Merge branch 'master' into testPeriodicBC

doc/Doxyfile.in

@@ -1988,6 +1988,9 @@ INCLUDE_FILE_PATTERNS  =
 # This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
 
 PREDEFINED             = "__attribute__(x)= "
+PREDEFINED             += HAVE_HDF5
+PREDEFINED             += WITH_MPI
+PREDEFINED             += WITH_VECTORIZATION
 
 # If the MACRO_EXPANSION and EXPAND_ONLY_PREDEF tags are set to YES then this
 # tag can be used to specify a list of macro names that should be expanded. The

src/collectgroup.c

@@ -170,7 +170,7 @@ static void doreduce1(struct mpicollectgroup1 *mpigrp11,
 }
 
 /**
- * @brief MPI reduce operator for #mpicollectgroup structures.
+ * @brief MPI reduce operator for #mpicollectgroup1 structures.
  */
 static void mpicollectgroup1_reduce(void *in, void *inout, int *len,
                                     MPI_Datatype *datatype) {

src/runner.c

@@ -1818,13 +1818,8 @@ void *runner_main(void *data) {
           break;
 
         case task_type_pair:
-          if (t->subtype == task_subtype_density) {
-#if defined(WITH_VECTORIZATION) && defined(GADGET2_SPH)
-            runner_dopair1_density_vec(r, ci, cj);
-#else
-            runner_dopair1_density(r, ci, cj);
-#endif
-          }
+          if (t->subtype == task_subtype_density)
+            runner_dopair1_branch_density(r, ci, cj);
 #ifdef EXTRA_HYDRO_LOOP
           else if (t->subtype == task_subtype_gradient)
             runner_dopair1_gradient(r, ci, cj);

src/runner_doiact.h

@@ -884,8 +884,11 @@ void DOSELF_SUBSET(struct runner *r, struct cell *restrict ci,
  * @param r The #runner.
  * @param ci The first #cell.
  * @param cj The second #cell.
+ * @param sid The direction of the pair
+ * @param shift The shift vector to apply to the particles in ci.
  */
-void DOPAIR1(struct runner *r, struct cell *ci, struct cell *cj) {
+void DOPAIR1(struct runner *r, struct cell *ci, struct cell *cj, const int sid,
+             const double *shift) {
 
   const struct engine *restrict e = r->e;
 
@@ -900,22 +903,6 @@ void DOPAIR1(struct runner *r, struct cell *ci, struct cell *cj) {
 
   TIMER_TIC;
 
-  /* Anything to do here? */
-  if (!cell_is_active(ci, e) && !cell_is_active(cj, e)) return;
-
-  if (!cell_are_part_drifted(ci, e) || !cell_are_part_drifted(cj, e))
-    error("Interacting undrifted cells.");
-
-  /* Get the sort ID. */
-  double shift[3] = {0.0, 0.0, 0.0};
-  const int sid = space_getsid(e->s, &ci, &cj, shift);
-
-  /* Have the cells been sorted? */
-  if (!(ci->sorted & (1 << sid)) || ci->dx_max_sort > space_maxreldx * ci->dmin)
-    runner_do_sort(r, ci, (1 << sid), 1);
-  if (!(cj->sorted & (1 << sid)) || cj->dx_max_sort > space_maxreldx * cj->dmin)
-    runner_do_sort(r, cj, (1 << sid), 1);
-
   /* Get the cutoff shift. */
   double rshift = 0.0;
   for (int k = 0; k < 3; k++) rshift += shift[k] * runner_shift[sid][k];
@@ -1116,6 +1103,51 @@ void DOPAIR1(struct runner *r, struct cell *ci, struct cell *cj) {
   TIMER_TOC(TIMER_DOPAIR);
 }
 
+/**
+ * @brief Determine which version of DOPAIR1 needs to be called depending on the
+ * orientation of the cells or whether DOPAIR1 needs to be called at all.
+ *
+ * @param r #runner
+ * @param ci #cell ci
+ * @param cj #cell cj
+ *
+ */
+void DOPAIR1_BRANCH(struct runner *r, struct cell *ci, struct cell *cj) {
+
+  const struct engine *restrict e = r->e;
+
+  /* Anything to do here? */
+  if (!cell_is_active(ci, e) && !cell_is_active(cj, e)) return;
+
+  /* Check that cells are drifted. */
+  if (!cell_are_part_drifted(ci, e) || !cell_are_part_drifted(cj, e))
+    error("Interacting undrifted cells.");
+
+  /* Get the sort ID. */
+  double shift[3] = {0.0, 0.0, 0.0};
+  const int sid = space_getsid(e->s, &ci, &cj, shift);
+
+  /* Have the cells been sorted? */
+  if (!(ci->sorted & (1 << sid)) || ci->dx_max_sort > space_maxreldx * ci->dmin)
+    runner_do_sort(r, ci, (1 << sid), 1);
+  if (!(cj->sorted & (1 << sid)) || cj->dx_max_sort > space_maxreldx * cj->dmin)
+    runner_do_sort(r, cj, (1 << sid), 1);
+
+  /* Have the cells been sorted? */
+  if (!(ci->sorted & (1 << sid)) || !(cj->sorted & (1 << sid)))
+    error("Trying to interact unsorted cells.");
+
+#if defined(WITH_VECTORIZATION) && defined(GADGET2_SPH) && \
+    (DOPAIR1_BRANCH == runner_dopair1_density_branch)
+  if (!sort_is_corner(sid))
+    runner_dopair1_density_vec(r, ci, cj, sid, shift);
+  else
+    DOPAIR1(r, ci, cj, sid, shift);
+#else
+  DOPAIR1(r, ci, cj, sid, shift);
+#endif
+}
+
 /**
  * @brief Compute the interactions between a cell pair (symmetric)
  *
@@ -2290,13 +2322,8 @@ void DOSUB_PAIR1(struct runner *r, struct cell *ci, struct cell *cj, int sid,
         cj->dx_max_sort > cj->dmin * space_maxreldx)
       runner_do_sort(r, cj, (1 << sid), 1);
 
-/* Compute the interactions. */
-#if (DOPAIR1 == runner_dopair1_density) && defined(WITH_VECTORIZATION) && \
-    defined(GADGET2_SPH)
-    runner_dopair1_density_vec(r, ci, cj);
-#else
-    DOPAIR1(r, ci, cj);
-#endif
+    /* Compute the interactions. */
+    DOPAIR1_BRANCH(r, ci, cj);
   }
 
   if (gettimer) TIMER_TOC(TIMER_DOSUB_PAIR);

src/runner_doiact_vec.c

@@ -20,13 +20,12 @@
 /* Config parameters. */
 #include "../config.h"
 
-#include "swift.h"
-
-#include "active.h"
-
 /* This object's header. */
 #include "runner_doiact_vec.h"
 
+/* Local headers. */
+#include "active.h"
+
 #ifdef WITH_VECTORIZATION
 /**
  * @brief Compute the vector remainder interactions from the secondary cache.
@@ -262,40 +261,40 @@ __attribute__((always_inline)) INLINE static void storeInteractions(
   }
 }
 
-/* @brief Populates the arrays max_di and max_dj with the maximum distances of
+/**
+ * @brief Populates the arrays max_di and max_dj with the maximum distances of
  * particles into their neighbouring cells. Also finds the first pi that
  * interacts with any particle in cj and the last pj that interacts with any
  * particle in ci.
+ *
  * @param ci #cell pointer to ci
  * @param cj #cell pointer to cj
  * @param sort_i #entry array for particle distance in ci
  * @param sort_j #entry array for particle distance in cj
- * @param ci_cache #cache for cell ci
- * @param cj_cache #cache for cell cj
  * @param dx_max maximum particle movement allowed in cell
  * @param rshift cutoff shift
+ * @param hi_max Maximal smoothing length in cell ci
+ * @param hj_max Maximal smoothing length in cell cj
+ * @param di_max Maximal position on the axis that can interact in cell ci
+ * @param dj_min Minimal position on the axis that can interact in cell ci
  * @param max_di array to hold the maximum distances of pi particles into cell
  * cj
  * @param max_dj array to hold the maximum distances of pj particles into cell
  * cj
  * @param init_pi first pi to interact with a pj particle
  * @param init_pj last pj to interact with a pi particle
+ * @param e The #engine.
  */
 __attribute__((always_inline)) INLINE static void populate_max_d_no_cache(
     const struct cell *ci, const struct cell *cj,
     const struct entry *restrict sort_i, const struct entry *restrict sort_j,
-    const float dx_max, const float rshift, float *max_di, float *max_dj,
-    int *init_pi, int *init_pj, const struct engine *e) {
+    const float dx_max, const float rshift, const double hi_max,
+    const double hj_max, const double di_max, const double dj_min,
+    float *max_di, float *max_dj, int *init_pi, int *init_pj,
+    const struct engine *e) {
 
