Add network communication through another GPU connected with NVLink
(PXN).
Add aggregation of messages coming from different local GPUs through
PXN and going to the same destination.
Add new v5 plugin API with grouped receives and tags.
Add compat for v4 plugins.
Add naming of NCCL threads to help debugging.
Fix NVLink detection and avoid data corruption when some NVLinks are
down.
Add support for Relaxed Ordering for IB.
Add profiling and timing infrastructure.


[ROCm/rccl commit: 3c223c105a]
This commit is contained in:
Sylvain Jeaugey
2022-01-07 06:39:55 -08:00
parent ed02fb8993
commit f8886d8687
70 changed files with 6453 additions and 3430 deletions
+9 -1
View File
@@ -1,5 +1,5 @@
/*************************************************************************
* Copyright (c) 2019-2021, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2019-2022, NVIDIA CORPORATION. All rights reserved.
*
* See LICENSE.txt for license information
************************************************************************/
@@ -94,6 +94,14 @@ static ncclResult_t xmlGetAttrInt(struct ncclXmlNode* node, const char* attrName
return ncclSuccess;
}
static ncclResult_t xmlGetAttrIntDefault(struct ncclXmlNode* node, const char* attrName, int* value, int defaultValue) {
const char* str;
NCCLCHECK(xmlGetAttr(node, attrName, &str));
*value = str ? strtol(str, NULL, 0) : defaultValue;
return ncclSuccess;
}
static ncclResult_t xmlGetAttrFloat(struct ncclXmlNode* node, const char* attrName, float* value) {
const char* str;
NCCLCHECK(xmlGetAttrStr(node, attrName, &str));