Cache the current program pointer

Cache the current program pointer

Hash table lookups are somewhat expensive, so cache a pointer to the current program for on a context. This gains about 3 fps (from 58) on a Native Client demo.


BUG=
TEST=webgl conformance tests

Committed: http://code.google.com/p/angleproject/source/detail?r=786

[nicolas, 2011-10-06 13:22:57]

Looks ok to me, aside from a potential missed optimization. We actually had a
patch for the exact same optimization a while back, but decided against applying
it due to low gain and slightly more complex code. Now that the rest of ANGLE is
more optimized I guess it's the right time to deal with this hotspot.

http://codereview.appspot.com/5206042/diff/1/src/libGLESv2/Context.cpp
File src/libGLESv2/Context.cpp (right):

http://codereview.appspot.com/5206042/diff/1/src/libGLESv2/Context.cpp#newcod...
src/libGLESv2/Context.cpp:1040: mCachedCurrentProgram = NULL;
This probably only has to be reset to NULL when priorProgram != program ?

[John Bauman, 2011-10-06 13:28:24]

On 2011/10/06 13:22:57, nicolas wrote:
> Looks ok to me, aside from a potential missed optimization. We actually had a
> patch for the exact same optimization a while back, but decided against
applying
> it due to low gain and slightly more complex code. Now that the rest of ANGLE
is
> more optimized I guess it's the right time to deal with this hotspot.
> 
Yeah, unfortunately most of the remaining hotspots aren't particularly hot. I
need about 5 patches to get from 60fps -> 70fps on this benchmark, and ANGLE is
still taking a bunch of time. Luckily, at 70fps other things start to become
bottlenecks, so this won't have to be optimized more.

> http://codereview.appspot.com/5206042/diff/1/src/libGLESv2/Context.cpp
> File src/libGLESv2/Context.cpp (right):
> 
>
http://codereview.appspot.com/5206042/diff/1/src/libGLESv2/Context.cpp#newcod...
> src/libGLESv2/Context.cpp:1040: mCachedCurrentProgram = NULL;
> This probably only has to be reset to NULL when priorProgram != program ?

[John Bauman, 2011-10-06 13:28:33]

http://codereview.appspot.com/5206042/diff/1/src/libGLESv2/Context.cpp
File src/libGLESv2/Context.cpp (right):

http://codereview.appspot.com/5206042/diff/1/src/libGLESv2/Context.cpp#newcod...
src/libGLESv2/Context.cpp:1040: mCachedCurrentProgram = NULL;
On 2011/10/06 13:22:58, nicolas wrote:
> This probably only has to be reset to NULL when priorProgram != program ?

Sure, fixed.

[dgkoch, 2011-10-07 04:16:31]

LGTM