dADP

contact: francesco.tomba@phd.units.it

USAGE

Clone the repository and compile with make The compilation process produces an executable called main, run it with mpirun The suggestion is to run it with one mpi task per socket.

• -h --help show help message
• -i --in-data (required) path of the input file
• -t --in-type (required) datatype of the input file, allowed choices f32, f64
• -d --in-dims (required) number of dimensions of the data file, dadp expects something of the form N x d where N is inferred from the lenght of the data file
• -o --out-data (optional) output path for the data, the datafile is shuffled between mpi tasks and datapoints are ordered default is out_data
• -a --out-assignment (optional) output path for the cluster assignment output ranges [0 ... Nc - 1] for core points halo points have indices [-Nc ... -1] conversion of idx for an halo point is cluster_idx = -halo_idx - 1, default is out_assignment

TODO

• arugment parser
• prettify overall stdout
• H1: implementation of lock free centers elimination
• kdtree: implement slim heap
• kdtree: optimization an profiling