Fix clipping problem in GPU Gaussian blur

The FBOs used for the fast Gaussian blur were being clipped to the size of the output window, causing the blurred primitive to be truncated if the FBO was larger than the output window.  Clip them to the srcRect instead, and restore when done.

[bsalomon, 2011-07-13 21:25:30]

http://codereview.appspot.com/4695057/diff/1/src/gpu/SkGpuDevice.cpp
File src/gpu/SkGpuDevice.cpp (right):

http://codereview.appspot.com/4695057/diff/1/src/gpu/SkGpuDevice.cpp#newcode881
src/gpu/SkGpuDevice.cpp:881: kRenderTarget_GrTextureFlagBit |
kNoStencil_GrTextureFlagBit,
I think we have to query whether the path renderer requires the stencil buffer
for the path. (Might need to make GrContext::getPathRenderer public)

http://codereview.appspot.com/4695057/diff/1/src/gpu/SkGpuDevice.cpp#newcode903
src/gpu/SkGpuDevice.cpp:903: GrClip oldClip = context->getClip();
Should we use an AutoRestore? (or a future work comment)

[Stephen White, 2011-07-13 22:15:06]

http://codereview.appspot.com/4695057/diff/1/src/gpu/SkGpuDevice.cpp
File src/gpu/SkGpuDevice.cpp (right):

http://codereview.appspot.com/4695057/diff/1/src/gpu/SkGpuDevice.cpp#newcode881
src/gpu/SkGpuDevice.cpp:881: kRenderTarget_GrTextureFlagBit |
kNoStencil_GrTextureFlagBit,
On 2011/07/13 21:25:30, bsalomon wrote:
> I think we have to query whether the path renderer requires the stencil buffer
> for the path. (Might need to make GrContext::getPathRenderer public)

Argh, you're right.  Since this should ideally be only done for the FBO used for
paths (and not the other two), I'll save this work for a future patch.

http://codereview.appspot.com/4695057/diff/1/src/gpu/SkGpuDevice.cpp#newcode903
src/gpu/SkGpuDevice.cpp:903: GrClip oldClip = context->getClip();
On 2011/07/13 21:25:30, bsalomon wrote:
> Should we use an AutoRestore? (or a future work comment)

It turns out there is an AutoClipRestore already, but it works on GrDrawTarget,
not GrContext.  I've put in a comment, and when this code moves into GrContext,
I'll use the AutoClipRestore.

[bsalomon, 2011-07-14 12:51:35]

Hey sorry I didn't notice your reply yesterday. I'm OK with committing as is
with those two things noted.


On 2011/07/13 22:15:06, Stephen White wrote:
> http://codereview.appspot.com/4695057/diff/1/src/gpu/SkGpuDevice.cpp
> File src/gpu/SkGpuDevice.cpp (right):
> 
>
http://codereview.appspot.com/4695057/diff/1/src/gpu/SkGpuDevice.cpp#newcode881
> src/gpu/SkGpuDevice.cpp:881: kRenderTarget_GrTextureFlagBit |
> kNoStencil_GrTextureFlagBit,
> On 2011/07/13 21:25:30, bsalomon wrote:
> > I think we have to query whether the path renderer requires the stencil
buffer
> > for the path. (Might need to make GrContext::getPathRenderer public)
> 
> Argh, you're right.  Since this should ideally be only done for the FBO used
for
> paths (and not the other two), I'll save this work for a future patch.
> 
>
http://codereview.appspot.com/4695057/diff/1/src/gpu/SkGpuDevice.cpp#newcode903
> src/gpu/SkGpuDevice.cpp:903: GrClip oldClip = context->getClip();
> On 2011/07/13 21:25:30, bsalomon wrote:
> > Should we use an AutoRestore? (or a future work comment)
> 
> It turns out there is an AutoClipRestore already, but it works on
GrDrawTarget,
> not GrContext.  I've put in a comment, and when this code moves into
GrContext,
> I'll use the AutoClipRestore.

[reed1, 2011-07-14 13:03:39]

SWDIK -- sure, what do I know

[Stephen White, 2011-07-14 15:08:51]

revert stencil change, add comment

[bsalomon, 2011-07-14 15:20:19]

On 2011/07/14 15:08:51, Stephen White wrote:
> revert stencil change, add comment

LGTM

[Stephen White, 2011-07-14 15:41:52]

Landed as r1859; closing.