Cell Questions

[Gholbine]

So I'm writing a paper on the Cell, and I'm having trouble finding information about certain things.

Namely, the arithmetic units in the SPU. Does anybody know where to find detailed information on them? In particular, when it comes to single-precision floating-point, what are the derivations from the IEEE754 standard and how does it increase performance? How was the double-precision performance sacrificed for increased single-precision performance?

Thanks in advance.

 

[drpepper]

Start here: http://www.research.ibm.com/cell/

This may also help: http://www-128.ibm.com/developerworks/power/cell/

 

[3dcgi]

I don't know how the SPE's deviate from the IEEE standard, but if they do deviate it's not directly to increase performance but rather to save area and complexity. Saving area is also the likely reason to sacrifice double precision performance. If most apps are expected to need only single precision then having full speed double precision wastes area and doesn't make sense.