Gradient Fixes

Added GM and Bench 

Added GM's and Bench


Added GM file nate_gradients.cpp

BUG=

Committed: http://code.google.com/p/skia/source/detail?r=9938

[dierk, 2013-07-03 20:56:06]

Made a GM that illustrates the 3 big issues we've discussed about gradients: 

Column 1) Many-colored gradients and noticeable color-banding. Very apparant
when zoomed in.
--fix by doing CPU lerp

Column 2) Fast-transitioning gradients and pre-computed table. Causes jumping
around when zooming.
--fix by evaluating directly instead of pre-computing

Column 3) Shallow gradient with large flat spans color-banding on GPU.
--fix by adding dithering to GPU.


There's also a bench that essentially just runs the GM and times it. How best to
start modifying the gradient code?

[dierk, 2013-07-08 18:51:00]

Removed the additional bench file I had made (and the include from .gypi) and
renamed the GM to gradientDirtyLaundry.cpp. I've also deleted the unused code
from the existing gradientBench file.

Planning on adding the additional bench tests for gradients with many colors
next.

Is there anyone else who has knowledge of the gradients code who could also
review (for official evaluation purposes?)

[bsalomon, 2013-07-08 19:28:46]

There already are shallow gray gradient GMs (one for each gradient type). Not
that it hurts to have them here, too.

https://codereview.chromium.org/18259012/diff/10001/gm/gradientDirtyLaundry.cpp
File gm/gradientDirtyLaundry.cpp (right):

https://codereview.chromium.org/18259012/diff/10001/gm/gradientDirtyLaundry.c...
gm/gradientDirtyLaundry.cpp:3: * Copyright 2011 Google Inc.
2013

https://codereview.chromium.org/18259012/diff/10001/gm/gradientDirtyLaundry.c...
gm/gradientDirtyLaundry.cpp:83: virtual SkISize onISize() { return
make_isize(640, 615); }
SK_OVERRIDE

https://codereview.chromium.org/18259012/diff/10001/gm/gradientDirtyLaundry.c...
gm/gradientDirtyLaundry.cpp:85: virtual void onDraw(SkCanvas* canvas) {
SK_OVERRIDE

https://codereview.chromium.org/18259012/diff/10001/gm/gradientDirtyLaundry.c...
gm/gradientDirtyLaundry.cpp:99: paint.setShader(shader);
you can just unref here instead of below (paint will take a ref).
paint.setShader(shader)->unref();

https://codereview.chromium.org/18259012/diff/10001/gyp/gmslides.gypi
File gyp/gmslides.gypi (right):

https://codereview.chromium.org/18259012/diff/10001/gyp/gmslides.gypi#newcode72
gyp/gmslides.gypi:72: '../gm/nate_gradients.cpp',
?

[dierk, 2013-07-09 17:52:52]

Applied suggested fixes as well as slimmed the GM to only include the many
colored gradient case as that is the issue I'll be working on first. Will add
other GM's as needed or use existing GM's (like GM:shallow_gradient_*)

https://codereview.chromium.org/18259012/diff/10001/gm/gradientDirtyLaundry.cpp
File gm/gradientDirtyLaundry.cpp (right):

https://codereview.chromium.org/18259012/diff/10001/gm/gradientDirtyLaundry.c...
gm/gradientDirtyLaundry.cpp:3: * Copyright 2011 Google Inc.
On 2013/07/08 19:28:46, bsalomon wrote:
> 2013

Done.

https://codereview.chromium.org/18259012/diff/10001/gm/gradientDirtyLaundry.c...
gm/gradientDirtyLaundry.cpp:83: virtual SkISize onISize() { return
make_isize(640, 615); }
On 2013/07/08 19:28:46, bsalomon wrote:
> SK_OVERRIDE

Done.

https://codereview.chromium.org/18259012/diff/10001/gm/gradientDirtyLaundry.c...
gm/gradientDirtyLaundry.cpp:85: virtual void onDraw(SkCanvas* canvas) {
On 2013/07/08 19:28:46, bsalomon wrote:
> SK_OVERRIDE

Done.

https://codereview.chromium.org/18259012/diff/10001/gm/gradientDirtyLaundry.c...
gm/gradientDirtyLaundry.cpp:99: paint.setShader(shader);
On 2013/07/08 19:28:46, bsalomon wrote:
> you can just unref here instead of below (paint will take a ref).
> paint.setShader(shader)->unref();

Done.

https://codereview.chromium.org/18259012/diff/10001/gyp/gmslides.gypi
File gyp/gmslides.gypi (right):

https://codereview.chromium.org/18259012/diff/10001/gyp/gmslides.gypi#newcode72
gyp/gmslides.gypi:72: '../gm/nate_gradients.cpp',
On 2013/07/08 19:28:46, bsalomon wrote:
> ?

Done.

[bsalomon, 2013-07-09 17:59:30]

I think our gms tend_to_be_named_with_underscores (the name string not the
class). Otherwise lgtm (by which I mean if you are only changing the string then
I don't need to see another patchset :) )

[commit-bot: I haz the power, 2013-07-09 19:35:30]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/dierk@google.com/18259012/23001

[commit-bot: I haz the power, 2013-07-09 19:43:31]

Change committed as 9938