68040 emulation using a 68020

[jotd]

Hi,

I have some 68040 code that I want to emulate using Musashi, and there are a lot of 040 specific / FPU instructions in it.

I don't really care about speed, I can use a fast machine to run it, so I was wondering if some native 68020 code existed to emulate those FPU/040 instructions, maybe possibly an amiga program, and even better, some source.

I guess the code would install itself on the ILLEGAL instruction vector (like MOVEP handler does on 68060 library) to manually decode 040 instructions and do something similar, slowly, but would work.

I liked the musashi design. Simple to understand and documented for integration in another project, but if you know of a 68040/FPU emu that can be adapted to a project, even better (not WinUAE which is very good but emulation core is not designed to be re-used but maybe I'm mistaken...)


thx

[chb]

Hi,

there was Pseud040 for the Macintosh, which did exactly what you want, but (to my knowledge) the source was never released. Maybe you could contact the author and try to convince him to release the source?

[meynaf]

Originally Posted by jotd

I guess the code would install itself on the ILLEGAL instruction vector (like MOVEP handler does on 68060 library) to manually decode 040 instructions and do something similar, slowly, but would work.

68060 library doesn't use ILLEGAL vector to handle MOVEP and friends
(there is a specific emu vector for this)

Aside of the FPU there isn't much 040 specific stuff (move16, cpush, cinv).
AFAIK everything 040-only goes to line-F, fpu or not.
I guess i could do the non-fpu stuff quite easily if i knew exactly what to do.

I remember having tried an FPU emulator for Mac under Shapeshifter but i couldn't make it work (at least not for Amiga apps).
No source for it of course.

I may be interested in fpu emulation as well, but so far all my researching on the matter remained fruitless.

[jotd]

you're right, it would be exactly what I need. I could also use some native C source too. I'm not making it run on a real 68020. If FPU 040 instructions were emulated by C code it would be just fine!

[pandy71]

perhaps? http://aranym.sourceforge.net/

[jotd]

Maybe. Basilik looks better to me. I think the author wrote Shapeshifter at the time, and he explains how his emulator works in a nice document. I think I'll settle for that. Thanks.

BTW I found this opcode: 50fa 00a2 => traptw #162
(HRTMon sees TRAPT.W)

Never heard about this trap instruction. I only knew TRAP, TRAPcc, did not find it in the ref manuals... anyone?

[Leffmann]

It's just TRAPcc with a condition of True.

[matthey]

Originally Posted by jotd

BTW I found this opcode: 50fa 00a2 => traptw #162
(HRTMon sees TRAPT.W)

Never heard about this trap instruction. I only knew TRAP, TRAPcc, did not find it in the ref manuals... anyone?

It looks to me like a TRAPcc.W with cc=True. ADIS, D68k and IRA all decode it as "TRAPT #$a2" for 68020. The 68000PRM encoding documentation is incorrect for TRAPcc which may have resulted in your confusion. There are several incorrect encodings documented in the 68000PRM but most are FPU instruction encodings.

[Leffmann]

Originally Posted by jotd

Hi,

I have some 68040 code that I want to emulate using Musashi, and there are a lot of 040 specific / FPU instructions in it.

Did you get the DLL for Musashi to work that I attached in the other thread? If it didn't work then you must probably build it yourself on the computer you intend to use it on, but I can write it all up if you don't mind installing Visual C++ and a bunch of other tools.

[jotd]

Thanks leffman but musashi wont do me any good for this coee which heavily uses 68040 fpu. The most affordable seems to be Basilik, and there is even some documentation on how to use uae cpu core.

[jotd]

Hi,

I butchered Basilik sources and got a bare 040+FPU emulation (not tested yet, but will do)

I have one question: when CPU encounters a TRAPTW (or TRAPEQW, or any W-type TRAP) what vector is called? (not the conventional $80->$BC TRAP vectors for sure)

thanks

[Leffmann]

It should be vector 7.

[jotd]

it is!! VBR+$1C is the offset. Thanks.

So how to make the difference between, say TRAPV and TRAPT.W ? Do I have to read the stack and check for the instruction that triggered the trap?
There should be some info better than that in the stackframe right?

[meynaf]

Originally Posted by jotd

There should be some info better than that in the stackframe right?

There is AFAIK no difference. TRAPV is just a special case of TRAPcc, this means TRAPV and TRAPVS are exactly the same. TRAPV is still there for 68000 compatibility.