Blur refactoring

This CL pushes the radius -> sigma fudging up to the boundary of Skia's API. All the action is in SkBlurMask.cpp, SkBlurMaskFilter.cpp and SkEmbossMaskFilter.cpp. The next steps will be:
  1) push the fudging across the API and into our tests
  2) move the "unused" param from new to old methods
  3) push the fudging across the API and into Blink 
  4) remove old methods (which now have the "unused" param)
With 3 having to happen in the same roll as 2.

[robertphillips, 2013-08-05 18:11:27]

https://codereview.chromium.org/21835004/diff/6001/src/effects/SkBlurMask.cpp
File src/effects/SkBlurMask.cpp (right):

https://codereview.chromium.org/21835004/diff/6001/src/effects/SkBlurMask.cpp...
src/effects/SkBlurMask.cpp:502: // Force high quality off for small radii
(performance)
This should really be 2.232 to be completely consistent with the old code but
using 2 doesn't introduce a lot of differences.

https://codereview.chromium.org/21835004/diff/6001/src/effects/SkBlurMask.cpp...
src/effects/SkBlurMask.cpp:506: 
If we were willing to live with this code here:

   SkScalar passRadius = (kHigh_Quality == quality) ? sigma - 0.5f :
                     (sigma - 0.5f) / SkBlurMask::kBLUR_SIGMA_SCALE;

The only GM image differences we would have would be in the GPU path. The
downside of this is that it lets kBLUR_SIGMA_SCALE out of its box.

https://codereview.chromium.org/21835004/diff/6001/src/effects/SkBlurMask.cpp...
src/effects/SkBlurMask.cpp:873: 
This is a bit of a radical switch in how sigma is computed. The old method was:
  rad*K/2 
while the new is:
  rad*K + .5

[reed1, 2013-08-05 20:23:11]

lgtm

[bsalomon, 2013-08-05 20:45:37]

https://codereview.chromium.org/21835004/diff/6001/include/effects/SkBlurDraw...
File include/effects/SkBlurDrawLooper.h (right):

https://codereview.chromium.org/21835004/diff/6001/include/effects/SkBlurDraw...
include/effects/SkBlurDrawLooper.h:38: SkBlurDrawLooper(SkScalar sigma, SkScalar
dx, SkScalar dy, SkColor color,
What's the plan for removing the unused bool? Could we maybe change the order of
params (rather than adding one) so that when the deprecated cons is removed we
don't need another change to clients (to remove faux bool params)?

[robertphillips, 2013-08-06 17:13:46]

PTAL. This patch changes the method signatures (rather than add the "unused"
parameter) to differentiate the radius & sigma -based entry points.

[bsalomon, 2013-08-06 17:29:38]

On 2013/08/06 17:13:46, robertphillips wrote:
> PTAL. This patch changes the method signatures (rather than add the "unused"
> parameter) to differentiate the radius & sigma -based entry points.

lgtm

[reed1, 2013-08-06 17:30:30]

lgtm

https://codereview.chromium.org/21835004/diff/16001/include/effects/SkBlurDra...
File include/effects/SkBlurDrawLooper.h (right):

https://codereview.chromium.org/21835004/diff/16001/include/effects/SkBlurDra...
include/effects/SkBlurDrawLooper.h:39: uint32_t flags = kNone_BlurFlag);
consider not having the flags have a def value... defaults are funny, and it
will also help you catch callers for the new api. Just an idea

[robertphillips, 2013-08-27 16:09:46]

This change will require rebaselines/supressions of the following 67 layout
tests (at least on Linux):

