Skip to content

Use x86 assembly for Pasta field addition and subtraction #435

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.

Sign up for GitHub

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

Merged
ValarDragon merged 3 commits into from
Sep 13, 2026
Merged

Use x86 assembly for Pasta field addition and subtraction #435

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
645ba98
Add x86 assembly field addition and subtraction
ValarDragon Sep 12, 2026
e15ce38
Benchmark selected field doubling backend
ValarDragon Sep 12, 2026
e2340ce
Add changelog for x86 field arithmetic
ValarDragon Sep 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion crates/pasta_curves/benches/fp.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,7 +78,7 @@ fn bench_fp_double(b: &mut Bencher) {
let mut count = 0;
b.iter(|| {
let mut tmp = v[count];
tmp = tmp.double();
tmp = <Fp as ff::Field>::double(&tmp);
count = (count + 1) % SAMPLES;
tmp
});
Expand Down
2 changes: 1 addition & 1 deletion crates/pasta_curves/benches/fq.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,7 +78,7 @@ fn bench_fq_double(b: &mut Bencher) {
let mut count = 0;
b.iter(|| {
let mut tmp = v[count];
tmp = tmp.double();
tmp = <Fq as ff::Field>::double(&tmp);
count = (count + 1) % SAMPLES;
tmp
});
Expand Down
99 changes: 75 additions & 24 deletions crates/pasta_curves/src/fields/fp.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,12 +159,27 @@ impl<'a, 'b> Sub<&'b Fp> for &'a Fp {
{
Fp(super::aarch64_asm::sub(&self.0, &rhs.0, &MODULUS.0))
}
#[cfg(not(all(
feature = "aarch64-asm",
target_arch = "aarch64",
any(target_family = "unix", target_os = "none"),
target_pointer_width = "64",
target_endian = "little",
#[cfg(all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
))]
{
Fp(super::x86_64_asm::sub(&self.0, &rhs.0, &X86_64_ASM_PARAMS))
}
#[cfg(not(any(
all(
feature = "aarch64-asm",
target_arch = "aarch64",
any(target_family = "unix", target_os = "none"),
target_pointer_width = "64",
target_endian = "little",
),
all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
)
)))]
{
self.sub(rhs)
Expand All @@ -187,12 +202,27 @@ impl<'a, 'b> Add<&'b Fp> for &'a Fp {
{
Fp(super::aarch64_asm::add(&self.0, &rhs.0, &MODULUS.0))
}
#[cfg(not(all(
feature = "aarch64-asm",
target_arch = "aarch64",
any(target_family = "unix", target_os = "none"),
target_pointer_width = "64",
target_endian = "little",
#[cfg(all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
))]
{
Fp(super::x86_64_asm::add(&self.0, &rhs.0, &X86_64_ASM_PARAMS))
}
#[cfg(not(any(
all(
feature = "aarch64-asm",
target_arch = "aarch64",
any(target_family = "unix", target_os = "none"),
target_pointer_width = "64",
target_endian = "little",
),
all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
)
)))]
{
self.add(rhs)
Expand Down Expand Up @@ -779,12 +809,27 @@ impl ff::Field for Fp {
{
Self(super::aarch64_asm::add(&self.0, &self.0, &MODULUS.0))
}
#[cfg(not(all(
feature = "aarch64-asm",
target_arch = "aarch64",
target_vendor = "apple",
target_pointer_width = "64",
target_endian = "little",
#[cfg(all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
))]
{
Self(super::x86_64_asm::add(&self.0, &self.0, &X86_64_ASM_PARAMS))
}
#[cfg(not(any(
all(
feature = "aarch64-asm",
target_arch = "aarch64",
target_vendor = "apple",
target_pointer_width = "64",
target_endian = "little",
),
all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
)
)))]
{
self.double()
Expand Down Expand Up @@ -1837,11 +1882,11 @@ fn constants_are_canonical() {
)
))]
#[test]
fn asm_mul_and_square_canonical_sweep_match_portable() {
fn asm_arithmetic_canonical_sweep_matches_portable() {
use rand::{Rng, SeedableRng};

