Module device_selection

This module provides functionality to automatically pair compute devices with the best available RDMA NICs based on PCI topology distance.

Structs

PCIDevice

Functions

create_cuda_to_rdma_mapping

Creates a mapping from CUDA PCI addresses to optimal RDMA devices

get_all_rdma_devices

get_cuda_pci_address

get_nic_pci_address

get_numa_pci_address

parse_device_string

parse_pci_topology

resolve_rdma_device

Resolves RDMA device using auto-detection logic when needed

select_optimal_rdma_device

Step 1: Parse device string into prefix and postfix Step 2: Get PCI address from compute device Step 3: Get PCI address for all RDMA NIC devices Step 4: Calculate PCI distances and return closest RDMA NIC device