Sandia OpenSHMEM v1.4.0 Alpha 1

The Sandia OpenSHMEM development team is pleased to announce a pre-release of SOS v1.4.0 (Alpha 1). Below are some of the changes included in this release:

• Full support for the OpenSHMEM 1.4 Specification. Here is the latest release candidate for the 1.4 specification document: OpenSHMEM-1.4-rc3.pdf
• New features include: Thread safety, communication management API (contexts), test routines, sync routines, calloc for symmetric objects, bitwise atomic operations, and updated C11 generic selection bindings.
• Deprecations include: Fortran API, shmem_wait routines, mpp header directory, and SMA-prefixed environment variables.
• This is an alpha release; full API support is provided, but the implementation may contain bugs and is not yet optimized for performance.

Sandia OpenSHMEM v1.3.4

The Sandia OpenSHMEM development team is pleased to announce the release of SOS v1.3.4. Below are some of the changes included in this release:

• Added manpages for OpenSHMEM API routines.
• Added polling limit control variables SMA_OFI_TX_POLL_LIMIT and SMA_OFI_RX_POLL_LIMIT (see README for details) to allow control over quiet and wait polling limits, respectively.
• Rewrite of environment variable management code and improved info output.
• Added support for processor atomics in thread-safe build.
• Added support for local communication optimization (--enable-memcpy).
• Added support for libfabric FI_MR_RMA_EVENT memory registration mode.
• Added experimental support for the PMIx process management interface (--with-pmix).
• Fixed completion race in multithreaded mode.
• Fixed bugs in support for zero-length point-to-point and collective communication.
• Multiple bugfixes to test suite, including GUPS benchmark

Sandia OpenSHMEM v1.3.3

The Sandia OpenSHMEM development team is pleased to announce the release of SOS v1.3.3. Below are some of the changes included in this release:

• Fixed dissemination barrier race that could lead to incorrect synchronization.
• Improved collect and all-to-all collectives performance.
• Improved error reporting and debugging output.
• Added support for proposed bitwise atomics (see shmemx.h).
• Added shmemx_pcontrol function to profiling interface extension (see shmemx.h).
• Added symbol hiding support in SHMEM library.
• Fixed symmetric heap corruption bug in GUPS benchmark (see test/apps/gups.c).
• Added target-side measurement and asymmetric configuration support to performance suite benchmarks.

Sandia OpenSHMEM v1.3.3 RC1

The Sandia OpenSHMEM development team is pleased to announce the release of SOS v1.3.3 release candidate 1 (RC1). Below are some of the changes included in this release:

• Fixed dissemination barrier race that could lead to incorrect synchronization.
• Improved collect and all-to-all collectives performance.
• Improved error reporting and debugging output.
• Added support for proposed bitwise atomics (see shmemx.h).
• Added shmemx_pcontrol function to profiling interface extension (see shmemx.h).
• Added symbol hiding support in SHMEM library.
• Fixed symmetric heap corruption bug in GUPS benchmark (see test/apps/gups.c).
• Added target-side measurement and asymmetric configuration support to performance suite benchmarks.

Sandia OpenSHMEM v1.3.2

• Improved support for proposed multithreading extension (see shmemx.h), enabled at compile time via --enable-threads.
• Enabled single-process, direct execution of SOS binaries with simple PMI.
• Added argument error checking for all SHMEM routines, enabled at compile time via --enable-error-checking.
• Multiple build system improvements, including support for VPATH builds.
• Added new C and Fortran bindings generator that generates all headers and bindings, including profiling interfaces.
• Added support for Fortran complex reductions API.
• Updated Fortran bindings to use short (OpenSHMEM style) header by default.
• Updated SHMEM_DEBUG output to include detailed build information.
• Added --enable-completion-polling build option to poll in quiet/fence operations rather than waiting. This can improve performance for libfabric providers that require software-generated progress.
• Added --disable-bounce-buffers build option to disable bounce buffering by default (i.e. set SHMEM_BOUNCE_SIZE to 0 by default).
• Improved library path propagation (rpath) in compiler wrappers.
• Improved PMI simple build and fixed integration of libpmi_simple library.
• Update symmetric heap allocator to dlmalloc v2.8.6.
• Update PMI-1 client library from MPICH.
• Improved bandwidth efficiency and fixed bug in collect routines.
• Fixed several bugs in tree-based collectives when using PE active sets.
• Fixed several bugs in recursive-doubling reduction routine when using PE active sets and when source and target buffers overlap.
• Fixed synchronization bug in memory management routines.

