Elbox Dragon compatibility?

[motorollin]

I wonder how feasible it would be to patch these functions? And what kind of performance could we expect?

--
moto

[Piru]

Ah yes, kudos to AmiDog for the diffs.

I wonder how feasible it would be to patch these functions?

The thing is, these are CPU instructions, and they do not generate exception (or generate it too late), thus there is no way to catch them with the emulation library.

The only way to handle these transparently (without need to patch the binaries) is to run some sort of full emulation all the time, and this is slow, and surely beats the whole purpose of the ColdFire CPU.

And what kind of performance could we expect?

For full compatibility (those missing user instructions, the user instructions behaving differently and supervisor emulation), I'd still estimate slower than 68060 @ 50.

If compatibility would be lowered, and the code would mostly construct out of supported instructions, then it would certainly be faster. But again, IMO this beats the purpose.

[motorollin]

No point emulating anything. Might as well use Peg or A1 and run UAE :roll: Dammit, I was really excited about the Dragon thinking it would mean a superfast 68k compatible CPU upgrade :-(

--
moto

[MrZammler]

Piru wrote:
The only way to handle these transparently (without need to patch the binaries) is to run some sort of full emulation all the time, and this is slow, and surely beats the whole purpose of the ColdFire CPU.

And I guess that would work for OS-compliant apps only?

[alexh]

The creators of these coldfire products cannot solve the 68k incompatibility (at least yet). If they could the parts would be onsale.

Realistically the only upgrade for classic Amiga's I could see feasible would be a new 060 based card that used the new v6 060 which can go upto 100MHz with SDRAM. (Similar to Centurbo 60 for the Atari Falcon)

[motorollin]

alexh wrote:
The creators of these coldfire products cannot solve the 68k incompatibility (at least yet). If they could the parts would be onsale.

From the sounds of it, it can't be made compatible without (a) designing a new chip, or (b) emulating 68k (which is pointless IMO).

alexh wrote:
Realistically the only upgrade for classic Amiga's I could see feasible would be a new 060 based card that used the new v6 060 which can go upto 100MHz with SDRAM. (Similar to Centurbo 60 for the Atari Falcon)

Yes please :-)

Is there a reason why a superfast (>200MHz) 680x0 chip couldn't be designed? I mean a technical reason, not an economic one :roll:

--
moto

[Oliver]

I spoke with a couple of engineers from Freescale at my uni last year, and they told me there had been great difficulty creating 68k compatibility in a modern chip.  The documentation was poor or lost, and none of the original engineers were available.  If I understood them correctly, they were very happy with what they have already achieved.  From the look of things, I think they were aiming at people using a faster chip similar to what they were familiar with, but not trying to make a drop in replacement part.  I'm thinking a fully compatible chip is very unlikely to appear.

[CLS2086]

not an economic one

yes it is !
A full new 060RC/50 is 350e at the FreeScale shop.
Who can buy a 2000e cpu today slower than the lowest x86 avaible today for 20/30e... ?
If there could be a market for that, it would already exist.
Sorry.
For me the best card are stil the Phase5/DCE 060+PPC+SCSI that will give you the max of performance and access to PPC OS (which can execute faster WB friendly and non chip dependands apps).

[Lando]

A lot of these differences are real show-stoppers - missing opcodes like ROR and ROL, instructions like ADD and SUB acting only on long-words, and missing addressing modes.  I really can't see how Elbox will make it work (but I'm still hoping).

Even discounting the CPU it still gives your A1200 an AGP slot and Radeon 9xxx drivers, DDR memory, 10/100 ethernet, fast serial ports and USB 2.0.

[motorollin]

Lando wrote:
Even discounting the CPU it still gives your A1200 an AGP slot and Radeon 9xxx drivers, DDR memory, 10/100 ethernet, fast serial ports and USB 2.0.

Yes, but at the expense of the 68k processor. If you could use the improved busboard with an 060 then it would be worth looking at, but I'm not prepared to switch my 060 for an incompatible Coldfire.

--
moto

[alexh]

The products are unfortunately vapourware so dont worry yourself about them. Even if they weren't, driver support is dire in the Amiga world.

[srg86]

Well this is dissapointing.

motorollin wrote:

MrZammler wrote:
It'd be cool to have a 266Mhz classic Amiga, but if it wont let me run all my 68k software, then it's of no use.

My thoughts exactly. We need Dennis to create a brand new processor which contains all of the instructions of a 68k processor, but running much, much faster :-)

--
moto

Indeed! all of the 68030+68882 plus any instructions added in the 040 and 060.

It seems to me that if the x86 could have been extended without dropping backwards compatibility then surely the 68k could have been.

The 486 had a full 5 stage pipeline'd version of the 386, plus new instructions, plus a full x87 FPU plus an 8KB cache all on one chip, then Motorolla could have done similar with the 68040. Although the 486 isn't as fast clock for clock as the 040, it definatly ran cooler at just 25MHz IMHO.

I have read that in the 68040 design they hit their transistor budget but maybe they shouldn't have tried to implement seperate MMUs for data and instructions, even AFAIK with their own TLBs (ATCs).

[alexh]

but maybe they shouldn't have tried to implement seperate MMUs for data and instructions, even AFAIK with their own TLBs (ATCs).

It's called the Harvard Architecture... look it up ;-)

[alexh]

We need Dennis to create a brand new processor which contains all of the instructions of a 68k processor, but running much, much faster :-)

Not going to do it, not on todays FPGA's.

[srg86]

alexh wrote:

but maybe they shouldn't have tried to implement seperate MMUs for data and instructions, even AFAIK with their own TLBs (ATCs).

It's called the Harvard Architecture... look it up ;-)

I've heard of the the harvard architecture

A true harvard machine IMHO would have completly seperate data and address busses for instructions and data. For example the PIC and AVR microcontrollers. This just seems to me a waste of transistors

[DamageX]

I think that if the Elbox folks are smart then this is what they are doing right now 8-) Having realized the hopelessness of getting coldfire to run 68k object code, they are redesigning the CPU board for a Pentium-M or some such thing with a JIT 68k emu in ROM. As long as they hide the CPU behind an impenetrable heatsink and don't tell anybody that it's really an x86 under there then it should be a popular product.

Anyway... one could say that the last "real" x86 chips were the Pentium MMX and the Cyrix 6x86MX (both of which only went up to 300MHz) because (if I'm not mistaken) all the newer CPUs are more or less RISC cores but with a part of the chip dedicated to translating x86 instructions. Naturally there is no technical reason that the same couldn't be done for 68k, only an economic reason.