Hui Zhou
Hui Zhou
## Pull Request Description Upgrade libfabric to mpich/v1.20.1. Fix libfabric API symbol visibility so the symbols are only hidden under embedded mode. Merge all mpich commits into a single commit...
## Pull Request Description Port the ofi GPU pipeline code into using MPIR_aysnc_things (https://github.com/pmodels/mpich/pull/6841). [skip warnings] ## Author Checklist * [x] **Provide Description** Particularly focus on _why_, not _what_. Reference...
## Pull Request Description Having multiple non-intercepting code paths to use the same structure fields in request makes code difficult to read. This PR separates the fields into distinct union...
## Pull Request Description While [OSU micro benchmarks](https://mvapich.cse.ohio-state.edu/benchmarks/) are commonly used, they still require separate builds and its comprehensive inclusions of many options make using and tuning the benchmarks non-trivial....
Raised by @jeffhammond during this week's MPICH development update call, currently, it is messy for users to figure out how to compile their GPU code with MPI. Using CUDA for...
We should add the PMI name publishing interface, e.g. `PMI_Publish_name`, `PMI_Unpublish_name`, and `PMI_Lookup_name` in `src/pm/util/pmiserv.c`, which `gforker` and `remshell` uses.
On 32bit nightly test using sockets provider, the test ``` export MPIR_CVAR_BCAST_OFI_INTRA_ALGORITHM=trigger_tree_tagged export MPIR_CVAR_CH4_OFI_ENABLE_TRIGGERED=1 export MPIR_CVAR_BCAST_TREE_TYPE=knomial_1 export MPIR_CVAR_BCAST_TREE_KVAL=2 mpirun -l -n 4 ./bcasttest ``` may get stuck in `fi_cntr_wait` after...
The configure option `--enable-fast=all` has been too slow to build -- likely due to forced inlining. The last Jenkins good build was on Jun 30, 2023.
The routine is defined in `mpir_request.h` as: ``` MPL_STATIC_INLINE_PREFIX int MPIR_Request_is_anysrc_mismatched(MPIR_Request * req_ptr) { return (MPIR_CVAR_ENABLE_FT && !MPIR_Request_is_complete(req_ptr) && MPID_Request_is_anysource(req_ptr) && !MPID_Comm_AS_enabled((req_ptr)->comm)); } ``` The idea is, for an anysource...
The following tests: ``` errors/comm/subcomm_abort 4 errors/comm/subcomm_abort2 6 errors/comm/intercomm_abort 4 ``` have processes call `MPI_Abort` on a subcomm (or an intercomm) and expect only the processes in the subcomm to...