Add support for bfloat16.
Add ncclAvg reduction operation.
Improve performance for aggregated operations.
Improve performance for tree.
Improve network error reporting.
Add NCCL_NET parameter to force a specific network.
Add NCCL_IB_QPS_PER_CONNECTION parameter to split IB traffic onto multiple queue pairs.
Fix topology detection error in WSL2.
Fix proxy memory elements affinity (improve alltoall performance).
Fix graph search on cubemesh topologies.
Fix hang in cubemesh during NVB connections.
Cette révision appartient à :
Ke Wen
2021-07-08 14:12:04 -07:00
Parent 3fec2fa5ee
révision 7e51592129
52 fichiers modifiés avec 3496 ajouts et 2469 suppressions
+1
Voir le fichier
@@ -1,3 +1,4 @@
include/nccl.h /usr/include
include/nccl_net.h /usr/include
lib/libnccl.so /usr/lib/${pkg:MultiArch}
lib/libnccl_static.a /usr/lib/${pkg:MultiArch}