3/4..4/4).
hi_quarter2
.iter()
.step_by(2)
.zip(hi_eighth6.iter_mut().zip(hi_eighth7))
.for_each(|(&source, (target_lo, target_hi))| {
// Copy every other in quarter #2 (from 2/4..3/4) to eighth #6 (6/8..7/8).
*target_lo = source;
// Copy every other in quarter #2 (from 2/4..3/4) to eighth #6 (7/8..8/8).
// and multiply by the second from the last element in point.
*target_hi = source * point[2 * nu - 3];
});
}
}
}
}

#[cfg(test)]
mod tests {
use super::{super::dynamic_dory_structure::row_start_index, *};

#[test]
fn we_can_compute_dynamic_standard_basis_vecs_from_length_0_point() {
let mut lo_vec = vec![F::ZERO; 2];
let mut hi_vec = vec![F::ZERO; 2];
lo_vec[0] = F::from(2);
hi_vec[0] = F::from(3);
let point = vec![];
let lo_vec_expected = vec![F::from(2), F::ZERO];
let hi_vec_expected = vec![F::from(3), F::from(3)];
compute_dynamic_standard_basis_vecs(&point, &mut lo_vec, &mut hi_vec);
assert_eq!(lo_vec, lo_vec_expected);
assert_eq!(hi_vec, hi_vec_expected);
}
#[test]
fn we_can_compute_dynamic_standard_basis_vecs_from_length_1_point() {
let mut lo_vec = vec![F::ZERO; 2];
let mut hi_vec = vec![F::ZERO; 2];
lo_vec[0] = F::from(2);
hi_vec[0] = F::from(3);
let point = vec![F::from(5)];
let lo_vec_expected = vec![F::from(2), F::from(2 * 5)];
let hi_vec_expected = vec![F::from(3), F::from(3)];
compute_dynamic_standard_basis_vecs(&point, &mut lo_vec, &mut hi_vec);
assert_eq!(lo_vec, lo_vec_expected);
assert_eq!(hi_vec, hi_vec_expected);
}
#[test]
fn we_can_compute_dynamic_standard_basis_vecs_from_length_2_point() {
let mut lo_vec = vec![F::ZERO; 4];
let mut hi_vec = vec![F::ZERO; 4];
lo_vec[0] = F::from(2);
hi_vec[0] = F::from(3);
let point = vec![F::from(5), F::from(7)];
let lo_vec_expected = vec![
F::from(2),
F::from(2 * 5),
F::from(2 * 7),
F::from(2 * 5 * 7),
];
let hi_vec_expected = vec![F::from(3), F::from(3), F::from(3 * 7), F::ZERO];
compute_dynamic_standard_basis_vecs(&point, &mut lo_vec, &mut hi_vec);
assert_eq!(lo_vec, lo_vec_expected);
assert_eq!(hi_vec, hi_vec_expected);
}
#[test]
fn we_can_compute_dynamic_standard_basis_vecs_from_length_3_point() {
let mut lo_vec = vec![F::ZERO; 4];
let mut hi_vec = vec![F::ZERO; 4];
lo_vec[0] = F::from(2);
hi_vec[0] = F::from(3);
let point = vec![F::from(5), F::from(7), F::from(11)];
let lo_vec_expected = vec![
F::from(2),
F::from(2 * 5),
F::from(2 * 7),
F::from(2 * 5 * 7),
];
let hi_vec_expected = vec![F::from(3), F::from(3), F::from(3 * 7), F::from(3 * 11)];
compute_dynamic_standard_basis_vecs(&point, &mut lo_vec, &mut hi_vec);
assert_eq!(lo_vec, lo_vec_expected);
assert_eq!(hi_vec, hi_vec_expected);
}
#[test]
fn we_can_compute_dynamic_standard_basis_vecs_from_length_4_point() {
let mut lo_vec = vec![F::ZERO; 8];
let mut hi_vec = vec![F::ZERO; 8];
lo_vec[0] = F::from(2);
hi_vec[0] = F::from(3);
let point = vec![F::from(5), F::from(7), F::from(11), F::from(13)];
let lo_vec_expected = vec![
F::from(2),
F::from(2 * 5),
F::from(2 * 7),
F::from(2 * 5 * 7),
F::from(2 * 11),
F::from(2 * 5 * 11),
F::from(2 * 7 * 11),
F::from(2 * 5 * 7 * 11),
];
let hi_vec_expected = vec![
F::from(3),
F::from(3),
F::from(3 * 7),
F::from(3 * 11),
F::from(3 * 13),
F::from(3 * 11 * 13),
F::ZERO,
F::ZERO,
];
compute_dynamic_standard_basis_vecs(&point, &mut lo_vec, &mut hi_vec);
assert_eq!(lo_vec, lo_vec_expected);
assert_eq!(hi_vec, hi_vec_expected);
}
#[test]
fn we_can_compute_dynamic_standard_basis_vecs_from_length_5_point() {
let mut lo_vec = vec![F::ZERO; 8];
let mut hi_vec = vec![F::ZERO; 8];
lo_vec[0] = F::from(2);
hi_vec[0] = F::from(3);
let point = vec![
F::from(5),
F::from(7),
F::from(11),
F::from(13),
F::from(17),
];
let lo_vec_expected = vec![
F::from(2),
F::from(2 * 5),
F::from(2 * 7),
F::from(2 * 5 * 7),
F::from(2 * 11),
F::from(2 * 5 * 11),
F::from(2 * 7 * 11),
F::from(2 * 5 * 7 * 11),
];
let hi_vec_expected = vec![
F::from(3),
F::from(3),
F::from(3 * 7),
F::from(3 * 11),
F::from(3 * 13),
F::from(3 * 11 * 13),
F::from(3 * 17),
F::from(3 * 13 * 17),
];
compute_dynamic_standard_basis_vecs(&point, &mut lo_vec, &mut hi_vec);
assert_eq!(lo_vec, lo_vec_expected);
assert_eq!(hi_vec, hi_vec_expected);
}

