First pass at Rect Effect

This CL only adds 'filled' rects as an effect. I would like to deliver it as a checkpoint but not turn it on. There are still some worrying performance issues with it; namely:

BEFORE:

aarects   GPU: cmsecs =  77.60 gmsecs =  48.37  NULLGPU: cmsecs =  63.88
drawpointsdash_5_5_aa   GPU: cmsecs =  54.73 gmsecs =  53.64  NULLGPU: cmsecs =  28.01
drawpointsdash_3_1_aa   GPU: cmsecs =  71.82 gmsecs =  67.50  NULLGPU: cmsecs =  41.81
drawpointsdash_1_1_aa   GPU: cmsecs = 128.37 gmsecs = 111.93  NULLGPU: cmsecs = 116.00

AFTER: 

aarects   GPU: cmsecs =  83.21 gmsecs =   7.50  NULLGPU: cmsecs =  75.41
drawpointsdash_5_5_aa   GPU: cmsecs =  64.03 gmsecs =  57.15  NULLGPU: cmsecs =  35.45
drawpointsdash_3_1_aa   GPU: cmsecs =  81.66 gmsecs =  73.17  NULLGPU: cmsecs =  50.60
drawpointsdash_1_1_aa   GPU: cmsecs = 152.18 gmsecs =  14.32  NULLGPU: cmsecs = 142.01

The regressions seem to be mainly due to increased CPU overhead.

[jvanverth1, 2013-04-08 14:58:56]

I think we may need to look into optimizing how Effects are set up. I ran into
similar problems with the oval/circle changes.

[bsalomon, 2013-04-08 15:48:38]

https://codereview.chromium.org/13521006/diff/9001/include/core/SkMatrix.h
File include/core/SkMatrix.h (right):

https://codereview.chromium.org/13521006/diff/9001/include/core/SkMatrix.h#ne...
include/core/SkMatrix.h:97: bool notSkewOrPersp(SkScalar tol =
SK_ScalarNearlyZero) const;
isOrthogonal()?

https://codereview.chromium.org/13521006/diff/9001/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

https://codereview.chromium.org/13521006/diff/9001/src/gpu/GrContext.cpp#newc...
src/gpu/GrContext.cpp:690: bool stroked) {
I think the width param already tells you if its stroked (-1 is fill and
otherwise its the stroke width IIRC). Maybe rename to strokeWidth?

https://codereview.chromium.org/13521006/diff/9001/src/gpu/effects/GrRectEffe...
File src/gpu/effects/GrRectEffect.cpp (right):

https://codereview.chromium.org/13521006/diff/9001/src/gpu/effects/GrRectEffe...
src/gpu/effects/GrRectEffect.cpp:45: builder->fsCodeAppendf("\tfloat wScale =
max(1.0, 2.0/(0.5+%s.x));\n", fsWidthHeightName);
Do you think it makes sense to do this in the VS?

https://codereview.chromium.org/13521006/diff/9001/src/gpu/effects/GrRectEffe...
File src/gpu/effects/GrRectEffect.h (right):

https://codereview.chromium.org/13521006/diff/9001/src/gpu/effects/GrRectEffe...
src/gpu/effects/GrRectEffect.h:13: class GrGLRectEffect;
What do you think of putting this in the AARectRenderer's cpp? We've done that
for ovals, etc. Maybe there are other callers for this? But otherwise it seems
nice to have the vert setup and shader code in the same file (unless the size
gets out of control).

https://codereview.chromium.org/13521006/diff/9001/src/gpu/effects/GrRectEffe...
src/gpu/effects/GrRectEffect.h:21: *      Half height + 0.5
Maybe comment here on what varying that this effect takes?

[robertphillips, 2013-04-08 19:33:20]

https://codereview.chromium.org/13521006/diff/9001/include/core/SkMatrix.h
File include/core/SkMatrix.h (right):

https://codereview.chromium.org/13521006/diff/9001/include/core/SkMatrix.h#ne...
include/core/SkMatrix.h:97: bool notSkewOrPersp(SkScalar tol =
SK_ScalarNearlyZero) const;
Done - although I'm not sure is 'isOrthogonal' is clear enough.

https://codereview.chromium.org/13521006/diff/9001/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

https://codereview.chromium.org/13521006/diff/9001/src/gpu/GrContext.cpp#newc...
src/gpu/GrContext.cpp:690: bool stroked) {
On 2013/04/08 15:48:39, bsalomon wrote:
> I think the width param already tells you if its stroked (-1 is fill and
> otherwise its the stroke width IIRC). Maybe rename to strokeWidth?

Done.

https://codereview.chromium.org/13521006/diff/9001/src/gpu/effects/GrRectEffe...
File src/gpu/effects/GrRectEffect.cpp (right):

https://codereview.chromium.org/13521006/diff/9001/src/gpu/effects/GrRectEffe...
src/gpu/effects/GrRectEffect.cpp:45: builder->fsCodeAppendf("\tfloat wScale =
max(1.0, 2.0/(0.5+%s.x));\n", fsWidthHeightName);
Yes but I would like to delay to the next revision. I have added a reminder
TODO.

https://codereview.chromium.org/13521006/diff/9001/src/gpu/effects/GrRectEffe...
File src/gpu/effects/GrRectEffect.h (right):

https://codereview.chromium.org/13521006/diff/9001/src/gpu/effects/GrRectEffe...
src/gpu/effects/GrRectEffect.h:13: class GrGLRectEffect;
Done - we can always move it out if line drawing needs it elsewhere.

https://codereview.chromium.org/13521006/diff/9001/src/gpu/effects/GrRectEffe...
src/gpu/effects/GrRectEffect.h:21: *      Half height + 0.5
On 2013/04/08 15:48:39, bsalomon wrote:
> Maybe comment here on what varying that this effect takes?

Done.

[bsalomon, 2013-04-08 21:48:41]

"orthogonal matrix" actually means ortho*normal* vectors, so my name was a bad
idea.

[bsalomon, 2013-04-09 13:39:18]

lgtm (with name change)

[robertphillips, 2013-04-09 14:02:21]

Renaming done.

Committed as r8571

[robertphillips, 2013-04-25 15:47:24]

Here are the regressions as of 4/25 with removal of the SkMatrix::reset
overhead. These result from running all the bench marks (at repeat=200 on
MacPro) and removing all the ones with <= .5ms or <=3% timing differences:

               aarects GPU c  46.03  52.28   -6.25 -13.6%
 drawpointsdash_1_1_aa GPU c  81.90  92.93  -11.03 -13.5%
 drawpointsdash_3_1_aa GPU c  37.12  41.73   -4.61 -12.4%
 drawpointsdash_5_5_aa GPU c  26.01  28.90   -2.89 -11.1%
          circles_fill GPU c  15.66  17.21   -1.55  -9.9%
        circles_stroke GPU c  21.68  23.52   -1.84  -8.5%
            lines_1_BW GPU c  10.23  10.90   -0.67  -6.5%
         dash_1_noclip GPU c  16.25  16.84   -0.59  -3.6%
               readpix GPU c  79.22  81.97   -2.75  -3.5%