ig we push these

lurk-lab · Jan 12, 2024 · 3f183e6 · 3f183e6
1 parent ef54fc2
commit 3f183e6
Show file tree

Hide file tree

Showing 28 changed files with 450 additions and 394 deletions.
diff --git a/.gitignore b/.gitignore
@@ -1,3 +1,2 @@
 /target
-Cargo.lock
-/plots
+Cargo.lock
diff --git a/benches/data.md b/benches/data.md
@@ -0,0 +1,51 @@
+# Benchmarks
+
+## Table of Contents
+
+- [Benchmark Results](#benchmark-results)
+    - [GPU](#GPU)
+    - [lurkrs](#lurkrs)
+
+## Benchmark Results
+
+### GPU
+
+|                                      |  `random`                  | `lurkrs`                          |
+|:-------------------------------------|:---------------------------|:----------------------------------|
+| **`witness 1, 7941351 scalars`**     | `119.79 ms` (✅ **1.00x**) | `552.09 ms` (❌ *4.60x slower*)   |
+| **`T 2, 9699051 scalars`**           | `144.74 ms` (✅ **1.00x**) | `143.87 ms` (✅ **1.00x faster**) |
+| **`witness 5, 7941351 scalars`**     | `118.40 ms` (✅ **1.00x**) | `560.24 ms` (❌ *4.74x slower*)   |
+| **`T 6, 9699051 scalars`**           | `143.99 ms` (✅ **1.00x**) | `323.89 ms` (❌ *2.25x slower*)   |
+| **`witness 9, 7941351 scalars`**     | `120.42 ms` (✅ **1.00x**) | `561.95 ms` (❌ *4.67x slower*)   |
+| **`T 10, 9699051 scalars`**          | `143.02 ms` (✅ **1.00x**) | `350.17 ms` (❌ *2.44x slower*)   |
+| **`witness 13, 7941351 scalars`**    | `111.75 ms` (✅ **1.00x**) | `560.65 ms` (❌ *5.04x slower*)   |
+| **`T 14, 9699051 scalars`**          | `119.85 ms` (✅ **1.00x**) | `468.23 ms` (❌ *3.93x slower*)   |
+| **`witness 17, 7941351 scalars`**    | `119.13 ms` (✅ **1.00x**) | `560.10 ms` (❌ *4.66x slower*)   |
+| **`T 18, 9699051 scalars`**          | `143.52 ms` (✅ **1.00x**) | `564.23 ms` (❌ *3.94x slower*)   |
+| **`witness 21, 7941351 scalars`**    | `121.13 ms` (✅ **1.00x**) | `558.80 ms` (❌ *4.65x slower*)   |
+| **`T 22, 9699051 scalars`**          | `145.63 ms` (✅ **1.00x**) | `614.23 ms` (❌ *4.23x slower*)   |
+| **`witness 25, 7941351 scalars`**    | `118.84 ms` (✅ **1.00x**) | `557.14 ms` (❌ *4.72x slower*)   |
+| **`T 26, 9699051 scalars`**          | `141.11 ms` (✅ **1.00x**) | `679.76 ms` (❌ *4.81x slower*)   |
+| **`witness 29, 7941351 scalars`**    | `119.40 ms` (✅ **1.00x**) | `557.92 ms` (❌ *4.68x slower*)   |
+| **`T 30, 9699051 scalar`**           | `142.80 ms` (✅ **1.00x**) | `702.32 ms` (❌ *4.94x slower*)   |
+
+### lurkrs
+
+|                                      |  `lurkrs/fibonacci`        | `pasta-msm/benches/lurkrs`        |
+|:-------------------------------------|:---------------------------|:----------------------------------|
+| **`witness 1, 7941351 scalars`**     | `559.07 ms` (✅ **1.00x**) | `552.09 ms` (✅ **0.98x faster**) |
+| **`T 2, 9699051 scalars`**           | `135.27 ms` (✅ **1.00x**) | `143.87 ms` (✅ **1.05x slower**) |
+| **`witness 5, 7941351 scalars`**     | `547.40 ms` (✅ **1.00x**) | `560.24 ms` (✅ **1.02x slower**) |
+| **`T 6, 9699051 scalars`**           | `316.52 ms` (✅ **1.00x**) | `323.89 ms` (✅ **1.02x slower**) |
+| **`witness 9, 7941351 scalars`**     | `544.42 ms` (✅ **1.00x**) | `561.95 ms` (✅ **1.03x slower**) |
+| **`T 10, 9699051 scalars`**          | `344.99 ms` (✅ **1.00x**) | `350.17 ms` (✅ **1.01x slower**) |
+| **`witness 13, 7941351 scalars`**    | `552.01 ms` (✅ **1.00x**) | `560.65 ms` (✅ **1.01x slower**) |
+| **`T 14, 9699051 scalars`**          | `457.50 ms` (✅ **1.00x**) | `468.23 ms` (✅ **1.02x slower**) |
+| **`witness 17, 7941351 scalars`**    | `545.15 ms` (✅ **1.00x**) | `560.10 ms` (✅ **1.02x slower**) |
+| **`T 18, 9699051 scalars`**          | `556.29 ms` (✅ **1.00x**) | `564.23 ms` (✅ **1.01x slower**) |
+| **`witness 21, 7941351 scalars`**    | `548.88 ms` (✅ **1.00x**) | `558.80 ms` (✅ **1.02x slower**) |
+| **`T 22, 9699051 scalars`**          | `607.14 ms` (✅ **1.00x**) | `614.23 ms` (✅ **1.01x slower**) |
+| **`witness 25, 7941351 scalars`**    | `550.40 ms` (✅ **1.00x**) | `557.14 ms` (✅ **1.01x slower**) |
+| **`T 26, 9699051 scalars`**          | `680.36 ms` (✅ **1.00x**) | `679.76 ms` (✅ **1.00x slower**) |
+| **`witness 29, 7941351 scalars`**    | `549.54 ms` (✅ **1.00x**) | `557.92 ms` (✅ **1.01x slower**) |
+| **`T 30, 9699051 scalar`**           | `692.39 ms` (✅ **1.00x**) | `702.32 ms` (✅ **1.02x slower**) |
diff --git a/benches/msm.rs b/benches/msm.rs
@@ -2,160 +2,169 @@
 // Licensed under the Apache License, Version 2.0, see LICENSE for details.
 // SPDX-License-Identifier: Apache-2.0
 
