store n-bits of luminance in fontcache rec

[reed1, 2011-11-22 19:07:00]

question: how can I better handle the logic in SkScalerContext.h
getLuminanceByte() when I don't know the value of kLuminance_Bits?

[TomH, 2011-11-22 19:14:12]

On 2011/11/22 19:07:00, reed1 wrote:
> question: how can I better handle the logic in SkScalerContext.h
> getLuminanceByte() when I don't know the value of kLuminance_Bits?

Simple option: why not store it as an 8b quantity? There's an extra shift on the
set (8-kLuminance_Bits?), but no conditional on the get?
Or is it important that you fill in the lower order bits with a copy of the
higher-order bits?

[caryclark1, 2011-11-22 20:07:47]

lgtm

http://codereview.appspot.com/5431045/diff/4001/include/core/SkScalerContext.h
File include/core/SkScalerContext.h (right):

http://codereview.appspot.com/5431045/diff/4001/include/core/SkScalerContext....
include/core/SkScalerContext.h:180: kLuminance_Bits           = 3,  // ensure
Flags doesn't exceed 16bits
Does the 'ensure...' comment mean there's a compile-time assert checking that
 kLuminance_Shift + kLuminance_Bits <= sizeof(fFlags) * 8
?

http://codereview.appspot.com/5431045/diff/4001/src/core/SkPaint.cpp
File src/core/SkPaint.cpp (right):

http://codereview.appspot.com/5431045/diff/4001/src/core/SkPaint.cpp#newcode1293
src/core/SkPaint.cpp:1293: int luminance = r * 2 + g * 5 + b;
Wikipedia's linear luminance from RGB is:
Y = 0.2126 R + 0.7152 G + 0.0722 B
Your x/8 approximates:
    0.25 R + 0.625 G + 0.125 B
Would x/16 be worth it? (4/11/1)>>1
    0.25 R + 0.6875 G + 0.0625 B

Worth a comment?