[hatari-devel] New STE DMA audio emulation

Laurent Sallafranque laurent.sallafranque at free.fr
Sat Jul 2 12:38:10 CEST 2011


There are 2 fifos in the Falcon DMA

1 32 bytes for DMA play
1 32 bytes for DMA record.

That point is a certitude (I've seen it in the compendium, STMagazine, ...)

The way these 2 buffers are filled is described nowhere. I guess it 
works in the HBL like the STe.


Until now, I thought I could bypass the 2 falcon DMA buffers (as we did 
for the STe).
But as your adding of the STE buffer improves sound quality, I think I 
should add the 2 buffers to the falcon too.

Regards

Laurent


Le 02/07/2011 12:17, Nicolas Pomarède a écrit :
> Le 02/07/2011 10:09, Laurent Sallafranque a écrit :
>> Hi Nicolas,
>>
>> Great Job.
>>
>> I have no time at the moment to test things or code for hatari.
>> But I guess some of the things you implemented for the STE DMA should
>> also be applied to the falcon DMA.
>>
>> I mean the 64 bits precision and the 2*32 octets DMA buffer (one buffer
>> for DMA play and one for DMA Record) .
>>
>> I guess (without clue) that the Falcon DMA bahaves like the STE one (I
>> mean HBL refresh).
>>
>> Should I apply these changes ? (Falcon sound with DMA emulation is not
>> nice sometimes).
>>
>> Regards
>>
>> Laurent
>
>
> Hello
>
> I really have no idea on how the Falcon DMA behaves ; the fact that 
> it's able to run in "STE compatible" mode doesn't necessarily mean 
> that the Falcon DMA works in a similar way with a FIFO.
>
> I'm not sure trying to fix dma emulation is the good order to do 
> things : this could lead to some limitations that are not "compatible" 
> with the real sound core of the Falcon. I think STE mode can be left 
> aside for now.
>
> Before going to the HBL level to see what it does, I think it's 
> essential to fix demos that do double buffer audio in falcon mode and 
> where sound is still not good. This is clearly an indication that some 
> low level logic are missing for now (or it could be a small but very 
> well hidden bug)
>
> Using 64 bits for arithmetic is still a good idea though, but it 
> wouldn't explain really "strange" audio result.
>
> Nicolas
>




More information about the hatari-devel mailing list