Push sigma-based blur interface into our GMs/benches/tests/samplecode

This makes all Skia-side tools use the new sigma-based blur interface. It is the first step towards truly deprecating the radius-based interface. (Next stop: Chromium).

[robertphillips, 2013-09-05 12:28:40]

It looks like the blur refactoring is going to stick so ... ping.

[bsalomon, 2013-09-05 13:09:09]

https://codereview.chromium.org/23701006/diff/3001/bench/BlurBench.cpp
File bench/BlurBench.cpp (right):

https://codereview.chromium.org/23701006/diff/3001/bench/BlurBench.cpp#newcode16
bench/BlurBench.cpp:16: #define SMALL   SkFloatToScalar(1.6547f)
This looks really weird, as though someone chose this particular float for some
reason. Could we have a static inline in SkBlurMaskFilter that converts?
Something like

SkMaskFilter* mf = SkBlurMaskFilter::Create(fStyle,
SkBlurMaskRadius::LegacyRadiusToSigma(fRadius), fFlags);

Also below we have the raiduses again as constructor params called offset. It's
obvious at all what their relationship is.

[robertphillips, 2013-09-05 14:00:14]

https://codereview.chromium.org/23701006/diff/3001/bench/BlurBench.cpp
File bench/BlurBench.cpp (right):

https://codereview.chromium.org/23701006/diff/3001/bench/BlurBench.cpp#newcode16
bench/BlurBench.cpp:16: #define SMALL   SkFloatToScalar(1.6547f)
I considered doing that (using a conversion method) but felt it wasn't really in
the spirit of the new API. If we did switch to a conversion method I would
prefer that it be in the testing/bench/GM code rather then in core Skia. I hope
to completely remove the deprecated entry points and the
SkBlurMask::ConvertRadiusToSigma function that currently does the conversion, so
having SkBlurMaskFilter::LegacyRadiusToSigma would keep that 0.57735 constant in
play.

The weird offset thing is to make the images consistent before and after this
change. Since this is bench it might not be important to preserve going forward,
but it was a useful check during the conversion process.

[bsalomon, 2013-09-05 14:10:54]

https://codereview.chromium.org/23701006/diff/3001/bench/BlurBench.cpp
File bench/BlurBench.cpp (right):

https://codereview.chromium.org/23701006/diff/3001/bench/BlurBench.cpp#newcode16
bench/BlurBench.cpp:16: #define SMALL   SkFloatToScalar(1.6547f)
On 2013/09/05 14:00:14, robertphillips wrote:
> I considered doing that (using a conversion method) but felt it wasn't really
in
> the spirit of the new API. If we did switch to a conversion method I would
> prefer that it be in the testing/bench/GM code rather then in core Skia. I
hope
> to completely remove the deprecated entry points and the
> SkBlurMask::ConvertRadiusToSigma function that currently does the conversion,
so
> having SkBlurMaskFilter::LegacyRadiusToSigma would keep that 0.57735 constant
in
> play.
> 
> The weird offset thing is to make the images consistent before and after this
> change. Since this is bench it might not be important to preserve going
forward,
> but it was a useful check during the conversion process.

It just seems like a future editor of this file (and BlurRectBench.cpp) will be
confused about the radius/offset correspondences such as the one between #define
REAL SkFloatToScalar(1.366025f) and 1.5. It's not obvious how they're related.

[robertphillips, 2013-09-05 19:35:28]

PTAL

[reed1, 2013-09-05 20:23:10]

lgtm

[bsalomon, 2013-09-05 22:57:12]

On 2013/09/05 20:23:10, reed1 wrote:
> lgtm

lgtm (I would've been fine with just changing up the benches to use a local func
but this works for me).

[robertphillips, 2013-09-06 14:28:17]

committed as r11129