speed up A8

speed up A8 by creating a new entry-point in SkDraw that blits the path's coverage directly into an A8 target, regardless of the previous pixel values.

R=bsalomon@google.com, mtklein@google.com

Committed: https://code.google.com/p/skia/source/detail?r=12118

[mtklein, 2013-11-01 21:55:45]

would be nice to know by how much :)

https://codereview.chromium.org/56453002/diff/40001/gm/aaclip.cpp
File gm/aaclip.cpp (right):

https://codereview.chromium.org/56453002/diff/40001/gm/aaclip.cpp#newcode13
gm/aaclip.cpp:13: canvas->drawColor(SK_ColorBLUE);
Can you add this as a bench too?

https://codereview.chromium.org/56453002/diff/40001/gm/aaclip.cpp#newcode23
gm/aaclip.cpp:23: canvas->drawPath(path, paint);
{} ?

https://codereview.chromium.org/56453002/diff/40001/src/gpu/GrSWMaskHelper.cpp
File src/gpu/GrSWMaskHelper.cpp (right):

https://codereview.chromium.org/56453002/diff/40001/src/gpu/GrSWMaskHelper.cp...
src/gpu/GrSWMaskHelper.cpp:104: if (!fBM.allocPixels()) {
Nice spot for calloc.  Shall we follow up with adding (bool zeroed) to
allocPixels?

[bsalomon, 2013-11-04 14:50:13]

https://codereview.chromium.org/56453002/diff/40001/include/core/SkDraw.h
File include/core/SkDraw.h (right):

https://codereview.chromium.org/56453002/diff/40001/include/core/SkDraw.h#new...
include/core/SkDraw.h:81: void drawPathCoverage(const SkPath& src, const
SkPaint& paint) const {
Should it take a paint? What other than the AA flag is used?

[reed1, 2013-11-04 20:00:37]

https://codereview.chromium.org/56453002/diff/40001/include/core/SkDraw.h
File include/core/SkDraw.h (right):

https://codereview.chromium.org/56453002/diff/40001/include/core/SkDraw.h#new...
include/core/SkDraw.h:81: void drawPathCoverage(const SkPath& src, const
SkPaint& paint) const {
On 2013/11/04 14:50:13, bsalomon wrote:
> Should it take a paint? What other than the AA flag is used?

stroking parameters are support, including pathEffects. maskfilters are not
supported, but they could be logically (as they affect coverage).

For now, I can simplify it to just be AA, and add those others in later
(assuming GPU doesn't want to call this for stroking ...?)

https://codereview.chromium.org/56453002/diff/40001/src/gpu/GrSWMaskHelper.cpp
File src/gpu/GrSWMaskHelper.cpp (right):

https://codereview.chromium.org/56453002/diff/40001/src/gpu/GrSWMaskHelper.cp...
src/gpu/GrSWMaskHelper.cpp:104: if (!fBM.allocPixels()) {
On 2013/11/01 21:55:46, mtklein wrote:
> Nice spot for calloc.  Shall we follow up with adding (bool zeroed) to
> allocPixels?

We should be working on a calloc version of allocPixels, definitely...
orthogonal to this CL.

[reed1, 2013-11-04 20:00:47]

/skia/trunk> rbench --match draw_coverage
skia bench: mode=normal alpha=0xFF antialias=1 filter=0 dither=default rotate=0
scale=0 clip=0 scalar=float system=MAC
running bench [640 480]  draw_coverage_true    8888: cmsecs =    210.65
running bench [640 480] draw_coverage_false    8888: cmsecs =    319.58

[mtklein, 2013-11-04 20:09:22]

lgtm

[bsalomon, 2013-11-04 20:10:15]

On 2013/11/04 20:09:22, mtklein wrote:
> lgtm

lgtm

[reed1, 2013-11-04 20:10:44]

Committed patchset #5 manually as r12118 (presubmit successful).

[robertphillips, 2013-11-05 17:17:46]

lgtm + nit

https://codereview.chromium.org/56453002/diff/160001/src/core/SkCoreBlitters.h
File src/core/SkCoreBlitters.h (right):

https://codereview.chromium.org/56453002/diff/160001/src/core/SkCoreBlitters....
src/core/SkCoreBlitters.h:51: virtual const SkBitmap*
justAnOpaqueColor(uint32_t*) SK_OVERRIDE;
private:
INHERITED?