On 2013/09/11 13:53:45, titzer wrote:
On 2013/09/11 10:34:07, rossberg wrote:
> https://codereview.chromium.org/23609020/diff/1/src/hydrogen.h
> File src/hydrogen.h (right):
>
>
https://codereview.chromium.org/23609020/diff/1/src/hydrogen.h#newcode2309
> src/hydrogen.h:2309: class HUnique V8_FINAL {
> I think this should live somewhere else (e.g. handles.h or perhaps its
own
> file), and be named Unique, since it isn't really specific to Hydrogen.
Yeah I agree that we might want it to live in a more general place later,
but
I
don't want to go too far down the path of generalizing it just yet. So, I
want
to introduce it and start using in hydrogen, and when we find other uses
for
it,
we can move it out.
It's already perfectly general. That's why this is the wrong place to put
it. ;)
And hydrogen.h is far too long already anyway.
Oh alright. You fancy kids and your && constructs.
I'd like to propose a 5€ kitty tax for unjustified uses of goto (a.k.a.
break).
:)
>
https://codereview.chromium.org/23609020/diff/1/src/hydrogen.h#newcode2518
> src/hydrogen.h:2518: int new_capacity = capacity_ + capacity_ + size;
//
> 2*current + needed
> That seems like a lot, and can cause up to a 200% space overhead.
What's the
> rationale?
I've used this growth heuristic before and it works pretty nicely. The
rationale
is that most sets will be 1 element, probably 99.9% after that will be <=
4
elements (in fact, SmallMapList is limited to 4 elements, IIRC). Other
sets
might be big, so you want to grow aggressively to avoid copying.
Note that on average, you copy half of the array anyway. If minimizing
copying
was a concern, it might be worth avoiding the duplicate copying. But I agree
that for the sizes we are dealing with here, it probably doesn't matter
either
way.
https://codereview.chromium.org/23609020/
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" 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.
