Документ взят из кэша поисковой машины. Адрес
оригинального документа
: http://sp.cs.msu.ru/dvm/dvmhtm1107/downloade/versione_history.html
Дата изменения: Mon Feb 13 12:59:04 2006 Дата индексирования: Mon Oct 1 23:11:59 2012 Кодировка: Windows-1251 |
DVM-system. Version history
First version of DVM-system appeared in October 1998. It contained the following components:
Fortran-DVM and C-DVM languages supported data distribution by equal blocks and corresponding distribution of parallel loop iterations (data parallelism).
Second version of DVM-system appeared in April 1999. A new toolkit - DVM-program performance analyzer was included in the version.
Third version of DVM-system appeared in November 1999. Another toolkit - DVM-program performance predictor was included in the version.
Fortran-DVM and C-DVM languages support now data distribution by non-equal blocks and parallel executed program section specification (task parallelism).
The version 3.15 (11.07.2000) was first version of DVM-system in source files and with complete set of documentation in Russian and English accessible via INTERNET.
The version 3.30 (27.12.2000) is differ from version 3.15 in the following:
The version 3.60 (9.07.2001) is differ from version 3.30 in the following:
The version 3.82 (13.02.2002) is differ from the version 3.60 in the following:
The
version 3.93
(30.09.2003) is differ from the version 3.82 in the following:
The exchange by packed messages is implemented.
To increase efficiency of cash-memory usage the tools of automatic correction to allocate in memory the multi-dimensional array elements has been provided.
Performance
of DVM-programs has been increased on Alpha 21264, Intel Itanium 2 and IBM
POWER4 processors.
The
version 3.95
(14.09.2004) is differ from the version 3.93 in the following:
Fortran-DVM
is extended by some facilities of Fortran 90 (dynamic arrays, array
operations, derivative types, modules and ets.).
Fortran-DVM
and C-DVM languages supports new type of data distribution - distribution by
blocks which size is multiple to specified integer.
The
number of inaccuracies complicating the system implementation on
Linux-clusters, is eliminated.