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

Merge remote-tracking branch 'remotes/nccl/master' into rccl_2.5.6

Browse Source
This commit is contained in:
Wenkai Du
2019-11-21 13:41:10 -08:00
parent b1ed4b7fa8 299c554dcc
commit 6648c81dc6
76 changed files with 5119 additions and 3071 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/transport.cc
+7 -4
View File
@@ -5,7 +5,8 @@
* See LICENSE.txt for license information
************************************************************************/
#include "core.h"
#include "comm.h"
#include "info.h"
extern struct ncclTransport p2pTransport;
extern struct ncclTransport shmTransport;
@@ -120,13 +121,13 @@ ncclResult_t transportSaveProxies(struct ncclProxyArgs* args, int pattern, int r
}
if (pattern == ncclPatternTreeUp || pattern == ncclPatternTreeUpDown) {
// Tree up
struct ncclTree* tree = &args->channel->tree;
struct ncclTree* tree = &args->channel->treeUp;
for (int i=0; i<NCCL_MAX_TREE_ARITY; i++) NCCLCHECK(SaveProxy<proxyRecv>(tree->down[i], args));
NCCLCHECK(SaveProxy<proxySend>(tree->up, args));
}
if (pattern == ncclPatternTreeDown || pattern == ncclPatternTreeUpDown) {
// Tree down
struct ncclTree* tree = &args->channel->tree;
struct ncclTree* tree = &args->channel->treeDn;
for (int i=0; i< NCCL_MAX_TREE_ARITY; i++) NCCLCHECK(SaveProxy<proxySend>(tree->down[i], args));
NCCLCHECK(SaveProxy<proxyRecv>(tree->up, args));
}
@@ -158,7 +159,9 @@ void* persistentThread(void *comm_) {
}
} while (op == NULL);
op->idle = 0;
if (op->state != ncclProxyOpNone) ret = op->progress(op);
// opCount >= lastOpCount are part of an ongoing GroupStart/GroupEnd that hasn't started
// yet and might be cancelled before they even start. Hold on on those.
if (op->state != ncclProxyOpNone && op->opCount < comm->lastOpCount) ret = op->progress(op);
if (ret != ncclSuccess) {
comm->fatalError = ret;
INFO(NCCL_ALL,"%s:%d -> %d [Proxy Thread]", __FILE__, __LINE__, ret);