diff --git a/src/engine.c b/src/engine.c
index 41103b06253eed6de3046d5efc82de2a3c242290..91e69aaba5ece90dd36f41895fae4fc88cb8d327 100644
--- a/src/engine.c
+++ b/src/engine.c
@@ -1420,14 +1420,6 @@ void engine_maketasks(struct engine *e) {
/* Re-set the scheduler. */
scheduler_reset(sched, s->tot_cells * engine_maxtaskspercell);
- /* Add the space sorting tasks. */
- for (int i = 0; i < e->nr_threads; i++) {
- scheduler_addtask(sched, task_type_part_sort, task_subtype_none, i, 0, NULL,
- NULL, 0);
- scheduler_addtask(sched, task_type_gpart_sort, task_subtype_none, i, 0,
- NULL, NULL, 0);
- }
-
/* Construct the firt hydro loop over neighbours */
engine_make_hydroloop_tasks(e);
@@ -2676,15 +2668,6 @@ void engine_init(struct engine *e, struct space *s,
scheduler_init(&e->sched, e->s, nr_tasks, nr_queues, scheduler_flag_steal,
e->nodeID, &e->threadpool);
- /* Create the sorting tasks. */
- for (int i = 0; i < e->nr_threads; i++) {
- scheduler_addtask(&e->sched, task_type_part_sort, task_subtype_none, i, 0,
- NULL, NULL, 0);
-
- scheduler_addtask(&e->sched, task_type_gpart_sort, task_subtype_none, i, 0,
- NULL, NULL, 0);
- }
-
scheduler_ranktasks(&e->sched);
/* Allocate and init the threads. */
diff --git a/src/runner.c b/src/runner.c
index fe86e65b879ba77300d1e140246d6e0b77c08893..7a2bcd8563c62250f44183aef3f4b1457623633e 100644
--- a/src/runner.c
+++ b/src/runner.c
@@ -1395,12 +1395,6 @@ void *runner_main(void *data) {
case task_type_grav_external:
runner_do_grav_external(r, t->ci, 1);
break;
- case task_type_part_sort:
- space_do_parts_sort();
- break;
- case task_type_gpart_sort:
- space_do_gparts_sort();
- break;
case task_type_comm_root:
break;
default:
diff --git a/src/scheduler.c b/src/scheduler.c
index 2b61c4ea73ae403b70d96e294e82fc2b3486575d..ca455d91ad58e3b83c88885e78d8402fa9f3691f 100644
--- a/src/scheduler.c
+++ b/src/scheduler.c
@@ -123,11 +123,6 @@ void scheduler_splittasks(struct scheduler *s) {
break;
}
- /* Skip sorting tasks. */
- if (t->type == task_type_part_sort) continue;
-
- if (t->type == task_type_gpart_sort) continue;
-
/* Empty task? */
if (t->ci == NULL || (t->type == task_type_pair && t->cj == NULL)) {
t->type = task_type_none;
diff --git a/src/space.c b/src/space.c
index 75c66d7a7476aa88315deab82e9bb18669315b67..a36906f6818c6bab0dbef717626fe537289f1715 100644
--- a/src/space.c
+++ b/src/space.c
@@ -50,9 +50,6 @@
#include "threadpool.h"
#include "tools.h"
-/* Shared sort structure. */
-struct parallel_sort space_sort_struct;
-
/* Split size. */
int space_splitsize = space_splitsize_default;
int space_subsize = space_subsize_default;
@@ -548,7 +545,7 @@ void space_rebuild(struct space *s, double cell_max, int verbose) {
for (size_t k = 0; k < nr_parts; k++)
if (s->parts[k].gpart != NULL) s->parts[k].gpart->part = &s->parts[k];
- /* Verify space_sort_struct. */
+ /* Verify sort_struct. */
/* for ( k = 1 ; k < nr_parts ; k++ ) {
if ( ind[k-1] > ind[k] ) {
error( "Sort failed!" );
@@ -681,31 +678,34 @@ void space_parts_sort(struct space *s, int *ind, size_t N, int min, int max,
const ticks tic = getticks();
- /*Populate the global parallel_sort structure with the input data */
- space_sort_struct.parts = s->parts;
- space_sort_struct.xparts = s->xparts;
- space_sort_struct.ind = ind;
- space_sort_struct.stack_size = 2 * (max - min + 1) + 10 + s->e->nr_threads;
- if ((space_sort_struct.stack = malloc(sizeof(struct qstack) *
- space_sort_struct.stack_size)) == NULL)
+ /* Populate a parallel_sort structure with the input data */
+ struct parallel_sort sort_struct;
+ sort_struct.parts = s->parts;
+ sort_struct.xparts = s->xparts;
+ sort_struct.ind = ind;
+ sort_struct.stack_size = 2 * (max - min + 1) + 10 + s->e->nr_threads;
+ if ((sort_struct.stack = malloc(sizeof(struct qstack) *
+ sort_struct.stack_size)) == NULL)
