primeorder: faster lincomb_vartime when alloc enabled

Cargo.toml

@@ -26,3 +26,5 @@ ed448-goldilocks = { path = "ed448-goldilocks" }
 hash2curve = { path = "hash2curve" }
 primefield = { path = "primefield" }
 primeorder = { path = "primeorder" }
+
+rustcrypto-group = { git = "https://github.com/RustCrypto/group" }

p256/benches/point.rs

@@ -1,18 +1,73 @@
 //! p256 `ProjectivePoint` benchmarks
 
-use criterion::{Criterion, criterion_group, criterion_main};
-use p256::ProjectivePoint;
+use criterion::{
+    BenchmarkGroup, Criterion, criterion_group, criterion_main, measurement::Measurement,
+};
+use hex_literal::hex;
+use p256::{
+    ProjectivePoint, Scalar,
+    elliptic_curve::{
+        Group, PrimeField,
+        ops::{LinearCombination, MulByGeneratorVartime, MulVartime},
+    },
+};
 use primefield::subtle::ConstantTimeEq;
+use std::hint::black_box;
 
-fn bench_projective(c: &mut Criterion) {
-    let mut group = c.benchmark_group("ProjectivePoint constant time operations");
+fn test_scalar() -> Scalar {
+    Scalar::from_repr(
+        hex!("519b423d715f8b581f4fa8ee59f4771a5b44c8130b4e3eacca54a56dda72b464").into(),
+    )
+    .unwrap()
+}
+
+fn bench_point_lincomb<M: Measurement>(group: &mut BenchmarkGroup<'_, M>) {
+    let p = ProjectivePoint::GENERATOR;
+    let s = test_scalar();
+
+    group.bench_function("point-scalar lincomb", |b| {
+        b.iter(|| ProjectivePoint::lincomb(&[(p, s), (p, s), (p, s)]))
+    });
+    group.bench_function("point-scalar lincomb (variable-time)", |b| {
+        b.iter(|| ProjectivePoint::lincomb_vartime(&[(p, s), (p, s), (p, s)]))
+    });
+}
+
+fn bench_point_mul<M: Measurement>(group: &mut BenchmarkGroup<'_, M>) {
+    let p = ProjectivePoint::GENERATOR;
+    let s = test_scalar();
+    group.bench_function("point-scalar mul", |b| {
+        b.iter(|| black_box(p) * black_box(s))
+    });
+    group.bench_function("point-scalar mul (variable-time)", |b| {
+        b.iter(|| black_box(p).mul_vartime(&black_box(s)))
+    });
+}
+
+fn bench_point_mul_by_generator<M: Measurement>(group: &mut BenchmarkGroup<'_, M>) {
+    let m = test_scalar();
+    let s = Scalar::from_repr(m.into()).unwrap();
+    group.bench_function("generator-scalar mul", |b| {
+        b.iter(|| ProjectivePoint::mul_by_generator(&black_box(s)))
+    });
+    group.bench_function("generator-scalar mul (variable-time)", |b| {
+        b.iter(|| ProjectivePoint::mul_by_generator_vartime(&black_box(s)))
+    });
+}
+
+fn bench_point(c: &mut Criterion) {
+    let mut group = c.benchmark_group("ProjectivePoint operations");
+
+    bench_point_lincomb(&mut group);
+    bench_point_mul(&mut group);
+    bench_point_mul_by_generator(&mut group);
 
-    group.bench_function("point_generator_ct_eq", |b| {
+    group.bench_function("ct_eq", |b| {
         b.iter(|| ProjectivePoint::GENERATOR.ct_eq(&ProjectivePoint::GENERATOR))
     });
 
     group.finish();
 }
 
-criterion_group!(benches, bench_projective);
+criterion_group!(benches, bench_point);
 criterion_main!(benches);

p256/benches/scalar.rs

@@ -4,14 +4,7 @@ use criterion::{
     BenchmarkGroup, Criterion, criterion_group, criterion_main, measurement::Measurement,
 };
 use hex_literal::hex;
-use p256::{
-    ProjectivePoint, Scalar,
-    elliptic_curve::{
-        Group,
-        group::ff::PrimeField,
-        ops::{MulByGeneratorVartime, MulVartime},
-    },
-};
+use p256::{Scalar, elliptic_curve::PrimeField};
 use std::hint::black_box;
 
 fn test_scalar_x() -> Scalar {
@@ -28,29 +21,6 @@ fn test_scalar_y() -> Scalar {
     .unwrap()
 }
 
-fn bench_point_mul<M: Measurement>(group: &mut BenchmarkGroup<'_, M>) {
-    let p = ProjectivePoint::GENERATOR;
-    let m = test_scalar_x();
-    let s = Scalar::from_repr(m.into()).unwrap();
-    group.bench_function("point-scalar mul", |b| {
-        b.iter(|| black_box(p) * black_box(s))
-    });
-    group.bench_function("point-scalar mul (variable-time)", |b| {
-        b.iter(|| black_box(p).mul_vartime(&black_box(s)))
-    });
-}
-
-fn bench_point_mul_by_generator<M: Measurement>(group: &mut BenchmarkGroup<'_, M>) {
-    let m = test_scalar_x();
-    let s = Scalar::from_repr(m.into()).unwrap();
-    group.bench_function("generator-scalar mul", |b| {
-        b.iter(|| ProjectivePoint::mul_by_generator(&black_box(s)))
-    });
-    group.bench_function("generator-scalar mul (variable-time)", |b| {
-        b.iter(|| ProjectivePoint::mul_by_generator_vartime(&black_box(s)))
-    });
-}
-
 fn bench_scalar_sub<M: Measurement>(group: &mut BenchmarkGroup<'_, M>) {
     let x = test_scalar_x();
     let y = test_scalar_y();
@@ -79,22 +49,15 @@ fn bench_scalar_invert<M: Measurement>(group: &mut BenchmarkGroup<'_, M>) {
     group.bench_function("invert", |b| b.iter(|| x.invert()));
 }
 
