Fixed bug(s) in batching of rects

The clip flag was causing a problem since it was set but, since the clip was wide open, the clipstack was reporting that it did not result from a intersection of rects.

I believe the batching test was just incorrect.

Before these changes: 

tablebench  
8888: cmsecs =  13.88  
GPU: cmsecs =  69.11 gmsecs =  65.64  
NULLGPU: cmsecs =  52.42 gmsecs =   0.14

After these changes:

tablebench  
8888: cmsecs =  13.88   
GPU: cmsecs =  52.73 gmsecs =  49.56  
NULLGPU: cmsecs =  42.28 gmsecs =   0.12

Altering the bench to always use the same paint (i.e., maximal batching) yields:

tablebench  
8888: cmsecs =  15.29    
GPU: cmsecs =  18.72 gmsecs =  13.25  
NULLGPU: cmsecs =  16.54 gmsecs =   0.09

[bsalomon, 2012-10-04 15:23:16]

The difference between w/ the fix and maximal batching is still huge. Any idea
what the extra cost is? Is it malloc/free  cost in GrInOrderDrawBuffer?  If so,
maybe we should preallocate more space.

https://codereview.appspot.com/6608043/diff/1/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

https://codereview.appspot.com/6608043/diff/1/src/gpu/GrContext.cpp#newcode573
src/gpu/GrContext.cpp:573: if (!clipData->fClipStack->isWideOpen()) {
Should we make this disable the bit otherwise?

https://codereview.appspot.com/6608043/diff/1/src/gpu/GrInOrderDrawBuffer.cpp
File src/gpu/GrInOrderDrawBuffer.cpp (left):

https://codereview.appspot.com/6608043/diff/1/src/gpu/GrInOrderDrawBuffer.cpp...
src/gpu/GrInOrderDrawBuffer.cpp:193: kDraw_Cmd != fCmds.back() &&
yowzers

[robertphillips, 2012-10-04 16:27:56]

On 2012/10/04 15:23:16, bsalomon wrote:
> The difference between w/ the fix and maximal batching is still huge. Any idea
> what the extra cost is? Is it malloc/free  cost in GrInOrderDrawBuffer?  If
so,
> maybe we should preallocate more space.

On the Mac, 25% of the time in the non-maximally-batched case is spent in:
mach_msg_trap, sk_atomic_inc/dec, SkMatrix::reset and the driver. This time
basically disappears in the maximally batched case.

[robertphillips, 2012-10-04 17:50:49]

https://codereview.appspot.com/6608043/diff/1/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

https://codereview.appspot.com/6608043/diff/1/src/gpu/GrContext.cpp#newcode573
src/gpu/GrContext.cpp:573: if (!clipData->fClipStack->isWideOpen()) {
On 2012/10/04 15:23:16, bsalomon wrote:
> Should we make this disable the bit otherwise?

Done.

[bsalomon, 2012-10-04 17:55:09]

On 2012/10/04 17:50:49, robertphillips wrote:
> https://codereview.appspot.com/6608043/diff/1/src/gpu/GrContext.cpp
> File src/gpu/GrContext.cpp (right):
> 
> https://codereview.appspot.com/6608043/diff/1/src/gpu/GrContext.cpp#newcode573
> src/gpu/GrContext.cpp:573: if (!clipData->fClipStack->isWideOpen()) {
> On 2012/10/04 15:23:16, bsalomon wrote:
> > Should we make this disable the bit otherwise?
> 
> Done.

lgtm

[robertphillips, 2012-10-04 17:57:29]

committed as r5811