runner.c

/*******************************************************************************
 * This file is part of SWIFT.
 * Copyright (c) 2012 Pedro Gonnet (pedro.gonnet@durham.ac.uk)
 *                    Matthieu Schaller (matthieu.schaller@durham.ac.uk)
 *               2015 Peter W. Draper (p.w.draper@durham.ac.uk)
 *               2016 John A. Regan (john.a.regan@durham.ac.uk)
 *                    Tom Theuns (tom.theuns@durham.ac.uk)
 *
 * This program is free software: you can redistribute it and/or modify
 * it under the terms of the GNU Lesser General Public License as published
 * by the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public License
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 *
 ******************************************************************************/

/* Config parameters. */
#include "../config.h"

/* Some standard headers. */
#include <float.h>
#include <limits.h>
#include <stdlib.h>

/* MPI headers. */
#ifdef WITH_MPI
#include <mpi.h>
#endif

/* This object's header. */
#include "runner.h"

/* Local headers. */
#include "active.h"
#include "approx_math.h"
#include "atomic.h"
#include "cell.h"
#include "chemistry.h"
#include "const.h"
#include "cooling.h"
#include "debug.h"
#include "drift.h"
#include "engine.h"
#include "error.h"
#include "gravity.h"
#include "hydro.h"
#include "hydro_properties.h"
#include "kick.h"
#include "logger.h"
#include "minmax.h"
#include "runner_doiact_vec.h"
#include "scheduler.h"
#include "sort_part.h"
#include "space.h"
#include "space_getsid.h"
#include "stars.h"
#include "task.h"
#include "timers.h"
#include "timestep.h"
#include "timestep_limiter.h"
#include "tracers.h"

#define TASK_LOOP_DENSITY 0
#define TASK_LOOP_GRADIENT 1
#define TASK_LOOP_FORCE 2
#define TASK_LOOP_LIMITER 3

/* Import the density loop functions. */
#define FUNCTION density
#define FUNCTION_TASK_LOOP TASK_LOOP_DENSITY
#include "runner_doiact.h"
#undef FUNCTION
#undef FUNCTION_TASK_LOOP

/* Import the gradient loop functions (if required). */
#ifdef EXTRA_HYDRO_LOOP
#define FUNCTION gradient
#define FUNCTION_TASK_LOOP TASK_LOOP_GRADIENT
#include "runner_doiact.h"
#undef FUNCTION
#undef FUNCTION_TASK_LOOP
#endif

/* Import the force loop functions. */
#define FUNCTION force
#define FUNCTION_TASK_LOOP TASK_LOOP_FORCE
#include "runner_doiact.h"
#undef FUNCTION
#undef FUNCTION_TASK_LOOP

/* Import the limiter loop functions. */
#define FUNCTION limiter
#define FUNCTION_TASK_LOOP TASK_LOOP_LIMITER
#include "runner_doiact.h"
#undef FUNCTION
#undef FUNCTION_TASK_LOOP

/* Import the gravity loop functions. */
#include "runner_doiact_grav.h"

/* Import the stars density loop functions. */
#define FUNCTION density
#include "runner_doiact_stars.h"
#undef FUNCTION

/* Import the stars feedback loop functions. */
#define FUNCTION feedback
#include "runner_doiact_stars.h"
#undef FUNCTION

/**
 * @brief Intermediate task after the density to check that the smoothing
 * lengths are correct.
 *
 * @param r The runner thread.
 * @param c The cell.
 * @param timer Are we timing this ?
 */
void runner_do_stars_ghost(struct runner *r, struct cell *c, int timer) {

  struct spart *restrict sparts = c->stars.parts;
  const struct engine *e = r->e;
  const struct cosmology *cosmo = e->cosmology;
  const struct stars_props *stars_properties = e->stars_properties;
  const float stars_h_max = stars_properties->h_max;
  const float eps = stars_properties->h_tolerance;
  const float stars_eta_dim = pow_dimension(stars_properties->eta_neighbours);
  const int max_smoothing_iter = stars_properties->max_smoothing_iterations;
  int redo = 0, scount = 0;

  TIMER_TIC;

  /* Anything to do here? */
  if (!cell_is_active_stars(c, e)) return;

  /* Recurse? */
  if (c->split) {
    for (int k = 0; k < 8; k++)
      if (c->progeny[k] != NULL) runner_do_stars_ghost(r, c->progeny[k], 0);
  } else {

    /* Init the list of active particles that have to be updated. */
    int *sid = NULL;
    if ((sid = (int *)malloc(sizeof(int) * c->stars.count)) == NULL)
      error("Can't allocate memory for sid.");
    for (int k = 0; k < c->stars.count; k++)
      if (spart_is_active(&sparts[k], e)) {
        sid[scount] = k;
        ++scount;
      }

    /* While there are particles that need to be updated... */
    for (int num_reruns = 0; scount > 0 && num_reruns < max_smoothing_iter;
         num_reruns++) {

      /* Reset the redo-count. */
      redo = 0;

      /* Loop over the remaining active parts in this cell. */
      for (int i = 0; i < scount; i++) {

        /* Get a direct pointer on the part. */
        struct spart *sp = &sparts[sid[i]];

#ifdef SWIFT_DEBUG_CHECKS
        /* Is this part within the timestep? */
        if (!spart_is_active(sp, e))
          error("Ghost applied to inactive particle");
#endif

        /* Get some useful values */
        const float h_old = sp->h;
        const float h_old_dim = pow_dimension(h_old);
        const float h_old_dim_minus_one = pow_dimension_minus_one(h_old);
        float h_new;
        int has_no_neighbours = 0;

        if (sp->density.wcount == 0.f) { /* No neighbours case */

          /* Flag that there were no neighbours */
          has_no_neighbours = 1;

          /* Double h and try again */
          h_new = 2.f * h_old;
        } else {

          /* Finish the density calculation */
          stars_end_density(sp, cosmo);

          /* Compute one step of the Newton-Raphson scheme */
          const float n_sum = sp->density.wcount * h_old_dim;
          const float n_target = stars_eta_dim;
          const float f = n_sum - n_target;
          const float f_prime =
              sp->density.wcount_dh * h_old_dim +
              hydro_dimension * sp->density.wcount * h_old_dim_minus_one;

          /* Avoid floating point exception from f_prime = 0 */
          h_new = h_old - f / (f_prime + FLT_MIN);
#ifdef SWIFT_DEBUG_CHECKS
          if ((f > 0.f && h_new > h_old) || (f < 0.f && h_new < h_old))
            error(
                "Smoothing length correction not going in the right direction");
#endif

          /* Safety check: truncate to the range [ h_old/2 , 2h_old ]. */
          h_new = min(h_new, 2.f * h_old);
          h_new = max(h_new, 0.5f * h_old);
        }

        /* Check whether the particle has an inappropriate smoothing length */
        if (fabsf(h_new - h_old) > eps * h_old) {

          /* Ok, correct then */
          sp->h = h_new;

          /* If below the absolute maximum, try again */
          if (sp->h < stars_h_max) {

            /* Flag for another round of fun */
            sid[redo] = sid[i];
            redo += 1;

            /* Re-initialise everything */
            stars_init_spart(sp);

            /* Off we go ! */
            continue;

          } else {

            /* Ok, this particle is a lost cause... */
            sp->h = stars_h_max;

            /* Do some damage control if no neighbours at all were found */
            if (has_no_neighbours) {
              stars_spart_has_no_neighbours(sp, cosmo);
            }
          }
        }

        /* We now have a particle whose smoothing length has converged */

        /* Compute the stellar evolution  */
        stars_evolve_spart(sp, stars_properties, cosmo);
      }

      /* We now need to treat the particles whose smoothing length had not
       * converged again */

      /* Re-set the counter for the next loop (potentially). */
      scount = redo;
      if (scount > 0) {

        /* Climb up the cell hierarchy. */
        for (struct cell *finger = c; finger != NULL; finger = finger->parent) {

          /* Run through this cell's density interactions. */
          for (struct link *l = finger->stars.density; l != NULL; l = l->next) {

#ifdef SWIFT_DEBUG_CHECKS
            if (l->t->ti_run < r->e->ti_current)
              error("Density task should have been run.");
#endif

            /* Self-interaction? */
            if (l->t->type == task_type_self)
              runner_doself_subset_branch_stars_density(r, finger, sparts, sid,
                                                        scount);

            /* Otherwise, pair interaction? */
            else if (l->t->type == task_type_pair) {

              /* Left or right? */
              if (l->t->ci == finger)
                runner_dopair_subset_branch_stars_density(
                    r, finger, sparts, sid, scount, l->t->cj);
              else
                runner_dopair_subset_branch_stars_density(
                    r, finger, sparts, sid, scount, l->t->ci);
            }

            /* Otherwise, sub-self interaction? */
            else if (l->t->type == task_type_sub_self)
              runner_dosub_subset_stars_density(r, finger, sparts, sid, scount,
                                                NULL, -1, 1);

            /* Otherwise, sub-pair interaction? */
            else if (l->t->type == task_type_sub_pair) {

              /* Left or right? */
              if (l->t->ci == finger)
                runner_dosub_subset_stars_density(r, finger, sparts, sid,
                                                  scount, l->t->cj, -1, 1);
              else
                runner_dosub_subset_stars_density(r, finger, sparts, sid,
                                                  scount, l->t->ci, -1, 1);
            }
          }
        }
      }
    }

    if (scount) {
      error("Smoothing length failed to converge on %i particles.", scount);
    }

    /* Be clean */
    free(sid);
  }

  if (timer) TIMER_TOC(timer_dostars_ghost);
}

/**
 * @brief Calculate gravity acceleration from external potential
 *
 * @param r runner task
 * @param c cell
 * @param timer 1 if the time is to be recorded.
 */
void runner_do_grav_external(struct runner *r, struct cell *c, int timer) {

  struct gpart *restrict gparts = c->grav.parts;
  const int gcount = c->grav.count;
  const struct engine *e = r->e;
  const struct external_potential *potential = e->external_potential;
  const struct phys_const *constants = e->physical_constants;
  const double time = r->e->time;

  TIMER_TIC;

  /* Anything to do here? */
  if (!cell_is_active_gravity(c, e)) return;

  /* Recurse? */
  if (c->split) {
    for (int k = 0; k < 8; k++)
      if (c->progeny[k] != NULL) runner_do_grav_external(r, c->progeny[k], 0);
  } else {

    /* Loop over the gparts in this cell. */
    for (int i = 0; i < gcount; i++) {

      /* Get a direct pointer on the part. */
      struct gpart *restrict gp = &gparts[i];

      /* Is this part within the time step? */
      if (gpart_is_active(gp, e)) {
        external_gravity_acceleration(time, potential, constants, gp);
      }
    }
  }

  if (timer) TIMER_TOC(timer_dograv_external);
}

/**
 * @brief Calculate gravity accelerations from the periodic mesh
 *
 * @param r runner task
 * @param c cell
 * @param timer 1 if the time is to be recorded.
 */
void runner_do_grav_mesh(struct runner *r, struct cell *c, int timer) {

  struct gpart *restrict gparts = c->grav.parts;
  const int gcount = c->grav.count;
  const struct engine *e = r->e;

#ifdef SWIFT_DEBUG_CHECKS
  if (!e->s->periodic) error("Calling mesh forces in non-periodic mode.");
#endif

  TIMER_TIC;

  /* Anything to do here? */
  if (!cell_is_active_gravity(c, e)) return;

  /* Recurse? */
  if (c->split) {
    for (int k = 0; k < 8; k++)
      if (c->progeny[k] != NULL) runner_do_grav_mesh(r, c->progeny[k], 0);
  } else {

    /* Get the forces from the gravity mesh */
    pm_mesh_interpolate_forces(e->mesh, e, gparts, gcount);
  }

  if (timer) TIMER_TOC(timer_dograv_mesh);
}

/**
 * @brief Calculate change in thermal state of particles induced
 * by radiative cooling and heating.
 *
 * @param r runner task
 * @param c cell
 * @param timer 1 if the time is to be recorded.
 */
void runner_do_cooling(struct runner *r, struct cell *c, int timer) {

  const struct engine *e = r->e;
  const struct cosmology *cosmo = e->cosmology;
  const int with_cosmology = (e->policy & engine_policy_cosmology);
  const struct cooling_function_data *cooling_func = e->cooling_func;
  const struct phys_const *constants = e->physical_constants;
  const struct unit_system *us = e->internal_units;
  const struct hydro_props *hydro_props = e->hydro_properties;
  const double time_base = e->time_base;
  const integertime_t ti_current = e->ti_current;
  struct part *restrict parts = c->hydro.parts;
  struct xpart *restrict xparts = c->hydro.xparts;
  const int count = c->hydro.count;

  TIMER_TIC;

  /* Anything to do here? */
  if (!cell_is_active_hydro(c, e)) return;

  /* Recurse? */
  if (c->split) {
    for (int k = 0; k < 8; k++)
      if (c->progeny[k] != NULL) runner_do_cooling(r, c->progeny[k], 0);
  } else {

    /* Loop over the parts in this cell. */
    for (int i = 0; i < count; i++) {

      /* Get a direct pointer on the part. */
      struct part *restrict p = &parts[i];
      struct xpart *restrict xp = &xparts[i];

      if (part_is_active(p, e)) {

        double dt_cool, dt_therm;
        if (with_cosmology) {
          const integertime_t ti_step = get_integer_timestep(p->time_bin);
          const integertime_t ti_begin =
              get_integer_time_begin(ti_current - 1, p->time_bin);

          dt_cool =
              cosmology_get_delta_time(cosmo, ti_begin, ti_begin + ti_step);
          dt_therm = cosmology_get_therm_kick_factor(e->cosmology, ti_begin,
                                                     ti_begin + ti_step);

        } else {
          dt_cool = get_timestep(p->time_bin, time_base);
          dt_therm = get_timestep(p->time_bin, time_base);
        }

        /* Let's cool ! */
        cooling_cool_part(constants, us, cosmo, hydro_props, cooling_func, p,
                          xp, dt_cool, dt_therm);
      }
    }
  }

  if (timer) TIMER_TOC(timer_do_cooling);
}

/**
 *
 */
void runner_do_star_formation(struct runner *r, struct cell *c, int timer) {

  struct engine *e = r->e;
  const struct cosmology *cosmo = e->cosmology;
  const int count = c->hydro.count;
  struct part *restrict parts = c->hydro.parts;
  struct xpart *restrict xparts = c->hydro.xparts;

  TIMER_TIC;

  /* Anything to do here? */
  if (!cell_is_active_hydro(c, e)) return;

  /* Recurse? */
  if (c->split) {
    for (int k = 0; k < 8; k++)
      if (c->progeny[k] != NULL) runner_do_star_formation(r, c->progeny[k], 0);
  } else {

    /* Loop over the gas particles in this cell. */
    for (int k = 0; k < count; k++) {

      /* Get a handle on the part. */
      struct part *restrict p = &parts[k];
      struct xpart *restrict xp = &xparts[k];

      if (part_is_active(p, e)) {

        const float rho = hydro_get_physical_density(p, cosmo);

        // MATTHIEU: Temporary star-formation law
        // Do not use this at home.
        if (rho > 1.7e7 && e->step > 2) {
          message("Removing particle id=%lld rho=%e", p->id, rho);

          struct spart *sp = cell_convert_part_to_spart(e, c, p, xp);

          /* Did we run out of fresh particles? */
          if (sp == NULL) continue;

          /* Set everything to a valid state */
          stars_init_spart(sp);
        }
      }
    }
  }

  if (timer) TIMER_TOC(timer_do_star_formation);
}

/**
 * @brief Sort the entries in ascending order using QuickSort.
 *
 * @param sort The entries
 * @param N The number of entries.
 */
void runner_do_sort_ascending(struct entry *sort, int N) {

  struct {
    short int lo, hi;
  } qstack[10];
  int qpos, i, j, lo, hi, imin;
  struct entry temp;
  float pivot;

  /* Sort parts in cell_i in decreasing order with quicksort */
  qstack[0].lo = 0;
  qstack[0].hi = N - 1;
  qpos = 0;
  while (qpos >= 0) {
    lo = qstack[qpos].lo;
    hi = qstack[qpos].hi;
    qpos -= 1;
    if (hi - lo < 15) {
      for (i = lo; i < hi; i++) {
        imin = i;
        for (j = i + 1; j <= hi; j++)
          if (sort[j].d < sort[imin].d) imin = j;
        if (imin != i) {
          temp = sort[imin];
          sort[imin] = sort[i];
          sort[i] = temp;
        }
      }
    } else {
      pivot = sort[(lo + hi) / 2].d;
      i = lo;
      j = hi;
      while (i <= j) {
        while (sort[i].d < pivot) i++;
        while (sort[j].d > pivot) j--;
        if (i <= j) {
          if (i < j) {
            temp = sort[i];
            sort[i] = sort[j];
            sort[j] = temp;
          }
          i += 1;
          j -= 1;
        }
      }
      if (j > (lo + hi) / 2) {
        if (lo < j) {
          qpos += 1;
          qstack[qpos].lo = lo;
          qstack[qpos].hi = j;
        }
        if (i < hi) {
          qpos += 1;
          qstack[qpos].lo = i;
          qstack[qpos].hi = hi;
        }
      } else {
        if (i < hi) {
          qpos += 1;
          qstack[qpos].lo = i;
          qstack[qpos].hi = hi;
        }
        if (lo < j) {
          qpos += 1;
          qstack[qpos].lo = lo;
          qstack[qpos].hi = j;
        }
      }
    }
  }
}

#ifdef SWIFT_DEBUG_CHECKS
/**
 * @brief Recursively checks that the flags are consistent in a cell hierarchy.
 *
 * Debugging function. Exists in two flavours: hydro & stars.
 */
#define RUNNER_CHECK_SORTS(TYPE)                                               \
  void runner_check_sorts_##TYPE(struct cell *c, int flags) {                  \
                                                                               \
    if (flags & ~c->TYPE.sorted) error("Inconsistent sort flags (downward)!"); \
    if (c->split)                                                              \
      for (int k = 0; k < 8; k++)                                              \
        if (c->progeny[k] != NULL && c->progeny[k]->TYPE.count > 0)            \
          runner_check_sorts_##TYPE(c->progeny[k], c->TYPE.sorted);            \
  }
#else
#define RUNNER_CHECK_SORTS(TYPE)                                       \
  void runner_check_sorts_##TYPE(struct cell *c, int flags) {          \
    error("Calling debugging code without debugging flag activated."); \
  }
#endif

RUNNER_CHECK_SORTS(hydro)
RUNNER_CHECK_SORTS(stars)

/**
 * @brief Sort the particles in the given cell along all cardinal directions.
 *
 * @param r The #runner.
 * @param c The #cell.
 * @param flags Cell flag.
 * @param cleanup If true, re-build the sorts for the selected flags instead
 *        of just adding them.
 * @param clock Flag indicating whether to record the timing or not, needed
 *      for recursive calls.
 */
void runner_do_hydro_sort(struct runner *r, struct cell *c, int flags,
                          int cleanup, int clock) {

  struct entry *fingers[8];
  const int count = c->hydro.count;
  const struct part *parts = c->hydro.parts;
  struct xpart *xparts = c->hydro.xparts;
  float buff[8];

  TIMER_TIC;

  /* We need to do the local sorts plus whatever was requested further up. */
  flags |= c->hydro.do_sort;
  if (cleanup) {
    c->hydro.sorted = 0;
  } else {
    flags &= ~c->hydro.sorted;
  }
  if (flags == 0 && !c->hydro.do_sub_sort) return;

  /* Check that the particles have been moved to the current time */
  if (flags && !cell_are_part_drifted(c, r->e))
    error("Sorting un-drifted cell c->nodeID=%d", c->nodeID);

#ifdef SWIFT_DEBUG_CHECKS
  /* Make sure the sort flags are consistent (downward). */
  runner_check_sorts_hydro(c, c->hydro.sorted);

  /* Make sure the sort flags are consistent (upard). */
  for (struct cell *finger = c->parent; finger != NULL;
       finger = finger->parent) {
    if (finger->hydro.sorted & ~c->hydro.sorted)
      error("Inconsistent sort flags (upward).");
  }

  /* Update the sort timer which represents the last time the sorts
     were re-set. */
  if (c->hydro.sorted == 0) c->hydro.ti_sort = r->e->ti_current;
#endif

  /* start by allocating the entry arrays in the requested dimensions. */
  for (int j = 0; j < 13; j++) {
    if ((flags & (1 << j)) && c->hydro.sort[j] == NULL) {
      if ((c->hydro.sort[j] = (struct entry *)malloc(sizeof(struct entry) *
                                                     (count + 1))) == NULL)
        error("Failed to allocate sort memory.");
    }
  }

  /* Does this cell have any progeny? */
  if (c->split) {

    /* Fill in the gaps within the progeny. */
    float dx_max_sort = 0.0f;
    float dx_max_sort_old = 0.0f;
    for (int k = 0; k < 8; k++) {
      if (c->progeny[k] != NULL && c->progeny[k]->hydro.count > 0) {
        /* Only propagate cleanup if the progeny is stale. */
        runner_do_hydro_sort(r, c->progeny[k], flags,
                             cleanup && (c->progeny[k]->hydro.dx_max_sort_old >
                                         space_maxreldx * c->progeny[k]->dmin),
                             0);
        dx_max_sort = max(dx_max_sort, c->progeny[k]->hydro.dx_max_sort);
        dx_max_sort_old =
            max(dx_max_sort_old, c->progeny[k]->hydro.dx_max_sort_old);
      }
    }
    c->hydro.dx_max_sort = dx_max_sort;
    c->hydro.dx_max_sort_old = dx_max_sort_old;

    /* Loop over the 13 different sort arrays. */
    for (int j = 0; j < 13; j++) {

      /* Has this sort array been flagged? */
      if (!(flags & (1 << j))) continue;

      /* Init the particle index offsets. */
      int off[8];
      off[0] = 0;
      for (int k = 1; k < 8; k++)
        if (c->progeny[k - 1] != NULL)
          off[k] = off[k - 1] + c->progeny[k - 1]->hydro.count;
        else
          off[k] = off[k - 1];

      /* Init the entries and indices. */
      int inds[8];
      for (int k = 0; k < 8; k++) {
        inds[k] = k;
        if (c->progeny[k] != NULL && c->progeny[k]->hydro.count > 0) {
          fingers[k] = c->progeny[k]->hydro.sort[j];
          buff[k] = fingers[k]->d;
          off[k] = off[k];
        } else
          buff[k] = FLT_MAX;
      }

      /* Sort the buffer. */
      for (int i = 0; i < 7; i++)
        for (int k = i + 1; k < 8; k++)
          if (buff[inds[k]] < buff[inds[i]]) {
            int temp_i = inds[i];
            inds[i] = inds[k];
            inds[k] = temp_i;
          }

      /* For each entry in the new sort list. */
      struct entry *finger = c->hydro.sort[j];
      for (int ind = 0; ind < count; ind++) {

        /* Copy the minimum into the new sort array. */
        finger[ind].d = buff[inds[0]];
        finger[ind].i = fingers[inds[0]]->i + off[inds[0]];

        /* Update the buffer. */
        fingers[inds[0]] += 1;
        buff[inds[0]] = fingers[inds[0]]->d;

        /* Find the smallest entry. */
        for (int k = 1; k < 8 && buff[inds[k]] < buff[inds[k - 1]]; k++) {
          int temp_i = inds[k - 1];
          inds[k - 1] = inds[k];
          inds[k] = temp_i;
        }

      } /* Merge. */

      /* Add a sentinel. */
      c->hydro.sort[j][count].d = FLT_MAX;
      c->hydro.sort[j][count].i = 0;

      /* Mark as sorted. */
      atomic_or(&c->hydro.sorted, 1 << j);

