Re: [hatari-devel] Blitter emulation corner case

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 20/05/2017 à 21:04, Christian Zietz a écrit :
> Nicolas Pomarède schrieb:
>
> > I think it's properly emulated, it's just that the blitter copy in that
> > case takes less than 64 bus cycles, so there's no need to restart the
> > blitter a 2nd time in non hog mode
>
> No, what I meant is that even when according to Atari docs I had to
> restart the Blitter but didn't, the result on Hatari should match the
> one real hardware. Which it currently doesn't as you know.

I agree your code doesn't restart the blitter, but in the case of those 
tests, it would not be needed anyway, whether you're in hog mode or not 
: from what I see in Hatari traces, the blitter in your test cases take 
at most ~200 cpu cycles (ie 50 bus accesses).

So, Hatari seems correct : the blitter is started, it copies in 50 bus 
accesses (or less depending on how many src words are read if prefetch 
or not), then the cpu gets control of the bus and when it tests busy 
bit, it's clear and you exit your while loop.
There's no need to restart the blitter a 2nd time, because everything 
was copied on the 1st run.

We see Hatari doesn't match the sequence of word reads and this needs to 
be improved, but the behaviour of the busy/hog bits seems correct to me 
at least.

Nicolas