From fb5fde28cf99d56da7edcbba5c4b4b6108d962bd Mon Sep 17 00:00:00 2001
From: Paul Miller <paul@paulmillr.com>
Date: Wed, 18 Dec 2024 21:09:11 +0000
Subject: [PATCH] Add deps-less bench script

---
 README.md              | 99 ++++++++++++++----------------------------
 benchmark/README.md    |  2 +-
 benchmark/_utils.js    |  9 ++--
 benchmark/noble.js     | 47 ++++++++++++++++++++
 test/crosstest.test.js |  2 +-
 5 files changed, 87 insertions(+), 72 deletions(-)
 create mode 100644 benchmark/noble.js

diff --git a/README.md b/README.md
index 8679fde..5a02c7d 100644
--- a/README.md
+++ b/README.md
@@ -470,73 +470,40 @@ uint8array into encrypt / decrypt methods.
 Benchmark results on Apple M2 with node v22:
 
 ```
-encrypt (64B)
-├─xsalsa20poly1305 x 485,908 ops/sec @ 2μs/op
-├─chacha20poly1305 x 414,250 ops/sec @ 2μs/op
-├─xchacha20poly1305 x 331,674 ops/sec @ 3μs/op
-├─aes-256-gcm x 144,237 ops/sec @ 6μs/op
-└─aes-256-gcm-siv x 121,373 ops/sec @ 8μs/op
-encrypt (1KB)
-├─xsalsa20poly1305 x 136,574 ops/sec @ 7μs/op
-├─chacha20poly1305 x 136,017 ops/sec @ 7μs/op
-├─xchacha20poly1305 x 126,008 ops/sec @ 7μs/op
-├─aes-256-gcm x 40,149 ops/sec @ 24μs/op
-└─aes-256-gcm-siv x 37,420 ops/sec @ 26μs/op
-encrypt (8KB)
-├─xsalsa20poly1305 x 22,517 ops/sec @ 44μs/op
-├─chacha20poly1305 x 23,187 ops/sec @ 43μs/op
-├─xchacha20poly1305 x 22,837 ops/sec @ 43μs/op
-├─aes-256-gcm x 7,993 ops/sec @ 125μs/op
-└─aes-256-gcm-siv x 7,836 ops/sec @ 127μs/op
-encrypt (1MB)
-├─xsalsa20poly1305 x 186 ops/sec @ 5ms/op
-├─chacha20poly1305 x 191 ops/sec @ 5ms/op
-├─xchacha20poly1305 x 191 ops/sec @ 5ms/op
-├─aes-256-gcm x 71 ops/sec @ 14ms/op
-└─aes-256-gcm-siv x 75 ops/sec @ 13ms/op
-```
-
-Unauthenticated encryption:
+64B
+xsalsa20poly1305 x 501,756 ops/sec @ 1μs/op
+chacha20poly1305 x 428,082 ops/sec @ 2μs/op
+xchacha20poly1305 x 343,170 ops/sec @ 2μs/op
+aes-256-gcm x 147,492 ops/sec @ 6μs/op
+aes-256-gcm-siv x 122,085 ops/sec @ 8μs/op
+# Unauthenticated encryption
+salsa20 x 1,288,659 ops/sec @ 776ns/op
