Re: [hatari-devel] DSP question

[ Thread Index | Date Index | More Archives ]

Hi Laurent - thanks for replying.

TBH I have now checked more carefully and it seems the L bit is working at least in most cases. I'm not sure yet if the rest of the problem is me, or something like the RTS/RTI case you describe. I'll try to find out more.

However for the simple case of clearing the L bit, forcing an overflow, then doing another store without overflow, then testing L bit, then it is set as it should be - so that much is definitely working.

If I can narrow down a bit more I'll report.


On 31 July 2015 at 14:28, <laurent.sallafranque@xxxxxxx> wrote:
Hi again Doug,

I throw a quick eye into the hatari DSP code, and I think there could be a problem with the L bit of the SR with RTS and RTI instructions.

This should be verified, but I think that if you store a 56 bits accu to some 24bit destination with overflow, the L bit is set correctly, but later if there's a RTS or RTI instruction, the L bit is lost (no sticking behaviour in this special case).

I should verify this on a real falcon and on hatari, but this may be a bug to fix.
Is it what you encontered this afternoon ?


----- Mail original -----
De: "laurent sallafranque" <laurent.sallafranque@xxxxxxx>
À: hatari-devel@xxxxxxxxxxxxxxxxxxx
Envoyé: Vendredi 31 Juillet 2015 15:09:28
Objet: Re: [hatari-devel] DSP question

Hi Doug,

It is implemented into hatari and it should work OK.
So if you encounter a problem here, it's because of a bug into hatari.

I'll look at this in a few days if you really encountered a bug (just tell me).



----- Mail original -----
De: "Douglas Little" <doug694@xxxxxxxxxxxxxx>
À: hatari-devel@xxxxxxxxxxxxxxxxxxx
Envoyé: Vendredi 31 Juillet 2015 14:53:47
Objet: [hatari-devel] DSP question

Does anyone know if the CCR saturation arithmetic / limit flag is implemented in Hatari?

i.e. when a 56bit accumulator is stored to some 24bit destination, overflows result in saturation and the L bit is set in the CCR and remains sticky until cleared.

I'm seeing some indication that the L flag is not being set - or if it is set, the JLS jump condition is ignoring it (although I'll need to test it more carefully to be 100% sure).


Mail converted by MHonArc 2.6.19+