Add luminance mask transfer modes.

Add luminance mask transfer modes.

This adds kSrcInLum_Mode and kDstInLum_Mode, to support CSS and SVG luminance masks (http://www.w3.org/TR/css-masking/#MaskValues , http://www.w3.org/TR/SVG/masking.html#Masking ).

The transfer coefficient is computed according to http://www.w3.org/TR/2011/REC-SVG11-20110816/filters.html#feColorMatrixElement "luminance-to-alpha":

  luma = 0.2125 * r + 0.7154 * g + 0.0721 * b


R=reed@google.com,robertphillips@google.com,bsalomon@google.com

Committed: http://code.google.com/p/skia/source/detail?r=10887

[f(malita), 2013-08-20 16:24:06]

Some bench numbers:

running bench [640 480]            Xfermode_DstInLum  8888: cmsecs =   2.02  
565: cmsecs =   4.00   GPU: cmsecs =   2.08 gmsecs =   1.13  NULLGPU: cmsecs =  
0.17
running bench [640 480]            Xfermode_SrcInLum  8888: cmsecs =   1.94  
565: cmsecs =   4.19   GPU: cmsecs =   2.10 gmsecs =   1.14  NULLGPU: cmsecs =  
0.18
running bench [640 480]          Xfermode_Luminosity  8888: cmsecs =  10.17  
565: cmsecs =  12.04   GPU: cmsecs =   2.15 gmsecs =   1.20  NULLGPU: cmsecs =  
0.17
running bench [640 480]               Xfermode_Color  8888: cmsecs =  25.86  
565: cmsecs =  27.64   GPU: cmsecs =   2.19 gmsecs =   1.24  NULLGPU: cmsecs =  
0.17
running bench [640 480]          Xfermode_Saturation  8888: cmsecs =   9.20  
565: cmsecs =  11.09   GPU: cmsecs =   2.30 gmsecs =   1.31  NULLGPU: cmsecs =  
0.18
running bench [640 480]                 Xfermode_Hue  8888: cmsecs =  14.31  
565: cmsecs =  14.92   GPU: cmsecs =   2.19 gmsecs =   1.34  NULLGPU: cmsecs =  
0.16
running bench [640 480]            Xfermode_Multiply  8888: cmsecs =   4.57  
565: cmsecs =   5.89   GPU: cmsecs =   2.01 gmsecs =   1.12  NULLGPU: cmsecs =  
0.16
running bench [640 480]           Xfermode_Exclusion  8888: cmsecs =   4.67  
565: cmsecs =   6.35   GPU: cmsecs =   1.95 gmsecs =   1.11  NULLGPU: cmsecs =  
0.15
running bench [640 480]          Xfermode_Difference  8888: cmsecs =   5.72  
565: cmsecs =   7.32   GPU: cmsecs =   1.96 gmsecs =   1.11  NULLGPU: cmsecs =  
0.15
running bench [640 480]           Xfermode_SoftLight  8888: cmsecs =  37.71  
565: cmsecs =  39.47   GPU: cmsecs =   2.20 gmsecs =   1.37  NULLGPU: cmsecs =  
0.15
running bench [640 480]           Xfermode_HardLight  8888: cmsecs =   5.39  
565: cmsecs =   6.87   GPU: cmsecs =   1.97 gmsecs =   1.14  NULLGPU: cmsecs =  
0.15
running bench [640 480]           Xfermode_ColorBurn  8888: cmsecs =   4.13  
565: cmsecs =   5.69   GPU: cmsecs =   2.07 gmsecs =   1.26  NULLGPU: cmsecs =  
0.15
running bench [640 480]          Xfermode_ColorDodge  8888: cmsecs =   7.64  
565: cmsecs =   9.31   GPU: cmsecs =   2.21 gmsecs =   1.31  NULLGPU: cmsecs =  
0.15
running bench [640 480]             Xfermode_Lighten  8888: cmsecs =   4.03  
