More blur fixes

- clamp the max. radius to 128, as the software path does
- use a more accurate radius-to-sigma conversion (0.6 instead of 0.6666)
- make SampleBlur derive from SampleView, not SkView, so benchmark mode ('f') works
- implement a new BigBlur sample to test large radii

[Stephen White, 2011-07-14 20:42:08]

Upload the right patch set this time..

[bsalomon, 2011-07-14 20:43:39]

On 2011/07/14 20:42:08, Stephen White wrote:
> Upload the right patch set this time..

LGTM

[reed1, 2011-07-14 20:47:26]

http://codereview.appspot.com/4726043/diff/3001/src/gpu/SkGpuDevice.cpp
File src/gpu/SkGpuDevice.cpp (right):

http://codereview.appspot.com/4726043/diff/3001/src/gpu/SkGpuDevice.cpp#newco...
src/gpu/SkGpuDevice.cpp:53: // Should probably be put in a common header
someplace.
Lets name it something blur-specific, and make it a scalar.

e.g.
#define MAX_BLUR_RADIUS  SkIntToScalar(128)

Same naming comment for MAX_SIGMA

http://codereview.appspot.com/4726043/diff/3001/src/gpu/SkGpuDevice.cpp#newco...
src/gpu/SkGpuDevice.cpp:844: radius = SkMinScalar(radius, MAX_RADIUS);
Where does 0.6 come from? Maybe its clear to gauss experts, but not me :(

[Stephen White, 2011-07-14 21:00:56]

http://codereview.appspot.com/4726043/diff/3001/src/gpu/SkGpuDevice.cpp
File src/gpu/SkGpuDevice.cpp (right):

http://codereview.appspot.com/4726043/diff/3001/src/gpu/SkGpuDevice.cpp#newco...
src/gpu/SkGpuDevice.cpp:53: // Should probably be put in a common header
someplace.
On 2011/07/14 20:47:26, reed1 wrote:
> Lets name it something blur-specific, and make it a scalar.
> 
> e.g.
> #define MAX_BLUR_RADIUS  SkIntToScalar(128)
> 
> Same naming comment for MAX_SIGMA

Done.

http://codereview.appspot.com/4726043/diff/3001/src/gpu/SkGpuDevice.cpp#newco...
src/gpu/SkGpuDevice.cpp:844: radius = SkMinScalar(radius, MAX_RADIUS);
On 2011/07/14 20:47:26, reed1 wrote:
> Where does 0.6 come from? Maybe its clear to gauss experts, but not me :(

It comes from me tweaking the result until it looks the same as the software
path.  :)  Upon further investigation, I think it actually should be 1/sqrt(3),
which is what junov added when he implemented the "high quality" mode in the
software path.  IMHO, it actually should be 1:  we blur "less" than we should do
according to the CSS and canvas specs, simply because Safari does the same. 
Firefox used to do the same too, until 4.0 where they fixed it.  So at some
point we should probably get rid of these scaling constants and rebaseline all
the blur tests.

[Stephen White, 2011-07-14 21:03:33]

Fixes from mike's comments

[reed1, 2011-07-15 12:40:09]

My question about the 0.6 was hoping to get your (good) response as a comment in
the code, for the next reader who wonders about it.

[reed1, 2011-07-15 12:43:48]

On 2011/07/15 12:40:09, reed1 wrote:
> My question about the 0.6 was hoping to get your (good) response as a comment
in
> the code, for the next reader who wonders about it.

LGTM if we can paste your 0.6 comment somewhere in the code.

[Stephen White, 2011-07-15 14:28:29]

On 2011/07/15 12:43:48, reed1 wrote:
> LGTM if we can paste your 0.6 comment somewhere in the code.

Done.  Landed as r1871.  Closing.