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

2.4.2-1

Browse Source 
Add tree algorithms for allreduce to improve performance at scale.
Add ncclCommAbort() and ncclCommGetAsyncError() to properly handle
network errors and be permit recover.
Detect initial CPU affinity and no longer escape it.
This commit is contained in:
Sylvain Jeaugey
2018-12-13 15:56:12 -08:00
parent 4861e197fd
commit 1450d42675
66 changed files with 3746 additions and 3251 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/include/enqueue.h
+1 -6
View File
@@ -10,12 +10,7 @@
#include "core.h"
#include "group.h"
typedef ncclResult_t(*ncclFunc_t)(const void* sendbuff, void* recvbuff, size_t count,
ncclDataType_t type, ncclRedOp_t op, int root, ncclComm_t comm, cudaStream_t stream);
ncclResult_t ncclEnqueueCheck(ncclFunc_t func, const char* primName, const void* sendbuff,
void* recvbuff, size_t count, ncclDataType_t type, ncclRedOp_t op, int root,
ncclComm_t comm, cudaStream_t stream);
ncclResult_t ncclEnqueueCheck(struct ncclInfo* info);
ncclResult_t ncclCpuBarrierIn(ncclComm_t comm, int* isLast);
ncclResult_t ncclCpuBarrierLast(ncclComm_t comm);
ncclResult_t ncclCpuBarrierOut(ncclComm_t comm);