-use criterion::{criterion_group, criterion_main, Criterion};
-use pasta_msm::utils::{gen_points, gen_scalars};
+#![allow(dead_code)]
+#![allow(unused_imports)]
+#![allow(unused_mut)]
+
+use std::io::Read;
+
+use abomonation::Abomonation;
+use criterion::{criterion_group, criterion_main, Criterion, BenchmarkId};
+
+use pasta_curves::{group::ff::{PrimeField, Field}, pallas};
+use pasta_msm::{self, utils::CommitmentKey};
+use rand::thread_rng;
 
 #[cfg(feature = "cuda")]
-use pasta_msm::cuda_available;
-use pasta_curves::{pallas, group::ff::Field};
+extern "C" {
+    fn cuda_available() -> bool;
+}
 
-fn criterion_benchmark(c: &mut Criterion) {
-    let bench_npow: usize = std::env::var("BENCH_NPOW")
-        .unwrap_or("18".to_string())
-        .parse()
-        .unwrap();
-    let npoints: usize = 1 << bench_npow;
+fn read_abomonated<T: Abomonation + Clone>(name: String) -> std::io::Result<T> {
+    use std::fs::OpenOptions;
+    use std::io::BufReader;
 
-    // println!("generating {} random points, just hang on...", npoints);
-    let mut points = gen_points(npoints);
-    let mut scalars = gen_scalars(npoints);
+    let arecibo = home::home_dir().unwrap().join(".arecibo_witness");
 
