| Re: [hatari-devel] WinUAE CPU core zero cycles |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/hatari-devel Archives
]
Hi,
On keskiviikko 20 helmikuu 2013, Laurent Sallafranque wrote:
> Just a little question before I start to search, do you use cycle exact
> + MMU ?
No, as cycles counter doesn't work with MMU core. This was with
the default Falcon/WinUAE CPU core.
- Eero
> If yes, I think when Thomas added the MMU in a new core, he didn't take
> into account the cycles values I added for the 68030 CPU.
>
> If that's the case, would you please make the test again but without the
> MMU core ?
>
> Regards
>
> Laurent
>
> Le 19/02/2013 22:08, Eero Tamminen a écrit :
> > Hi,
> >
> > On tiistai 19 helmikuu 2013, laurent.sallafranque@xxxxxxx wrote:
> >> Eero, can you tell me which instructions return 0 cycles ?
> >>
> >> I know I didn't finish the job for Fpu and MMU instructions, but if
> >> the instructions you speak about are in the 68030 CPU, I'd be
> >> interrested.
> >
> > This is from TOS v4.x bootup:
> > $e3ddc4 : 3047 movea.w d7,a0
> > inf% (3, 0, 0)
> > $e4929c : c042 and.w d2,d0
> > inf% (9, 0, 0)
> > $e4929c : c042 and.w d2,d0
> > inf% (10, 0, 0)
> > $e4929c : c042 and.w d2,d0
> > inf% (11, 0, 0)
> > $e4929c : c042 and.w d2,d0
> > inf% (12, 0, 0)
> > $e490ec : 7200 moveq #0,d1
> > inf% (1, 0, 0)
> > $e490f4 : 7400 moveq #0,d2
> > inf% (1, 0, 0)
> > $e25bf4 : 7218 moveq #$18,d1
> > inf% (3, 0, 0)
> > $e25c20 : c041 and.w d1,d0
> > inf% (3, 0, 0)
> > $e1fb30 : 200d move.l a5,d0
> > inf% (5, 0, 0)
> > $e1fa54 : 200c move.l a4,d0
> > inf% (5, 0, 0)
> > $e1fb30 : 200d move.l a5,d0
> > inf% (6, 0, 0)
> > $e25bf4 : 7218 moveq #$18,d1
> > inf% (4, 0, 0)
> > $e25c20 : c041 and.w d1,d0
> > inf% (4, 0, 0)
> > $e1fa54 : 200c move.l a4,d0
> > inf% (6, 0, 0)
> >
> > This from latest EmuTOS CVS snapshot bootup on Falcon mode:
> > $e0b912 : 5242 addq.w #1,d2
> > 0.00% (6, 0, 0)
> > $e0dd1c : b042 cmp.w d2,d0
> > 0.00% (1, 0, 0)
> > $e0dc1c : 3801 move.w d1,d4
> > 0.00% (1, 0, 0)
> > $e0dc44 : 3608 move.w a0,d3
> > 0.00% (1, 0, 0)
> > $e05854 : c001 and.b d1,d0
> > 0.00% (1, 0, 0)
> > $e0589e : 8001 or.b d1,d0
> > 0.00% (1, 0, 0)
> > $e05f64 : d082 add.l d2,d0
> > 0.00% (1, 0, 0)
> > $e05f62 : 2002 move.l d2,d0
> > 0.00% (2, 2, 1)
> > ...
> > $e0dd1c : b042 cmp.w d2,d0
> > 0.00% (56, 28, 0)
> > $e139d4 : 2245 movea.l d5,a1
> > 0.00% (1, 0, 0)
> > $e136e4 : 3800 move.w d0,d4
> > 0.00% (1, 0, 0)
> > $e13a48 : 4443 neg.w d3
> > 0.00% (75, 0, 0)
> >
> > I.e. it's the basic instructions which would seem to have problems
> >
> > with cycles counter:
> > move, neg. or, and, add, cmp...
> >
> > The interesting thing is that some of them (at least add, move, cmp)
> > have cycles sometimes.
> >
> > As debugger is called right before DSP and DSP uses the same call to
> > get cycles, CPU vs. DSP ratio is most likely pretty bogus because of
> > this. :-/
> >
> >
> > If you want to see the same output, update to latest Hatari version,
> >
> > change DEBUG in start of src/debug/profile.c to 1 and do:
> > echo "profiling on" > profile.ini
> > hatari --parse profile.ini --machine falcon --tos tos404.img
> >
> >
> > - Eero