perf: improve Buffer.from(buf) by 29x

[littledivy]

Improve Buffer.from(buffer) performance by ~29x

$ deno run -A bench.mjs
7987.390458 ms

$ node bench.mjs
773.420042 ms

$ bun bench.mjs
737.7943750 ms

$ target/release/deno run -A bench.mjs
275.3805 ms

import { Buffer } from "node:buffer";

const buffer = Buffer.alloc(1024 * 10, "latin1");

const start = performance.now();
for (let i = 0; i < 1000000; i++) {
  Buffer.from(buffer);
}
console.log(performance.now() - start);

Ref #24323

[0f-0b]

bench.mjs does not involve base64 decoding because buffer is not a string.

[littledivy]

Nice catch, updated the description

[0f-0b]

Suggested change

	const buf = new Uint8Array(u8.buffer, u8.byteOffset, u8.byteLength);
	Object.setPrototypeOf(buf, Buffer.prototype);
	return buf.slice();
	const buf = new Uint8Array(u8);
	Object.setPrototypeOf(buf, Buffer.prototype);
	return buf;

buf.slice refers to Buffer.prototype.slice which has undesired behavior.

$ cat a.mjs
import { Buffer } from "node:buffer";

const buf = new Uint8Array(1);
Buffer.from(buf)[0] = 1;
console.log(buf[0]);
$ node a.mjs
0
$ deno run a.mjs
0
$ ./deno-canary run a.mjs
1

[dsherret]

@0f-0b would you be willing to open a PR with that fix?

[0f-0b]

would you be willing to open a PR with that fix?

Opened #24352.