diff options
author | Sylvain Jeaugey <sjeaugey@nvidia.com> | 2018-12-05 01:42:28 +0300 |
---|---|---|
committer | Sylvain Jeaugey <sjeaugey@nvidia.com> | 2018-12-05 01:42:28 +0300 |
commit | 4b39a4cf9113e01d80377236020de418a0381358 (patch) | |
tree | ebf715b678eaf4fcf73f7c2c66116284ebac51ca | |
parent | b8a9a32ccb889d00ffd060601f6c825a9fb05a2f (diff) |
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.
-rw-r--r-- | src/transport/net.cu | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/transport/net.cu b/src/transport/net.cu index 165187c..9c366b3 100644 --- a/src/transport/net.cu +++ b/src/transport/net.cu @@ -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; |