From 325c29dc7204f95242c598feec9c382ddc15ebe7 Mon Sep 17 00:00:00 2001
From: Hanting Zhang <winston@lurk-lab.com>
Date: Fri, 9 Feb 2024 02:00:56 +0000
Subject: [PATCH] bechmarks

---
 benches/grumpkin_msm.rs  | 15 ++++++++++++---
 examples/grumpkin_msm.rs |  2 +-
 src/lib.rs               |  4 ++++
 src/pasta.rs             |  4 ++++
 4 files changed, 21 insertions(+), 4 deletions(-)

diff --git a/benches/grumpkin_msm.rs b/benches/grumpkin_msm.rs
index 993d5e4..72c6d12 100644
--- a/benches/grumpkin_msm.rs
+++ b/benches/grumpkin_msm.rs
@@ -65,15 +65,24 @@ 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| {
@@ -81,7 +90,7 @@ fn criterion_benchmark(c: &mut Criterion) {
                     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()),
                     );
                 })
             },
diff --git a/examples/grumpkin_msm.rs b/examples/grumpkin_msm.rs
index 208183f..1ab89c5 100644
--- a/examples/grumpkin_msm.rs
+++ b/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;
diff --git a/src/lib.rs b/src/lib.rs
index 2582d55..e7b7a24 100644
--- a/src/lib.rs
+++ b/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],
diff --git a/src/pasta.rs b/src/pasta.rs
index 007b9de..be8682d 100644
--- a/src/pasta.rs
+++ b/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],