Cold Fusion accelerator

[CannonFodder]

What ever happened to it?

http://www.cdtv.org.uk/coldfire/1207.html

[McVenco]

The same that probably happens to all post-1997 Amiga hardware. It gets announced and then it dies.

On a more recent subject: I also wonder what's happening with the PowerVixxen LT. Haven't heard from that in a while as well, not even the slightest rumours....

[Oli_hd]

You rang my lord? ;-)
Well after taking the last prototype to the bigbash I set to work on testing it, basically it didnt work again.
So its paused at the moment while I build up some more experience, I now have a few simpler projects on the go which should help.... that said I dont think your going to see a Coldfire card from me, although I will keep working on one.
In my honest opinion the I dont think a standard V4 coldfire can be interfaced to the Amiga's bus, except in the same piggyback way the PPC was done on the Cyberstorm. The only way I see forward is to use the V4e and its built in PCI bus and stick a PCI slave controller on the Amiga's side... not a clean system and not without its problems (for example trying to keep the address's of the motherboard mounted chips and ports in the same place)

So anyway at the moment Im playing with 10 minor projects, most just Zorro test boards and the like. (Placed an order just last night for the chips needed, the PCB's arrived two weeks ago... also I ordered an old V2 Coldfire which uses more 68K like bus signals, but is less 68K compatiable, just to play with)

Oh and by the way guys I reuploaded the CDTV website (along with the coldfire site) as I think Ive finally sorted out the server problems I had been having. I am also working on a redesign of the CDTV website, with lots of new info.

Thats basically it, I wish it was quicker, the project is getting rather old now and the costs have been high yet I still have nothing to show for it which is a bit of a bummer but I wont give up... but its probably best if you look at other solutions.

[Louis Dias]

would a dual-cpu accelerator be possible?

one where most instruction would go to the coldfire but when an instuction not supported was needed, the appropriate register values are copied to the normal (040 or 060) cpu and executed there and the changed registers copied back to the coldfire...

In that sense you'd still have 90+% useage of the coldfire cpu...

[Zac67]

You'd have to keep all registers/states synchronized across both CPUs - which raises the problem that an '060 can't keep up with the CF's pace, so the latter would have to be slowed down to the same speed...

[Rob]

Could the missing instuctions be added using an FPGA and then interface that to the Coldfire.

[Louis Dias]

Zac67 wrote:
You'd have to keep all registers/states synchronized across both CPUs - which raises the problem that an '060 can't keep up with the CF's pace, so the latter would have to be slowed down to the same speed...

No, only the 1 or 2 a particular instuction affects and only on the incompatible instructions.  If the following instruction is compatible and doesn't use either of those registers, then it can be executed on the coldfire while waiting for the prior one to be returned from the legacy cpu.  In this situation, programmable cache would come in handy...

[AJCopland]

That's a shame I was really hoping that you'd be the one to pull it off. Good luck with your other projects.

Andy

[AJCopland]

lou_dias, it really really ain't that simple.
Besides which the CF has an emulation layer precisely for the purpose of emulating unsupported instructions (ignoring the remaining compatibility issues that itself causes).

The problem that i think Oli_hd is referring too is actually getting the existing data & address bus used in the Amiga to talk to a Coldfire V4 or V4e cpu as they must be somewhat different. That might require some kind of bridge that converts and syncs the signals between them. It could be done in an FPGA or something similar but it's possibly a lot of work and quite detailed at a technical level.

I'll leave that for someone else to discuss as I'm meandering on none coding land adn getting scared :-D

Andy

[Piru]

one where most instruction would go to the coldfire but when an instuction not supported was needed, the appropriate register values are copied to the normal (040 or 060) cpu and executed there and the changed registers copied back to the coldfire...

For this to work you'd need transfer the whole state of the CPU (and with FPU and MMU it gets even more complicated). It would be even slower than emulating the unsupported instructions with coldfire itself.

I'm sure with some heavy coding it might be possible to have such solution, but it would be pointless: Slower than coldfire alone, and more expensive.

[Oli_hd]

The problem that i think Oli_hd is referring too is actually getting the existing data & address bus used in the Amiga to talk to a Coldfire V4 or V4e cpu as they must be somewhat different.

Yeah there were a few things, getting all the interupts out of the V4 was a pig (hacking the serial port to output int2 for example) and a lot of other similar things which may not have stopped the card working on its own but overall it just didnt work.

If you were to make a Coldfire co-pro for a normal 68K then a PCI or Bridgeboard solution would be the quickest, and just like the PPC it could lead to a CF native motherboard after a while... that wasnt my goal though.

[Zac67]

BTW: why hasn't anyone evaluated the use of a Crusoe line CPU? With its 'code morphing' capability, you could make it 100% 68k, '030 or '040 compatible. Too expensive? Docs not available?

[Donar]

Sorry i'm not too familiar, with all of this maybe i make false assumptions...

Oli_hd wrote:
Yeah there were a few things, getting all the interupts out of the V4 was a pig (hacking the serial port to output int2 for example) and a lot of other similar things which may not have stopped the card working on its own but overall it just didnt work.

So you need someone "fixing"/re- designing your FPGA glue logic, or wasn't there any? I think there is something in this direction on the Coldfusion card is it?

Did you try to "initialize"/boot up the card with the Freescale CF68k lib? Otherwise the Design may actually be fine, but the whole thing could have become stuck in Kickstart.

Would be interesting to hear something more specific, i really like the idea of ColdFusion, but unfortunately it seems to be a big problem to realise it.

[Donar]

lou_dias wrote:
would a dual-cpu accelerator be possible?
...

It surely is possible (at least it was with PPC...) but it is an expensive solution as you need an additional 68k CPU to the Coldfire and will end up with more traces, a bigger board or a PCB with more layers. :-(

[AJCopland]

Half of what you said made no sense to me :-D

So, making it as a bridgeboard (sticking it on PCI just means dealing with add PCI prob's to my mind) and using it as a co-pro makes more sense? If you, or someone else, could do such a thing would that mean that the built-in 68k would still be the main processor and software/libs would need to be written specifically for it rather like the normal custom chips and/or PPC?

Seems a bit of a shame... so when can we expect that version :-D

Andy

*edit: I jest I hasten to add! Before someone takes my idea the wrong way.*

Oli_hd wrote:
Yeah there were a few things, getting all the interupts out of the V4 was a pig (hacking the serial port to output int2 for example) and a lot of other similar things which may not have stopped the card working on its own but overall it just didnt work.

If you were to make a Coldfire co-pro for a normal 68K then a PCI or Bridgeboard solution would be the quickest, and just like the PPC it could lead to a CF native motherboard after a while... that wasnt my goal though.