Add GrDrawState::reset() function

[bsalomon, 2012-01-14 19:40:59]

This is mostly done to put reset in the "right" place as a member. But it also
very slightly (but consistently) improved the perf of the chrome_scrollGmail
NULLGPU bench. I played with making fields in GrDrawState bitfields and not. Not
was a definite win over bitfields.

[TomH, 2012-01-17 15:17:45]

LGTM

http://codereview.appspot.com/5543059/diff/2003/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

http://codereview.appspot.com/5543059/diff/2003/src/gpu/GrContext.cpp#newcode776
src/gpu/GrContext.cpp:776: *drawState->viewMatrix() = vm;
Do we want to note that calling this->setPaint() with a default paint is
equivalent to calling this->reset() & setting the viewMatrix? Or is it?

Is it significant that sometimes we do both and sometimes we do only one? Or I
note that we frequently call reset() and setRenderTarget() as a pair; should
those be combined? Here we set the viewMatrix to something different, which
seems like an unusual use case?

http://codereview.appspot.com/5543059/diff/2003/src/gpu/GrDrawState.h
File src/gpu/GrDrawState.h (right):

http://codereview.appspot.com/5543059/diff/2003/src/gpu/GrDrawState.h#newcode56
src/gpu/GrDrawState.h:56: void reset() {
Could we not inline this?

http://codereview.appspot.com/5543059/diff/2003/src/gpu/GrDrawState.h#newcode748
src/gpu/GrDrawState.h:748: // @{ Initialized to values other than zero
Awesome commentary.

http://codereview.appspot.com/5543059/diff/2003/src/gpu/GrDrawState.h#newcode773
src/gpu/GrDrawState.h:773: (unsigned char*)&fBlendConstant);
Yep, this was necessary, and the fragile part.

[bsalomon, 2012-01-17 16:02:05]

Closed with r3043.

http://codereview.appspot.com/5543059/diff/2003/src/gpu/GrContext.cpp
File src/gpu/GrContext.cpp (right):

http://codereview.appspot.com/5543059/diff/2003/src/gpu/GrContext.cpp#newcode776
src/gpu/GrContext.cpp:776: *drawState->viewMatrix() = vm;
On 2012/01/17 15:17:45, TomH wrote:
> Do we want to note that calling this->setPaint() with a default paint is
> equivalent to calling this->reset() & setting the viewMatrix? Or is it?

They aren't really equivalent. The GrPaint is a subset of the GrDrawState. There
are several differences (that happen not to matter here). This code was written
before there was any way to reset the GrDrawState (or even a GrDrawState at all)
and so used a default GrPaint to get to a known baseline state for most of the
draw state.

> 
> Is it significant that sometimes we do both and sometimes we do only one? Or I
> note that we frequently call reset() and setRenderTarget() as a pair; should
> those be combined? Here we set the viewMatrix to something different, which
> seems like an unusual use case?

We might want to have a reset that takes a rt and vm since these frequently set
right after reset.  I experimented with a flag on reset to preserve the rt and
vm but the conditional added a measurable slow down.

http://codereview.appspot.com/5543059/diff/2003/src/gpu/GrDrawState.h
File src/gpu/GrDrawState.h (right):

http://codereview.appspot.com/5543059/diff/2003/src/gpu/GrDrawState.h#newcode56
src/gpu/GrDrawState.h:56: void reset() {
On 2012/01/17 15:17:45, TomH wrote:
> Could we not inline this?

Isn't it inline? (this is a header file)