Merge pull request #85 from mir-protocol/efficient_insert

Efficient insert
This commit is contained in:
Nicholas Ward 2021-07-01 17:43:44 -07:00 committed by GitHub
commit dcde599377
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
8 changed files with 140 additions and 42 deletions

View File

@ -310,16 +310,16 @@ impl<F: Extendable<D>, const D: usize> CircuitBuilder<F, D> {
input: w,
}),
});
self.add_generator(CopyGenerator {
src: Target::Wire(Wire {
self.generate_copy(
Target::Wire(Wire {
gate: gate_1,
input: w,
}),
dst: Target::Wire(Wire {
Target::Wire(Wire {
gate: gate_2,
input: w,
}),
});
);
}
}

View File

@ -6,7 +6,10 @@ use crate::util::{log2_strict, reverse_index_bits};
// TODO: Should really do some "dynamic" dispatch to handle the
// different FFT algos rather than C-style enum dispatch.
enum FftStrategy { Classic, Unrolled }
enum FftStrategy {
Classic,
Unrolled,
}
const FFT_STRATEGY: FftStrategy = FftStrategy::Classic;
@ -33,7 +36,6 @@ fn fft_classic_root_table<F: Field>(n: usize) -> FftRootTable<F> {
root_table
}
fn fft_unrolled_root_table<F: Field>(n: usize) -> FftRootTable<F> {
// Precompute a table of the roots of unity used in the main
// loops.
@ -67,18 +69,20 @@ fn fft_unrolled_root_table<F: Field>(n: usize) -> FftRootTable<F> {
fn fft_dispatch<F: Field>(
input: Vec<F>,
zero_factor: Option<usize>,
root_table: Option<FftRootTable<F>>
root_table: Option<FftRootTable<F>>,
) -> Vec<F> {
let n = input.len();
match FFT_STRATEGY {
FftStrategy::Classic
=> fft_classic(input,
zero_factor.unwrap_or(0),
root_table.unwrap_or_else(|| fft_classic_root_table(n))),
FftStrategy::Unrolled
=> fft_unrolled(input,
zero_factor.unwrap_or(0),
root_table.unwrap_or_else(|| fft_unrolled_root_table(n)))
FftStrategy::Classic => fft_classic(
input,
zero_factor.unwrap_or(0),
root_table.unwrap_or_else(|| fft_classic_root_table(n)),
),
FftStrategy::Unrolled => fft_unrolled(
input,
zero_factor.unwrap_or(0),
root_table.unwrap_or_else(|| fft_unrolled_root_table(n)),
),
}
}
@ -91,10 +95,12 @@ pub fn fft<F: Field>(poly: PolynomialCoeffs<F>) -> PolynomialValues<F> {
pub fn fft_with_options<F: Field>(
poly: PolynomialCoeffs<F>,
zero_factor: Option<usize>,
root_table: Option<FftRootTable<F>>
root_table: Option<FftRootTable<F>>,
) -> PolynomialValues<F> {
let PolynomialCoeffs { coeffs } = poly;
PolynomialValues { values: fft_dispatch(coeffs, zero_factor, root_table) }
PolynomialValues {
values: fft_dispatch(coeffs, zero_factor, root_table),
}
}
#[inline]
@ -105,7 +111,7 @@ pub fn ifft<F: Field>(poly: PolynomialValues<F>) -> PolynomialCoeffs<F> {
pub fn ifft_with_options<F: Field>(
poly: PolynomialValues<F>,
zero_factor: Option<usize>,
root_table: Option<FftRootTable<F>>
root_table: Option<FftRootTable<F>>,
) -> PolynomialCoeffs<F> {
let n = poly.len();
let lg_n = log2_strict(n);
@ -136,7 +142,7 @@ pub fn ifft_with_options<F: Field>(
pub(crate) fn fft_classic<F: Field>(
input: Vec<F>,
r: usize,
root_table: FftRootTable<F>
root_table: FftRootTable<F>,
) -> Vec<F> {
let mut values = reverse_index_bits(input);
@ -144,7 +150,11 @@ pub(crate) fn fft_classic<F: Field>(
let lg_n = log2_strict(n);
