Re: [hatari-devel] Regression in Falcon sound

[ Thread Index | Date Index | More lists.tuxfamily.org/hatari-devel Archives ]


Le 31/03/2024 à 14:16, Anders Eriksson a écrit :

Hello,

after all of Nicolas recent Falcon fixes, I'm sorry to report some regressions and some old standing bugs.


1. DUMP v1.71 (DSPMOD) module player + Hyperbased module (or any other module)

- Old bug: When starting a second time, there's old sample data played for a frame or two, VGA+RGB (also present in 2.4.1)
- Old bug: Each time starting the modplayer the left/right channels can be swapped around, VGA+RGB (also present in 2.4.1)
- Regression: Big stuttering in VGA mode after some time (like skipping forward or backwards, the moment when it happen differs each test run (OK in 2.4.1)
- Regression: Same as above in RGB mode, but much more subtle and more often (OK in 2.4.1)

DUMP: https://demozoo.org/productions/95633/
Hyperbased: https://demozoo.org/music/2233/


2. MP2 Audio (direct from RAM playback to avoid any drive streaming issues)

Same problems as with DSPMOD, except I haven't verified the left/right channel mixup.

Test package: http://ae.dhs.nu/tmp/hatari250mp2test.zip
(music from New Beats "That time of the year" demo). Set machine to 14 MB so it can fit the file in RAM.



Hatari version:
https://api.cirrus-ci.com/v1/artifact/github/hatari/hatari/macos12/hatari/hatari-snapshot.zip
INFO : Hatari v2.5.0-devel (Mar 29 2024), compiled on:  Mar 29 2024, 01:39:05

Host system: macOS 13.6.5, ARM64 (M1)


Double checked against a real Falcon (RGB+VGA) and none of these problems appear there.


Hi

thanks for the download link and the detailed description.

I confirm I get the same result as you did, there's some kind of randomness that makes it a little more difficult to reproduce exactly on every run.

For now, I don't know why there's this regression but as it involves DSP it's not sure that it's related to the very recent changes I made to DMA audio.

I will try to get some time to bisect, but even so this could imply quite some changes, as Falcon audio still uses some "float" types to mix at different frequencies and this can give too much rounding errors that accumulate over time and can cause drift or "jump" in the sound buffer. The plan is to replace this with integer math that "self adjusts" over time to keep rounding unnoticable (as done for YM2149 and STE/TT audio)

It's quite possible it will be fixed after Hatari 2.5 is released, else I'm afraid this would postpone the release for too long :(

Nicolas



Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/