MPFR 2.2.0 Release candidate
Jim White
mathimagics at yahoo.co.uk
Tue Sep 13 18:54:28 CEST 2005
I'm happy to advise successful build and tests of MPFR 2.2.0 release candidate as shared library (DLL) on Windows XP, with MSys.
Notes:
1. Win DLL building no longer requires manual gcc linkage step. The configure script supports --disable-static --enable-shared and that works just like it does for GMP (hmmm, is this really new? Or has it been supported anyway - if so, forgive my being slow to discover it!)
2. For version 2.1.2, I had manually patched the mpfr_const_log2 and mpfr_const_pi functions to cache the results, otherwise functions like mpfr_log became distinctively sluggish at higher precisions (when the proportional cost of calculating log2 and pi on the fly reached and passed a critical level). This has been fixed in 2.2.0, and even when re-evaluation of log2 does have to be done, I see that this calculation has also been greatly improved in the new version (by a factor of 3 to 4 on my system at 50,000 digits precision).
3. On my system, the new mpfr_sin() runs 1.75 to 2 times faster than before, and although that's still slower than a GMP (mpf) version that I have written, the gap has been reduced - I was 2 to 3 times faster before at 10,000 to 100,000 digits, but now I'm only 1.2 to 1.6 times faster.
4. mpfr_exp is more or less the same speed - I have an mpf version that runs 1.5 times faster, although this is in a narrower range (10,000 to 20,000) digits. MPFR changes from a Taylor series to a Brent algorithm at this stage, and my mpf version (which always uses a Taylor-based method) runs at about the same speed at 50,000 digits, and starts to fall behind at 100,000 digits.
Jim White
14 Sep 2005
Canberra
gmp-discuss-request at swox.com wrote:
Today's Topics:
3. MPFR 2.2.0 Release Candidate (Vincent Lefevre)
Date: Tue, 13 Sep 2005 10:13:27 +0200
From: Vincent Lefevre
Subject: MPFR 2.2.0 Release Candidate
To: mpfr at loria.fr
Cc: gmp-discuss at swox.com
The release of MPFR 2.2.0 is imminent. Please help to make this
release as good as possible by downloading and testing the
release candidate:
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://gmplib.org/list-archives/gmp-discuss/attachments/20050913/817db3ba/attachment.html
More information about the gmp-discuss
mailing list