/// Computes the evaluation of a basis monomial at the given point.
///
/// In other words, the result is `prod point[i]^(b[i])` where
/// `index = sum 2^i*b[i]` and `b[i]` is `0` or `1`. (i.e. `b` is the binary representation of `index`.)
/// Note: `point` is padded with zeros as needed.
///
/// This method is primarily to test the `build_standard_basis_vecs` method.
fn get_binary_eval(index: usize, point: &[F]) -> F {
core::iter::successors(Some(index), |&k| match k >> 1 {
0 => None,
k => Some(k),
})
.enumerate()
.filter_map(|(i, b)| match b % 2 == 0 {
true => None,
false => Some(point.get(i).copied().unwrap_or(F::ZERO)),
})
.product()
}

#[test]
fn we_can_compute_dynamic_random_standard_basis_vecs() {
use ark_std::{test_rng, UniformRand};
use itertools::Itertools;
let mut rng = test_rng();
for num_vars in 0..10 {
let point = core::iter::repeat_with(|| F::rand(&mut rng))
.take(num_vars)
.collect_vec();
let alpha = F::rand(&mut rng);
let beta = F::rand(&mut rng);
let nu = point.len() / 2 + 1;
let mut lo_vec = vec![F::ZERO; 1 << nu];
let mut hi_vec = vec![F::ZERO; 1 << nu];
lo_vec[0] = alpha;
hi_vec[0] = beta;
compute_dynamic_standard_basis_vecs(&point, &mut lo_vec, &mut hi_vec);
for i in 0..1 << nu {
assert_eq!(lo_vec[i], alpha * get_binary_eval(i, &point));
assert_eq!(
hi_vec[i],
beta * get_binary_eval(row_start_index(i), &point)
);
}
}
}
}
Loading

0 comments on commit 301674c

Please sign in to comment.