PSM2_12.0.1

What's Changed

• Fix memory leak in psmi_shm_create by @Kamalheib in #66
• Disable rpmbuild pulling the build date from libspm2.spec's changelog, by @gregkresge-cornelisnetworks in #68
• Reset rpm_release_extension to "1" for psm-v12.0 series by @BrendanCunningham in #67

New Contributors

• @Kamalheib made their first contribution in #66
• @gregkresge-cornelisnetworks made their first contribution in #68

Full Changelog: PSM2_11.2.230...PSM2_12.0.1

PSM2_11.2.230

What's Changed

• Fix silent data error on FECN-marked TID packet for unaligned receive. by @BrendanCunningham in #65

Full Changelog: PSM2_11.2.229...PSM2_11.2.230

PSM2_11.2.228

Full Changelog: PSM2_11.2.205...PSM2_11.2.228

PSM2_11.2.206

Fix for build problem with newer versions of gcc.

PSM2_11.2.205

Updates the PSM2 library to match release 11.2.205

Corresponds to IFS release OPENIB_INSTALL_LINUX_opa-10_11_0_1_10_11_0_1_2 with additional post-release fix.

Prior PSM2 11.2.203 release should have had a version greater than 11.2.204. This oversight meant that packages built with 11.2.203 would not install as newer than 11.2.204 even though code was newer.

PSM2_11.2.202

Corresponds to IFS release OPENIB_INSTALL_LINUX_opa-10_11_10_11_0_0_578
with additional post-release fix.

Changes:

• Check PSM2_GPUDIRECT_RDMA_SEND, PSM2_GPUDIRECT_RDMA_RECV before
  disabling GDRCopy when PSM2_GPUDIRECT=0.
• Remove bashisms from Makefile. H/t Brian T. Smith.
• Fix syntax errors in Makefile. H/t Chuck Cranor.
• Replace CUDA events with cuStreamSynchronize. H/t Jonas Hahnfeld.
• Lazily allocate buffers, events, and streams for GPUs. This allows
  PSM2 to be initialized before GPUs are chosen. H/t Jonas Hahnfeld.
• Remove ad_context from psmi_mpool_create_for_cuda; ad_context was not
  used. H/t Jonas Hahnfeld.
• Other CUDA code cleanup and improvements to allow CUDA deferred
  initialization.
• Fix crash on uninitialized CUDA IPC variable. H/t Rémi Lacroix.
• Other minor fixes.
• Update contributors.

Signed-off-by: Brendan Cunningham [email protected]

PSM2 w/changes for PSM2-NCCL plugin support

Interim release for users of PSM2-NCCL plugin.

Not needed if you are not using the PSM2-NCCL plugin.

PSM2_11.2.185

Corresponds to IFS release OPENIB_INSTALL_LINUX_opa-10_10_10_10_3_0_12

Changes:

• Add support for clang.
• Add support for gcc 10.
• Change default value of MQ_HFI_THRESH_EGR_SDMA_SQ_XEON to 16384.
• am_cuda_memhandle_cache: add cache clear counter.
• Other minor fixes to am_cuda_memhandle_cache code.
• Update contributors.

Signed-off-by: Brendan Cunningham [email protected]

PSM2_11.2.166

Updates the PSM2 library to PSM2 11.2.166.

Corresponds to IFS release
OPENIB_INSTALL_LINUX_opa-10_10_2_2_10_10_2_2_1. Also corresponds to PSM2_11.2.166.

am_cuda_memhandle_cache changes:

• Reduce CUDA IPC cache evictions by using EPID as part of cache key.
  Only applicable when multi-EP is disabled.
• Fix problem where cache did not consider that new cache enrtry
  could span/collide with more than one existing entry.
• Fix incorrect linked-list updates that could cause an entry other than
  the least-recently-used to be evicted when the cache is full.
• Handle CUDA IPC open function returning CUDA_ALREADY_MAPPED error by
  clearing cache and retrying open call.
• Always compile cache counters; add cache counters for evictions,
  collisions, and cache clears.
• Refactor, hide internal data and interfaces.

Other changes:

• Fix problem that could occur when multiple HFIs are present and
  lower-ordered HFI(s) are inactive.

Note this release has the CUDA fixes from CudaMemhandleCacheImprovement. This release also has the multi-HFI fix from PSM2 11.2.173. This release does not have the libpsm2 reference counting implementation from 11.2.173.

Signed-off-by: Brendan Cunningham [email protected]

PSM2_11.2.173

This is an interim release of PSM2 needed to support Open-MPI 5.x development.

This release has no effect on other applications.