I was thinking about this after issue #10867 and what appears to be an error in the SAMD vs other CPU code. In a few places in synth, audioeffects and audiomixer (and likely others) values are converted from float to do fixed point math or vice versa. It may be better to standardize on one way and if we need exceptions for some advanced processing in an effect or similar have that as a one off.
If this sounds like a good idea it does not have to be an immediate change for everything but a goal to hit. I'd also suggest (which I could do) a quick writeup on fixed-point math for anyone coming in cold to it.
There are pros and cons. Floating point can be more accurate but of course most the controllers like a FPU so that would hurt the speed quite a lot.
Just looking for thoughts at this point.
I was thinking about this after issue #10867 and what appears to be an error in the SAMD vs other CPU code. In a few places in synth, audioeffects and audiomixer (and likely others) values are converted from float to do fixed point math or vice versa. It may be better to standardize on one way and if we need exceptions for some advanced processing in an effect or similar have that as a one off.
If this sounds like a good idea it does not have to be an immediate change for everything but a goal to hit. I'd also suggest (which I could do) a quick writeup on fixed-point math for anyone coming in cold to it.
There are pros and cons. Floating point can be more accurate but of course most the controllers like a FPU so that would hurt the speed quite a lot.
Just looking for thoughts at this point.