Re: [hatari-devel] WinUAE CPU core cycle-exact option impact for 68000

[Eero Tamminen <oak@xxxxxxxxxxxxxx>]

Hi,

On 1/16/20 12:01 AM, Nicolas Pomarède wrote:
> Le 15/01/2020 à 22:49, Eero Tamminen a écrit :
> > I'd like to clarify in the Hatari manual what
> > the impact of cycle-exact option is for 68000.
> >
> > I see that Hatari's cycle-exact option enables
> > both CPU and memory cycle-exact options from
> > the WinUAE CPU core.
> >
> > WinUAE CPU core uses different functions for
> > instruction and exception handling when cycle-
> > exact is enabled, but what's the actual impact
> > of those for the speed of the emulated code?
>
> that's sthg I'm planning to measure one day. The impact on exception 
> should not be noticable (they don't happen that often compared to 
> "normal" instructions).
>
> Note that I have a devel tree not committed yet with even more CE 
> behaviour, including very accurate bus error emulation (that Toni/WinUAE 
> improved extensively in the latest weeks/months).

I'm interested more in actual details of how
the cycle-ccurate mode differs on 68000 from
non-cycle-accurate one, rather than measurements
from some specific tests.


> This might have an impact too, maybe not that much in the end, but I'm 
> planning to measure it too.

Btw. On EmuTOS devel there was comment that VDI
vr_recfl() tester managed 30 loops in cycle-
accurate mode, and 39 nine otherwise, so
perf impact 68000 cycle-accuracy option can
apparently be quite noticeable.

This was for code that spent most of it's time
in loops of few instructions (not doing any exceptions).


	- Eero