[hatari-devel] DSP Instruction cycle count

[Laurent Sallafranque]

Hello,

>> I've noticed that BadMood seems to run much faster.
>>     
>
> Which version?  BadMood 2.14 seems to work fine, but
> BadMood 3.x graphics look broken.
>
>   
I've tested many times BadMood 3.07A.
It's still working well for me, and much faster than before.
Anyone 's got problems with BadMood ?


I've tested with programs wich are timing sensitives like Rot3dDSP, 
Bounds3 falcon, bound2 falcon, anf Hmmm.
They all still work.

By the way, I think I'll have to customize some timings (lostly the 
external X, Y and P memory access).

Does anybody knows a tool or have examples of DSP programs with cycle 
timings that I could use to compare with the ones I added to hatari ?

If there are too much problems with my actual timings code, we can still 
use the 'i=7' test in dsp.c for now.

>> Dsp sound seems to be clearer (but still not perfect).
>>     
> Is it normal that when programs (demos) exit, there's still some
> background sound?
>   
I think sound is bufferized, that's probably why we can listen some 
music after a reset (Thomas ?)

>> emulation improvement ? 
>>     
> "Atari UK demos" and "Les Dinosaures" demo version work now.
>   
Good :)

>> Tell me if you detect regression with some programs that were running
>> before.
>>     
>
> "Built-in Obsolescence" earlier run a bit, now it stops at startup.  Hatari
> outputs a DSP stack overflow warning.
>   
Yes, I know for this one, but it is not relative to the new cycle code.
(I've uncommented some lines into the stack management code).
I'll have to search deeper that strange bug.

> Similar issue with with the "Are you sitting comfortably?" demo from RG.
>   
I'll have a look at this one, but all bugs relative to the stack are not 
relative to the new cycles counting.
In fact, Buil in obsolescence goes into an interrupt with a jmp (fast 
interrupt), but sub the stack pointer by 1.
The correct way to code it would be to jsr to the interrupt (long 
interrupt). Many programs are correctly coded, this one not.
> Btw.  Have you tried the conditional breakpoints yet & do they work OK for 
> DSP? :-)
>   

Not yet, sorry. I haven't got that much time these days.

:-/

Regards,

Laurent