Re: [hatari-devel] Tests with the cycle exact fix 68030

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 07/02/2017 à 21:47, Eero Tamminen a écrit :
> Hi,
>
> On 02/07/2017 12:38 AM, Nicolas Pomarède wrote:
> > Le 06/02/2017 à 23:09, Eero Tamminen a écrit :
> > > Do you have the same binary:
> > > $ md5sum bound2.prg
> > > bfccd5b37618f6434de2655f53f853ea  bound2.prg  (11729 bytes)
> > > ?
> > Yes, same md5.
> > The code above waits for DSP transmit empty ; just in case, are you sure
> > DSP is enabled in your case ?
>
> It is.  I removed my own hatari.cfg so that Hatari uses
> its own defaults and ran:
> src/hatari --trace os_base -s 4 --machine falcon --tos tos404.img
> bound/bound2.prg
>
> Same result as earlier, see attached screenshot.
>
> And disabling CE mode gets the ball bouncing.

My bad, I  was testing with my own dev version which includes better 
cycles accuracy.
If I test with the current dev version from mercurial, then I get the 
same bug as you, so I think this demo will just be fixed when I commit 
my changes later.

(note that the demo also works in 68020 CE mode, for some reason the 
68020 cycles seem to work better in the case of the DSP loop you saw)

Nicolas