chore: tune signals for better runtime perf #9529
Conversation
🦋 Changeset detectedLatest commit: 40e2c92 The changes in this PR will be included in the next version bump. This PR includes changesets to release 1 package
Not sure what this means? Click here to learn what changesets are. Click here if you're a maintainer who wants to add another changeset to this PR |
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
|
| @@ -744,12 +734,9 @@ export function get(signal) { | |||
| current_dependencies_index++; | |||
| } else if (current_dependencies === null) { | |||
| current_dependencies = [signal]; | |||
| } else if (signal.read !== current_consumer_read_clock) { | |||
| } else if (signal !== current_dependencies.at(-1)) { | |||
There was a problem hiding this comment.
isn't at like really slow compared to [.length - 1]?
There was a problem hiding this comment.
I thought it was faster?
There was a problem hiding this comment.
There was a problem hiding this comment.
Hmm, .at is faster on that site, but I got 2.5s for .at and 670ms for .length - 1, when running this snippet in my console,
{
let N = 1000000;
const arr = Array.from({length: 100}).map(Math.random);
let acc= 0;
console.log(arr.at(-1), arr[arr.length - 1])
for (var j = 1000; j--;) acc += arr.at(-1);
for (var j = 1000; j--;) acc += arr[arr.length - 1];
console.time();
for (var i = N; i--;) { acc = 0; for (var j = 1000; j--;) acc += arr.at(-1); }
console.timeEnd();
console.time();
for (var i = N; i--;) { acc = 0; for (var j = 1000; j--;) acc += arr[arr.length - 1]; }
console.timeEnd();
console.log(acc);
}I read about it being slower somewhere because the method had to also consider ArrayLike objects
There was a problem hiding this comment.
ESBench results with the same code from measurethat
There was a problem hiding this comment.
I'll put that change in the other PR.
There was a problem hiding this comment.
On the left side is at() function, and on the right side is length() function.
function at(i) {
return i.at(-1);
}
function length(i) {
return i[i.length - 1]
}There was a problem hiding this comment.
Looks like the left side is slightly better.
There was a problem hiding this comment.
I think that length() function should be faster, it will execute 0-6e instructions if nothing goes wrong like OOB, wrong object shape, etc. And at() function will execute 0-5d, jmp to 84-8c, jmp to 6a-78. But I am not 100% sure about it, just quickly looked at it :)
We can improve some areas of the signals runtime, credit to @localvoid for pointing these out to me. :)