By using this site, you agree to our Privacy Policy and our Terms of Use. Close
alephnull said:
klystron said:
alephnull said:

It's just a conference paper so he's not spending a whole lot of time on it, but with all these things most of the time is spent on tweaking parameters, measuring op counts, cache misses, runtime, etc.

I've not run Linux on my PS3 (not enough ram to bother) so I'm not sure what profiling tools are available for its architecture. I guess it'd be PPC so probably all the traditional stuff is supported.

There's always gprof or manually reading from this cycle counter-like thing the cell has, but you probably want to use the cell specific CPC which is included in IBM's SDK.

There isn't much ram, but it has high bandwidth; so as long as your working sets fit it's not the end of the world. It's like a cheap Tesla box in that regard.

This is the argument that always kills me.  Not enough RAM?  Maybe not as much as you want, but there is plenty there.  Just remove what you do not want (out of gnome/kde/fluxbox whatever X you are using) and everything will run great.  Yes it takes awhile, but once I had everything up and running it runs wonderfully.