-fn bench_point(c: &mut Criterion) {
-    let mut group = c.benchmark_group("point operations");
-    bench_point_mul(&mut group);
-    group.finish();
-}
-
 fn bench_scalar(c: &mut Criterion) {
     let mut group = c.benchmark_group("scalar operations");
     bench_scalar_sub(&mut group);
     bench_scalar_add(&mut group);
     bench_scalar_mul(&mut group);
-    bench_point_mul_by_generator(&mut group);
     bench_scalar_negate(&mut group);
     bench_scalar_invert(&mut group);
     group.finish();
 }
 
-criterion_group!(benches, bench_point, bench_scalar);
+criterion_group!(benches, bench_scalar);
 criterion_main!(benches);

p256/tests/projective.rs

@@ -105,9 +105,11 @@ proptest! {
         s1 in scalar(),
         p2 in projective(),
         s2 in scalar(),
+        p3 in projective(),
+        s3 in scalar(),
     ) {
-        let reference = p1 * s1 + p2 * s2;
-        let test = ProjectivePoint::lincomb(&[(p1, s1), (p2, s2)]);
+        let reference = p1 * s1 + p2 * s2 + p3 * s3;
+        let test = ProjectivePoint::lincomb(&[(p1, s1), (p2, s2), (p3, s3)]);
         assert_eq!(reference, test);
     }
 
@@ -118,9 +120,26 @@ proptest! {
         s1 in scalar(),
         p2 in projective(),
         s2 in scalar(),
+                p3 in projective(),
+        s3 in scalar(),
     ) {
-        let reference = p1 * s1 + p2 * s2;
-        let test = ProjectivePoint::lincomb(vec![(p1, s1), (p2, s2)].as_slice());
+        let reference = p1 * s1 + p2 * s2 + p3 * s3;
+        let test = ProjectivePoint::lincomb(vec![(p1, s1), (p2, s2), (p3, s3)].as_slice());
+        assert_eq!(reference, test);
+    }
+
+    #[test]
+    #[cfg(feature = "alloc")]
+    fn lincomb_vartime(
+        p1 in projective(),
+        s1 in scalar(),
+        p2 in projective(),
+        s2 in scalar(),
+        p3 in projective(),
+        s3 in scalar(),
+    ) {
+        let reference = p1 * s1 + p2 * s2 + p3 * s3;
+        let test = ProjectivePoint::lincomb_vartime(vec![(p1, s1), (p2, s2), (p3, s3)].as_slice());
         assert_eq!(reference, test);
     }
 }

primeorder/src/projective.rs

@@ -31,7 +31,7 @@ use elliptic_curve::{
 #[cfg(feature = "alloc")]
 use {
     alloc::vec::Vec,
-    elliptic_curve::group::{Wnaf, WnafGroup},
+    elliptic_curve::group::{Wnaf, WnafBase, WnafGroup, WnafScalar},
 };
 
 #[cfg(all(feature = "alloc", feature = "basepoint-table"))]
@@ -494,6 +494,23 @@ where
 
         lincomb::<C>(&mut ks, &mut pcs)
     }
+
+    #[cfg(feature = "alloc")]
+    fn lincomb_vartime(points_and_scalars: &[(Self, Scalar<C>)]) -> Self {
+        // TODO(tarcieri): make this customizable?
+        const WINDOW_SIZE: usize = 4;
+        let points = points_and_scalars
+            .iter()
+            .map(|(point, _)| WnafBase::<Self, WINDOW_SIZE>::new(*point))
+            .collect::<Vec<_>>();
+
+        let scalars = points_and_scalars
+            .iter()
+            .map(|(_, scalar)| WnafScalar::<Scalar<C>, WINDOW_SIZE>::new(scalar))
+            .collect::<Vec<_>>();
+
+        WnafBase::multiscalar_mul(scalars.iter(), points.iter())
+    }
 }
 
 impl<C, const N: usize> LinearCombination<[(Self, Scalar<C>); N]> for ProjectivePoint<C>
@@ -508,6 +525,11 @@ where
 
         lincomb::<C>(&mut ks, &mut pcs)
     }
+
+    #[cfg(feature = "alloc")]
+    fn lincomb_vartime(points_and_scalars: &[(Self, Scalar<C>); N]) -> Self {
+        Self::lincomb_vartime(points_and_scalars.as_slice())
+    }
 }
 
 fn lincomb<C: PrimeCurveParams>(
@@ -950,6 +972,7 @@ where
 
     // When we're guaranteed *not* to have basepoint tables available (because they need `alloc`)
     // use linear combinations for this computation, but they're slower when they are available
+    // TODO(tarcieri): `WnafBase::multiscalar_mul` w\ basepoint table when `alloc` *is* available
     #[cfg(not(feature = "alloc"))]
     fn mul_by_generator_and_mul_add_vartime(
         a: &Self::Scalar,