Id buy a coldfire

[jdiffend]

bloodline wrote:
I think the point here is, buy a Cold Fire if you have apps on your Amiga now, which need more power... ray tracing, music processing and image processing springs to mind.

I spent quite a bit of time looking into this a couple years ago and well.... the problem with the Coldfire is compatibility.  Even with the emulation libs there is stuff that just won't run without a recompile, rewrite or patch.  

The supervisor stack is NOT compatible with the 68K.  You really need to rewrite *at least* the exec.

Software must use the math libs if floating point is used.

There needs to be Coldfire math libs.

You need to patch any software with unemulatable instructions like mpy/div with illegal instructions that can be trapped and add emulation for them to the emulation libs.

The list goes on  and on.  

Just having the hardware isn't going to cut it.  As it is I'm not sure Oli's emulation approach will work.

[jdiffend]

Only if you replace the Kickstart ROM!
And no... supervisor mode isn't the only problem... but I assume you were being funny.

[jdiffend]

V4 core fixes NONE of the things I mentioned but it does support a few more address modes than the V3 so less instructions need emulated.

The V4e supposedly adds a supervisor stack pointer which will speed up emulation of the exec.  It does not fix ANY of the things I mentioned... nor will a V5 core or probably even a V6 core!

The supervisor stack is different on the Coldfire CPUs for speed.  It's a much faster way to implement the supervisor stack in a RISC cpu.  It's highly unlikely they will try to add more 68K compatibility after the V4e unless it's with additional address modes.

The mpy/div instructions don't set certain status bits and any software that tests those without first executing an instruction that will set them will not function the same on the Coldfire as a 680x0.  Patching these instructions with an illegal instruction that can be trapped allows exact emulation where needed... but a program to do that still needs to be written.

The FPU on future chips isn't fully compatible but is similar to the one on the 060.  

MMU?  I really haven't looked at that.

Certain address modes can't be emulated because of the way a pipelined CPU executes instructions.  These should not be a problem for any software that was compiled or written for 68040 and above.  68020 code... could fail.  The only solution short of a rewrite or recompile is to patch those instructions just like I suggested for the mpy/div.

The nice thing about the patched software is it could still run on 680x0 processors if you have a program similar to decigel to run on them.  The exception could be caught, the instruction decoded and the program patched to the original instruction on the fly.

[jdiffend]

He seems to think he can get it working with just emulation... I have my doubts.  And I don't think an AROS port will solve some of his problems.

The AROS port is for my own project and I hope to start next week.  But like I said... it will be aimed at a V2 core.  Well see... it sounds like I have another job (finally) and that comes first.   I think Oli might have been planning his own port anyway.

I was working on a Coldfire version of the exec... but that bit the dust when my hard drive got wiped a few days back so I'd have to start over.

While I'm on the kickstart subject... the new commented kickstart disassembly I've been working on could be done in a day or two if I'd just stop being lazy.  (less than 1000 lines to comment) I've only put in about 5 days work total but I started a year ago. :-D  
Since Amiga Inc is legal happy :roll: it would be released as the disassembler, config files, a diff file and a batch file to run everything.  I'd still need to compare several ROM versions so it will work on more than just one.

[jdiffend]

Um... less than 1000 lines to comment on the exec... not the entire kickstart.  I'm not that fast!  
 :lol: