Return early for vec.remove(position, 0).

Return early for vec.remove(position, 0).

If the item range being removed has no extent, no item is removed
from the vector. Return early, if so.

Style invalidation uses this Vector remove() operation upon restoring
state (RecursionCheckpoint), with the length being zero often.
Optimize its handling.

The performance of blink_perf.css.ClassInvalidation and
blink_perf.css.ClassDescendantSelector is improved 15-20% as a result
(without Oilpan, windows + linux64.)

R=haraken
BUG=

Committed: https://crrev.com/38b1397ba7eace722265035ef6b584b2a8f450c8
Cr-Commit-Position: refs/heads/master@{#351029}

[sof, 2015-09-27 07:14:23]

sigbjornf@opera.com changed reviewers:
+ haraken@chromium.org, tkent@chromium.org

[sof, 2015-09-27 07:14:24]

please take a look.

For the tests mentioned in the description, the improvement seen with Oilpan
enabled is somewhat less (~8% on Windows.)

[haraken, 2015-09-27 14:30:38]

LGTM

> The performance of blink_perf.css.ClassInvalidation and
> blink_perf.css.ClassDescendantSelector is improved 15-20% as a result.

Great improvement, thanks!!

Is there any improvement in non-oilpan as well? If there is no improvement in
non-oilpan but 15-20% improvement in oilpan, does it mean that either of
TypeOperations::destruct, TypeOperations::moveOverlapping or clearUnusedSlots is
the performance bottleneck of oilpan?

[sof, 2015-09-27 14:34:11]

On 2015/09/27 14:30:38, haraken wrote:
> LGTM
> 
> > The performance of blink_perf.css.ClassInvalidation and
> > blink_perf.css.ClassDescendantSelector is improved 15-20% as a result.
> 
> Great improvement, thanks!!
> 
> Is there any improvement in non-oilpan as well? If there is no improvement in
> non-oilpan but 15-20% improvement in oilpan, does it mean that either of
> TypeOperations::destruct, TypeOperations::moveOverlapping or clearUnusedSlots
is
> the performance bottleneck of oilpan?

You're reading the description not as intended - 15-20% is the improvement on
trunk (non-Oilpan). ~8% with Oilpan.

[haraken, 2015-09-27 14:55:49]

On 2015/09/27 14:34:11, sof wrote:
> On 2015/09/27 14:30:38, haraken wrote:
> > LGTM
> > 
> > > The performance of blink_perf.css.ClassInvalidation and
> > > blink_perf.css.ClassDescendantSelector is improved 15-20% as a result.
> > 
> > Great improvement, thanks!!
> > 
> > Is there any improvement in non-oilpan as well? If there is no improvement
in
> > non-oilpan but 15-20% improvement in oilpan, does it mean that either of
> > TypeOperations::destruct, TypeOperations::moveOverlapping or
clearUnusedSlots
> is
> > the performance bottleneck of oilpan?
> 
> You're reading the description not as intended - 15-20% is the improvement on
> trunk (non-Oilpan). ~8% with Oilpan.

ah, ok :)

Either way, this is a good change.

[sof, 2015-09-27 15:05:13]

On 2015/09/27 14:55:49, haraken wrote:
> On 2015/09/27 14:34:11, sof wrote:
> > On 2015/09/27 14:30:38, haraken wrote:
> > > LGTM
> > > 
> > > > The performance of blink_perf.css.ClassInvalidation and
> > > > blink_perf.css.ClassDescendantSelector is improved 15-20% as a result.
> > > 
> > > Great improvement, thanks!!
> > > 
> > > Is there any improvement in non-oilpan as well? If there is no improvement
> in
> > > non-oilpan but 15-20% improvement in oilpan, does it mean that either of
> > > TypeOperations::destruct, TypeOperations::moveOverlapping or
> clearUnusedSlots
> > is
> > > the performance bottleneck of oilpan?
> > 
> > You're reading the description not as intended - 15-20% is the improvement
on
> > trunk (non-Oilpan). ~8% with Oilpan.
> 
> ah, ok :)
> 
> Either way, this is a good change.

Thanks; silly benchmarks they are, but the fast path has merit regardless, imo.

Why Oilpan doesn't benefit equally, I don't have a handle on - caching effects
would be a guess.

[sof, 2015-09-28 05:09:40]

The CQ bit was checked by sigbjornf@opera.com

[commit-bot: I haz the power, 2015-09-28 05:09:51]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1370933002/20001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1370933002/20001

[commit-bot: I haz the power, 2015-09-28 07:12:58]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2015-09-28 07:13:37]

Patchset 2 (id:??) landed as
https://crrev.com/38b1397ba7eace722265035ef6b584b2a8f450c8
Cr-Commit-Position: refs/heads/master@{#351029}