Hi all, I'm currently looking for fast ways to sum exponential-type Taylor series in fixed point arithmetic using mpz (see http://code.google.com/p/fastfunlib/)). I'm mainly interested in low precision (say a few hundred or thousand bits), so I don't think binary splitting is worthwhile (but I could be wrong here).
The most important missing ingredient is a function that does mpz_mul(z, x, y); mpz_tdiv_q_2exp(z, z, p); efficiently in a single step. For other purposes, it should ideally only use the necessary high limbs of x and y so that x can be a fixed point value at much higher precision, without the need to place a truncated copy in a temporary variable. If it helps, it would be possible to choose p so as to be a multiple of the limb size. Secondly, there are a lot of divisions by small integers (1, 2, 3, ...). I'm not sure how much they contribute in total, but if there is a way to speed up the first few with precomputation, it could help. A single-step x = x + y/k, k a small integer, would also be useful if it could be done faster. Finally, for fixed-point square root, it can probably be done faster than padding with zeros and performing an integer square root. Any other ideas? Fredrik --~--~---------~--~----~------------~-------~--~----~ 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 -~----------~----~----~----~------~----~------~--~---
