On Tuesday, November 19, 2013 2:04:45 AM UTC-8, Dmitry Lomov wrote:
>
> Here is the change: https://code.google.com/p/v8/source/detail?r=15205
> Instead of weak handles, I taught our GC to free array buffer backing
> store when array buffers die.
>
That's awesome. Something like that is exactly what I originally wanted as
an extension of Object::SetIndexedPropertiesToExternalArrayData() where a
user could pass in a fourth optional bool that told the Object to
automatically clean up the attached external memory when the Object was
GC'd.
Oh well. Not like it matters much now. :)
> Off the top of my head, this smells like GC kicking in, although it does
> not explain why this is not reproducible with a native callback below. Can
> you share code for these benchmarks?
>
Sure. It's pretty basic:
var ITER = 1e6;
var SIZE = 16 * 1024;
// This size runs over 4x faster in Node
//var SIZE = 32 * 1024;
var t = -Date.now();
for (var i = 0; i < ITER; i++) {
new Uint8Array(SIZE);
}
t += Date.now();
console.log((t * 1e6 / ITER).toFixed(1));
I can't exhibit the same problem in d8 that exists in Node. But in Node on
my machine the 16KB allocations run in ~1800ns/op while the 32KB
allocations run in ~500ns/op. Here's Node's ArrayBuffer::Allocator
implementation: http://git.io/bK_aAA
Also, I was incorrect. The issue does still occur if called using the C++
API. Testing your hypothesis about GC I ran the tests using --prof and got
the following:
16KB size:
[GC]:
ticks total nonlib name
536 29.1%
32KB size:
[GC]:
ticks total nonlib name
224 46.9%
Here are the full logs for both
runs: https://gist.github.com/trevnorris/7551538
Let me know if there's any other information that would be helpful. :)
--
--
v8-users mailing list
[email protected]
http://groups.google.com/group/v8-users
---
You received this message because you are subscribed to the Google Groups
"v8-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.
