Support a fixed-point build

[jwcullen]

It's common for this type of software to be configurable to work better on fixed-point systems.

• https://github.com/xiph/opus/blob/main/README#L166-L171

There may be other parts that are floating point, but at least demixing module and resampler only use floating point.

I think this could be a big task to fix, but it may make the software more useful for use on certain systems.

[yilun-zhangs]

@jwcullen
Yes, it may require fixed-point version reference for certain systems,
and this is a big task to support fixed-point option, we will enhance this next, it may take long time to proceed.

[yilun-zhangs]

Hi @jwcullen
Do you know how many bits dose libopus fixed-point reference support?
If we want libiamf to be configurable to work better on fixed-point systems, are there any requirement to the bits of fixed-point?

[jwcullen]

Do you know how many bits dose libopus fixed-point reference support?

Before version 1.5: 16-bit.
Version 1.5+: 16 or new 24-bit.

If we want libiamf to be configurable to work better on fixed-point systems, are there any requirement to the bits of fixed-point?

I filed this issue based on vendor feedback. I do not have specific requirements.

I believe it is understood that a fixed point configuration is intended to be faster on embedded systems without an FPU, at the cost of (potentially) reduced audio quality.