**Mail Archives: djgpp/2000/06/20/17:18:42**
I have finished transcribing my genprec benchmark (gnlindex.cpp
which is part of the genprec package available at genprec.home.att.net)
to gmp. This benchmark tests only floating point +,-,*, and /.
gmp produces an executable which is twice as fast as the executable
produced by genprec. Obviously, the author(s) know something about
arithmetic which I don't (most likely an algorithm for dividing two
multiple precision floating point numbers).
However, genprec deals with real and complex numbers and all
of the elementary transcendental functions (sin, exponential, atan,
and ln functions, etc) with real and complex arguments. Furthermore,
it contains a package of applications (such as the solution of
polynomials with real or complex coefficients). Lastly, the
interface is much more natural since operator overloading is
possible in C++, so that one can write things like
a=b+c+d+e;
(which requires three lines in gmp) and
a=sin(b)+cos(c)+ln(d)+atan(e);
which is not possible in gmp.
gmp contains things not contemplated in genprec, such as the
factorization of a large integer.
A factor 2 is quite valuable. I would like to see gmp
rewritten in C++, complex numbers introduced, and the elementary
transcendental functions included in the package. Maybe some-
body has already done something which I do not know about or is
doing something. Let me know.

*- Raw text -*