uccl-project/mKernel

Optimized primitives for collective multi-GPU communication

Distributed Compiler and Optimized Parallel Kernels

UCCL is an efficient communication library for GPUs, covering collectives, P2P (e.g., KV cache transfer, RL weight transfer), and EP (e.g., GPU-driven)

A fast communication-overlapping library for tensor/expert parallelism on GPUs.

MSCCL++: A GPU-driven communication stack for scalable AI applications

DeepEP: an efficient expert-parallel communication library

Tokamax: A GPU and TPU kernel library.

DeepGEMM: clean and efficient BLAS kernel library on GPU

Microsoft Collective Communication Library