Fix crash when setting NCCL_MAX_P2P_NCHANNELS below nchannels.
Fix hang during sendrecv dynamic NVB connection establishment on
cubemesh topologies.
Add environment variable to only use SHARP on communicators beyond
a given number of ranks.
Add debug subsystem to trace memory allocations.
Fix compilation with TRACE=1. (Issue #505)


[ROCm/rccl commit: 3fec2fa5ee]
Этот коммит содержится в:
Sylvain Jeaugey committed 2021-05-12 11:09:31 -07:00
1 родитель 780273774a
Коммит 3dd3a1ca66
14 files changed
+141 -51

No files matched your search

+3 -1
Просмотреть файл
@@ -1,5 +1,5 @@
/*************************************************************************
* Copyright (c) 2016-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2016-2021, NVIDIA CORPORATION. All rights reserved.
*
* See LICENSE.txt for license information
************************************************************************/
@@ -62,6 +62,8 @@ void ncclDebugInit() {
mask = NCCL_TUNING;
} else if (strcasecmp(subsys, "ENV") == 0) {
mask = NCCL_ENV;
} else if (strcasecmp(subsys, "ALLOC") == 0) {
mask = NCCL_ALLOC;
} else if (strcasecmp(subsys, "ALL") == 0) {
mask = NCCL_ALL;
}