on CUDA exact_sstp_cond 3 times faster than the default? #326
Description
Judging from the lgrngn_cond_substep tests, exact (per-particle) substepping is much faster than the old, per-cell substepping.
Why is it so? I expected the opposite.
{{ message }}