Performance benchmarks of the lenS²HAT code. From left to right, we show the memory, runtime, and total CPU time, summed over all MPI processes, as a function of the number of processors (or MPI tasks). In all cases the simulations have been made for three Stokes parameters and used a ECP grid of 32   768² points, and oversampling factor κ = 8. In each panel the dashed lines show a theoretical scaling as expected in the ideal circumstances. The thick lines show the indicative results from the LensPix code run on a HEALPix grid with nside = 4096, band limit ℓ_max = 8000 and an oversampling factor of 2. In the right panel, the LensPix alm2map curve includes the time required to generate the oversampled ECP grid used for interpolation and the interpolation time itself. See Sect. 4.8 for a detailed discussion.