I would like to hear various opinions on autotuners [http:// view.eecs.berkeley.edu/wiki/Autotuners ] in mathematical libraries/sw. One successful we already have - ATLAS. Also various approaches with evolutionary algorithms seem to be interesting and rewardful.
Since I lack a deep knowledge of EA`s, I`m asking why not give a program some basic algorithm to do some bignum operation, specify the resources it is allowed to use (CPU cores and/or GPU, RAM) with the task to improve overall computing time and wait for the result(s). As always, the devil is/could be in details, but the resources are certainly out there - http://fah-web.stanford.edu/cgi-bin/main.py?qtype=osstats . An example of a success story (in physics though): http://www.fmf.uni-lj.si/si/obvestila/12128/ They seem to work also for GPUs and Cell (some benchmarks): http://www.spiral.net/bench.html Some interesting papers: http://www.cs.rice.edu/~ken/autotuning/ Sincerely, Jernej --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "mpir-devel" group. To post to this group, send email to mpir-devel@googlegroups.com To unsubscribe from this group, send email to mpir-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/mpir-devel?hl=en -~----------~----~----~----~------~----~------~--~---