-    #[cfg(feature = "cuda")]
-    {
-        unsafe { pasta_msm::CUDA_OFF = true };
-    }
+    let data = OpenOptions::new()
+        .read(true)
+        .write(true)
+        .create(true)
+        .open(arecibo.join(name))?;
+    let mut reader = BufReader::new(data);
+    let mut bytes = vec![];
+    reader.read_to_end(&mut bytes)?;
 
-    let mut group = c.benchmark_group("CPU");
-    group.sample_size(10);
+    let (data, _) = unsafe { abomonation::decode::<T>(&mut bytes).unwrap() };
 
-    group.bench_function(format!("2**{} points", bench_npow), |b| {
-        b.iter(|| {
-            let _ = pasta_msm::pallas(&points, &scalars, npoints);
-        })
-    });
+    Ok(data.clone())
+}
+
+fn criterion_benchmark(c: &mut Criterion) {
+    let npoints: usize = 10_000_000;
+    let scalars = pasta_msm::utils::gen_scalars(npoints);
+    // let nonuniform_scalars = pasta_msm::utils::generate_nonuniform_scalars(npoints);
+    let points = pasta_msm::utils::gen_points(npoints);
 
-    group.finish();
+    let mut rng = thread_rng();
 
     #[cfg(feature = "cuda")]
     if unsafe { cuda_available() } {
         unsafe { pasta_msm::CUDA_OFF = false };
 
-        const EXTRA: usize = 5;
-        let bench_npow = bench_npow + EXTRA;
-        let npoints: usize = 1 << bench_npow;
-
-        while points.len() < npoints {
-            points.append(&mut points.clone());
-        }
-        scalars.append(&mut gen_scalars(npoints - scalars.len()));
-
         let mut group = c.benchmark_group("GPU");
         group.sample_size(20);
 
-        let context = pasta_msm::pallas_init(&points, npoints);
+        for i in 0..32 {
+            let witness_i = read_abomonated::<
+                Vec<<pallas::Scalar as PrimeField>::Repr>,
+            >(i.to_string())
+            .unwrap();
+            let mut witness_i = unsafe {
+                std::mem::transmute::<Vec<_>, Vec<pallas::Scalar>>(witness_i)
+            };
 
-        for nnz in [1.0, 0.75, 0.5, 0.25, 0.10, 0.01] {
-            let zeros = ((1.0 - nnz) * npoints as f64) as usize;
+            let witness_n = witness_i.len();
 
-            for i in 0..zeros {
-                scalars[i] = pallas::Scalar::ZERO;
+            if witness_n < 1_000_000 {
+                continue;
             }
 
-            group.bench_function(format!("2**{} points {}", bench_npow, nnz), |b| {
-                b.iter(|| {
-                    let _ = pasta_msm::pallas_with(&context, npoints, npoints - zeros, &scalars);
-                })
-            });
-
+            let name = format!("#{i}, {} scalars", witness_n);
+
+            // group.bench_function(BenchmarkId::new("random", &name), |b| {
+            //     b.iter(|| {
+            //         let _ = pasta_msm::pallas(
+            //             &points[..witness_n],
+            //             &scalars[..witness_n],
+            //             witness_n,
+            //         );
+            //     })
+            // });
+
+            group.bench_function(
+                BenchmarkId::new("lurkrs", &name),
+                |b| {
+                    b.iter(|| {
+                        let _ = pasta_msm::pallas(
+                            &points[..witness_n],
+                            &witness_i,
+                            witness_n,
+                        );
+                    })
+                },
+            );
+
+            for w_i in witness_i.iter_mut() {
+                if w_i.is_zero_vartime() {
+                    *w_i = pallas::Scalar::random(&mut rng);
+                }
+            }
+            group.bench_function(
+                BenchmarkId::new("lurkrs no zeros", &name),
+                |b| {
+                    b.iter(|| {
+                        let _ = pasta_msm::pallas(
+                            &points[..witness_n],
+                            &witness_i,
+                            witness_n,
+                        );
+                    })
+                },
+            );
         }
 
