No subject

CPU transform (Eric's patch) for 3 GPU transforms and 1/20th (worst case
in the examples) of a uniform upload (my branch).

The situation is not clear-cut as the results are going to be very
dependent on the platform (CPU/GPU) used and the quality of the drivers.
I am confident, though, that modern setups are more than capable of
winning this trade.

I believe that the cairo-gl backend should aim to utilize the GPU as
much as possible, especially considering that in the embedded world the
CPU/GPU gap is more discrete and power consumption is of prime
importance. Unfortunately, I cannot yet perform any cairo benchmarks on
an embedded device with GLES2.0, but that day is getting closer :)

Thanks,
Alexandros