add SkDeviceLooper to handle larger-than-fixedpoint

add SkDeviceLooper to handle larger-than-fixedpoint

BUG=

Committed: https://code.google.com/p/skia/source/detail?r=10943

[reed1, 2013-08-23 21:54:34]

First cut at a fix. If this idea works, might apply this to paths, which today
(a) turn-off AA above 8K, and above 16 or 32K, just don't draw.

[reed1, 2013-08-26 16:27:28]

ptal

[reed1, 2013-08-27 17:53:58]

Committed patchset #2 manually as r10943.

[tomhudson, 2013-08-28 10:53:05]

lgtm

Sorry, Monday was a Bank Holiday in the UK and I'm still catching up with
things. None of my comments were more than nits, anyway.

https://codereview.chromium.org/23392006/diff/6001/src/core/SkDeviceLooper.h
File src/core/SkDeviceLooper.h (right):

https://codereview.chromium.org/23392006/diff/6001/src/core/SkDeviceLooper.h#...
src/core/SkDeviceLooper.h:15: class SkDeviceLooper {
Since this has state that changes the class semantics, can we get a little
documentation here about the lifecycle?

https://codereview.chromium.org/23392006/diff/6001/src/core/SkDraw.cpp
File src/core/SkDraw.cpp (right):

https://codereview.chromium.org/23392006/diff/6001/src/core/SkDraw.cpp#newcod...
src/core/SkDraw.cpp:887: while (looper.next()) {
This injects another degree of indirection & bunch of allocations into every
drawRect() call. For large cases it's only incidental, but do we have any perf
tests that would capture impact on a page with a large number of small rectangle
draws?

[reed1, 2013-08-28 12:35:07]

https://codereview.chromium.org/23392006/diff/6001/src/core/SkDeviceLooper.h
File src/core/SkDeviceLooper.h (right):

https://codereview.chromium.org/23392006/diff/6001/src/core/SkDeviceLooper.h#...
src/core/SkDeviceLooper.h:15: class SkDeviceLooper {
On 2013/08/28 10:53:05, tomhudson wrote:
> Since this has state that changes the class semantics, can we get a little
> documentation here about the lifecycle?

Yes indeed, I will add dox.

https://codereview.chromium.org/23392006/diff/6001/src/core/SkDraw.cpp
File src/core/SkDraw.cpp (right):

https://codereview.chromium.org/23392006/diff/6001/src/core/SkDraw.cpp#newcod...
src/core/SkDraw.cpp:887: while (looper.next()) {
On 2013/08/28 10:53:05, tomhudson wrote:
> This injects another degree of indirection & bunch of allocations into every
> drawRect() call. For large cases it's only incidental, but do we have any perf
> tests that would capture impact on a page with a large number of small
rectangle
> draws?

There should be zero allocations in the (common) case where the rect fits inside
+- 16K. Is that not the case?