    } /* loop over sort arrays. */

  } /* progeny? */

  /* Otherwise, just sort. */
  else {

    /* Reset the sort distance */
    if (c->hydro.sorted == 0) {
#ifdef SWIFT_DEBUG_CHECKS
      if (xparts != NULL && c->nodeID != engine_rank)
        error("Have non-NULL xparts in foreign cell");
#endif

      /* And the individual sort distances if we are a local cell */
      if (xparts != NULL) {
        for (int k = 0; k < count; k++) {
          xparts[k].x_diff_sort[0] = 0.0f;
          xparts[k].x_diff_sort[1] = 0.0f;
          xparts[k].x_diff_sort[2] = 0.0f;
        }
      }
      c->hydro.dx_max_sort_old = 0.f;
      c->hydro.dx_max_sort = 0.f;
    }

    /* Fill the sort array. */
    for (int k = 0; k < count; k++) {
      const double px[3] = {parts[k].x[0], parts[k].x[1], parts[k].x[2]};
      for (int j = 0; j < 13; j++)
        if (flags & (1 << j)) {
          c->hydro.sort[j][k].i = k;
          c->hydro.sort[j][k].d = px[0] * runner_shift[j][0] +
                                  px[1] * runner_shift[j][1] +
                                  px[2] * runner_shift[j][2];
        }
    }

    /* Add the sentinel and sort. */
    for (int j = 0; j < 13; j++)
      if (flags & (1 << j)) {
        c->hydro.sort[j][count].d = FLT_MAX;
        c->hydro.sort[j][count].i = 0;
        runner_do_sort_ascending(c->hydro.sort[j], count);
        atomic_or(&c->hydro.sorted, 1 << j);
      }
  }

#ifdef SWIFT_DEBUG_CHECKS
  /* Verify the sorting. */
  for (int j = 0; j < 13; j++) {
    if (!(flags & (1 << j))) continue;
    struct entry *finger = c->hydro.sort[j];
    for (int k = 1; k < count; k++) {
      if (finger[k].d < finger[k - 1].d)
        error("Sorting failed, ascending array.");
      if (finger[k].i >= count) error("Sorting failed, indices borked.");
    }
  }

  /* Make sure the sort flags are consistent (downward). */
  runner_check_sorts_hydro(c, flags);

  /* Make sure the sort flags are consistent (upward). */
  for (struct cell *finger = c->parent; finger != NULL;
       finger = finger->parent) {
    if (finger->hydro.sorted & ~c->hydro.sorted)
      error("Inconsistent sort flags.");
  }
#endif

  /* Clear the cell's sort flags. */
  c->hydro.do_sort = 0;
  c->hydro.do_sub_sort = 0;
  c->hydro.requires_sorts = 0;

  if (clock) TIMER_TOC(timer_dosort);
}

/**
 * @brief Sort the stars particles in the given cell along all cardinal
 * directions.
 *
 * @param r The #runner.
 * @param c The #cell.
 * @param flags Cell flag.
 * @param cleanup If true, re-build the sorts for the selected flags instead
 *        of just adding them.
 * @param clock Flag indicating whether to record the timing or not, needed
 *      for recursive calls.
 */
void runner_do_stars_sort(struct runner *r, struct cell *c, int flags,
                          int cleanup, int clock) {

  struct entry *fingers[8];
  const int count = c->stars.count;
  struct spart *sparts = c->stars.parts;
  float buff[8];

  TIMER_TIC;

  /* We need to do the local sorts plus whatever was requested further up. */
  flags |= c->stars.do_sort;
  if (cleanup) {
    c->stars.sorted = 0;
  } else {
    flags &= ~c->stars.sorted;
  }
  if (flags == 0 && !c->stars.do_sub_sort) return;

  /* Check that the particles have been moved to the current time */
  if (flags && !cell_are_spart_drifted(c, r->e))
    error("Sorting un-drifted cell c->nodeID=%d", c->nodeID);

#ifdef SWIFT_DEBUG_CHECKS
  /* Make sure the sort flags are consistent (downward). */
  runner_check_sorts_stars(c, c->stars.sorted);

