Comment 2 for bug 1350674

Thanks Mirco!

> I don't think replacing a multiply with a branch will be faster. It most likely will be slower specially in embedded devices.
> Instead I would create two separate shaders, one that does not do alpha multiplies and the default one and activate
> the respective shader by testing renderable.alpha() == 1.0

It's difficult to tell. I remember from past benchmarks that a single if() in a shader didn't bring any performance penalty, although it could bring a slight power penalty, since the GPUs tended to handle this case by executing both paths concurrently.
It's also possible that the overhead of the multiplication is actually so small, that it doesn't matter if we just keep the current shader and use it all the time. Using separate shaders we get a simpler/better shader for the opaque case, but we also have the overhead of switching GL programs.

So, it's not immediately clear what's best, and my plan is to benchmark all of these approaches. In all approaches there are trade-offs, so we need to ensure that what we select the one that offers the best performance for our use cases and hardware.