Re: [hatari-devel] relative perf between 030 / DSP / CPU?

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 25/06/2025 à 22:34, Eero Tamminen a écrit :
> Hi,
>
> On 25.6.2025 10.56, Nicolas Pomarède wrote:
> > I would need to have a closer look for DSP,
>
> Laurent stated earlier (maybe few years back) that DSP side is accurate, 
> I guess the issue is relative speed between CPU vs DSP.

Hi,
yes, the cpu is the "clock" so if an instruction takes 4 cycles in 
Hatari instead of 6 for example, then you will run less dsp cycles 
during these 4 cycles and the benchmark becomes wrong.

> > As to why FPU would be slower when using MMU, either it's a different 
> > code path in WinUAE or there's sthg wrong in Hatari.
>
> Do you have some Amiga FPU test that could be used to test this?
>
> If it shows a clear perf difference, I think issue is definitely on 
> WinUAE side.

No, I don't have such test. Maybe Toni can have a look at this later fpu 
speed difference between mmu/no muu after WinUAE 6 is released.

> > Note that MMU emulation has no cycle exact behaviour, so if you want 
> > to compare with/without MMU, be sure to not use "cycle exact" mode in 
> > Hatari
>
> What exactly is not cycle-exact when MMU is enabled; just the virtual- 
>  >physical address translation or also something else?

address translation, page swap in case physical page is available or 
not. Also I think the read/write functions themselves are not the cycle 
accurate "plain 68030" version when MMU is used, because MMU is not 
cycle accurate at the time and using CE memory access functions would 
just slow down everything for no real accuracy gain


Nicolas