  /* Make sure the sort flags are consistent (upward). */
  for (struct cell *finger = c->parent; finger != NULL;
       finger = finger->parent) {
    if (finger->stars.sorted & ~c->stars.sorted)
      error("Inconsistent sort flags (upward).");
  }

  /* Update the sort timer which represents the last time the sorts
     were re-set. */
  if (c->stars.sorted == 0) c->stars.ti_sort = r->e->ti_current;
#endif

  /* start by allocating the entry arrays in the requested dimensions. */
  for (int j = 0; j < 13; j++) {
    if ((flags & (1 << j)) && c->stars.sort[j] == NULL) {
      if ((c->stars.sort[j] = (struct entry *)malloc(sizeof(struct entry) *
                                                     (count + 1))) == NULL)
        error("Failed to allocate sort memory.");
    }
  }

  /* Does this cell have any progeny? */
  if (c->split) {

    /* Fill in the gaps within the progeny. */
    float dx_max_sort = 0.0f;
    float dx_max_sort_old = 0.0f;
    for (int k = 0; k < 8; k++) {
      if (c->progeny[k] != NULL && c->progeny[k]->stars.count > 0) {
        /* Only propagate cleanup if the progeny is stale. */
        runner_do_stars_sort(r, c->progeny[k], flags,
                             cleanup && (c->progeny[k]->stars.dx_max_sort_old >
                                         space_maxreldx * c->progeny[k]->dmin),
                             0);
        dx_max_sort = max(dx_max_sort, c->progeny[k]->stars.dx_max_sort);
        dx_max_sort_old =
            max(dx_max_sort_old, c->progeny[k]->stars.dx_max_sort_old);
      }
    }
    c->stars.dx_max_sort = dx_max_sort;
    c->stars.dx_max_sort_old = dx_max_sort_old;

    /* Loop over the 13 different sort arrays. */
    for (int j = 0; j < 13; j++) {

      /* Has this sort array been flagged? */
      if (!(flags & (1 << j))) continue;

      /* Init the particle index offsets. */
      int off[8];
      off[0] = 0;
      for (int k = 1; k < 8; k++)
        if (c->progeny[k - 1] != NULL)
          off[k] = off[k - 1] + c->progeny[k - 1]->stars.count;
        else
          off[k] = off[k - 1];

      /* Init the entries and indices. */
      int inds[8];
      for (int k = 0; k < 8; k++) {
        inds[k] = k;
        if (c->progeny[k] != NULL && c->progeny[k]->stars.count > 0) {
          fingers[k] = c->progeny[k]->stars.sort[j];
          buff[k] = fingers[k]->d;
          off[k] = off[k];
        } else
          buff[k] = FLT_MAX;
      }

      /* Sort the buffer. */
      for (int i = 0; i < 7; i++)
        for (int k = i + 1; k < 8; k++)
          if (buff[inds[k]] < buff[inds[i]]) {
            int temp_i = inds[i];
            inds[i] = inds[k];
            inds[k] = temp_i;
          }

      /* For each entry in the new sort list. */
      struct entry *finger = c->stars.sort[j];
      for (int ind = 0; ind < count; ind++) {

        /* Copy the minimum into the new sort array. */
        finger[ind].d = buff[inds[0]];
        finger[ind].i = fingers[inds[0]]->i + off[inds[0]];

        /* Update the buffer. */
        fingers[inds[0]] += 1;
        buff[inds[0]] = fingers[inds[0]]->d;

        /* Find the smallest entry. */
        for (int k = 1; k < 8 && buff[inds[k]] < buff[inds[k - 1]]; k++) {
          int temp_i = inds[k - 1];
          inds[k - 1] = inds[k];
          inds[k] = temp_i;
        }

      } /* Merge. */

      /* Add a sentinel. */
      c->stars.sort[j][count].d = FLT_MAX;
      c->stars.sort[j][count].i = 0;

      /* Mark as sorted. */
      atomic_or(&c->stars.sorted, 1 << j);

    } /* loop over sort arrays. */

  } /* progeny? */

  /* Otherwise, just sort. */
  else {