-        group.finish();
+        // group.bench_function(format!("biased {} scalars", npoints), |b| {
+        //     b.iter(|| {
+        //         let _ =
+        //             pasta_msm::pallas(&points, &nonuniform_scalars, npoints);
+        //     })
+        // });
+
+        // for npoints in [7941351, 9699051] {
+        //     group.bench_function(format!("random {} scalars", npoints), |b| {
+        //         b.iter(|| {
+        //             let _ = pasta_msm::pallas(&points, &scalars, npoints);
+        //         })
+        //     });
+        // }
+
+        // let context = pasta_msm::pallas_init(&points, npoints);
+
+        // group.bench_function(format!("preallocated lurkrs {} scalars", npoints), |b| {
+        //     b.iter(|| {
+        //         let _ = pasta_msm::pallas_with(&context, npoints, npoints, &witness_primary);
+        //     })
+        // });
+
+        // group.bench_function(
+        //     format!("preallocated biased {} scalars", npoints),
+        //     |b| {
+        //         b.iter(|| {
+        //             let _ = pasta_msm::pallas_with(
+        //                 &context,
+        //                 npoints,
+        //                 npoints,
+        //                 &nonuniform_scalars,
+        //             );
+        //         })
+        //     },
+        // );
+
+        // group.bench_function(
+        //     format!("preallocated random {} scalars", npoints),
+        //     |b| {
+        //         b.iter(|| {
+        //             let _ = pasta_msm::pallas_with(
+        //                 &context, npoints, npoints, &scalars,
+        //             );
+        //         })
+        //     },
+        // );
+
+        // group.finish();
     }
 }
 
