Fix clipping of ovals on top & right

The oval fragment shader does a smooth step +/-0.5 about the outer radius. The actual drawn geometry thus has to be 1.0f greater than the bounding box containing outer radius to capture all the pixels.

[reed1, 2012-06-19 21:39:23]

Would a comment to this effect be good in the code, to document why we're adding
1?

[robertphillips, 2012-06-20 13:43:47]

Brian - I must have made some error when testing the first go round. Outsetting
half a pixel all around works correctly.

Note that there is no built in linear equivalent to smoothstep so we would have
to write our own (& add it to the FS) or do it inline - which would be messy. It
would essentially be the first step of the smoothstep calculation (t = clamp((x
- edge0) / (edge1 - edge0), 0.0, 1.0)).

[bsalomon, 2012-06-20 14:05:44]

LGTM

On 2012/06/20 13:43:47, robertphillips wrote:
> Brian - I must have made some error when testing the first go round.
Outsetting
> half a pixel all around works correctly.
> 

Phew, sanity is restored!

> Note that there is no built in linear equivalent to smoothstep so we would
have
> to write our own (& add it to the FS) or do it inline - which would be messy.
It
> would essentially be the first step of the smoothstep calculation (t =
clamp((x
> - edge0) / (edge1 - edge0), 0.0, 1.0)).

We can defer changing the ramp to another CL (and inserting pre-canned functions
to an even later CL).

http://codereview.appspot.com/6294091/diff/2002/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

http://codereview.appspot.com/6294091/diff/2002/src/gpu/GrContext.cpp#newcode...
src/gpu/GrContext.cpp:1114: SkScalar L = center.fX - outerRadius -
SkFloatToScalar(0.5f);
Might be a bit easier to read with SK_ScalarHalf rather than
SkFloatToScalar(0.5f).

[robertphillips, 2012-06-22 13:15:47]

committed as r4303

This delivery will require re-baselining of:

twopointconical, verttext, imagefiltersbase, convexpaths, hairmodes, circles,
aarectmodes, shadows