QUOTE(ConteZero76 @ Apr 18 2007, 05:29 AM)

SPU aren't made for multitasking, you've got to use cooperative multitasking (i.e. the SPU has to switch himself from one task to another using some sort of embedded OS) or via a "brute" switch.
About the rest Linux uses PS3 like every other application, only RSX is half shut down, nothing that should bother the OS.
As for the rest I still doesn't see what's the reason to use two SPU (if you could be kind enough to explain me).
And you should believe me, I'm Claus, Santa Claus.
Actually the SPU can easily be forced to do a context switch, and it takes around 20 microseconds according to this interview with a senior IBM programmer. As 20 microseconds isn't much time, it is possible that Sony does use a second SPU at times.
Also, the PPU can easily control context switches on the SPU, via the MMIO registers. This is public knowledge in the IBM docs.
Even if the SPU is running in isolation mode you can replace the currently running context, assuming the ppu task is in the same LPAR as the SPU.
QUOTE
dW: Looking at the multitasking model -- which has the pre-emptive scheduling -- when it comes to a context change, what's the order of magnitude versus some of the effects like cache misses?
Chow: The context switch for the whole SPE takes about 20 microseconds. The majority of the latency of the context switch save and restore is because you have to swap out the whole local store and then replace that with the new context. It is very expensive compared to the native PowerPC® threads. For the PPE core, you don't have to save and restore the whole execution memory image of the PowerPC task. To context switch an SPE program properly, a lot of activities need to happen.We have to wait for the outstanding DMAs to complete, copy out of all of the entries in the DMA queues, copy out of all the local store and register content... By the way, it is a huge register file. Then you have to bring in the previous context and other context to be swapped back in. It takes a relative long time to complete. That's the reason we usually favor the "run to completion" model.
http://www-128.ibm.c...ary/pa-expert8/