+xsalsa20 x 1,055,966 ops/sec @ 947ns/op
+chacha20 x 1,506,024 ops/sec @ 664ns/op
+xchacha20 x 1,064,962 ops/sec @ 939ns/op
+chacha8 x 1,683,501 ops/sec @ 594ns/op
+chacha12 x 1,628,664 ops/sec @ 614ns/op
+aes-256-ecb x 775,193 ops/sec @ 1μs/op
+aes-256-cbc x 738,552 ops/sec @ 1μs/op
+aes-256-ctr x 737,463 ops/sec @ 1μs/op
+
+1MB
+xsalsa20poly1305 x 205 ops/sec @ 4ms/op
+chacha20poly1305 x 213 ops/sec @ 4ms/op
+xchacha20poly1305 x 213 ops/sec @ 4ms/op
+aes-256-gcm x 77 ops/sec @ 12ms/op
+aes-256-gcm-siv x 81 ops/sec @ 12ms/op
+# Unauthenticated encryption
+salsa20 x 498 ops/sec @ 2ms/op
+xsalsa20 x 493 ops/sec @ 2ms/op
+chacha20 x 506 ops/sec @ 1ms/op
+xchacha20 x 506 ops/sec @ 1ms/op
+chacha8 x 956 ops/sec @ 1ms/op
+chacha12 x 735 ops/sec @ 1ms/op
+aes-256-ecb x 229 ops/sec @ 4ms/op
+aes-256-cbc x 110 ops/sec @ 9ms/op
+aes-256-ctr x 115 ops/sec @ 8ms/op
 
-```
-encrypt (64B)
-├─salsa x 1,221,001 ops/sec @ 819ns/op
-├─chacha x 1,373,626 ops/sec @ 728ns/op
-├─xsalsa x 1,019,367 ops/sec @ 981ns/op
-└─xchacha x 1,019,367 ops/sec @ 981ns/op
-encrypt (1KB)
-├─salsa x 349,162 ops/sec @ 2μs/op
-├─chacha x 372,717 ops/sec @ 2μs/op
-├─xsalsa x 327,868 ops/sec @ 3μs/op
-└─xchacha x 332,446 ops/sec @ 3μs/op
-encrypt (8KB)
-├─salsa x 55,178 ops/sec @ 18μs/op
-├─chacha x 51,535 ops/sec @ 19μs/op
-├─xsalsa x 54,274 ops/sec @ 18μs/op
-└─xchacha x 55,645 ops/sec @ 17μs/op
-encrypt (1MB)
-├─salsa x 451 ops/sec @ 2ms/op
-├─chacha x 464 ops/sec @ 2ms/op
-├─xsalsa x 455 ops/sec @ 2ms/op
-└─xchacha x 462 ops/sec @ 2ms/op
-
-AES
-encrypt (64B)
-├─ctr-256 x 679,347 ops/sec @ 1μs/op
-├─cbc-256 x 699,300 ops/sec @ 1μs/op
-└─ecb-256 x 717,875 ops/sec @ 1μs/op
-encrypt (1KB)
-├─ctr-256 x 93,423 ops/sec @ 10μs/op
-├─cbc-256 x 95,721 ops/sec @ 10μs/op
-└─ecb-256 x 154,726 ops/sec @ 6μs/op
-encrypt (8KB)
-├─ctr-256 x 12,908 ops/sec @ 77μs/op
-├─cbc-256 x 13,411 ops/sec @ 74μs/op
-└─ecb-256 x 22,681 ops/sec @ 44μs/op
-encrypt (1MB)
-├─ctr-256 x 105 ops/sec @ 9ms/op
-├─cbc-256 x 108 ops/sec @ 9ms/op
-└─ecb-256 x 181 ops/sec @ 5ms/op
 ```
 
