Add new bench to compare drawSprite() and drawBitmap().

Add new bench to compare drawSprite() and drawBitmap().

Currently, drawSprite() is faster when an image filter is attached.

BUG=skia:1062

[Stephen White, 2015-10-14 16:53:00]

senorblanco@chromium.org changed reviewers:
+ reed@google.com

[Stephen White, 2015-10-14 16:53:10]

reed@: PTAL. Thanks!

[Stephen White, 2015-10-14 16:56:56]

On 2015/10/14 16:53:10, Stephen White wrote:
> reed@: PTAL. Thanks!

mtklein: I've just realized this may be triggering the optimization in
SkOffsetImageFilter that just passes the offset through to the final restore.
You might want to check that the copy is still getting triggered, and set
SK_DISABLE_OFFSETIMAGEFILTER_OPTIMIZATION 1 if so.

(or use another filter)

[mtklein, 2015-10-14 17:00:51]

On 2015/10/14 16:56:56, Stephen White wrote:
> On 2015/10/14 16:53:10, Stephen White wrote:
> > reed@: PTAL. Thanks!
> 
> mtklein: I've just realized this may be triggering the optimization in
> SkOffsetImageFilter that just passes the offset through to the final restore.
> You might want to check that the copy is still getting triggered, and set
> SK_DISABLE_OFFSETIMAGEFILTER_OPTIMIZATION 1 if so.
> 
> (or use another filter)

Looks like a good demo.  I see sprite (5) < bitmap (12), and with the cache copy
both regress, sprite by ~80% to 9, bitmap by 33% to 16.  (Which makes sense...
the copy costs a flat 4.)

[reed1, 2015-10-14 19:33:01]

With applyFilter (once we have a gpu-version of colorfilterimagefilter), there
will be no callers of drawSprite.

[Stephen White, 2015-10-14 19:36:51]

On 2015/10/14 19:33:01, reed1 wrote:
> With applyFilter (once we have a gpu-version of colorfilterimagefilter), there
> will be no callers of drawSprite.

Sadly, it turns out I don't think we can use an applyFilter()-type approach
after all. I actually need the pad-to-black behaviour of drawBitmap() 
in order to finish implementing filter effects region (http://crbug.com/240827).
So I think the best way forward is to optimize drawBitmap()
and stop calling drawSprite(). Happy to explain via the comms of your choice. :)

[Stephen White, 2015-10-15 03:13:56]

reed@: Could we land the drawBitmap() part of this, at a minimum? It would be
useful to prevent regressions, such as the one it
just detected in mtklein@'s change.

[Stephen White, 2015-12-07 21:02:29]

Closing this in favour of reed@'s https://codereview.chromium.org/1497843003
which has already landed.