Re: [hatari-devel] New cycle exact bus sharing for blitter

[Eero Tamminen <oak@xxxxxxxxxxxxxx>]

Hi,

On 08/11/2017 05:23 PM, Nicolas Pomarède wrote:
> I just committed some new blitter code that will handle bus sharing 
> between blitter and CPU as a real STE (or STF with a blitter) would do.
>
> This cycle accurate blitter mode will be enabled when using the 68000 in 
> CE mode (as for emulating an STE)
>
> This allows for example to run part of a cpu instruction during a 
> blitter transfer if this instruction doesn't need to access the bus
>
> For example, see this thread with a test program by C. Konador :
>
> http://www.atari-forum.com/viewtopic.php?p=96197#p96197
>
> As can be seen, it's possible to run costly instruction like MUL or DIV 
> during the blitter transfer, thus saving a lot of CPU cycles (but the 
> code is of course more complicated)
>
> This new blitter code will also accurately count 64 or 63 bus accesses 
> for the blitter in non-HOG mode (which seems to be a bug in the real 
> blitter).
>
> This also required to be able to suspend/resume the blitter at any time, 
> not just after writing the current word as it was the case before.
>
> Despite the rather small changes in Hatari sources to achieve this CE 
> mode, those changes took in fact a lot of time to complete, due to doing 
> lots of tests in different situations to really understand the timing of 
> the blitter and to ensure nothing gets broken.
>
> The case where blitter sometimes runs for 63 accesses instead of 64 was 
> quite an headache to solve :)

I checked few blitter demos.

E.g. Agression's RGBeast worked fine, but startup overscan plasma
screen looks still partly broken in ICE's Jam-Cols:
	http://www.pouet.net/prod.php?which=31452


	- Eero