|
Many many improvements and bug fixes since the last packaged
release. A partial list is:
atlc should no longer fail any test on a multi-processor
system configured with --with-threads. The algorithm used
for both single processor and multiple processor (i.e.
threaded code is the same).
The -t option to atlc, which sets the number of threads
when configured for multiple processors can be set to 0
to use entirely the single-threaded algorithm. If set to
one, it will use the multi-threaded algoritm, but use
only one thread. If set to some other number, it will use
that number of threads and be optimal for the same number
of cpus.
Added some support for gathering hardware data under Linux.
Removed MPI support, as its not working at all.
Changes made to the code to remove the need for a type long long,
which should make the code more portable.
Some changes have been made to the bests so that when
the benchmark runs it should not produce junk for
the hardware information on any system. Previosly
is could create a lot of junk, that was all wrong.
uname is now only called once from try_portable.c
and the splatform specific stuff computed elsewhere
atlc now runs on anything from a toy to a supercomuter! Yes, that is
right. An early version of atlc has been run on a Sony Playstation 2
games console and version 4.4.0 has been tested on the Cray Y-MP
supercomputer!! It has also been run on a very large number of other
UNIX systems, so is hopefully very portable.
An option that was previously reccommended --enable-hardware-info
has been removed. It is now enabled by default, but can be
over-ridden with --disable-hardware-info.
Added a system call to get the number of configured
processors online in Linux. This seems to be undocumented
so it not without its risks, but it seems to work okay on the
limited number of systems tested on.
Someone has done a Windoze port of atlc. Appently it took just 5
minutes, from start to finish. A single bug was found that prevented
atlc compiling, but that was fixed - it needed a left brace removed.
This had never been seen on a UNIX system, since the offending code
was between a couple of #define's.
It has been bought to my attention that bitmaps created with Photoshop
prior to version 7.01 could not be read by atlc. This was not a
fault of Photoshop, but of atlc, so that bug has been fixed.
NEWS for realease 4.6.0 Nobember 2003.
This is a very different from the last release (4.5.1) in
two very important ways.
1) The basic accuracy for single dielectrics has been improved.
Now typical errors are only around 0.1%
2) I have re-enabled the calculation of multiple dielectrics
which were disabled due to accuracy concerns. I'm still not
100% happy with the algorithms, but on tests with a dual
coaxial cable with two dielectrics shows errors of under 2%,
I hope to improve this further at a later date.
|