if root_table.len() != lg_n {
panic!("Expected root table of length {}, but it was {}.", lg_n, root_table.len());
panic!(
"Expected root table of length {}, but it was {}.",
lg_n,
root_table.len()
);
}
// After reverse_index_bits, the only non-zero elements of values
@ -154,7 +164,8 @@ pub(crate) fn fft_classic<F: Field>(
// element i*2^r with the value at i*2^r. This corresponds to the
// first r rounds of the FFT when there are 2^r zeros at the end
// of the original input.
if r > 0 { // if r == 0 then this loop is a noop.
if r > 0 {
// if r == 0 then this loop is a noop.
let mask = !((1 << r) - 1);
for i in 0..n {
values[i] = values[i & mask];
@ -162,7 +173,7 @@ pub(crate) fn fft_classic<F: Field>(
}
let mut m = 1 << (r + 1);
for lg_m in (r+1)..=lg_n {
for lg_m in (r + 1)..=lg_n {
let half_m = m / 2;
for k in (0..n).step_by(m) {
for j in 0..half_m {
@ -185,11 +196,7 @@ pub(crate) fn fft_classic<F: Field>(
/// The parameter r signifies that the first 1/2^r of the entries of
/// input may be non-zero, but the last 1 - 1/2^r entries are
/// definitely zero.
fn fft_unrolled<F: Field>(
input: Vec<F>,
r_orig: usize,
root_table: FftRootTable<F>
) -> Vec<F> {
fn fft_unrolled<F: Field>(input: Vec<F>, r_orig: usize, root_table: FftRootTable<F>) -> Vec<F> {
let n = input.len();
let lg_n = log2_strict(input.len());
@ -197,7 +204,7 @@ fn fft_unrolled<F: Field>(
// FFT of a constant polynomial (including zero) is itself.
if n < 2 {
return values
return values;
}
// The 'm' corresponds to the specialisation from the 'm' in the
@ -206,7 +213,8 @@ fn fft_unrolled<F: Field>(
// (See comment in fft_classic near same code.)
let mut r = r_orig;
let mut m = 1 << r;
if r > 0 { // if r == 0 then this loop is a noop.
if r > 0 {
// if r == 0 then this loop is a noop.
let mask = !((1 << r) - 1);
for i in 0..n {
values[i] = values[i & mask];
@ -225,11 +233,15 @@ fn fft_unrolled<F: Field>(
}
if n == 2 {
return values
return values;
}
if root_table.len() != (lg_n - 1) {
panic!("Expected root table of length {}, but it was {}.", lg_n, root_table.len());
panic!(
"Expected root table of length {}, but it was {}.",
lg_n,
root_table.len()
);
}
// m = 2
@ -253,7 +265,7 @@ fn fft_unrolled<F: Field>(
// m >= 4
for lg_m in r..lg_n {
for k in (0..n).step_by(2*m) {
for k in (0..n).step_by(2 * m) {
// Unrolled the commented loop by groups of 4 and
// rearranged the lines. Improves runtime by about
// 10%.
@ -294,11 +306,10 @@ fn fft_unrolled<F: Field>(
values
}
#[cfg(test)]
mod tests {
use crate::field::crandall_field::CrandallField;
use crate::field::fft::{fft, ifft, fft_with_options};
use crate::field::fft::{fft, fft_with_options, ifft};
use crate::field::field::Field;
use crate::polynomial::polynomial::{PolynomialCoeffs, PolynomialValues};
use crate::util::{log2_ceil, log2_strict};
@ -328,7 +339,10 @@ mod tests {
for r in 0..4 {
// expand ceofficients by factor 2^r by filling with zeros
let zero_tail = coefficients.clone().lde(r);
assert_eq!(fft(zero_tail.clone()), fft_with_options(zero_tail, Some(r), None));
assert_eq!(
fft(zero_tail.clone()),
fft_with_options(zero_tail, Some(r), None)
);
}
}

