Amiga Coldfire project dead?

[Piru]

Coldfire V4 is an interesting CPU, at 266mhz it runs at a nice and efficient 400 MIPS.

...running coldfire code. It won't be anywhere near that when running typical 68k code. Instruction emulation is possible of course, but it introduces huge performance penalty due to exception processing. Even then there are certain opcodes that are incompatible with the 68k, and cannot be patched on the fly. If you instead go for JIT route then why bother with coldfire in the first place, just use faster and cheaper solutions instead.

As far as I can remember Elbox went as far as to actually build a working prototype only to kill the project. I would guess the reasons were performance and compatibility issues.

Of course anyone who actually read the coldfire documentation could easily see that and predict this result.

[Piru]

So how does one explain this then?

http://acp.atari.org

Explain what in particular?

Reading that page gives no indication that this board actually has the 68k emulation implemented, or how fast it would be in real life. All it says that such emulation it would be 680x0 compatible, and that the emulation will be added later. Of course they're wrong here, even with the emulation there still are instructions that won't work correctly.

[Piru]

The Atari Coldfire team have managed to create a Coldfire based motherboard that runs 68k Atari binaries with no apparent loss of performance or compatibility.

Well, nothing on that page indicates that. Care to point out any benchmark results in running some actual 68k code?

[Piru]

I'm waiting for an email reply.  As I said "apparently", not definitely.

Well, you could as well just ask them how they're going to handle any code that depends on:

• 68020 multiply/divide instructions behaving like they do on 68k
• MULU and MULS setting the overflow bit
• ASL and ASR setting the overflow bit
• MOVE.B ,-(A7) and MOVE.B (A7)+, changing ea by 2

[Piru]

Have a look at the ST/TT section's Firebee thread on atariage.com/forums/

Might be someone with technical info or some more leads to other sites there.

I am not that interested. Now, if someone can find the actual 68k emulation performance claims/data I can comment on those.

[Piru]

Why not build hardware-based dynamic recompilation (or translation) into the execution pipeline, i.e. an MC68000 compatible ISA that translates only the necessary bits to the ColdFire ISA (the "microcode," so to speak).

Sound simple enough! Let us know when you're done.

[Piru]

PPC laptops don't exist

[youtube]V89YwHFoXyw[/youtube]
[youtube]tYeZCsKvjSg[/youtube]

[Piru]

If I meant Mac I would have said Mac, I meant PPC as in OS4 compatible or any embedded controller type setup people might want to make useable with OS4/AROS/MoS ie non-Apple Mac hardware.

You're right, there's no such hardware if you exclude Apple. But I wonder why you would do that?

Apple PowerPC hardware is still the best there is, and it has great support and repair services. Why go for some rare (tiny production runs), prohibitively expensive custom HW with substandard support when there are better options around?

Try to get support, spare parts and/or repairs for AmigaONE SE/XE or Micro-A1 today and you get the idea.

[Piru]

@Piro/Bloodline
How does Akiko work? I take it you mean C2P conversion.

There's a single 32bit register. In order to convert 32 pixels you write 8 32-bit (4 byte) chunky pixels. Then reading back the 32bit register 8 times you get the planar data out. IIRC you get the least significant bits out first so that you can omit reading some planes if you use say 128 or 64 colours.

[Piru]

That's not quite how I remembered it (I thought there were actually 8 registers), but your recollection is probably better than mine. Either way, it was a write and read back mechanism that meant you still had to push the data back to the bitplanes yourself.

Well either way, it sucks the same

[Piru]

What is amazing is that a 500Mhz PC from 6 years ago can run youtube fine, but a G4 mac mini chokes to death on youtube.

Well it's no secret that windoze flash is better accelerated.

[Piru]

I think the point of Akiko was for reading bitmaps from disk to facilitate ports.

Rendering dynamic graphics in planar is way too slow, especially if you need to do transparency. Also, blitter is really slow in 256 color mode. Such rendering is done in chunky mode and then the framebuffer is converted to planar with c2p conversion graphics.library/WriteChunkyPixels. This function was added specifically to AmigaOS 3.1 and uses Akiko chip when available. With Akiko the otherwise underpowered CD32 was able to have games that would have been otherwise impossible.

[Piru]

[citation needed]

I doubt Microcosm would be possible without Akiko.