sglang/sgl-kernel/csrc/cpu/shm.h

#include <torch/all.h>

#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>

#ifndef __SHM_COLLECTIVES__
#define __SHM_COLLECTIVES__
#define VECTOR_LENGTH_IN_BYTES 32
void shm_initialize(int size, int rank, const char* addr_string, const char* port_string);
void all_reduce_outer_loop(torch::Tensor& data, size_t numel, int data_size);
torch::Tensor& all_gather(torch::Tensor& result, torch::Tensor& data, int dim, size_t numel, int data_size);
#endif