View File

@ -323,7 +323,7 @@ macro_rules! test_arithmetic {
let v = <F as Field>::PrimeField::TWO_ADICITY;
for e in [0, 1, 2, 3, 4, v - 2, v - 1, v, v + 1, v + 2, 123*v] {
for e in [0, 1, 2, 3, 4, v - 2, v - 1, v, v + 1, v + 2, 123 * v] {
let x = F::TWO.exp(e as u64).inverse();
let y = F::inverse_2exp(e);
assert_eq!(x, y);

View File

@ -1,9 +1,42 @@
use crate::circuit_builder::CircuitBuilder;
use crate::field::extension_field::target::ExtensionTarget;
use crate::field::extension_field::Extendable;
use crate::generator::NonzeroTestGenerator;
use crate::target::Target;
impl<F: Extendable<D>, const D: usize> CircuitBuilder<F, D> {
/// Evaluates to 0 if `x` equals zero, 1 otherwise.
/// From section 2 of https://github.com/mir-protocol/r1cs-workshop/blob/master/workshop.pdf,
/// based on an idea from https://eprint.iacr.org/2012/598.pdf.
pub fn is_nonzero(&mut self, x: Target) -> Target {
// Dummy variable.
let m = self.add_virtual_target();
// The prover sets this the dummy variable to 1/x if x != 0, or to an arbitrary value if
// x == 0.
self.add_generator(NonzeroTestGenerator {
to_test: x,
dummy: m,
});
// Evaluates to (0) * (0) = 0 if x == 0 and (x) * (1/x) = 1 otherwise.
let y = self.mul(x, m);
// Enforce that (1 - y) * x == 0.
let prod = self.arithmetic(F::NEG_ONE, x, y, F::ONE, x);
self.assert_zero(prod);
y
}
/// Evaluates to 1 if `x` and `y` are equal, 0 otherwise.
pub fn is_equal(&mut self, x: Target, y: Target) -> Target {
let difference = self.sub(x, y);
let not_equal = self.is_nonzero(difference);
let one = self.one();
self.sub(one, not_equal)
}
/// Inserts a `Target` in a vector at a non-deterministic index. This is done by rotating to the
/// left, inserting at 0 and then rotating to the right.
/// Note: `index` is not range-checked.
@ -13,9 +46,29 @@ impl<F: Extendable<D>, const D: usize> CircuitBuilder<F, D> {
element: ExtensionTarget<D>,
v: Vec<ExtensionTarget<D>>,
) -> Vec<ExtensionTarget<D>> {
let mut v = self.rotate_left(index, &v);
v.insert(0, element);
self.rotate_right(index, &v)
let mut already_inserted = self.zero();
let mut new_list = Vec::new();
for i in 0..v.len() {
let one = self.one();
let cur_index = self.constant(F::from_canonical_usize(i));
let insert_here = self.is_equal(cur_index, index);
let mut new_item = self.zero_extension();
new_item = self.scalar_mul_add_extension(insert_here, element, new_item);
if i > 0 {
new_item = self.scalar_mul_add_extension(already_inserted, v[i - 1], new_item);
}
already_inserted = self.add(already_inserted, insert_here);
let not_already_inserted = self.sub(one, already_inserted);
new_item = self.scalar_mul_add_extension(not_already_inserted, v[i], new_item);
new_list.push(new_item);
}
new_list
}
}
#[cfg(test)]

View File

@ -130,3 +130,27 @@ impl<F: Field> SimpleGenerator<F> for RandomValueGenerator {
PartialWitness::singleton_target(self.target, random_value)
}
}
/// A generator for testing if a value equals zero
pub(crate) struct NonzeroTestGenerator {
pub(crate) to_test: Target,
pub(crate) dummy: Target,
}
impl<F: Field> SimpleGenerator<F> for NonzeroTestGenerator {
fn dependencies(&self) -> Vec<Target> {
vec![self.to_test]
}
fn run_once(&self, witness: &PartialWitness<F>) -> PartialWitness<F> {
let to_test_value = witness.get_target(self.to_test);
let dummy_value = if to_test_value == F::ZERO {
F::ONE
} else {
to_test_value.inverse()
};
PartialWitness::singleton_target(self.dummy, dummy_value)
}
}

View File

@ -1,6 +1,7 @@
use std::cmp::max;
use std::iter::Sum;
use std::ops::{Add, AddAssign, Mul, MulAssign, Sub, SubAssign};
use std::time::Instant;
use anyhow::{ensure, Result};

View File

@ -7,11 +7,15 @@ use crate::wire::Wire;
#[derive(Copy, Clone, Eq, PartialEq, Hash, Debug)]
pub enum Target {
Wire(Wire),
PublicInput { index: usize },
PublicInput {
index: usize,
},
/// A target that doesn't have any inherent location in the witness (but it can be copied to
/// another target that does). This is useful for representing intermediate values in witness
/// generation.
VirtualTarget { index: usize },
VirtualTarget {
index: usize,
},
}
impl Target {

View File

@ -77,7 +77,9 @@ pub(crate) fn reverse_bits(n: usize, num_bits: usize) -> usize {
// to plain '>>' is to accommodate the case n == num_bits == 0,
// which would become `0 >> 64`. Rust thinks that any shift of 64
// bits causes overflow, even when the argument is zero.
n.reverse_bits().overflowing_shr(usize::BITS - num_bits as u32).0
n.reverse_bits()
.overflowing_shr(usize::BITS - num_bits as u32)
.0
}
#[cfg(test)]