Re: [hatari-devel] Yanartas demo doesn't work in ST mode

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 09/02/2015 23:30, Troed Sångberg a écrit :
> Your testprograms testbot/testtop.prg gives me the same values (on STE)
> as you posted, but I'm still not sure about the videocounter not making
> a difference. The reason I added removal of right border (which is cycle
> exact) to my test was to not rely on synclock being positioned the same
> on different machines (since it depends on the videocounter, back to
> pre-fetch uncertainty).

Maybe that's where we don't measure the same thing.

By first synchronizing with the known point to remove border, your 
program self calibrate itself to be in STE mode if we can say so.
So, from there your next switches will match the STE timings.

Back to Yanartas demo, my point was more to show that the *same program* 
that doesn't know where the switch should be made (and that doesn't know 
if the machine is a STF or a STE) will self calibrate differently in STE 
and STF, with a 1 NOP difference, which could explain why Yanartas has 
stable border removal on STE, but not on STF (although I should check it 
on my real STF to be sure), and why I saw several demo doing the switch 
to close to cycle 500/504 that worked only in STE mode and not STF (or 
the opposite)


> (Also I would be wary about a 120 cycle wide switch - on STE you'd risk
> creating a +2 line already at cycle 36 while on ST you're safe up until
> cycle 52. But that depends on how you do all the detection etc when
> adding cycles of course)

That was a mistake in my previous mail. I read ff8209 120 cycles after 
switch to 60 Hz, but there're only 40 cycles between 60 and 50 Hz 
switch, so this should be good.
But in all cases, this test just try to remove the border, it doesn't 
try to remove possible glitches during the next line.
Once the position is known in a real demo, the switch would be shorter.


Nicolas