SkColorCubeFilter_opts: rounding is actually free here.

SkColorCubeFilter_opts: rounding is actually free here.

(Sk4f(float) is statically initializable, unlike the old SkPMFlor(SkPMColor).)

BUG=skia:4117
Committed: https://skia.googlesource.com/skia/+/c5281e5b103247a7d5cd8c266a2269d114081a47

[mtklein_C, 2015-09-01 13:20:01]

mtklein@chromium.org changed reviewers:
+ noel@chromium.org

[mtklein_C, 2015-09-01 13:20:05]

The CQ bit was checked by mtklein@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-09-01 13:20:12]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1317593007/1
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1317593007/1

[commit-bot: I haz the power, 2015-09-01 13:24:53]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-09-01 13:24:54]

Dry run: This issue passed the CQ dry run.

[Noel Gordon, 2015-09-01 14:08:22]

LGTM, speed and precision (we get a float round and clamp when converting to
premult color bytes at the end of the loop).

[mtklein, 2015-09-01 14:08:44]

mtklein@google.com changed reviewers:
+ mtklein@google.com

[mtklein, 2015-09-01 14:08:44]

The CQ bit was checked by mtklein@google.com

[mtklein, 2015-09-01 14:08:45]

lgtm

[commit-bot: I haz the power, 2015-09-01 14:08:50]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1317593007/1
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1317593007/1

[commit-bot: I haz the power, 2015-09-01 14:09:25]

Committed patchset #1 (id:1) as
https://skia.googlesource.com/skia/+/c5281e5b103247a7d5cd8c266a2269d114081a47

[mtklein_C, 2015-09-01 15:24:06]

On 2015/09/01 14:09:25, commit-bot: I haz the power wrote:
> Committed patchset #1 (id:1) as
> https://skia.googlesource.com/skia/+/c5281e5b103247a7d5cd8c266a2269d114081a47

https://perf.skia.org/#4002

Welcome to the mad world of man vs. compiler vs. machine.  This looks like an
~8% regression on the Nexus Player, an Atom compiled with, I think, GCC 4.9.
Everyone else that's finished so far looks okay.  I think we should keep it. 
The images are all returning to their previous values:

https://gold.skia.org/search?q=1&pos=true&query=config%3D8888%26config%3Dpipe...

[mtklein_C, 2015-09-01 15:26:24]

On 2015/09/01 15:24:06, mtklein_C wrote:
> On 2015/09/01 14:09:25, commit-bot: I haz the power wrote:
> > Committed patchset #1 (id:1) as
> >
https://skia.googlesource.com/skia/+/c5281e5b103247a7d5cd8c266a2269d114081a47
> 
> https://perf.skia.org/#4002
> 
> Welcome to the mad world of man vs. compiler vs. machine.  This looks like an
> ~8% regression on the Nexus Player, an Atom compiled with, I think, GCC 4.9.
> Everyone else that's finished so far looks okay.  I think we should keep it. 
> The images are all returning to their previous values:
> 
>
https://gold.skia.org/search?q=1&pos=true&query=config%3D8888%26config%3Dpipe...

(Shorter link:)
https://gold.skia.org/search?q=1&pos=true&query=cpu_or_gpu%3DCPU%26name%3Dcol...

[Noel Gordon, 2015-09-02 08:02:26]

On 2015/09/01 15:26:24, mtklein_C wrote:
> On 2015/09/01 15:24:06, mtklein_C wrote:
> > On 2015/09/01 14:09:25, commit-bot: I haz the power wrote:
> > > Committed patchset #1 (id:1) as
> > >
> https://skia.googlesource.com/skia/+/c5281e5b103247a7d5cd8c266a2269d114081a47
> > 
> > https://perf.skia.org/#4002
> > 
> > Welcome to the mad world of man vs. compiler vs. machine.  This looks like
an
> > ~8% regression on the Nexus Player, an Atom compiled with, I think, GCC 4.9.

Not a regression since ...

> > Everyone else that's finished so far looks okay.  I think we should keep it.

> > The images are all returning to their previous values:
> > 
> >
>
https://gold.skia.org/search?q=1&pos=true&query=config%3D8888%26config%3Dpipe...
> 
> (Shorter link:)
>
https://gold.skia.org/search?q=1&pos=true&query=cpu_or_gpu%3DCPU%26name%3Dcol...

... if I'm reading that link correctly (new tool for me), the gm results
recovered to those produced by SkPMFloat.round() in earlier patches.  Precision
was restored.  +1 to that.

[mtklein, 2015-09-02 12:55:46]

On 2015/09/02 08:02:26, noel gordon wrote:
> On 2015/09/01 15:26:24, mtklein_C wrote:
> > On 2015/09/01 15:24:06, mtklein_C wrote:
> > > On 2015/09/01 14:09:25, commit-bot: I haz the power wrote:
> > > > Committed patchset #1 (id:1) as
> > > >
> >
https://skia.googlesource.com/skia/+/c5281e5b103247a7d5cd8c266a2269d114081a47
> > > 
> > > https://perf.skia.org/#4002
> > > 
> > > Welcome to the mad world of man vs. compiler vs. machine.  This looks like
> an
> > > ~8% regression on the Nexus Player, an Atom compiled with, I think, GCC
4.9.
> 
> Not a regression since ...
> 
> > > Everyone else that's finished so far looks okay.  I think we should keep
it.
> 
> > > The images are all returning to their previous values:
> > > 
> > >
> >
>
https://gold.skia.org/search?q=1&pos=true&query=config%3D8888%26config%3Dpipe...
> > 
> > (Shorter link:)
> >
>
https://gold.skia.org/search?q=1&pos=true&query=cpu_or_gpu%3DCPU%26name%3Dcol...
> 
> ... if I'm reading that link correctly (new tool for me), the gm results
> recovered to those produced by SkPMFloat.round() in earlier patches. 
Precision
> was restored.  +1 to that.

Right, still 9-10% faster than where we started.

[Noel Gordon, 2015-09-02 13:23:08]

On 2015/09/02 12:55:46, mtklein wrote:
>
https://gold.skia.org/search?q=1&pos=true&query=cpu_or_gpu%3DCPU%26name%3Dcol...
> > 
> > ... if I'm reading that link correctly (new tool for me), the gm results
> > recovered to those produced by SkPMFloat.round() in earlier patches. 
> Precision
> > was restored.  +1 to that.
> 
> Right, still 9-10% faster than where we started.

Yes, but even more than that.  Is there a way in https://perf.skia.org to widen
the time-line (x-axis) so we can see the total improvement, aka since before we
added the new code to SkColorCubeFilter till today?  Damned if I could find a
way, tried help etc ...

[mtklein, 2015-09-02 13:27:35]

That is perf.skia.org's #1 feature request.  Looks like the first two steps down
have fallen off the left of our chart.  Some of those bots were in the 1000s.

[Noel Gordon, 2015-09-02 13:59:23]

Yeap, 1000's are way lower now.  The overall improvement, since before Radu
suggested the initial change, is ~3x.  In absolute terms,
qcms_transform_data_tetra_clut_rgba_sse2 corrects a 2880 x 1800 image in ~149ms
on win32.  GM:colorcube, a tri-linear interpolator, with sse does the same in
~125ms on win32.  That's an excellent result.

[Noel Gordon, 2015-09-02 14:04:29]

On 2015/09/02 13:27:35, mtklein wrote:
> That is perf.skia.org's #1 feature request.  Looks like the first two steps
down
> have fallen off the left of our chart.

Shame. Woulda loved to snap a picture of the overall improvement over this
series of patches.