Index of /dist/other/ammpi

[ICO]NameLast modifiedSize

[PARENTDIR]Parent Directory  -
[   ]Makefile.am2018-07-19 12:45 5.0K
[   ]Makefile.tests2018-07-19 12:45 3.6K
[   ]Makefile.titanium.in2018-07-19 12:45 794
[TXT]README2018-07-19 12:45 5.0K
[TXT]ammpi.h2018-07-19 12:45 48K
[TXT]ammpi_ep.c2018-07-19 12:45 52K
[TXT]ammpi_internal.h2018-07-19 12:45 17K
[TXT]ammpi_reqrep.c2018-07-19 12:45 50K
[TXT]ammpi_spmd.c2018-07-19 12:45 19K
[TXT]ammpi_spmd.h2018-07-19 12:45 4.6K
[TXT]license.txt2018-07-19 12:45 1.9K
[   ]Makefile.in2018-07-19 13:22 32K

AMMPI Documentation
===================
Author: Dan Bonachea
Contact email: gasnet-devel@lbl.gov
Home page: http://gasnet.lbl.gov/ammpi

AMMPI is an implementation of the AM-2 specification over MPI-1.

AMMPI is a portable implementation of the AM-2 specification that runs on MPI-1.1
message-passing operations. It does not currently use MPI-RMA in any way.

The intent is a fully-portable implementation that will work on any system with
a compliant MPI implementation.  We don't expect to achieve latency performance
competitive with a native implementation of AM optimized for special-purpose
hardware, instead we seek to provide a compatibility layer that will allow
AM-based systems to quickly get up and running on any HPC platform. 

AMMPI is a production-quality AM layer that provides a number of features in
addition to the standard AM-2 API, including a helper library for SPMD job
startup and an API to access built-in networking utilization and performance
statistics. It has been successfully ported to a number of parallel
architectures, and porting to new architectures generally requires minimal
effort. AMMPI performs all its MPI calls on a separate, private MPI
communicator, which is strongly guaranteed by the MPI spec to isolate it from
any other MPI communication in the application, so there's never a possibility
of deadlock or hard-limit resource contention between the two.

AMMPI is used to implement the GASNet communication system (http://gasnet.lbl.gov). 
Further documentation about using AMMPI is available in the mpi-conduit
documentation: http://gasnet.lbl.gov/dist/mpi-conduit/README

Requirements
------------

* MPI 1.1 or later implementation, with a C99 compiler

* GNU Make and basic UNIX tools for the provided Makefiles

Limitations
-----------

AMMPI is a purely polling-based implementation of Active Messages, in that incoming
active messages are only serviced inside synchronous calls to the AMMPI library from
the local process. This can lead to increased latency for those requests when the local
node is inattentive to the network.

AMMPI has a few notable departures from the AM-2 specification:

* The client must call AM_SetExpectedResources() exactly once on an endpoint
  after setting up the translation table and before making any calls to the
  transport functions (AM_Poll, AM_Reply* or AM_Request*). It is also an error to
  call AM_Map, AM_MapAny or AM_Unmap (which change the translation table) after
  the call to AM_SetExpectedResources().

* AM_PAR bundle/endpoint access is not implemented - this means AMMPI does
  not provide thread safety, but can be used in a thread-funnelled mode with
  locking at the client level.

* The nbytes argument to Medium and Long AM handlers has type `size_t`

* AM_GetXferM is not implemented.

* Clients should not take the address of AM_ entry point "functions", because
  many are implemented as macros.

Change Log
----------
AMMPI 2.9 (06/2018)
- Medium and Long AM handlers change in signature on 64-bit platforms: 
  the nbytes argument now has type size_t instead of int
- New AMMPI_LIBRARY_VERSION_{MAJOR,MINOR} defines
- Significant rearrangement of internal files and utilities
- AM Handler dispatching code is now strictly spec-compliant
- Several pseudo-public programmatic knobs renamed into the AMX_ namespace
- AMX_ProcessLabel now supported for labelling per-node library output
- Bug 2774: source_addr to Med/Long injection is now ignored for nbytes=0
- Fix an unreported cosmetic output bug in AMX_RETURN error reporting

AMMPI 2.8 (03/2017)
- Add AMMPI_FS_SYNC envvar to control exit-time filesystem sync()
- Add AMMPI_MPI_THREAD to control MPI threading mode used to init MPI 2+ in AMMPI_SPMDSetThreadMode
- Fix AM_{Get,Set}NumHandlers and AM_MoveEndpoint
- Export AMMPI_enEqual
- Expose AMX_SetTranslationTag
- Add C++ compatibility to the public header (ie extern "C")
- Renamed portable_platform.h
- Rename some platform Makefile fragments
- Remove support for: Cray T3E, Catamount, native Win32 (use cygwin instead)
- Remove some obsolete files
- make run-tests now outputs summary results at the end
- Fix some harmless check-exports failures, and allow the test to be skipped via CHECK_EXPORTS=0
- Fix some harmless warnings on amxdist
- Respect MANUAL_MPICFLAGS in stand-alone mode
- Remove the debug/opt workaround
- Fix possible syntax errors with AM expression arguments
- Silence the annoying MPICH/Linux _FORTIFY_SOURCE warning
- Use MPI_Comm_set_errhandler for MPI 2+
- Use realloc when appropriate to manage internal state
- Use C99 __func__ when available in error messages
- Fix compile errors for AMMPI_NONBLOCKING_SENDS==0 or AMMPI_PREPOST_RECVS=0
- Remove dead code
- Update URL and contact email

AMMPI 2.7 (10/2010)
- Test for MPI_VERSION >= 2 in threaded utilities
- Fix a harmless warning

AMMPI 2.6 (05/2010)
- Fix an assertion failure on PREDICT_PACKET_LENGTH_

AMMPI 2.5 (10/2009)
- Implement strict prototypes, ie: fn(void)
- Handle compilers that don't support __attribute__((__format__ (__printf__))

AMMPI 2.4 (10/2007)
- Implement AMMPI_SPMDSetThreadMode
- Implement AMMPI_VERIFY_MPI_ORDERING