Re: [hatari-devel] New WinUAE core Falcon compatibility

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


Le 02/01/2015 12:57, Eero Tamminen a écrit :
Hi,

On perjantai 02 tammikuu 2015, Nicolas Pomarède wrote:
Le 02/01/2015 12:01, Eero Tamminen a écrit :
On torstai 01 tammikuu 2015, Nicolas Pomarède wrote:
Le 01/01/2015 19:33, Eero Tamminen a écrit :
One thing that has regressed in Falcon emulation (not 060),
is GT2 playback with GemPlay. Now GT2 format playback bombs
out instead of just having bad sound.

never tried this. Is this with a particular module ? with or without
mmu?

Using the yescrew.gt2 tune coming with GemPlay. Without MMU or TT-RAM.

I tested it more and it happens also with old WinUAE version
before you upgraded the CPU core.

GT2 playback works fine when using old UAE core though
(whereas with that, AM playback sounds just noise).

OK, I will try it myself. Does it crash always at the same point ?

It's not that deterministic because one can use the GEM
GUI at the same time as module is playing.


Hi

I can get the crash too with v1.95.

First,if gemplay doesn't have enough memory to load the module, it seems it will crash with a bus error (that's what I get with 2MB falcon and yescrew.gt2). This looks like a bug in the program.

With 8 MB, yescrew.gt2 will give an address after a brief moment of playing some sound. I tried some smaller modules from "ftp://ftp.modland.com/pub/modules/Graoumf Tracker 2" (rage2.gt2 for example), and I get the same crash.

Looking at the disasm before the crash, I can't see anything obvious, so I think this could be due to a desync between the dsp and the cpu ; maybe the new cpu core has some incorrect cycles count (due to ram access time) and this cause some buffer overflow if sound data are not processed at the correct pace by the DSP.

For now, it seems rather complex to investigate, so I'd rather leave it for later, maybe fixing some other cpu / dsp timings will help fix gemplay+gt2.

As gemplay leaves the OS running, it's harder to trace. If we could get the same crash from a demo or a program that "shuts down" the OS first, it would be easier to follow.

Nicolas






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