+compositing/shadows/shadow-drawing.html
+fast/borders/border-radius-split-inline.html
+fast/borders/border-radius-with-box-shadow-01.html
+fast/borders/border-radius-with-box-shadow.html
+fast/box-shadow/basic-shadows.html
+fast/box-shadow/box-shadow-radius.html
+fast/box-shadow/box-shadow-transformed.html
+fast/box-shadow/inset-box-shadow-radius.html
+fast/box-shadow/inset-box-shadows.html
+fast/box-shadow/inset.html
+fast/box-shadow/scaled-box-shadow.html
+fast/box-shadow/shadow-buffer-partial.html
+fast/box-shadow/shadow-tiling-artifact.html
+fast/box-shadow/single-pixel-shadow.html
+fast/canvas/canvas-incremental-repaint.html
+fast/css/box-shadow-and-border-radius.html
+fast/css/color-correction-on-text-shadow.html
+fast/css/shadow-multiple.html
+fast/dom/HTMLMeterElement/meter-styles.html
+fast/forms/calendar-picker/calendar-picker-appearance-ar.html
+fast/forms/calendar-picker/calendar-picker-appearance-required-ar.html
+fast/forms/calendar-picker/calendar-picker-appearance-required.html
+fast/forms/calendar-picker/calendar-picker-appearance-ru.html
+fast/forms/calendar-picker/calendar-picker-appearance-step.html
+fast/forms/calendar-picker/calendar-picker-appearance.html
+fast/forms/calendar-picker/month-picker-appearance-step.html
+fast/forms/calendar-picker/month-picker-appearance.html
+fast/forms/calendar-picker/week-picker-appearance-step.html
+fast/forms/calendar-picker/week-picker-appearance.html
+fast/forms/checkbox/checkbox-appearance-basic.html
+fast/forms/color/color-suggestion-picker-appearance.html
+fast/forms/color/color-suggestion-picker-one-row-appearance.html
+fast/forms/color/color-suggestion-picker-two-row-appearance.html
+fast/forms/color/color-suggestion-picker-with-scrollbar-appearance.html
+fast/forms/radio/radio-appearance-basic.html
+fast/forms/range/range-appearance-basic.html
+fast/forms/search/search-appearance-basic.html
+fast/forms/select/listbox-appearance-basic.html
+fast/forms/select/menulist-appearance-basic.html
+fast/forms/submit/submit-appearance-basic.html
+fast/forms/text/text-appearance-basic.html
+fast/forms/textarea/textarea-appearance-basic.html
+fast/multicol/shadow-breaking.html
+fast/text/shadow-translucent-fill.html
+fast/text/stroking-decorations.html
+fast/text/stroking.html
+fast/transforms/shadows.html
+fast/writing-mode/box-shadow-horizontal-bt.html
+fast/writing-mode/box-shadow-vertical-lr.html
+fast/writing-mode/box-shadow-vertical-rl.html
+fast/writing-mode/english-lr-text.html
+ietestcenter/css3/text/textshadow-001.htm
+ietestcenter/css3/text/textshadow-002.htm
+ietestcenter/css3/text/textshadow-003.htm
+ietestcenter/css3/text/textshadow-004.htm
+ietestcenter/css3/text/textshadow-010.htm
+inspector/debugger/live-edit.html
+svg/css/text-gradient-shadow.svg
+svg/css/text-shadow-multiple.xhtml
+svg/custom/repaint-shadow.svg
+transitions/svg-text-shadow-transition.html
+virtual/gpu/fast/canvas/canvas-draw-canvas-on-canvas-shadow.html
+virtual/gpu/fast/canvas/canvas-incremental-repaint.html
+virtual/gpu/fast/canvas/canvas-render-layer.html
+virtual/gpu/fast/canvas/canvas-scale-strokePath-shadow.html
+virtual/gpu/fast/canvas/canvas-transforms-fillRect-shadow.html
+virtual/softwarecompositing/shadows/shadow-drawing.html

[robertphillips, 2013-08-27 16:15:01]

committed as r10936

[robertphillips, 2013-08-28 12:38:26]

This did require rebaselining theverge baselines (because it got slower). Most
likely the new cutoff tolerances choose a more expensive path for some of the
blurs.