 criterion_group!(benches, criterion_benchmark);
 criterion_main!(benches);
-
-
-
-// // Copyright Supranational LLC
-// // Licensed under the Apache License, Version 2.0, see LICENSE for details.
-// // SPDX-License-Identifier: Apache-2.0
-
-// #![allow(dead_code)]
-// #![allow(unused_imports)]
-// #![allow(unused_mut)]
-
-// use std::io::Read;
-
-// use abomonation::Abomonation;
-// use criterion::{criterion_group, criterion_main, Criterion};
-
-// use pasta_curves::{pallas, group::ff::PrimeField};
-// use pasta_msm::{self, utils::CommitmentKey};
-
-// #[cfg(feature = "cuda")]
-// extern "C" {
-//     fn cuda_available() -> bool;
-// }
-
-// include!("../src/tests.rs");
-
-// fn read_abomonated<T: Abomonation + Clone>(name: String) -> std::io::Result<T> {
-//     use std::fs::OpenOptions;
-//     use std::io::BufReader;
-
-//     let arecibo = home::home_dir().unwrap().join(".arecibo");
-
-//     let data = OpenOptions::new()
-//         .read(true)
-//         .write(true)
-//         .create(true)
-//         .open(arecibo.join(name))?;
-//     let mut reader = BufReader::new(data);
-//     let mut bytes = vec![];
-//     reader.read_to_end(&mut bytes)?;
-
-//     let (data, _) = unsafe { abomonation::decode::<T>(&mut bytes).unwrap() };
-
-//     Ok(data.clone())
-// }
-
-// fn criterion_benchmark(c: &mut Criterion) {
-//     let witness_primary = read_abomonated::<
-//         Vec<<pallas::Scalar as PrimeField>::Repr>,
-//     >("witness_primary".into())
-//     .unwrap();
-//     let witness_primary = unsafe {
-//         std::mem::transmute::<Vec<_>, Vec<pallas::Scalar>>(witness_primary)
-//     };
-//     let npoints: usize = witness_primary.len();
-
-//     let scalars = crate::tests::gen_scalars(npoints);
-//     let points = crate::tests::gen_points(npoints);
-
-//     #[cfg(feature = "cuda")]
-//     if unsafe { cuda_available() } {
-//         unsafe { pasta_msm::CUDA_OFF = false };
-
-//         let mut group = c.benchmark_group("GPU");
-//         group.sample_size(20);
-
-//         group.bench_function(format!("lurkrs {} scalars", npoints), |b| {
-//             b.iter(|| {
-//                 let _ = pasta_msm::pallas(&points, &witness_primary);
-//             })
-//         });
-
-//         group.bench_function(format!("random {} scalars", npoints), |b| {
-//             b.iter(|| {
-//                 let _ = pasta_msm::pallas(&points, &scalars);
-//             })
-//         });
-
-//         group.finish();
-//     }
-// }
-
-// criterion_group!(benches, criterion_benchmark);
-// criterion_main!(benches);
diff --git a/benchmark_cmp_zeros.md b/benchmark_cmp_zeros.md
@@ -0,0 +1,33 @@
+# Benchmarks
+
+## Table of Contents
+
+- [Benchmark Results](#benchmark-results)
+    - [GPU](#gpu)
+
+## Benchmark Results
+
+### GPU
+
+|                            | `lurkrs`                  | `lurkrs no zeros`                 |
+|:---------------------------|:--------------------------|:--------------------------------- |
+| **`#1, 7941351 scalars`**  | `550.19 ms` (✅ **1.00x**) | `566.97 ms` (✅ **1.03x slower**)  |
+| **`#2, 9699051 scalars`**  | `148.77 ms` (✅ **1.00x**) | `176.85 ms` (❌ *1.19x slower*)    |
+| **`#5, 7941351 scalars`**  | `558.14 ms` (✅ **1.00x**) | `572.18 ms` (✅ **1.03x slower**)  |
+| **`#6, 9699051 scalars`**  | `330.08 ms` (✅ **1.00x**) | `363.44 ms` (✅ **1.10x slower**)  |
+| **`#9, 7941351 scalars`**  | `559.80 ms` (✅ **1.00x**) | `571.99 ms` (✅ **1.02x slower**)  |
+| **`#10, 9699051 scalars`** | `350.76 ms` (✅ **1.00x**) | `394.22 ms` (❌ *1.12x slower*)    |
+| **`#13, 7941351 scalars`** | `557.43 ms` (✅ **1.00x**) | `569.30 ms` (✅ **1.02x slower**)  |
+| **`#14, 9699051 scalars`** | `465.93 ms` (✅ **1.00x**) | `501.17 ms` (✅ **1.08x slower**)  |
+| **`#17, 7941351 scalars`** | `556.75 ms` (✅ **1.00x**) | `569.25 ms` (✅ **1.02x slower**)  |
+| **`#18, 9699051 scalars`** | `562.11 ms` (✅ **1.00x**) | `596.44 ms` (✅ **1.06x slower**)  |
+| **`#21, 7941351 scalars`** | `557.79 ms` (✅ **1.00x**) | `572.85 ms` (✅ **1.03x slower**)  |
+| **`#22, 9699051 scalars`** | `615.48 ms` (✅ **1.00x**) | `645.91 ms` (✅ **1.05x slower**)  |
+| **`#25, 7941351 scalars`** | `558.32 ms` (✅ **1.00x**) | `569.35 ms` (✅ **1.02x slower**)  |
+| **`#26, 9699051 scalars`** | `681.24 ms` (✅ **1.00x**) | `708.16 ms` (✅ **1.04x slower**)  |
+| **`#29, 7941351 scalars`** | `556.30 ms` (✅ **1.00x**) | `570.89 ms` (✅ **1.03x slower**)  |
+| **`#30, 9699051 scalars`** | `704.70 ms` (✅ **1.00x**) | `731.01 ms` (✅ **1.04x slower**)  |
+
+---
+Made with [criterion-table](https://github.com/nu11ptr/criterion-table)
+