Re: [hatari-devel] Falcon tests -> sound issues

On Fri, Jun 17, 2022 at 8:52 AM Nicolas Pomarède <npomarede@xxxxxxxxxxxx> wrote:

Le 06/06/2022 à 18:29, Nicolas Pomarède a écrit :
> Le 05/06/2022 à 23:10, Nicolas Pomarède a écrit :
>
>>>
>>> Hello,
>>>
>>> I found some code from a few years back;
>>> http://ae.dhs.nu/tmp/falcsmp.zip
>>>
>>> I'm a noob about setting up the Falcon sound system but it does play
>>> 16 bit stereo 49170 Hz OK on my Falcon in 030 and 060 modes.
>>> In Hatari it will get distorted sound that comes and goes.
>>>
>>
>> Hi
>>
>> thanks, that's just what I needed, it will be easier to fix/improve
>> falcon sound with a "simple" program that just play sound and wait for
>> a key.
>>
>
> One bug was fixed, related to the new CycInt code ; sound is now better,
> at least during 20 sec when using your falcsmp.tos test.
>
> After that, it becomes noisy again, some bugs remain to be fixed :)

Hi

good news, the sound problem in Falcon mode was finally fixed.

This was quite hard to debug because the whole crossbar code requires to
mix several input source at different freq and to output sound at
another freq too (44.1 kHz for example), all that with possible rounding
errors that must be compensated to keep a constant flow of bytes.

After much tracing, the problem was in fact that this code used a wrong
value for the Falcon's cpu freq since several years :-( This was not
noticable before because sound mixing was done in a different (less
accurate) way, but now with cycle accurate YM2149 we require all the
clocks (cpu, dma, ym) to be as accurate as possible.

Anyway, sound quality should be as good as before ; only remaining
rounding error can cause a small drift that will mess sound again after
2 hours of playing sound :(

So not perfect yet, but it will be good enough for Hatari 2.4 and should
be fixed later.

I tested with 4musiK, the falcsmp example program from Ander and 030
demo by DML. Please check the demos that didn't sound good recently.

Nicolas