AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

2.10.3-1

Browse Source 
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.


[ROCm/rccl commit: 7e51592129]
This commit is contained in:
Ke Wen
2021-07-08 14:12:04 -07:00
parent 3dd3a1ca66
commit 58a9f19dc3
52 changed files with 3496 additions and 2469 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/include/core.h
+4 -1
View File
@@ -1,5 +1,5 @@
/*************************************************************************
* Copyright (c) 2015-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2015-2021, NVIDIA CORPORATION. All rights reserved.
*
* See LICENSE.txt for license information
************************************************************************/
@@ -36,6 +36,9 @@ static __inline__ int ncclTypeSize(ncclDataType_t type) {
case ncclUint8:
return 1;
case ncclFloat16:
#if defined(__CUDA_BF16_TYPES_EXIST__)
case ncclBfloat16:
#endif
return 2;
case ncclInt32:
case ncclUint32: