AUDIO: Add support for sample rates >65kHz.

[fuzzie]

This is needed for 96kHz ogg files in the AGS game The Blackwell Legacy (1.7).

The ARM code isn't updated by this yet.

[sev-]

I have a feeling that you may be losing precision. Did you consider employing a running average instead of silently throwing the samples out?

[fuzzie]

Yes, this is losing precision, and although it's not audible with my test case, it's not a nice solution.. After some more consideration I think a better solution might be to just reduce the accuracy of the interpolation of the linear converter slightly (to 15-bit precision). I'll update this pull request with that alternative later.

[fuzzie]

This new commit (reducing the fractional bits used in the rate code to 15) works fine for my test case, and I think/hope it's fine, although we should sync up the ARM code too.

[sev-]

Looks fine now.

[lordhoto]

Looks fine to me too. But I would rename the pull request before merging it. The "hack" bit could be a confusion source later on.

[lordhoto]

@fuzzie Do you think we should fix the ARM code before merging this or afterwards?

[bluegr]

IMHO, adapting the ARM code could be done afterwards. I mean, there's no reason to hold this working solution in queue while waiting for someone to adapt the other relevant ARM code

[lordhoto]

fuzzie wants to have the ARM code fixed first. I think that makes sense since it avoids any chances of the ARM code being forgotten. Furthermore, this change is only needed for code not in-tree. Thus, delaying the merge is really no issue at all.

[bluegr]

Just a note: This is needed by the Wintermute game Shaban, which is trying to use a 96000Hz sample

[somaen]

Any news on this?

[digitall]

This is now needed for the Wintermute game "Helga Deep in Trouble". See:
http://sourceforge.net/p/scummvm/bugs/6580/

Will take a look at the required ARM asm changes.

[digitall]

Have made the required changes as far as I understand them... based on comparing fuzzie's changes and applying them across to the ARM code. Totally untested, so any review of the changes is good... If these can get incorporated and tested, then we can merge and close this:
https://github.com/digitall/scummvm/compare/rate-hack
digitall@9003ce5

[bluegr]

I've checked @digitall's changes, and they seem correct, at a first glance.

The only thing I could spot is the FIXME that @digitall added...

In the C++ code, we got:

ilast0 = ilast1 = 0;

but in the ARM code, we got:

lr.ilast[0] = lr.ilast[1] = 32768;

Shouldn't this be 0?

[digitall]

@fuzzie Could you cherry pick digitall@9003ce5 on top of your Pull Request branch and then deal with the FIXME @bluegr indicates please? :)

[bluegr]

I've just talked with Robin Watts, who wrote the ARM code, and he said that @digitall's changes are OK. The different value in the ARM code was added by him in commit b9c8c6c. That value is holding a rounding bit, and he said that this could be changed to 16384, but it's way too small for anyone to hear a difference.

[digitall]

@bluegr : Thanks for spending the time to review this. Hopefully now fuzzie can cherry pick, deal with the FIXME and we can merge! :)

[digitall]

@fuzzie : What is the status of this? If you are busy, if you add me as a contributer on this branch, I will do the change myself so this can get merged.

[bluegr]

Any updates on this? Can the changes from fuzzie and digitall be added?

Some feedback is needed from @fuzzie, @sev-, @wjp, @lordhoto

[sev-]

@digitall, could you please issue new pull request based on your branch, and we merge it and close this PR?

[digitall]

I have opened PR #625 to include my commit as well. The FIXME can be removed later by @bluegr or @fuzzie if they and Robin Watts are happy that the code is correct there, although it should probably just be changed to a comment to explain why it is different when compared to the C code equivalent.

[digitall]

@sev- : I leave whether to merge PR #625 and close this PR up to you.

[fuzzie]

Thanks both. We can discuss further in PR #625 if needed (but I don't think we need to).