Re: Supported Hardware ?
Hi,
On Mon, 24 Oct 2016, Lennart Sorensen wrote:
> That one would be powerpcspe, not powerpc, unless I have misunderstood
> the e500v2. e300, e500mc, e5500 and e6500 are all normal sane powerpc
> designs, while e500v1 and e500v2 are the SPE chips that don't have normal
> powerpc FPUs.
That is indeed the case.
> Whyever would they be playing with that chip?
That's what the sane part of the Amiga world has been asking for a while.
> I would expect them to hit software compatibility problems.
That's what the sane part of the Amiga world has been telling for a while.
:)
> I suppose since AmigaOS never really did worry about FPU (most Amigas
> never had one), a powerpc build that doesn't use the FPU much would
> run on either, and they could even use math libraries (which I seem to
> recall AmigaOS always did anyhow) to hide the difference in architecture
> of the hardware.
That doesn't apply to Amiga PowerPC software. All Amiga PowerPC systems
and expansion cards from the very early days had a full standard PPC FPU.
(We are talking about almost two decades of compatibility in this case.)
So this board will be an exception in every possible sense, and either
incompatible with existing software base, or running them with a huge
penalty hit. Heck, almost all performance-intensive PPC software is full
with hand-optimized stuff like memcopy with FPU, and so on, to squeeze
out all bits of performance from the previous low-end boards which were
the only ones available...
> The only way I see they could be running Debian on that board, is with
> a custom built kernel with MATH_EMULATION enabled. That might be what
> they have done. I do see one of the blog pages running powerpcspe on
> it instead, which makes more sense.
Yes, the A1222 is running powerpcspe, although some ran the normal powerpc
kernel as well with math emu, but it's "not recommended", AFAIK.
For AmigaOS, where most of the software, esp. legacy ones only exist in
binary distributions, they always make an analogy with the "software
supported" FPU of the 68040 and 68060 CPUs (in short for those unfamiliar:
those CPUs never implemented the full 68881 FPU instruction set, mainly
trigonometry-related instructions were missing, and they were supported by
a software library provided by Motorola for user space apps), meaning that
because that was working reasonably well, this should be working fine too
in the end.
Obviously anyone with that reasoning doesn't get the depth of the problem
in this case.
> But at least you can run powerpc on the X1000 and X5000, even if the
> A1222 can without a custom kernel and a performance hit on FPU code.
I'd say a "performance hit" is quite an understatement, given the existing
software base. But we'll see. At least not many of their betatesters is
so willing to post benchmarks, which is always an answer to many
questions...
> Too bad in the past at least they cost way more than I was willing to pay.
Yeah, same here.
Charlie
Reply to: