Options

Experiments in software for approximating non-amenable Markov process transitions

I've added a page Experiments in software for approximating non-amenable Markov process transitions. It's very, very much a work in progress: basically it's reached the stage where things are starting to become vaguely interesting to other people looking at programming. Hopefully I'll improve it and keep it up-to-date as the modeling software evolves, although since I'm forced to work mostly offline at the moment updates will probably be sporadic. (Category is software because it's describing the software side of things, but it's titled an experiment because the software is not finished yet, and may end up being dramatically changed if the current "design" (hah! we don't need no stinking design before we start) proves unable to accomplish certain things.

Comments

  • 1.

    FWIW, currently wondering what to do. The project is very heavily dependent on good floating point optimizations. On the one hand, apparentlly LLVM, which currenlty doesn't do floating point opt, will be getting some but it's maybe 3-4 months way from being in the develpment sources, 6-9 months avay from an end-user release. On the ohter hand, implementing floating point optimizations myself is reaching the point where the complexity is starting to get cumbersome and error prone, particularly since while they're important FP opts are not the "interesting" aspects of this stochastic simulation.

    Comment Source:FWIW, currently wondering what to do. The project is very heavily dependent on good floating point optimizations. On the one hand, apparentlly LLVM, which currenlty doesn't do floating point opt, will be getting some but it's maybe 3-4 months way from being in the develpment sources, 6-9 months avay from an end-user release. On the ohter hand, implementing floating point optimizations myself is reaching the point where the complexity is starting to get cumbersome and error prone, particularly since while they're important FP opts are not the "interesting" aspects of this stochastic simulation.
Sign In or Register to comment.