// Random canonical operands: the selected assembly `mul` must agree
// with the portable implementation and return a canonical residue.
// Random canonical operands: the selected assembly operations must agree
// with the portable implementations and return canonical residues.
let mut rng = rand_xorshift::XorShiftRng::from_seed([0x42; 16]);
let mut random = || {
let mut l = [0u64; 4];
Expand All @@ -1853,9 +1898,15 @@ fn asm_mul_and_square_canonical_sweep_match_portable() {
for _ in 0..200_000u32 {
let a = random();
let b = random();
let asm = a.mul_runtime(&b);
assert_eq!(asm, Fp::mul(&a, &b), "lhs {:x?} rhs {:x?}", a.0, b.0);
assert!(is_canonical(&asm));
let add = &a + &b;
let sub = &a - &b;
let mul = a.mul_runtime(&b);
assert_eq!(add, Fp::add(&a, &b), "lhs {:x?} rhs {:x?}", a.0, b.0);
assert_eq!(sub, Fp::sub(&a, &b), "lhs {:x?} rhs {:x?}", a.0, b.0);
assert_eq!(mul, Fp::mul(&a, &b), "lhs {:x?} rhs {:x?}", a.0, b.0);
assert!(is_canonical(&add));
assert!(is_canonical(&sub));
assert!(is_canonical(&mul));
assert_eq!(a.square_runtime(), a.square(), "value {:x?}", a.0);
}
}
Expand Down
99 changes: 75 additions & 24 deletions crates/pasta_curves/src/fields/fq.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,12 +159,27 @@ impl<'a, 'b> Sub<&'b Fq> for &'a Fq {
{
Fq(super::aarch64_asm::sub(&self.0, &rhs.0, &MODULUS.0))
}
#[cfg(not(all(
feature = "aarch64-asm",
target_arch = "aarch64",
any(target_family = "unix", target_os = "none"),
target_pointer_width = "64",
target_endian = "little",
#[cfg(all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
))]
{
Fq(super::x86_64_asm::sub(&self.0, &rhs.0, &X86_64_ASM_PARAMS))
}
#[cfg(not(any(
all(
feature = "aarch64-asm",
target_arch = "aarch64",
any(target_family = "unix", target_os = "none"),
target_pointer_width = "64",
target_endian = "little",
),
all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
)
)))]
{
self.sub(rhs)
Expand All @@ -187,12 +202,27 @@ impl<'a, 'b> Add<&'b Fq> for &'a Fq {
{
Fq(super::aarch64_asm::add(&self.0, &rhs.0, &MODULUS.0))
}
#[cfg(not(all(
feature = "aarch64-asm",
target_arch = "aarch64",
any(target_family = "unix", target_os = "none"),
target_pointer_width = "64",
target_endian = "little",
#[cfg(all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
))]
{
Fq(super::x86_64_asm::add(&self.0, &rhs.0, &X86_64_ASM_PARAMS))
}
#[cfg(not(any(
all(
feature = "aarch64-asm",
target_arch = "aarch64",
any(target_family = "unix", target_os = "none"),
target_pointer_width = "64",
target_endian = "little",
),
all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
)
)))]
{
self.add(rhs)
Expand Down Expand Up @@ -779,12 +809,27 @@ impl ff::Field for Fq {
{
Self(super::aarch64_asm::add(&self.0, &self.0, &MODULUS.0))
}
#[cfg(not(all(
feature = "aarch64-asm",
target_arch = "aarch64",
target_vendor = "apple",
target_pointer_width = "64",
target_endian = "little",
#[cfg(all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
))]
{
Self(super::x86_64_asm::add(&self.0, &self.0, &X86_64_ASM_PARAMS))
}
#[cfg(not(any(
all(
feature = "aarch64-asm",
target_arch = "aarch64",
target_vendor = "apple",
target_pointer_width = "64",
target_endian = "little",
),
all(
feature = "x86_64-asm",
target_arch = "x86_64",
target_pointer_width = "64"
)
)))]
{
self.double()
Expand Down Expand Up @@ -1836,11 +1881,11 @@ fn constants_are_canonical() {
)
))]
#[test]
fn asm_mul_and_square_canonical_sweep_match_portable() {
fn asm_arithmetic_canonical_sweep_matches_portable() {
use rand::{Rng, SeedableRng};

// Random canonical operands: the selected assembly `mul` must agree
// with the portable implementation and return a canonical residue.
// Random canonical operands: the selected assembly operations must agree
// with the portable implementations and return canonical residues.
let mut rng = rand_xorshift::XorShiftRng::from_seed([0x42; 16]);
let mut random = || {
let mut l = [0u64; 4];
Expand All @@ -1852,9 +1897,15 @@ fn asm_mul_and_square_canonical_sweep_match_portable() {
for _ in 0..200_000u32 {
let a = random();
let b = random();
let asm = a.mul_runtime(&b);
assert_eq!(asm, Fq::mul(&a, &b), "lhs {:x?} rhs {:x?}", a.0, b.0);
assert!(is_canonical(&asm));
let add = &a + &b;
let sub = &a - &b;
let mul = a.mul_runtime(&b);
assert_eq!(add, Fq::add(&a, &b), "lhs {:x?} rhs {:x?}", a.0, b.0);
assert_eq!(sub, Fq::sub(&a, &b), "lhs {:x?} rhs {:x?}", a.0, b.0);
assert_eq!(mul, Fq::mul(&a, &b), "lhs {:x?} rhs {:x?}", a.0, b.0);
assert!(is_canonical(&add));
assert!(is_canonical(&sub));
assert!(is_canonical(&mul));
assert_eq!(a.square_runtime(), a.square(), "value {:x?}", a.0);
}
}
Expand Down
Loading
Loading