[hatari-devel] Could you all test this on hatari falcon new core and real hardware please ?

Laurent Sallafranque laurent.sallafranque at free.fr
Wed Aug 24 23:02:00 CEST 2011


So, if I compare the results :

My falcon (VGA, all resolutions)

$b9
$5df
$706
$75a4

Matthias :
Plain TOS4.02, clean boot:
$cd
$367
$6bc
$4599

George:
$c4
$395
$726
$46be


It seems that my Falcon reacts differently (I've got a Hd driver loaded, I don't know if this change anything).
I haven't got floppy disk, I'll try to find one.

Matthias and George have close results.


Theory (and Hatari) are a bit too far :

Hatari results:
$96
$2ac
$546
$3426


So, it seems that the 68030 in hatari runs too fast (because in theory, 
the DSP should add less). This is visible with the 10000 loop test.



Also, testing many times doesn't give the same value each time on the 
real hardware.
I wanted to use this little program to compute precisely cycles for 
68030, DSP and synchro between them.
I think this program will have to be improved before to give better 
results. (more interrupts to stop, as suggest Thomas for example)

Any other ideas ?

Are you running in VGA  or TV/monitor ? (I guess videl cycles may be 
involved here)

Thanks for the tests.

Regards

Laurent


Le 24/08/2011 21:00, Thomas Huth a écrit :
> Am Wed, 24 Aug 2011 10:12:13 +0200
> schrieb Laurent Sallafranque<laurent.sallafranque at free.fr>:
>
>> I've also noticed that my hatari results are constant if I try many
>> times, but the Falcon results are close but not constants after each
>> try.
>>
>>   From me, on the 68030 side :
>>       a 68030 NOP is 4 cycles, so, for 100 NOPs, 400 cycles at 16 Mhz.
>>
>> On the DSP side :
>>        p:005a  200050         (02 cyc)  add y0,a
>>        p:005b  0c005a         (04 cyc)  jmp p:$005a
>>
>> In theory, the formula for the result of the adds should be : 100
>> (loop)
>> * 4 (cycles fo NOP) * 2 (32 Mhz) / 6 (DSP cycles for the add+jmp).
>>
>> So, I should get :
>>
>> 100 NOP      : add = $85
>> 500 NOP      : add = $29a
>> 1 000 NOP   : add = $535
>> 10 000 NOP : add = $3415
>>
>> These results are really close to the Hatari results, but far from my
>> real falcon results.
> Just a though: Did you disable really all interrupts before starting
> your benchmark loop? I had a quick look at your sourcecode, and you
> disable the MFP interrupts there, but I did not see that you've
> increased the IPL to disable VBL interrupts? So if some VBL interrupts
> occur during your measurement, these might mess up the results. That
> might also be an explanation for the differences, since the VBLs are
> not emulated correctly in Falcon mode yet, are they?
>
>   Thomas
> _______________________________________________
> hatari-devel mailing list
> hatari-devel at lists.berlios.de
> https://lists.berlios.de/mailman/listinfo/hatari-devel
>




More information about the hatari-devel mailing list