export PATH=$PATH:/usr/local/cuda/bin export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64

export PATH=$PATH:/usr/include/mpich-3.2-x86_64 export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib64/mpich-3.2/lib export LD_LIBRARY_PATH=/usr/lib64/mpich-3.2/lib:$LD_LIBRARY_PATH

#NCCL Allreduce nvcc nccl-reduce.cu -o nccl-reduce -lnccl

#MPI结合NCCL nvcc nccl-reducempi.cu -o nccl-reducempi -lnccl -lmpi -L /usr/lib64/mpich-3.2/lib/ -I /usr/include/mpich-3.2-x86_64 mpiexec -n 4 ./nccl-reducempi

#分布式一维向量的softmax算子 nvcc nccl-softmax.cu -o nccl-softmax -lnccl -lmpi -L /usr/lib64/mpich-3.2/lib/ -I /usr/include/mpich-3.2-x86_64

##Cuda stream #jacobi迭代结合NCCL和MPI的多卡算法 nvcc nccl-mpi-jacobi.cu -o nccl-mpi-jacobi -lnccl -lmpi -L /usr/lib64/mpich-3.2/lib/ -I /usr/include/mpich-3.2-x86_64 mpiexec -n 2 ./nccl-mpi-jacobi，表示使用nranks=2个进程。

#jacobi迭代结合NCCL的多卡算法 nvcc nccl-jacobi.cu -o nccl-jacobi -lnccl -lmpi -L /usr/lib64/mpich-3.2/lib/ -I /usr/include/mpich-3.2-x86_64

#nccl-overlay nvcc nccl-overlay.cu -o nccl-overlay -lnccl -lmpi -L /usr/lib64/mpich-3.2/lib/ -I /usr/include/mpich-3.2-x86_64