565: cmsecs =   5.72   GPU: cmsecs =   1.95 gmsecs =   1.12  NULLGPU: cmsecs =  
0.15
running bench [640 480]              Xfermode_Darken  8888: cmsecs =   4.03  
565: cmsecs =   5.76   GPU: cmsecs =   1.99 gmsecs =   1.14  NULLGPU: cmsecs =  
0.15
running bench [640 480]             Xfermode_Overlay  8888: cmsecs =   4.97  
565: cmsecs =   6.53   GPU: cmsecs =   2.00 gmsecs =   1.17  NULLGPU: cmsecs =  
0.15
running bench [640 480]              Xfermode_Screen  8888: cmsecs =   3.54  
565: cmsecs =   5.23   GPU: cmsecs =   0.27 gmsecs =   0.22  NULLGPU: cmsecs =  
0.04
running bench [640 480]            Xfermode_Modulate  8888: cmsecs =   2.77  
565: cmsecs =   4.42   GPU: cmsecs =   0.26 gmsecs =   0.22  NULLGPU: cmsecs =  
0.04
running bench [640 480]                Xfermode_Plus  8888: cmsecs =   2.26  
565: cmsecs =   4.04   GPU: cmsecs =   0.27 gmsecs =   0.22  NULLGPU: cmsecs =  
0.04
running bench [640 480]                 Xfermode_Xor  8888: cmsecs =   4.31  
565: cmsecs =   5.91   GPU: cmsecs =   0.28 gmsecs =   0.23  NULLGPU: cmsecs =  
0.04
running bench [640 480]             Xfermode_DstATop  8888: cmsecs =   3.63  
565: cmsecs =   5.27   GPU: cmsecs =   0.27 gmsecs =   0.22  NULLGPU: cmsecs =  
0.04
running bench [640 480]             Xfermode_SrcATop  8888: cmsecs =   3.58  
565: cmsecs =   5.29   GPU: cmsecs =   0.28 gmsecs =   0.23  NULLGPU: cmsecs =  
0.05
running bench [640 480]              Xfermode_DstOut  8888: cmsecs =   0.68  
565: cmsecs =   2.85   GPU: cmsecs =   0.26 gmsecs =   0.22  NULLGPU: cmsecs =  
0.04
running bench [640 480]              Xfermode_SrcOut  8888: cmsecs =   1.14  
565: cmsecs =   2.96   GPU: cmsecs =   0.27 gmsecs =   0.22  NULLGPU: cmsecs =  
0.04
running bench [640 480]               Xfermode_DstIn  8888: cmsecs =   0.56  
565: cmsecs =   2.91   GPU: cmsecs =   0.26 gmsecs =   0.22  NULLGPU: cmsecs =  
0.04
running bench [640 480]               Xfermode_SrcIn  8888: cmsecs =   1.13  
565: cmsecs =   2.85   GPU: cmsecs =   0.27 gmsecs =   0.22  NULLGPU: cmsecs =  
0.04
running bench [640 480]             Xfermode_DstOver  8888: cmsecs =   1.22  
565: cmsecs =   0.01   GPU: cmsecs =   0.27 gmsecs =   0.22  NULLGPU: cmsecs =  
0.04
running bench [640 480]             Xfermode_SrcOver  8888: cmsecs =   0.29  
565: cmsecs =   0.50   GPU: cmsecs =   0.38 gmsecs =   0.23  NULLGPU: cmsecs =  
0.08
running bench [640 480]                 Xfermode_Dst  8888: cmsecs =   0.01  
565: cmsecs =   0.01   GPU: cmsecs =   0.03 gmsecs =   0.02  NULLGPU: cmsecs =  
0.01
running bench [640 480]                 Xfermode_Src  8888: cmsecs =   0.43  
565: cmsecs =   2.04   GPU: cmsecs =   0.26 gmsecs =   0.21  NULLGPU: cmsecs =  
0.04
running bench [640 480]               Xfermode_Clear  8888: cmsecs =   0.22  
565: cmsecs =   2.13   GPU: cmsecs =   0.32 gmsecs =   0.22  NULLGPU: cmsecs =  
0.04

