Re: [hatari-devel] Prefetch issue on 68000?

[Eero Tamminen <oak@xxxxxxxxxxxxxx>]

Hi,

On tiistai 19 helmikuu 2013, Nicolas Pomarède wrote:
> that's a known story, nothing mysterious about prefetch (it was used in
> many protection code more than 20 years ago) and the fact that memory
> accesses are made at different "sub-time" depending on the instruction
> and its micro code is also known.

The reason why asked was the claim about Hatari timings being wrong...


> I tried to fix the most common cases for this along the years using
> uae-cpu core (the old one) and winuae has direct support for nearly all
> read/write accesses inside any instruction in 68000 cycle exact mode.
> 
> For now, Hatari doesn't use the cleanest solution, which would be using
> a cycle exact mode as the winuae's 68000 mode, but most hardcoded cases
> are enough to handle all overscan effects so far (which are the most
> sensitive cases to these behaviours) and other real life cases.

Currently I wouldn't count so much on WinUAE core as its "cycle exact"
cycles returned by Cycles_GetCounter(CYCLES_COUNTER_CPU) are more bogus
(often there's zero diff from previous instruction) than the old UAE core
provided ones.  :-/


	- Eero