-  struct part *restrict parts_i = ci->parts;
-  struct part *restrict parts_j = cj->parts;
-  struct part *p = &parts_i[sort_i[0].i];
-
-  float h, d;
-
-  /* Get the distance of the last pi and the first pj on the sorted axis.*/
-  const float di_max = sort_i[ci->count - 1].d - rshift;
-  const float dj_min = sort_j[0].d;
+  const struct part *restrict parts_i = ci->parts;
+  const struct part *restrict parts_j = cj->parts;
 
   int first_pi = 0, last_pj = cj->count - 1;
 
@@ -304,19 +303,19 @@ __attribute__((always_inline)) INLINE static void populate_max_d_no_cache(
   /* Populate max_di with distances. */
   int active_id = ci->count - 1;
   for (int k = ci->count - 1; k >= 0; k--) {
-    p = &parts_i[sort_i[k].i];
-    h = p->h;
-    d = sort_i[k].d + h * kernel_gamma + dx_max - rshift;
+    const struct part *pi = &parts_i[sort_i[k].i];
+    const float d = sort_i[k].d + dx_max;
 
-    max_di[k] = d;
+    // max_di[k] = d + h * kernel_gamma - rshift;
+    max_di[k] = d + hi_max;
 
     /* If the particle is out of range set the index to
      * the last active particle within range. */
-    if (d < dj_min) {
+    if (d + hi_max < dj_min) {
       first_pi = active_id;
       break;
     } else {
-      if (part_is_active(p, e)) active_id = k;
+      if (part_is_active(pi, e)) active_id = k;
     }
   }
 
@@ -329,19 +328,20 @@ __attribute__((always_inline)) INLINE static void populate_max_d_no_cache(
   /* Populate max_dj with distances. */
   active_id = 0;
   for (int k = 0; k < cj->count; k++) {
-    p = &parts_j[sort_j[k].i];
-    h = p->h;
-    d = sort_j[k].d - h * kernel_gamma - dx_max - rshift;
+    const struct part *pj = &parts_j[sort_j[k].i];
+    const float d = sort_j[k].d - dx_max;
 
-    max_dj[k] = d;
+    /*TODO: don't think rshift should be taken off here, waiting on Pedro. */
+    // max_dj[k] = d - h * kernel_gamma - rshift;
+    max_dj[k] = d - hj_max;
 
     /* If the particle is out of range set the index to
      * the last active particle within range. */
-    if (d > di_max) {
+    if (d - hj_max > di_max) {
       last_pj = active_id;
       break;
     } else {
-      if (part_is_active(p, e)) active_id = k;
+      if (part_is_active(pj, e)) active_id = k;
     }
   }
 
@@ -611,9 +611,12 @@ __attribute__((always_inline)) INLINE void runner_doself1_density_vec(
  * @param r The #runner.
  * @param ci The first #cell.
  * @param cj The second #cell.
+ * @param sid The direction of the pair
+ * @param shift The shift vector to apply to the particles in ci.
  */
 void runner_dopair1_density_vec(struct runner *r, struct cell *ci,
-                                struct cell *cj) {
+                                struct cell *cj, const int sid,
+                                const double *shift) {
 
 #ifdef WITH_VECTORIZATION
   const struct engine *restrict e = r->e;
@@ -622,22 +625,6 @@ void runner_dopair1_density_vec(struct runner *r, struct cell *ci,
 
   TIMER_TIC;
 
