Jump to content

  • Log In with Google      Sign In   
  • Create Account


#ActualStroppy Katamari

Posted 29 December 2012 - 05:35 AM

How in-depth is this supposed to be? Are you expected to use an hour, a day, or a week on this? And how well are you expected to know assembler at this point?

Anyway, I don't know much assembler either, but to answer this on a superficial level you don't really need to.
- look at the relevant algorithm/situation (you mention cloth simulation, but if you have a choice, take a simpler particle system instead)
- on paper, roughly break it down to individual operations (multiplies, adds, ...)
- take some super-scalar instruction and figure out which steps you could replace with it
- look at the specs of some common core with that instruction; you want to know how many simple operations and how many of the new instructions it can carry out
- based on that, and your on-paper-optimized version of the algorithm, you can estimate the speedup

#1Stroppy Katamari

Posted 29 December 2012 - 05:31 AM

How in-depth is this supposed to be? Are you expected to use an hour, a day, or a week on this? And how well are you expected to know assembler at this point?

Anyway, I don't know much assembler either, but to answer this on a superficial level you don't really need to.
- look at the relevant algorithm/situation (if you have a choice, e.g. a simple particle system with no inter-particle interaction would be easier than a cloth simulation)
- on paper, roughly break it down to individual operations (multiplies, adds, ...)
- take some super-scalar instruction and figure out which steps you could replace with it
- look at the specs of some common core with that instruction; you want to know how many simple operations and how many of the new instructions it can carry out
- based on that, and your on-paper-optimized version of the algorithm, you can estimate the speedup

PARTNERS