Trim in some cases of Array.splice.

Trim in some cases of Array.splice.

Committed: http://code.google.com/p/v8/source/detail?r=4321

[antonm, 2010-03-29 14:33:27]

Mads, Lasse,

may you have a look?

[antonm, 2010-03-30 10:08:33]

Erik,

would you mind reviewing it then?

tia and yours,
anton.

[Erik Corry, 2010-03-30 12:05:41]

LGTM.

If splice is carving a big chunk out of an array it would make sense to reuse
the backing store for the returned array instead of for the receiver.  That's
just an idea for another change, not for this one.

http://codereview.chromium.org/1562001/diff/1/2
File src/builtins.cc (right):

http://codereview.chromium.org/1562001/diff/1/2#newcode352
src/builtins.cc:352: MemsetPointer(former_start, filler_map, to_trim);
It would be faster to put a byte array here in the case where the size is more
than a few words.  Then you only have to tough the start.

[antonm, 2010-03-30 12:15:50]

Very cool idea about not copying large result arrays, thanks a lot, I'll give it
a shot.

And thanks a lot for review!

http://codereview.chromium.org/1562001/diff/1/2
File src/builtins.cc (right):

http://codereview.chromium.org/1562001/diff/1/2#newcode352
src/builtins.cc:352: MemsetPointer(former_start, filler_map, to_trim);
On 2010/03/30 12:05:42, Erik Corry wrote:
> It would be faster to put a byte array here in the case where the size is more
> than a few words.  Then you only have to tough the start.

That's true, but I have another idea: when I need to grow an array at the start
(for unshift and some forms of splice like splice(0, 0, ...), I can check if
words before a single word fillers and reuse them.

That might be problematic if we ever store those maps in the objects (btw, do
you know if we do), so I'd like to see performance impact first.  And if it's
not worth it, then, yes, I'd switch to variable size fillers here.

[antonm, 2010-03-30 12:16:31]

http://codereview.chromium.org/1562001/diff/1/2
File src/builtins.cc (right):

http://codereview.chromium.org/1562001/diff/1/2#newcode352
src/builtins.cc:352: MemsetPointer(former_start, filler_map, to_trim);
Anyway, I think that for now I'd better put it---it'd be easier to remove later.

On 2010/03/30 12:15:50, antonm wrote:
> On 2010/03/30 12:05:42, Erik Corry wrote:
> > It would be faster to put a byte array here in the case where the size is
more
> > than a few words.  Then you only have to tough the start.
> 
> That's true, but I have another idea: when I need to grow an array at the
start
> (for unshift and some forms of splice like splice(0, 0, ...), I can check if
> words before a single word fillers and reuse them.
> 
> That might be problematic if we ever store those maps in the objects (btw, do
> you know if we do), so I'd like to see performance impact first.  And if it's
> not worth it, then, yes, I'd switch to variable size fillers here.
>