-  /* Anything to do here? */
-  if (!cell_is_active(ci, e) && !cell_is_active(cj, e)) return;
-
-  if (!cell_are_part_drifted(ci, e) || !cell_are_part_drifted(cj, e))
-    error("Interacting undrifted cells.");
-
-  /* Get the sort ID. */
-  double shift[3] = {0.0, 0.0, 0.0};
-  const int sid = space_getsid(e->s, &ci, &cj, shift);
-
-  /* Have the cells been sorted? */
-  if (!(ci->sorted & (1 << sid)) || ci->dx_max_sort > space_maxreldx * ci->dmin)
-    runner_do_sort(r, ci, (1 << sid), 1);
-  if (!(cj->sorted & (1 << sid)) || cj->dx_max_sort > space_maxreldx * cj->dmin)
-    runner_do_sort(r, cj, (1 << sid), 1);
-
   /* Get the cutoff shift. */
   double rshift = 0.0;
   for (int k = 0; k < 3; k++) rshift += shift[k] * runner_shift[sid][k];
@@ -726,8 +713,9 @@ void runner_dopair1_density_vec(struct runner *r, struct cell *ci,
   /* Find particles maximum distance into cj, max_di[] and ci, max_dj[]. */
   /* Also find the first pi that interacts with any particle in cj and the last
    * pj that interacts with any particle in ci. */
-  populate_max_d_no_cache(ci, cj, sort_i, sort_j, dx_max, rshift, max_di,
-                          max_dj, &first_pi, &last_pj, e);
+  populate_max_d_no_cache(ci, cj, sort_i, sort_j, dx_max, rshift, hi_max,
+                          hj_max, di_max, dj_min, max_di, max_dj, &first_pi,
+                          &last_pj, e);
 
   /* Find the maximum index into cj that is required by a particle in ci. */
   /* Find the maximum index into ci that is required by a particle in cj. */
@@ -770,13 +758,22 @@ void runner_dopair1_density_vec(struct runner *r, struct cell *ci,
 
   if (cell_is_active(ci, e)) {
 
-    /* Loop over the parts in ci. */
+    /* Loop over the parts in ci until nothing is within range in cj. */
     for (int pid = count_i - 1; pid >= first_pi_loop && max_ind_j >= 0; pid--) {
 
       /* Get a hold of the ith part in ci. */
       struct part *restrict pi = &parts_i[sort_i[pid].i];
       if (!part_is_active(pi, e)) continue;
 
+      /* Set the cache index. */
+      int ci_cache_idx = pid - first_pi_align;
+
+      /* Skip this particle if no particle in cj is within range of it. */
+      const float hi = ci_cache->h[ci_cache_idx];
+      const double di_test =
+          sort_i[pid].d + hi * kernel_gamma + dx_max - rshift;
+      if (di_test < dj_min) continue;
+
       /* Determine the exit iteration of the interaction loop. */
       dj = sort_j[max_ind_j].d;
       while (max_ind_j > 0 && max_di[pid] < dj) {
@@ -786,10 +783,6 @@ void runner_dopair1_density_vec(struct runner *r, struct cell *ci,
       }
       int exit_iteration = max_ind_j + 1;
 
-      /* Set the cache index. */
-      int ci_cache_idx = pid - first_pi_align;
-
-      const float hi = ci_cache->h[ci_cache_idx];
       const float hig2 = hi * hi * kernel_gamma2;
 
       vector pix, piy, piz;
@@ -903,13 +896,24 @@ void runner_dopair1_density_vec(struct runner *r, struct cell *ci,
   }
 
   if (cell_is_active(cj, e)) {
-    /* Loop over the parts in cj. */
+
+    /* Loop over the parts in cj until nothing is within range in ci. */
     for (int pjd = 0; pjd <= last_pj_loop && max_ind_i < count_i; pjd++) {
 
       /* Get a hold of the jth part in cj. */
       struct part *restrict pj = &parts_j[sort_j[pjd].i];
       if (!part_is_active(pj, e)) continue;
 
+      /* Set the cache index. */
+      int cj_cache_idx = pjd;
+
+      /*TODO: rshift term. */
+      /* Skip this particle if no particle in ci is within range of it. */
+      const float hj = cj_cache->h[cj_cache_idx];
+      const double dj_test =
+          sort_j[pjd].d - hj * kernel_gamma - dx_max - rshift;
+      if (dj_test > di_max) continue;
+
       /* Determine the exit iteration of the interaction loop. */
       di = sort_i[max_ind_i].d;
       while (max_ind_i < count_i - 1 && max_dj[pjd] > di) {
@@ -919,10 +923,6 @@ void runner_dopair1_density_vec(struct runner *r, struct cell *ci,
       }
       int exit_iteration = max_ind_i;
 
-      /* Set the cache index. */
-      int cj_cache_idx = pjd;
-
-      const float hj = cj_cache->h[cj_cache_idx];
       const float hjg2 = hj * hj * kernel_gamma2;
 
       vector pjx, pjy, pjz;

src/runner_doiact_vec.h

@@ -35,8 +35,8 @@
 
 /* Function prototypes. */
 void runner_doself1_density_vec(struct runner *r, struct cell *restrict c);
-void runner_doself1_density_vec_2(struct runner *r, struct cell *restrict c);
 void runner_dopair1_density_vec(struct runner *r, struct cell *restrict ci,
-                                struct cell *restrict cj);
+                                struct cell *restrict cj, const int sid,
+                                const double *shift);
 
