Re: [hatari-devel] WinUAE core freeze with ST emulation

[Laurent Sallafranque <laurent.sallafranque@xxxxxxx>]

Hi,

Doing some tests, I've noticed that a part of my code is probably too 
naive for now.
I mean in newcpu.c

        /* Laurent : if 68030 instr cache is on, cycles are computed 
with head / tail / and cache_cycles
         *           else, cycles are equal to non cache cycles.
         */
        falcon_instr_cycle = regs.ce030_instr_cycles;

        if ((currprefs.cpu_model == 68030) && ((r->cacr & 3) == 1)) { 
// not frozen and enabled
            if (falcon_instr_cycle.head < sav_tail)
                curr_cycles = (falcon_instr_cycle.cache_cycles - 
falcon_instr_cycle.head);
            else
                curr_cycles = (falcon_instr_cycle.cache_cycles - sav_tail);

            sav_tail = falcon_instr_cycle.tail;
        }
        else {
            curr_cycles = falcon_instr_cycle.noncache_cycles;
        }


In fact, I think that if the instruction is not in the cache, it may 
take some cycles to prefetch, which is not well coded here.
Actually, I only test if the cache is active and in this case, I use the 
"cache cycles" instead of the "non cache cycles"

If I remove all the code in the if condition and only keep the single line :
            curr_cycles = falcon_instr_cycle.noncache_cycles;

The Falcon works as if the cache was not activated.
In this mode, I have 2 or 3 more demos running (I don't remenber the 
names, but Eero, you noticed them in a previous thread.

Maybe we should comment these lines for now, and recode this better in 
the next release, no ?

Best regards

Laurent






Le 04/06/2012 23:32, Eero Tamminen a écrit :
> Hi,
>
> On maanantai 04 kesäkuu 2012, Thomas Huth wrote:
> > Ok, thanks for the extensive tests! I think I found and fixed the
> > problem with extended VDI/GEMDOS now: The problem was how the CPU cores
> > emulated the CPU internal prefetch ... the code changed between our old
> > core and WinUAE core.
> > Could you please do again some regression test to see whether it works
> > now as expected and take care of the TT FPU (Falcon was not shipped with
> > a FPU by default, so I think you only have to set it for TT mode).
> OK, attached is the winuae patch with CPU prefetch/exact stuff removed,
> does it look OK to apply?
>
> I tested your changes with it and now all TOS versions boot fine
> with the WinUAE core in all HW configs TOS tester checks!
>
>
> I guess also some tests with few cycle exact Falcon demos would also
> be needed to verify that the changes didn't affect the Falcon emulation
> side negatively.
>
>
> 	- Eero