E-mail
Last modified - Oct 28, 2002
Fast exponent functions
Using table-driven decomposition, it is possible to make exp() function,
which if 2-10 times faster, then native exp() of Pentium(R) processor.
As an example, the sample function t2exp() provides for 18-bit precision
within the argument range -64..0.
The function was developed in cooperation with
Phillip S. Pang, Ph.D. Biochemistry and Molecular Biophysics.
Columbia University. NYC.
The package also contains our implementation of algorithm
fexp() from Nicol N. Schraudolph, which does not use any tables.
This algorithm provides for 11-bit precision for the range -709..+709
Basic code is written on ANSI C for generic processor
and tested under Borland C++ 3.1 and 4.5.
Optimized code is written for Visual C++ 6.0 using
assembler-level instructions of Pentium(R) processor.
Download: Source code