{ "_comment": [ "GIN / rocSHMEM test matrix for run-gin-ci.sh (baremetal translation of", "docker_build_test.bash). Each test runs as:", " mpirun -np $NP <-x env> -x LD_LIBRARY_PATH ", "kind: 'rocshmem' -> $ROCSHMEM_TESTS_BIN_DIR/; 'rccl-tests' -> $RCCL_TESTS_BIN_DIR/", "{E} is replaced with NP*MSG_SIZE bytes. Tests are non-gating for now.", "-D 3 is the rocSHMEM-SDMA path the script labels 'Crash' (kept for visibility).", "debug_env: -x added to every test only when RCCL_CI_DEBUG=1.", " {LOGDIR} in debug_env is replaced by the runner with the per-job debug dir,", " so NCCL_DEBUG_FILE writes per-rank logs there instead of flooding stdout." ], "mca": "-mca pml ob1 -mca btl ^openib", "debug_env": ["NCCL_DEBUG=INFO", "NCCL_DEBUG_SUBSYS=INIT,NET", "NCCL_DEBUG_FILE={LOGDIR}/nccl-debug.%h.%p.log"], "tests": [ { "name": "rocshmem-a2a-sdma", "kind": "rocshmem", "bin": "rocshmem_functional_tests", "env": ["NCCL_DEBUG=VERSION", "ROCSHMEM_DEBUG_LEVEL=info:noversion", "ROCSHMEM_TEST_UUID=1"], "args": "-a 19 -w 1 -z 256 -v {E} -n 100 -noverif" }, { "name": "rccl-gin-hostproxy-d1", "kind": "rccl-tests", "bin": "alltoall_perf", "env": ["NCCL_CUMEM_ENABLE=1", "RCCL_ENABLE_INTRANET=1", "NCCL_DMABUF_ENABLE=1", "HSA_NO_SCRATCH_RECLAIM=1"], "args": "-b 128 -e {E} -f 2 -g 1 -R 2 -D 1 -A 1" }, { "name": "rccl-gin-rocshmem-sdma-d3", "kind": "rccl-tests", "bin": "alltoall_perf", "env": ["RCCL_ROCSHMEM_ENABLE=1", "ROCSHMEM_BACKEND=ipc", "ROCSHMEM_HEAP_SIZE=1073741824", "ROCSHMEM_SDMA_ENABLED=1", "NCCL_GIN_ENABLE=1", "NCCL_GIN_TYPE=4", "NCCL_CUMEM_ENABLE=1", "RCCL_ENABLE_INTRANET=1", "NCCL_DMABUF_ENABLE=1", "NCCL_MSCCL_ENABLE=0", "HSA_NO_SCRATCH_RECLAIM=1"], "args": "-b 128 -e {E} -f 2 -g 1 -R 2 -D 3 -A 1" } ] }