Ring weights distribution
Given that each proc stores in the memory only a part of the entire map corresponding to a subinterval of all rings as defined by two integer variables, first_ring to last_ring, see map distribution, it needs only last_ring-first_ring+1 weights corresponding to the rings from first_ring to last_ring. This is how the full weights vector is distributed over procs, with each processor keeping only a subset of weights corresponding to the map rings it is supposed to process. Different weights are allowed for different Stokes parameters but the same weights are assumed for all maps processed simultaneously. In Fortran 90 the local weights are stored as a two-dimensional array of double precision numbers with sizes given by (1:last_ring-first_ring+1,1:nstokes). In C this is stored as a vector in the column-wise order with the index running from 0 to (last_ring-first_ring+1)*nstokes-1. Next: Compilation   Previous: alm distribution   Up: Data distribution radek stompor 2009-10-15 |