Sandia OpenSHMEM v1.3.2 RC2

• Improved support for proposed multithreading extension (see shmemx.h)
• Enabled single-process, direct execution of SOS binaries with simple PMI
• Added argument error checking for all SHMEM routines, enabled at compile time via --enable-error-checking.
• Multiple build system improvements, including support for VPATH builds
• Added new C and Fortran bindings generator that generates all headers and bindings, including profiling interfaces.
• Added support for Fortran complex reductions API
• Updated Fortran bindings to use short (OpenSHMEM style) header by default
• Updated SHMEM_DEBUG output to include detailed build information
• Added --enable-completion-polling build option to poll in quiet/fence operations rather than waiting. This can improve performance for libfabric providers that require software-generated progress.
• Improved library path propagation (rpath) in compiler wrappers
• Improved PMI simple build and fixed integration of libpmi_simple library
• Update symmetric heap allocator to dlmalloc v2.8.6
• Update PMI-1 client library from MPICH
• Improved bandwidth efficiency and fixed bug in collect routines
• Fixed several bugs in tree-based collectives when using PE active sets
• Fixed several bugs in recursive-doubling reduction routine when using PE active sets and when source and target buffers overlap
• Fixed synchronization bug in memory management routines

Sandia OpenSHMEM v1.3.2 RC1

• Enabled single-process, direct execution of SOS binaries with simple PMI
• Enabled VPATH builds
• Added argument error checking for all SHMEM routines, enabled at compile time
  via --enable-error-checking.
• Added new C and Fortrans bindings generator that generates all headers and
  bindings, including profiling interfaces.
• Added support for Fortran complex reductions API
• Updated SHMEM_DEBUG output to include detailed build information
• Added --enable-completion-polling build option to poll in quiet/fence
  operations rather than waiting. This can improve performance for libfabric
  providers that require software-generated progress.
• Update symmetric heap allocator to dlmalloc v2.8.6
• Update PMI-1 client library from MPICH
• Improved bandwidth efficiency and fixed bug in collect routines
• Fixed several bugs in tree-based collectives when using PE active sets
• Fixed several bugs in recursive-doubling reduction routine when using PE
  active sets and when source and target buffers overlap
• Fixed synchronization bug in memory management routines

Sandia OpenSHMEM v1.3.1

• Support for allocating the symmetric heap using Linux huge pages.
• Improved support for Cray XC platforms.
• Improved support for variations in atomics support across OFI providers.
• Initial support for thread safety in the OFI transport.
• Fix several issues with C/C++ bindings and profiling interface.
• Added new benchmarks, unit tests, and examples.

Sandia OpenSHMEM v1.3.1 RC1

• Support for allocating the symmetric heap using Linux huge pages.
• Improved support for Cray XC platforms.
• Improved support for variations in atomics support across OFI providers.
• Initial support for thread safety in the OFI transport.
• Fix several issues with C/C++ bindings and profiling interface.
• Added new benchmarks, unit tests, and examples.

Sandia OpenSHMEM v1.3.0

• Support for OpenSHMEM 1.3 specification, including nonblocking communication, fetch/set atomics, all-to-all collectives, C11 generic bindings, and addition of const/volatile to C API.
• Improvements to error checking of OpenSHMEM calls and internal debugging enabled through --enable-error-checking configure option.
• Improvements to C++ compatibility, including support for type-generic bindings in C++.
• Support for fabric and domain selection in the OFI transport, see SMA_OFI_DOMAIN/FABRIC environment variables in the README file for details.
• Support for systems using the Aries network via the OFI GNI provider.
• Enabled XPMEM support in combination with the OFI transport.
• Support for PMI2-compliant process managers.
• The OFI transport automatically falls back to software reductions when the provider does not support the particular combination of atomic operation and datatype for a given OpenSHMEM reduction.
• Added multiple communication performance benchmarks to the test suite.
• Multiple bug fixes and improvements to stability and performance.