Quote:
* my impression is that compilers still love the ppc architecture - with all the popularity ARM currently enjoys, it's yet to reach the level of compiler support that ppc (well, book e, at least) has had for long now. it's because the ISA was designed as the sane middle-ground architecture - nothing 'common' is prohibitively complex to implement on it (you need cheap dp fp - sure. you want sane SIMD - got it).
I wouldn't say that. GCC enjoys just as good ARM support as any other compiler. The thing is that more people doing GCC development are more likely to have a PC or Mac - which basically puts them in the x86 or PPC camp.
There aren't a great number of desktop ARM systems in the world but this is changing really fast - especially as boards like Beagleboard, Gumstix, Sheeva etc. get more popular.
LLVM 2.6 will include pretty good ARM7a support and NEON intrinstics. If you hated GCC's code generation, you have an alternative now. By the end of the year it may be possible to build a kernel, and entire userspace using LLVM CLang.
Quote:
* how many non-x86 ISAs can boast with such excellent x86-borrowed pheripheral support?
Anything with a PCI slot. Marvell's ARM platform supports PCI Express. There is nothing stopping you - with the right compatibility technology for which with our firmware is done entirely in software, at a firmware level - plugging in a PCI Express graphics card here and enjoying that same support.
Quote:
thought the whole point of your effort was to cater to the ppc geek crowd, not to seek the greenest pastures (you could just as well re-sell atom boards to that end : )
Except Atom sucks. The great thing about the PPC SoCs is high integration. Atom does not have that. It also has higher performance, an out of order CPU core, AltiVec, and similar power levels.
With the Cortex-A8 you get Atom performance at realistic 1W power levels at the same price. With the Cortex-A9 you get Core Duo performance at Atom power levels for the same price. Several PPC engineers have basically said, this is the end of the platform - there is no reason not to move to Cortex-A9 now except to support existing customers. If your existing userbase is 1500 users with no coherent wants or needs, it's hardly worth designing new chips for them when you can subsist on the old ones. And designing new boards around old chips could be a waste of money when you could migrate them all to new technology (m68k->PPC, PPC->x86..) as Apple did.
I'd be all for a Cortex-A9 desktop board. All we need is a chip that has a couple PCI Express buses.