Make GPU linear gradients do bilinear filtering.

Make GPU linear gradients do bilinear filtering.

This will require GM (and webkit) rebaselines.
Committed: https://code.google.com/p/skia/source/detail?r=4680

[TomH, 2012-07-19 15:25:46]

I like - but it'll take willpower to push this through. Can you do a hack build
of chrome/DumpRenderTree with this and figure out just how many webkit
rebaselines you require? 50 may be doable (Elliot?), 500 isn't.

If you're only on a laptop, that's probably an unfair request; we can try it on
my workstation.

[epoger, 2012-07-19 15:43:51]

On 2012/07/19 15:25:46, TomH wrote:
> I like - but it'll take willpower to push this through. Can you do a hack
build
> of chrome/DumpRenderTree with this and figure out just how many webkit
> rebaselines you require? 50 may be doable (Elliot?), 500 isn't.

I think it's a good idea to check the webkit impact before committing.  You can
do this with trybots... see
https://sites.google.com/site/skiadocs/developer-documentation/managing-chrom...

Let's see what the impact is.  Unfortunately we are still in a bad spot for
large-scale rebaselines due to Skia rolls.  See
https://code.google.com/p/skia/issues/detail?id=712 ('we need a way to
rebaseline lots of webkit layout tests already')
> 
> If you're only on a laptop, that's probably an unfair request; we can try it
on
> my workstation.

[TomH, 2012-07-19 15:48:00]

On 2012/07/19 15:25:46, TomH wrote:
> I like - but it'll take willpower to push this through. Can you do a hack
build
> of chrome/DumpRenderTree with this and figure out just how many webkit
> rebaselines you require? 50 may be doable (Elliot?), 500 isn't.
> 
> If you're only on a laptop, that's probably an unfair request; we can try it
on
> my workstation.

Just to record for posterity: as I understand it, Brian's building
DumpRenderTree with this patch.

[bsalomon, 2012-07-19 17:36:29]

On 2012/07/19 15:48:00, TomH wrote:
> On 2012/07/19 15:25:46, TomH wrote:
> > I like - but it'll take willpower to push this through. Can you do a hack
> build
> > of chrome/DumpRenderTree with this and figure out just how many webkit
> > rebaselines you require? 50 may be doable (Elliot?), 500 isn't.
> > 
> > If you're only on a laptop, that's probably an unfair request; we can try it
> on
> > my workstation.
> 
> Just to record for posterity: as I understand it, Brian's building
> DumpRenderTree with this patch.

I'm going to build DRT with Riley's earlier patch at r4675 and asses the WK
impact of that. Then I will try this patch.

[bsalomon, 2012-07-19 19:40:15]

On 2012/07/19 17:36:29, bsalomon wrote:
> On 2012/07/19 15:48:00, TomH wrote:
> > On 2012/07/19 15:25:46, TomH wrote:
> > > I like - but it'll take willpower to push this through. Can you do a hack
> > build
> > > of chrome/DumpRenderTree with this and figure out just how many webkit
> > > rebaselines you require? 50 may be doable (Elliot?), 500 isn't.
> > > 
> > > If you're only on a laptop, that's probably an unfair request; we can try
it
> > on
> > > my workstation.
> > 
> > Just to record for posterity: as I understand it, Brian's building
> > DumpRenderTree with this patch.
> 
> I'm going to build DRT with Riley's earlier patch at r4675 and asses the WK
> impact of that. Then I will try this patch.

The only test that is changed by this CL is:
platform/chromium/virtual/gpu/fast/canvas/canvas-text-alignment.html

LGTM

[TomH, 2012-07-19 19:40:52]

On 2012/07/19 19:40:15, bsalomon wrote:
>
> The only test that is changed by this CL is:
> platform/chromium/virtual/gpu/fast/canvas/canvas-text-alignment.html

Sweet!  LGTM, then, too.