Fix GPU Direct RDMA detection.
Whether the network supported GPU Direct RDMA or not was ignored, causing sockets to break when cards were local enough that NCCL tried to use it.
This commit is contained in:
@@ -241,7 +241,7 @@ static ncclResult_t netGetGdrSupport(int dev, int distance, int read, int* useGd
|
||||
// Finally, check if the NIC supports it
|
||||
int flags;
|
||||
NCCLCHECK(ncclNetPtrSupport(dev, &flags));
|
||||
if (flags & NCCL_PTR_CUDA == 0) return ncclSuccess;
|
||||
if ((flags & NCCL_PTR_CUDA) == 0) return ncclSuccess;
|
||||
*useGdr = 1;
|
||||
INFO(NCCL_INIT|NCCL_NET,"NET/%s : GPU Direct RDMA Enabled for GPU %d / HCA %d (distance %d >= %d), read %d", ncclNetName(), cudaDev, dev, distance, netGdrLevel, read);
|
||||
return ncclSuccess;
|
||||
|
||||
Reference in New Issue
Block a user