Re: [hatari-devel] Basic cpu testsuite

[Thorsten Otto <admin@xxxxxxxxxxx>]

On Sonntag, 3. Juni 2018 18:09:39 CEST Thomas Huth wrote:

>However, after having a closer look at that "move SR" stuff, I think
>that's already too much for the minimalistic fake-TOS. 

It is currently used in one place only. It could be greatly simplified when it 
is ensured that this is the only place; all the address decoding logic could 
be removed then.

> 1) We could simply run such code in supervisor mode.

I did not want to rely on this, since the programs should also be runnable on 
real hardware, where you would have to explicitly switch to Supervisor mode 
first, and that Super() call would then be rejected by hatari's tos-less mode. 
Also, i currently run the tests on Aranym too, which does not have a tos-less 
mode (yet). Actually, i don't want to rely on any Atari specifics there, since 
i'm currently also cross-compiling them for amiga, running the binaries in 
WINUAE ;)

>We could set the _longframe (0x59e) system variable

Same problem. You would need to query the system variable, which would require 
Supervisor mode. Or implement MiNT's Ssystem() call in the gemdos emulation.

However, there are some infos that might be interesting when running the 
tests, namely
- the cpu type emulated (for things like the FPCR handling that was recently 
changed) and other minor differences in the emulated CPUs, so the tests do not 
generate false reports
- whether JIT is enabled (not Hatari maybe)
- whether FPU is enabled, so those tests could just be skipped
-maybe whether accurate softfloat emulation is in effect, so that minor 
differences in the results can be silently accepted.