obanite said:
That's no different to writing code for vector units on the PS2 though...

the real challenge with making code work quickly on the PS3 is getting CPU intensive stuff threaded and onto the SPU's, it's not really a compiler issue. e.g. here we have skinning on an SPU.
It might not be any harder than coding for the EE, but bit-twiddling of individual registers isn't fun in anyone's book; making all that shit happen in a multithreaded model is a fucking nightmare.
Neither is the issue just about getting a particular subroutine running on an SPE, cos I imagine that's impractial. The SPE's are SIMD units and would suck at heavy integer work. Any particular code path/thread is likely to have datasets that need to be worked on by both the PPC core and the SPE's, so you need to work out a way to copy the data into the SPE, wait for it to do its thing and then incorporate it back into the core, all the while having to think about locks, mutexes
and making your code run efficiently on an in-order architecture.
And yes, everyone does love to make everyone think they're an expert on low-level assembly

I'm not, but I'm hoping I know enough about the general topic to know when to say "this shit is hard"
Garf, I don't really give a crap about the PS3 or the games debate in general; I just like talking about techy shit. The PS3 just makes for the most interesting console to talk about cos it's the most "out there" in terms of its hardware (the 3-core CPU inside the Xbox 360 is interesting too, but it's nowhere near as radical as the Cell architecture). All I was doing originally was pointing out that a list of arbitary numbers does not a good console make, I didn't mean to turn this into another thread on the semantics of console design, but it has. I'm not even the sort of person who'd say "I'm not buying a console because the CPU is shit" (heck, I bought an N64 and that had design problems coming out of it's ears), I just find it interesting to talk about.