add entry-point to SkMaskFilter to fast-path rectangles.

add entry-point to SkMaskFilter to fast-path rectangles.
have blurmaskfilter override this new method.

#define SK_IGNORE_FAST_BLURRECT if you want to disable this (not sure if we'll need this)
Committed: https://code.google.com/p/skia/source/detail?r=6363

[caryclark1, 2012-11-08 21:13:31]

lgtm

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp
File src/core/SkMaskFilter.cpp (right):

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp#ne...
src/core/SkMaskFilter.cpp:63: int cy = mask.fBounds.top() +
mask.fBounds.bottom() >> 1;
Interesting that SkRect has centerX() and centerY() but not SkIRect. I wonder
how often this pattern shows up...

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp#ne...
src/core/SkMaskFilter.cpp:172: if (!clipper.done() && (bounder == NULL ||
bounder->doIRect(outerR))) {
is the skia convention NULL == bounder?

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp#ne...
src/core/SkMaskFilter.cpp:189: #endif
I always wonder about this convention because it prevents brace matching from
working

[bsalomon, 2012-11-08 21:18:21]

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp
File src/core/SkMaskFilter.cpp (right):

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp#ne...
src/core/SkMaskFilter.cpp:63: int cy = mask.fBounds.top() +
mask.fBounds.bottom() >> 1;
On 2012/11/08 21:13:31, caryclark1 wrote:
> Interesting that SkRect has centerX() and centerY() but not SkIRect. I wonder
> how often this pattern shows up...

I'd say add centerX() and centerY()... the avg by adding and shifting makes it
slightly harder to read.

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp#ne...
src/core/SkMaskFilter.cpp:189: #endif
On 2012/11/08 21:13:31, caryclark1 wrote:
> I always wonder about this convention because it prevents brace matching from
> working

Others may disagree, but I like having bool macros always defined to 0 or 1 and
then treating them like runtime checks, knowing the that the compiler can
optimize:

if (!SK_IGNORE_FAST_BLURRECT &&
    devPath.isRect(&rect) &&
    SkPaint::kFill_Style == style) {
...

[reed1, 2012-11-09 13:33:43]

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp
File src/core/SkMaskFilter.cpp (right):

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp#ne...
src/core/SkMaskFilter.cpp:63: int cy = mask.fBounds.top() +
mask.fBounds.bottom() >> 1;
On 2012/11/08 21:13:31, caryclark1 wrote:
> Interesting that SkRect has centerX() and centerY() but not SkIRect. I wonder
> how often this pattern shows up...

Done.

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp#ne...
src/core/SkMaskFilter.cpp:172: if (!clipper.done() && (bounder == NULL ||
bounder->doIRect(outerR))) {
On 2012/11/08 21:13:31, caryclark1 wrote:
> is the skia convention NULL == bounder? 

Done.

https://codereview.appspot.com/6815087/diff/7006/src/core/SkMaskFilter.cpp#ne...
src/core/SkMaskFilter.cpp:189: #endif
On 2012/11/08 21:18:21, bsalomon wrote:
> On 2012/11/08 21:13:31, caryclark1 wrote:
> > I always wonder about this convention because it prevents brace matching
from
> > working
> 
> Others may disagree, but I like having bool macros always defined to 0 or 1
and
> then treating them like runtime checks, knowing the that the compiler can
> optimize:
> 
> if (!SK_IGNORE_FAST_BLURRECT &&
>     devPath.isRect(&rect) &&
>     SkPaint::kFill_Style == style) {
> ...

Done.

[bsalomon, 2012-11-09 13:46:39]

LGTM. (Some minor Q/comments)

https://codereview.appspot.com/6815087/diff/5008/include/core/SkMaskFilter.h
File include/core/SkMaskFilter.h (right):

https://codereview.appspot.com/6815087/diff/5008/include/core/SkMaskFilter.h#...
include/core/SkMaskFilter.h:121: *  My convention, the caller will take the
center rol/col from the returned
My->By? Must the returned mask have an odd W/H?

https://codereview.appspot.com/6815087/diff/5008/src/core/SkMaskFilter.cpp
File src/core/SkMaskFilter.cpp (right):

https://codereview.appspot.com/6815087/diff/5008/src/core/SkMaskFilter.cpp#ne...
src/core/SkMaskFilter.cpp:16: #ifdef SK_IGNORE_FAST_BLURRECT
Why the second macro instead of making post config just define this to 0 if it
hasn't been defined by user config or by the build system?

[reed1, 2012-11-09 14:24:40]

https://codereview.appspot.com/6815087/diff/5008/include/core/SkMaskFilter.h
File include/core/SkMaskFilter.h (right):

https://codereview.appspot.com/6815087/diff/5008/include/core/SkMaskFilter.h#...
include/core/SkMaskFilter.h:121: *  My convention, the caller will take the
center rol/col from the returned
On 2012/11/09 13:46:39, bsalomon wrote:
> My->By? Must the returned mask have an odd W/H?

Done.

https://codereview.appspot.com/6815087/diff/5008/src/core/SkMaskFilter.cpp
File src/core/SkMaskFilter.cpp (right):

https://codereview.appspot.com/6815087/diff/5008/src/core/SkMaskFilter.cpp#ne...
src/core/SkMaskFilter.cpp:16: #ifdef SK_IGNORE_FAST_BLURRECT
On 2012/11/09 13:46:39, bsalomon wrote:
> Why the second macro instead of making post config just define this to 0 if it
> hasn't been defined by user config or by the build system?

Done.

[caryclark1, 2012-11-09 14:26:53]

lgtm

[bsalomon, 2012-11-09 14:29:16]

On 2012/11/09 14:26:53, caryclark1 wrote:
> lgtm

LGTM