Hi Bill, On Fri, May 14, 2010 at 3:28 PM, Bill Hart <goodwillh...@googlemail.com> wrote: > If I make a couple of simplifications, namely assume that the output > fits into two limbs, and that none of the polynomials has length > > 2^32 - 1, etc, I get pretty good times, certainly better than reported > in Francesco's paper. I also don't know if any assumptions about the > coefficients being unsigned are made in any of the benchmarks. That > would further reduce the times, etc.
Just for the record, in the paper the only integer coefficients I consider in the benchmarks are plain out-of-the-box GMP mpz types (well, the C++ interface to them to be precise :). So I imagine that such coefficients will incur in a lot of overhead, from heap allocation to function calling overhead, suboptimal cache usage from random memory accesses etc. For my own purposes, right now I'm mostly interested in double-precision floating point real and complex coefficients, but I do intend to perform benchmarks/optimisation on fixed-width multi-precision integers in the future. Cheers, Francesco. -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
