Mojo module

all_reduce

Aliases

all_reduce: Main entry point for performing all-reduce across multiple GPUs.
all_reduce_naive: Performs all-reduce across GPUs without using peer-to-peer access.
all_reduce_p2p: Performs all-reduce using peer-to-peer access between GPUs.
all_reduce_p2p_kernel: Kernel implementing all-reduce using peer-to-peer access between GPUs.
can_enable_p2p: Checks and enables peer-to-peer access between all GPU pairs.
multi_gpu_barrier: Implements a barrier synchronization across multiple GPUs.
naive_reduce_kernel: A simple reduction kernel that adds source buffer values to destination buffer.