No, the optimization problem is to speed up for a given set of parameters, <br>one single expression involving log/exp, addition/subtraction/multiplication/division with huge numbers. <br><br>That's all I need. Once the critical single expression has been optimized, then I can vary the parameters. The current focus is that single expression.
<br><br><div><span class="gmail_quote">On 1/2/07, <b class="gmail_sendername">Jürgen Kahrs</b> <<a href="mailto:Juergen.Kahrs@vr-web.de">Juergen.Kahrs@vr-web.de</a>> wrote:</span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Patrick Pelissier wrote:<br><br>>> (2) We<br>>> have to optimize the code and make it faster by a speedup of hundreds to<br>>> thousands times.<br>><br>> Such a speedup can only be reached by an algorithm optimization (if
<br>> possible).<br><br>That's right. When reading Michael's messages, I always<br>wonder why he insists on speeding up GMP and MPFR.<br>I bet that GMP and MPFR are those parts of his software<br>that already _are_ near the optimum run-time efficiency.
<br><br>As you (Patrick) suggested, Michael should concentrate on<br>speeding up the algorithm that makes use of MPFR. It sounds<br>like his algorithm follows a brute-force approach of trying<br>all possible combinations of values of some variables.
<br></blockquote></div><br>