bechmarks

benches/grumpkin_msm.rs

@@ -65,23 +65,32 @@ fn criterion_benchmark(c: &mut Criterion) {
         let mut group = c.benchmark_group("GPU");
         group.sample_size(20);
 
+        let indices = (0..(npoints as u32)).rev().collect::<Vec<_>>();
+
         group.bench_function(format!("2**{} points", bench_npow), |b| {
             b.iter(|| {
                 let _ = grumpkin_msm::bn256::msm_aux(&points, &scalars, None);
             })
         });
 
+        scalars.reverse();
+        group.bench_function(format!("2**{} points rev", bench_npow), |b| {
+            b.iter(|| {
+                let _ = grumpkin_msm::bn256::msm_aux(&points, &scalars, Some(indices.as_slice()));
+            })
+        });
+
         let context = grumpkin_msm::bn256::init(&points);
 
-        let indices = (0..(npoints as u32)).rev().collect::<Vec<_>>();
+        scalars.reverse();
         group.bench_function(
             format!("preallocate 2**{} points", bench_npow),
             |b| {
                 b.iter(|| {
                     let _ = grumpkin_msm::bn256::with_context_aux(
                         &context,
                         &scalars,
-                        Some(indices.as_slice()),
+                        None,
                     );
                 })
             },
@@ -93,7 +102,7 @@ fn criterion_benchmark(c: &mut Criterion) {
             |b| {
                 b.iter(|| {
                     let _ = grumpkin_msm::bn256::with_context_aux(
-                        &context, &scalars, None,
+                        &context, &scalars, Some(indices.as_slice()),
                     );
                 })
             },

examples/grumpkin_msm.rs

@@ -6,7 +6,7 @@ use grumpkin_msm::cuda_available;
 
 fn main() {
     let bench_npow: usize = std::env::var("BENCH_NPOW")
-        .unwrap_or("22".to_string())
+        .unwrap_or("23".to_string())
         .parse()
         .unwrap();
     let npoints: usize = 1 << bench_npow;

src/lib.rs

@@ -214,6 +214,10 @@ macro_rules! impl_msm {
 
         /// An indexed MSM. We do not check if the indices are valid, i.e.
         /// for i in 0..nscalars, 0 <= indices[i] < npoints
+        /// 
+        /// Also, this version carries a performance penalty that all the
+        /// points must be moved onto the GPU once instead of in batches.
+        /// If the points are to be reused, please use the [`MSMContext`] API
         pub fn indexed_msm(
             points: &[$affine],
             scalars: &[$scalar],

src/pasta.rs

@@ -232,6 +232,10 @@ macro_rules! impl_pasta {
 
         /// An indexed MSM. We do not check if the indices are valid, i.e.
         /// for i in 0..nscalars, 0 <= indices[i] < npoints
+        /// 
+        /// Also, this version carries a performance penalty that all the
+        /// points must be moved onto the GPU once instead of in batches.
+        /// If the points are to be reused, please use the [`MSMContext`] API
         pub fn indexed_msm(
             points: &[$affine],
             scalars: &[$scalar],