Bill
- removed reference to outdated machines (introduction)
Changed:
These include everything from tightly integrated MPPs such as the
Intel Paragon and the Meiko CS-2 to heterogeneous networks of
workstations.
To:
These include everything from tightly integrated MPPs
to heterogeneous networks of
workstations.
- deleted (section 4.2.2)
\begin{implementors}
A virtual machine abstraction is appropriate in most environments
and should be part of a ``high-quality'' implementation. In other
cases, providing the virtual machine abstraction may add too much
baggage (e.g., daemon processes) or inhibit portability.
\end{implementors}
- changed (section 4.2.2)
"An attribute variable MPI_UNIVERSE_SIZE"
to
"An attribute MPI_UNIVERSE_SIZE on MPI_COMM_WORLD
- added (section 4.3.2)
An implementation should document how executables are found and
working directories are determined.
- added (section 4.3.2)
The ordering of processes in the local and remote groups is the
same as the as the ordering of the group of the spawning communicator
and \const{MPI\_COMM\_WORLD} of the spawned processes, respectively.
[it should be possible to word this more clearly...]
- replaced (section 4.3.2)
MPI provides a predefined communicator \const{MPI\_COMM\_PARENT}.
If a process was started with \mpifunc{MPI\_SPAWN} or
\mpifunc{MPI\_SPAWN\_MULTIPLE}, \const{MPI\_COMM\_PARENT} is
the other side of the intercommunicator returned by those
calls. Otherwise, \const{MPI\_COMM\_PARENT} is an
intercommunicator with a remote group size of 0.
with
\MPI/ provides a predefined intercommunicator \const{MPI\_COMM\_PARENT}.
The local group of \const{MPI\_COMM\_PARENT} is identical
(\mpifunc{MPI\_GROUP\_COMPARE} returns \const{MPI\_IDENT}) to the group
of \const{MPI\_COMM\_WORLD}. If a process was spawned with \mpifunc{MPI\_SPAWN}
or \mpifunc{MPI\_SPAWN\_MULTIPLE}, the remote group is identical to the
group of the spawning parent communicator. Otherwise, the remote group
of \const{MPI\_COMM\_PARENT} has size zero (see \mpifunc{MPI\_COMM\_REMOTE\_SIZE}).
- fixed
Example 4.2 to say "SPAWN_MULTIPLE" instead of "SPAWN", and
use the correct first argument.
- deleted
Section 4.3.4: "Nonblocking requests" containing ISPAWN
and ISPAWN_MULTIPLE.
- fixed manager-worked example in section 4.3.6
Used MPI_UNIVERSE_SIZE attribute incorrectly.
Had an unnecessary and confusing comment
- fixed SPMD self-starting code example, page 60
Same UNIVERSE_SIZE problem.
- deleted
Section 4.4.5 "non-blocking routines".
- added
MPI_TRYACCEPT, for discussion.
- fixed
most of the examples (there is still a problem with MPI_NAMES_ARE_PUBLISHED)
- added
Not allowed to call MPI_COMM_DISCONNECT on MPI_COMM_SELF.
- revised
this text as requested by forum:
It is the user's responsibility to complete or free all outstanding
requests associated with a communicator before calling
\mpifunc{MPI\_COMM\_DISCONNECT}. This may be done by
by \mpifunc{MPI\_REQUEST\_FREE} or some variant
of \mpifunc{TEST} or \mpifunc{WAIT}.
- added
MPI_INFO proposal from Rolf.
- clarified
description of MPI_UNIVERSE_SIZE, to make it clear that
in C the attribute is a pointer to an integer.
- changed
all C++ bindings to new appendix C style.
Note: spawn and accept calls changed to methods on Comm::Intra, not Comm::