Make GrContext track the current matrix, render target, and clip directly rather than using GrDrawS…

Make GrContext track the current matrix, render target, and clip directly rather than using GrDrawState.

Committed: http://code.google.com/p/skia/source/detail?r=9297

[bsalomon, 2013-05-24 20:55:19]

Currently GrContext keeps track of its matrix and RT in its drawstate object and
its clip in its gpu object. The drawstate object gets whacked by various draw
calls and renderers so we always have to be careful to restore the rt and matrix
so that from an GrContext's client's view the RT and matrix are persistent.

Moreover, at the beginning of each GrContext draw it sets most of the rest of
the GrDrawState state from the passed in GrPaint. But not all of it (e.g. the
state flags, blend const color, ...). So again the drawing code paths that begin
in GrContext have to be careful to restore these extra bits of state to the
values GrContext expects.

After this change GrContext tracks its clip, matrix, and rt separately and fully
configures the GrDrawState for each draw. I think this will allow us to remove a
bunch of our auto-restore stuff in the future.

I ran bench before and after on the GPU config. I reran the top 10 "regressions"
with a high repeat count to verify that they were all noise.

No GMs change on my windows machine.

[robertphillips, 2013-05-28 13:33:30]

lgtm + a question.

https://codereview.chromium.org/15821008/diff/2001/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

https://codereview.chromium.org/15821008/diff/2001/src/gpu/GrContext.cpp#newc...
src/gpu/GrContext.cpp:1569: }
What happened to the WideOpen check?

[bsalomon, 2013-05-28 14:15:51]

https://codereview.chromium.org/15821008/diff/2001/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

https://codereview.chromium.org/15821008/diff/2001/src/gpu/GrContext.cpp#newc...
src/gpu/GrContext.cpp:1569: }
On 2013/05/28 13:33:30, robertphillips wrote:
> What happened to the WideOpen check?

Nice catch, added it back.

[jvanverth1, 2013-05-28 15:28:59]

lgtm + some comments

https://codereview.chromium.org/15821008/diff/6001/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

https://codereview.chromium.org/15821008/diff/6001/src/gpu/GrContext.cpp#newc...
src/gpu/GrContext.cpp:103: fOvalRenderer = NULL;
Any init for RenderTarget needed?

https://codereview.chromium.org/15821008/diff/6001/src/gpu/GrContext.cpp#newc...
src/gpu/GrContext.cpp:1567: *fDrawState->viewMatrix() = fViewMatrix;
While legal this seems a bit squirrely to me. Add a setViewMatrix method?

[bsalomon, 2013-05-28 15:33:29]

https://codereview.chromium.org/15821008/diff/6001/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

https://codereview.chromium.org/15821008/diff/6001/src/gpu/GrContext.cpp#newc...
src/gpu/GrContext.cpp:103: fOvalRenderer = NULL;
On 2013/05/28 15:28:59, JimVV wrote:
> Any init for RenderTarget needed?

SkAutoTUnref defaults to NULL.

https://codereview.chromium.org/15821008/diff/6001/src/gpu/GrContext.cpp#newc...
src/gpu/GrContext.cpp:1567: *fDrawState->viewMatrix() = fViewMatrix;
On 2013/05/28 15:28:59, JimVV wrote:
> While legal this seems a bit squirrely to me. Add a setViewMatrix method?

My next CL removes this viewMatrix() function and will pass the VM to reset. All
matrix changes on GrDrawState will go through functions/helper classes that
update the coord change matrix on GrEffectStages.

[commit-bot: I haz the power, 2013-05-28 17:08:52]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/bsalomon@google.com/15821008/6001

[commit-bot: I haz the power, 2013-05-28 17:25:52]

Change committed as 9297