module load gcc/9.3.0
module load cuda/11.4.1
module load openmpi/4.1.1-cuda11.4

export QUDA_RESOURCE_PATH=$(pwd)/tuning
export OMP_NUM_THREADS=4
export OMPI_MCA_btl=^uct,openib
export OMPI_MCA_pml=ucx # by fabian. no idea what this is
#export UCX_TLS=rc,rc_x,sm,cuda_copy,cuda_ipc,gdr_copy
export UCX_TLS=gdr_copy,rc,rc_x,sm,cuda_copy,cuda_ipc
export UCX_RNDV_THRESH=16384
export UCX_RNDV_SCHEME=put_zcopy
export UCX_IB_GPU_DIRECT_RDMA=yes
export UCX_MEMTYPE_CACHE=n

export OMPI_MCA_io=romio321
export OMPI_MCA_btl_openib_allow_ib=true
export OMPI_MCA_btl_openib_device_type=infiniband
export OMPI_MCA_btl_openib_if_exclude=mlx5_1,mlx5_2,mlx5_3

export QUDA_REORDER_LOCATION=GPU # this is the default anyway