 #endif /* SWIFT_RUNNER_VEC_H */

src/serial_io.c

@@ -59,19 +59,15 @@
  * @brief Reads a data array from a given HDF5 group.
  *
  * @param grp The group from which to read.
- * @param name The name of the array to read.
- * @param type The #DATA_TYPE of the attribute.
- * @param N The number of particles.
- * @param dim The dimension of the data (1 for scalar, 3 for vector)
- * @param part_c A (char*) pointer on the first occurrence of the field of
- *interest in the parts array
- * @param partSize The size in bytes of the particle structure.
- * @param importance If COMPULSORY, the data must be present in the IC file. If
- *OPTIONAL, the array will be zeroed when the data is not present.
+ * @param props The #io_props of the field to read
+ * @param N The number of particles to read on this rank.
+ * @param N_total The total number of particles on all ranks.
+ * @param offset The offset position where this rank starts reading.
+ * @param internal_units The #unit_system used internally
+ * @param ic_units The #unit_system used in the ICs
  *
  * @todo A better version using HDF5 hyper-slabs to read the file directly into
- *the part array
- * will be written once the structures have been stabilized.
+ * the part array will be written once the structures have been stabilized.
  */
 void readArray(hid_t grp, const struct io_props props, size_t N,
                long long N_total, long long offset,
@@ -274,16 +270,17 @@ void prepareArray(struct engine* e, hid_t grp, char* fileName, FILE* xmfFile,
  * @param fileName The name of the file in which the data is written
  * @param xmfFile The FILE used to write the XMF description
  * @param partTypeGroupName The name of the group containing the particles in
- *the HDF5 file.
- * @param name The name of the array to write.
- * @param type The #DATA_TYPE of the array.
+ * the HDF5 file.
+ * @param props The #io_props of the field to read
  * @param N The number of particles to write.
- * @param dim The dimension of the data (1 for scalar, 3 for vector)
- * @param part_c A (char*) pointer on the first occurrence of the field of
- *interest in the parts array
- * @param partSize The size in bytes of the particle structure.
- * @param us The unit_system currently in use
- * @param convFactor The UnitConversionFactor for this arrayo
+ * @param N_total The total number of particles on all ranks.
+ * @param offset The offset position where this rank starts writing.
+ * @param mpi_rank The MPI rank of this node
+ * @param internal_units The #unit_system used internally
+ * @param snapshot_units The #unit_system used in the snapshots
+ *
+ * @todo A better version using HDF5 hyper-slabs to write the file directly from
+ * the part array will be written once the structures have been stabilized.
  */
 void writeArray(struct engine* e, hid_t grp, char* fileName, FILE* xmfFile,
                 char* partTypeGroupName, const struct io_props props, size_t N,

src/single_io.c

@@ -64,8 +64,7 @@
  * @param ic_units The #unit_system used in the ICs
  *
  * @todo A better version using HDF5 hyper-slabs to read the file directly into
- *the part array
- * will be written once the structures have been stabilized.
+ * the part array will be written once the structures have been stabilized.
  */
 void readArray(hid_t h_grp, const struct io_props prop, size_t N,
                const struct unit_system* internal_units,

tests/Makefile.am

@@ -97,6 +97,6 @@ EXTRA_DIST = testReading.sh makeInput.py testPair.sh testPairPerturbed.sh \
 	     test27cells.sh test27cellsPerturbed.sh testParser.sh testPeriodicBC.sh \
 	     testPeriodicBCPerturbed.sh test125cells.sh test125cellsPerturbed.sh testParserInput.yaml \
 	     difffloat.py tolerance_125_normal.dat tolerance_125_perturbed.dat \
-             tolerance_27_normal.dat tolerance_27_perturbed.dat \
+             tolerance_27_normal.dat tolerance_27_perturbed.dat tolerance_27_perturbed_h.dat \
 	     tolerance_pair_normal.dat tolerance_pair_perturbed.dat \
 	     fft_params.yml tolerance_periodic_BC_normal.dat tolerance_periodic_BC_perturbed.dat