2.20.3-1

Add support for alternating rings, allow for cross-nic rings without
cross-rail communication.
Add support for user buffer registration for network send/recv.
Optimize aggregated operations to better utilize all channels.
Add flattening for BCM PCI gen5 switches.
Add support for inter-node NVLink communication
Add support for port fusion in NET/IB.
Add support for ReduceScatter and AllGather using Collnet.
Update net API to v8.
Fix hang during A2A connection.

src/include/alloc.h

@@ -85,13 +85,14 @@ static inline ncclResult_t ncclCuMemAlloc(void **ptr, CUmemGenericAllocationHand
   CUmemAllocationProp prop = {};
   CUmemAccessDesc accessDesc = {};
   CUmemGenericAllocationHandle handle;
+  CUmemAllocationHandleType type = ncclCuMemHandleType;
   int cudaDev;
   int flag = 0;
   CUDACHECK(cudaGetDevice(&cudaDev));
   CUCHECK(cuDeviceGet(&currentDev, cudaDev));
   prop.type = CU_MEM_ALLOCATION_TYPE_PINNED;
   prop.location.type = CU_MEM_LOCATION_TYPE_DEVICE;
-  prop.requestedHandleTypes = NCCL_P2P_HANDLE_TYPE; // So it can be exported
+  prop.requestedHandleTypes = type;
   prop.location.id = currentDev;
   // Query device to see if RDMA support is available
   CUCHECK(cuDeviceGetAttribute(&flag, CU_DEVICE_ATTRIBUTE_GPU_DIRECT_RDMA_SUPPORTED, currentDev));