 Compare to other implementations:
diff --git a/benchmark/README.md b/benchmark/README.md
index be63408..0c74289 100644
--- a/benchmark/README.md
+++ b/benchmark/README.md
@@ -7,4 +7,4 @@ node ciphers.js           # ChaCha, etc - all libs
 node ciphers.js noble     # ChaCha, etc - only noble
 node aes.js               # AES, SIV
 node poly.js              # Poly1305
-```
\ No newline at end of file
+```
diff --git a/benchmark/_utils.js b/benchmark/_utils.js
index d3d836b..28d2fff 100644
--- a/benchmark/_utils.js
+++ b/benchmark/_utils.js
@@ -1,6 +1,11 @@
 import { deepStrictEqual } from 'assert';
 import { compare } from 'micro-bmark';
 
+export const onlyNoble = process.argv[2] === 'noble';
+export function buf(n) {
+  return new Uint8Array(n).fill(n % 251);
+}
+
 // type Buffers = ({size: string, samples: number, data: Uint8Array})[];
 export async function crossValidate(buffers, ciphers) {
   // Verify that things we bench actually work
@@ -64,10 +69,6 @@ export async function validateHashes(buffers, HASHES) {
   console.log('Libraries cross-validated against each other correctly');
 }
 
-export const onlyNoble = process.argv[2] === 'noble';
-export function buf(n) {
-  return new Uint8Array(n).fill(n);
-}
 export async function benchmarkOnlyNoble(buffers, ciphers) {
   const nobleImpls = [];
   // chacha20_poly1305: {
diff --git a/benchmark/noble.js b/benchmark/noble.js
new file mode 100644
index 0000000..927b74b
--- /dev/null
+++ b/benchmark/noble.js
@@ -0,0 +1,47 @@
+import { mark } from 'micro-bmark';
+import { buf } from './_utils.js';
+import { concatBytes } from '@noble/ciphers/utils';
+import { xsalsa20poly1305 } from '@noble/ciphers/salsa';
+import { xchacha20poly1305, chacha20poly1305 } from '@noble/ciphers/chacha';
+import { salsa20, xsalsa20 } from '@noble/ciphers/salsa';
+import { chacha20, xchacha20, chacha8, chacha12 } from '@noble/ciphers/chacha';
+import { ecb, ctr, cbc, gcm, siv } from '@noble/ciphers/aes';
+
+const buffers = [
+  // { size: '16B', samples: 1_500_000, data: buf(16) }, // common block size
+  { size: '32B', samples: 1_500_000, data: buf(32) },
+  { size: '64B', samples: 1_000_000, data: buf(64) },
+  // { size: '1KB', samples: 50_000, data: buf(1024) },
+  // { size: '8KB', samples: 10_000, data: buf(1024 * 8) },
+  { size: '1MB', samples: 100, data: buf(1024 * 1024) },
+];
+
+async function main() {
+  const key = buf(32);
+  const nonce = buf(12);
+  const nonce8 = buf(8);
+  const nonce16 = buf(16);
+  const nonce24 = buf(24);
+  for (let i = 0; i < 100000; i++) xsalsa20poly1305(key, nonce24).encrypt(buf(64)); // warm-up
+  for (const { size, samples: i, data: buf } of buffers) {
+    console.log(size);
+    await mark('xsalsa20poly1305', i, () => xsalsa20poly1305(key, nonce24).encrypt(buf));
+    await mark('chacha20poly1305', i, () => chacha20poly1305(key, nonce).encrypt(buf));
+    await mark('xchacha20poly1305', i, () => xchacha20poly1305(key, nonce24).encrypt(buf));
+    await mark('aes-256-gcm', i, () => gcm(key, nonce).encrypt(buf));
+    await mark('aes-256-gcm-siv', i, () => siv(key, nonce).encrypt(buf));
+
+    console.log('# Unauthenticated encryption');
+    await mark('salsa20', i, () => salsa20(key, nonce8, buf));
+    await mark('xsalsa20', i, () => xsalsa20(key, nonce24, buf));
+    await mark('chacha20', i, () => chacha20(key, nonce, buf));
+    await mark('xchacha20', i, () => xchacha20(key, nonce24, buf));
+    await mark('chacha8', i, () => chacha8(key, nonce, buf));
+    await mark('chacha12', i, () => chacha12(key, nonce, buf));
+    await mark('aes-256-ecb', i, () => ecb(key).encrypt(buf));
+    await mark('aes-256-cbc', i, () => cbc(key, nonce16).encrypt(buf));
+    await mark('aes-256-ctr', i, () => ctr(key, nonce16).encrypt(buf));
+    console.log();
+  }
+}
+main();
diff --git a/test/crosstest.test.js b/test/crosstest.test.js
index 1b494a0..3d90d86 100644
--- a/test/crosstest.test.js
+++ b/test/crosstest.test.js
@@ -76,7 +76,7 @@ const nodeCipher = (name, pcks7 = true) => {
 };
 
 function buf(n) {
-  return new Uint8Array(n).fill(n);
+  return new Uint8Array(n).fill(n % 251);
 }
 // TODO: re-use in benchmarks?
 // There is more ciphers, also 192 versions