LCI v2.0.0-dev
For Asynchronous Multithreaded Communication
Loading...
Searching...
No Matches
Changelog v1.7.3

Change log for LCI v1.7.3

Major Changes

  • Integrate PAPI into LCI.
    • Use cmake option LCI_USE_PAPI to enable.
    • Use env var LCI_PAPI_EVENTS to specify events to monitor.
  • Add pmix support for LCI.
  • Update README, doxygen comments, and example codes.

Breaking Changes

  • Change uint32_t rank to int rank in sendl/recvl to match other communication operations.

Other Changes

  • Add cmake option LCI_IBV_ENABLE_TRY_LOCK_QP.
  • Use thread domain when try_lock is enabled.
  • Add more performance counters.
  • Only print performance counter data when LCI_USE_PERFORMANCE_COUNTERS is enabled.
  • Add cmake option LCI_ENABLE_SLOWDOWN and env var LCI_SEND_SLOW_DOWN_USEC and LCI_RECV_SLOW_DOWN_USEC: manually slow down the messages.
  • Change the putva rendezvous protocol: first send FIN and then deregister memory.
  • Make LCI_PACKET_SIZE configurable at runtime.
  • Add some log to help debug the hang (if there is any) in the initialization phase.
  • Only do fetch_and_add on g_next_rdma_key when FI_MR_PROV_KEY is not set.
  • Use the msg_comp_type as the remote default completion type instead of the hardcoded LCI_COMP_QUEUE.
  • Use CSV format for performance counter printing outputs.

Fixed Issues

  • Fix LCIU_timespec_diff.
  • Fix a few alignment bugs.
  • Fix a bug in ibv backend when td is not supported.
  • Fix a putva bug: putva send buffers might never get freed.
  • Fix LCI_TOUCH_LBUFFER when allocating long buffers.
  • Fix compiler non-void function no return error.
  • Provides workaround for libfabric/cxi fi_mr_bind bug.
  • Fix the handler completion mechanism.