|Re: [hatari-devel] Release 1.6.2 in mid May|
[ Thread Index |
| More lists.tuxfamily.org/hatari-devel Archives
In complement to my previous mail, I've tried to be as accurate as
possible with 68030 cycles, except for the cycles that are written as
"MAX" in the documentation.
Maybe adding the TT to the hatari_falcon exe could improve the TT
emulation, I really don't know.
I think it may be possible to choose a TT from the falcon exe emulator, no ?
Le 01/05/2012 21:52, Laurent Sallafranque a écrit :
I don't know the TT a lot, and I have tried it 2 minutes in hatari.
The problem with benchmarks is that they test the div a lot, and the
div in hatari (030) is really not accurate.
I've asked if someone knew the algo implemented in the 68030 but
Nicolas replied me it's secret (only motorola knows it).
So, no hope actually to have an accurate div (in cycles).
Another issue is emulated TT performance. GEMBench reports strange
results, e.g. "Integer division" shows 34% of real TT. Maybe adding TT
emulation to hatari_falcon.exe could fix that issue?
Laurent, any comments?