Add benchmarks for blits that test source alpha

Add benchmarks for blits that test source alpha
    
Overrides BitmapBench to generate four different types of
source images for the following blit cases
- completely opaque
- completely transparent
- 2 striped patterns that try to hit the worst case of
  state-machine based alpha prediction

BUG=
TEST=bench -match bitmap_8888_A_source -repeat 50

[Antti, 2012-08-29 16:28:05]

Derek, could you take a look?

I've also considered adding a benchmark that would do a=0.5, a=0, a=1, a=0.5,
a=0, a=1,... stripe pattern instead of just a=1, a=0, a=1, a=0,... That would
cause the state machine to visit every state and trash cache even more.

Your function pointer approach is more light weight, I could change this to do
that.

[DerekS, 2012-08-29 18:58:49]

On 2012/08/29 16:28:05, Antti wrote:
> Derek, could you take a look?
> 
> I've also considered adding a benchmark that would do a=0.5, a=0, a=1, a=0.5,
> a=0, a=1,... stripe pattern instead of just a=1, a=0, a=1, a=0,... That would
> cause the state machine to visit every state and trash cache even more.
> 
> Your function pointer approach is more light weight, I could change this to do
> that.

If you want to add the additional alpha changes that would be great, but other
than that I'm pretty much happy with this patch as is.  We are always glad when
someone contributes to our benchmarking suite!

[DerekS, 2012-08-29 18:59:21]

On 2012/08/29 16:28:05, Antti wrote:
> Derek, could you take a look?
> 
> I've also considered adding a benchmark that would do a=0.5, a=0, a=1, a=0.5,
> a=0, a=1,... stripe pattern instead of just a=1, a=0, a=1, a=0,... That would
> cause the state machine to visit every state and trash cache even more.
> 
> Your function pointer approach is more light weight, I could change this to do
> that.

If you want to add the additional alpha changes that would be great, but other
than that I'm pretty much happy with this patch as is.  We are always glad when
someone contributes to our benchmarking suite!

[Antti, 2012-08-30 11:37:31]

On 2012/08/29 18:59:21, DerekS wrote:
> If you want to add the additional alpha changes that would be great, but other
> than that I'm pretty much happy with this patch as is.

Results look like this. This is before the patches from:
http://codereview.appspot.com/6465075/

For source-independent blit function these score the same.

 running bench [640 480] bitmap_8888_A_source_stripes_evil
   8888: cmsecs =  72.71
 running bench [640 480] bitmap_8888_A_source_stripes
   8888: cmsecs =  72.57
 running bench [640 480] bitmap_8888_A_source_transparent
   8888: cmsecs =  72.49
 running bench [640 480]  bitmap_8888_A_source_opaque
   8888: cmsecs =  72.48

[Antti, 2012-08-30 11:40:10]

Derek, I don't quite like the evil name, but couldn't come up with better
either. At least it is hitting the slow case, if you compare the results before
and after of patch 6465075

bitmap_8888_A_source_stripes_evil 8888 c         72.71      83.60     -10.89  
-15.0% 
bitmap_8888_A_source_stripes 8888 c         72.57      76.59      -4.02    -5.5%

 bitmap_8888_A_source_opaque 8888 c         72.48      41.21     +31.27   +43.1%

bitmap_8888_A_source_transparent 8888 c         72.49      27.22     +45.27  
+62.4%

[Antti, 2012-08-30 11:44:30]

Btw, I don't have commit rights to the repo, so if this looks good to you, can
you help me commit it?

[DerekS, 2012-08-31 12:43:13]

On 2012/08/30 11:44:30, Antti wrote:
> Btw, I don't have commit rights to the repo, so if this looks good to you, can
> you help me commit it?

I've incorporated this change into https://codereview.appspot.com/6465075/ which
has been committed so you can close this issue.