NEON work to do.

[Jacob Bramley]

Hello all,

I was just about to jump in with some NEON optimizations for YCbCr-to-RGB with
scaling (bug 650651), when Jeff mentioned (on the bug) that GL layers will make
the NEON conversions redundant. I knew the GL stuff was coming but thought the
NEON conversions would still be used.

Firstly, is this really the case? If we have to copy codec output to graphics
memory anyway, why not do the YCbCr-to-RGB conversion at the same time? The
conversion (probably) has minimal cost compared to the memory overheads of the
copy. No scaling would be required in this case though, as the GPU will
definitely be better at that.

Secondly, are there any other NEON routines that need to be tackled? Also on my
list to look at are canvas and CSS3 rendering code, but I'm not sure how much of
that is about to change with GL and maybe Azure too. I don't really know much
OpenGL-ES and I don't yet have a board with GL drivers, so it'll be hard for me
to contribute there in the short term.

Any ideas?

Thanks,
Jacob