[reed1, 2013-08-20 17:21:35]

https://codereview.chromium.org/22918012/diff/1/include/core/SkXfermode.h
File include/core/SkXfermode.h (right):

https://codereview.chromium.org/22918012/diff/1/include/core/SkXfermode.h#new...
include/core/SkXfermode.h:134: kSrcInLum_Mode,
Can we initially introduce this xfermode w/o adding it to public SkXfermode.h's
enum? See SkArithmeticMode.h as an example.

[vandebo (ex-Chrome), 2013-08-20 17:24:47]

https://codereview.chromium.org/22918012/diff/1/src/pdf/SkPDFGraphicState.cpp
File src/pdf/SkPDFGraphicState.cpp (right):

https://codereview.chromium.org/22918012/diff/1/src/pdf/SkPDFGraphicState.cpp...
src/pdf/SkPDFGraphicState.cpp:51: case SkXfermode::kSrcInLum_Mode:
These can probably be supported in the current framework. If not, we should
seriously consider if they should be added to the public framework.

[f(malita), 2013-08-22 18:30:04]

Re-implemented as a custom transfer class - PTAL.

[reed1, 2013-08-22 18:35:51]

https://codereview.chromium.org/22918012/diff/12001/include/effects/SkLumaXfe...
File include/effects/SkLumaXfermode.h (right):

https://codereview.chromium.org/22918012/diff/12001/include/effects/SkLumaXfe...
include/effects/SkLumaXfermode.h:19: *  SrcIn or DstIn xfer:
Is this list (srcin, dstin) complete, or just e.g. ?

https://codereview.chromium.org/22918012/diff/12001/include/effects/SkLumaXfe...
include/effects/SkLumaXfermode.h:21: *    luma(C) = (0.2125 * C.r + 0.7154 * C.g
+ 0.0721 * C.b) * C.a
Can you document what happens to the resulting r,g,b?

https://codereview.chromium.org/22918012/diff/12001/include/effects/SkLumaXfe...
include/effects/SkLumaXfermode.h:27: static SkXfermode*
Create(SkXfermode::Mode);
Should you support another Create that takes SkXfermode* instead of an enum?

https://codereview.chromium.org/22918012/diff/12001/src/effects/SkLumaXfermod...
File src/effects/SkLumaXfermode.cpp (right):

https://codereview.chromium.org/22918012/diff/12001/src/effects/SkLumaXfermod...
src/effects/SkLumaXfermode.cpp:21: unsigned luma =
SkComputeLuminance(SkGetPackedR32(b),
Is the spec cool with computing luminance on the premul'd components, instead of
unpremultiplying them first?

[f(malita), 2013-08-22 19:25:52]

https://codereview.chromium.org/22918012/diff/12001/include/effects/SkLumaXfe...
File include/effects/SkLumaXfermode.h (right):

https://codereview.chromium.org/22918012/diff/12001/include/effects/SkLumaXfe...
include/effects/SkLumaXfermode.h:19: *  SrcIn or DstIn xfer:
On 2013/08/22 18:35:51, reed1 wrote:
> Is this list (srcin, dstin) complete, or just e.g. ?

It is complete with regard to SVG/CSS masking support. I did consider the more
general approach of wrapping arbitrary transfer modes, but decided to avoid it
initially for two reasons:

1) unclear semantics - for masking (SrcIn, DstOut) the luma operand is clearly
defined (the one with a zero color transfer coefficient); for arbitrary modes,
we would need a way to specify which of the inputs needs to be luma-transformed.

2) efficiency - conditionally applying luma to one/both operands and invoking
arbitrary mode procs in the transfer loop is possibly significantly slower than
the current inlined implementation. I don't have hard numbers though, so this
concern may be unwarranted.

Both of these are solvable/tolerable, but what tipped the balance for me was the
lack of a good use case for anything other than srcin/dst out - it's unclear
whether any of the other resulting transfer functions make sense/have any
practical use. I could certainly be convinced otherwise :)

https://codereview.chromium.org/22918012/diff/12001/include/effects/SkLumaXfe...
include/effects/SkLumaXfermode.h:21: *    luma(C) = (0.2125 * C.r + 0.7154 * C.g
+ 0.0721 * C.b) * C.a
On 2013/08/22 18:35:51, reed1 wrote:
> Can you document what happens to the resulting r,g,b?

Will do.

https://codereview.chromium.org/22918012/diff/12001/include/effects/SkLumaXfe...
include/effects/SkLumaXfermode.h:27: static SkXfermode*
Create(SkXfermode::Mode);
On 2013/08/22 18:35:51, reed1 wrote:
> Should you support another Create that takes SkXfermode* instead of an enum?

This ties into the decision to support arbitrary transfer modes. We can
certainly do that if needed, but we'll need a more flexible API and will
possibly take a perf hit.

https://codereview.chromium.org/22918012/diff/12001/src/effects/SkLumaXfermod...
File src/effects/SkLumaXfermode.cpp (right):

https://codereview.chromium.org/22918012/diff/12001/src/effects/SkLumaXfermod...
src/effects/SkLumaXfermode.cpp:21: unsigned luma =
SkComputeLuminance(SkGetPackedR32(b),
On 2013/08/22 18:35:51, reed1 wrote:
> Is the spec cool with computing luminance on the premul'd components, instead
of
> unpremultiplying them first?

Actually they explicitly mention non-premultiplied colors :(

But, algebraically, the results should be equivalent as un-premultiplication can
be factored and inverted by the final alpha channel multiplication. Are you
worried about numeric side effects?

[reed1, 2013-08-22 19:38:43]

All good explanations. As long as the dox are very clear under what conditions
the factory will return NULL, lgtm

[bsalomon, 2013-08-22 19:47:46]

I think it'd be good to get rid of the background texture stuff in the GrEffect
since we aren't testing it and don't anticipate using this xfermode as an image
filter.

[f(malita), 2013-08-22 20:48:21]

On 2013/08/22 19:47:46, bsalomon wrote:
> I think it'd be good to get rid of the background texture stuff in the
GrEffect
> since we aren't testing it and don't anticipate using this xfermode as an
image
> filter.

Done.

[bsalomon, 2013-08-22 20:52:20]

lgtm

[commit-bot: I haz the power, 2013-08-23 13:12:37]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/fmalita@chromium.org/22918012/29001

[commit-bot: I haz the power, 2013-08-23 13:13:25]

Change committed as 10887