Closed kabicm closed 3 years ago
This PR brings the following improvements:
igri
was accessed after clear()
, which is now fixed.push_pivots_up
is skipped when first_non_pivot_row >= n_rows
.if (n_local_active_rows < 0) continue
was causing MPI to hang. This is removed and the proper checks for n_local_active_rows
are added.push_pivots_up
is now much faster: vectors pivots, early_non_pivots, late_pivots
are now preallocated and analyzed only once per iteration. Previously, these arrays were allocated/deallocated 4 times per iteration.
This PR brings the following improvements:
igri
was accessed afterclear()
, which is now fixed.push_pivots_up
is skipped whenfirst_non_pivot_row >= n_rows
.if (n_local_active_rows < 0) continue
was causing MPI to hang. This is removed and the proper checks forn_local_active_rows
are added.push_pivots_up
is now much faster: vectorspivots, early_non_pivots, late_pivots
are now preallocated and analyzed only once per iteration. Previously, these arrays were allocated/deallocated 4 times per iteration.