Some benchmarking tools that were used in the development of earlier versions are now publicly available for testing of user-programmed algorithms. This version makes precision checks more admissible, in that the precision of a number can be adapted automatically if necessary. The handling of limit cases in the trigonometric routines has been improved and more carefully tested. As usual, there are also many minor bugfixes and enhancements of the lower level algorithms.
The new naming scheme that was introduced in
version 1.0 is now applied more consistenty. A new
routine, real_c::accumulate, provides a fast
method for implementation of series expansions,
and is likely to be emplyoed in many of the
exponential and trigonometric functions in
upcoming versions of the library. A critical bug
in sqrt has been identified and removed. A
critical bug in atan2 has been identified and
removed.
In this release, a new type naming scheme is employed for the first time. The source tree has been tidied up and the build process streamlined. A simple MPI interface for simplification of number transfer in parallel computations has been added, along with minor technical enhancements.
An improved source tree layout. Sign representation has been simplified. ++ and -- operators have been added to the "real" interface. New simple functions that are very handy, e.g. for comparison of absolute values. Significant improvements in log performance. Improvements to trigonometric functions. Bugfixes in boolean arithmetic and multiplication.