error("Failed to allocate sorting stack.");
- for (int i = 0; i < space_sort_struct.stack_size; i++)
- space_sort_struct.stack[i].ready = 0;
+ for (int i = 0; i < sort_struct.stack_size; i++)
+ sort_struct.stack[i].ready = 0;
/* Add the first interval. */
- space_sort_struct.stack[0].i = 0;
- space_sort_struct.stack[0].j = N - 1;
- space_sort_struct.stack[0].min = min;
- space_sort_struct.stack[0].max = max;
- space_sort_struct.stack[0].ready = 1;
- space_sort_struct.first = 0;
- space_sort_struct.last = 1;
- space_sort_struct.waiting = 1;
-
- /* Launch the sorting tasks. */
- engine_launch(s->e, s->e->nr_threads, (1 << task_type_part_sort), 0);
-
- /* Verify space_sort_struct. */
+ sort_struct.stack[0].i = 0;
+ sort_struct.stack[0].j = N - 1;
+ sort_struct.stack[0].min = min;
+ sort_struct.stack[0].max = max;
+ sort_struct.stack[0].ready = 1;
+ sort_struct.first = 0;
+ sort_struct.last = 1;
+ sort_struct.waiting = 1;
+
+ /* Launch the sorting tasks with a stride of zero such that the same
+ map data is passed to each thread. */
+ threadpool_map(&s->e->threadpool, space_parts_sort_mapper,
+ &sort_struct, s->e->threadpool.num_threads, 0, NULL);
+
+ /* Verify sort_struct. */
/* for (int i = 1; i < N; i++)
if (ind[i - 1] > ind[i])
error("Sorting failed (ind[%i]=%i,ind[%i]=%i), min=%i, max=%i.", i - 1,
@@ -714,37 +714,40 @@ void space_parts_sort(struct space *s, int *ind, size_t N, int min, int max,
message("Sorting succeeded."); */
/* Clean up. */
- free(space_sort_struct.stack);
+ free(sort_struct.stack);
if (verbose)
message("took %.3f %s.", clocks_from_ticks(getticks() - tic),
clocks_getunit());
}
-void space_do_parts_sort() {
+void space_parts_sort_mapper(void *map_data, void *extra_data) {
+
+ /* Unpack the mapping data. */
+ struct parallel_sort *sort_struct = (struct parallel_sort *)map_data;
/* Pointers to the sorting data. */
- int *ind = space_sort_struct.ind;
- struct part *parts = space_sort_struct.parts;
- struct xpart *xparts = space_sort_struct.xparts;
+ int *ind = sort_struct->ind;
+ struct part *parts = sort_struct->parts;
+ struct xpart *xparts = sort_struct->xparts;
/* Main loop. */
- while (space_sort_struct.waiting) {
+ while (sort_struct->waiting) {
/* Grab an interval off the queue. */
int qid =
- atomic_inc(&space_sort_struct.first) % space_sort_struct.stack_size;
+ atomic_inc(&sort_struct->first) % sort_struct->stack_size;
/* Wait for the entry to be ready, or for the sorting do be done. */
- while (!space_sort_struct.stack[qid].ready)
- if (!space_sort_struct.waiting) return;
+ while (!sort_struct->stack[qid].ready)
+ if (!sort_struct->waiting) return;
/* Get the stack entry. */
- ptrdiff_t i = space_sort_struct.stack[qid].i;
- ptrdiff_t j = space_sort_struct.stack[qid].j;
- int min = space_sort_struct.stack[qid].min;
- int max = space_sort_struct.stack[qid].max;
- space_sort_struct.stack[qid].ready = 0;
+ ptrdiff_t i = sort_struct->stack[qid].i;
+ ptrdiff_t j = sort_struct->stack[qid].j;
+ int min = sort_struct->stack[qid].min;
+ int max = sort_struct->stack[qid].max;
+ sort_struct->stack[qid].ready = 0;
/* Loop over sub-intervals. */
while (1) {
@@ -773,7 +776,7 @@ void space_do_parts_sort() {
}
}
- /* Verify space_sort_struct. */
+ /* Verify sort_struct. */
/* for (int k = i; k <= jj; k++)
if (ind[k] > pivot) {
message("sorting failed at k=%i, ind[k]=%i, pivot=%i, i=%i, j=%i.", k,
@@ -792,18 +795,18 @@ void space_do_parts_sort() {
/* Recurse on the left? */
if (jj > i && pivot > min) {
- qid = atomic_inc(&space_sort_struct.last) %
- space_sort_struct.stack_size;
- while (space_sort_struct.stack[qid].ready)
+ qid = atomic_inc(&sort_struct->last) %
+ sort_struct->stack_size;
+ while (sort_struct->stack[qid].ready)
;
- space_sort_struct.stack[qid].i = i;
- space_sort_struct.stack[qid].j = jj;
- space_sort_struct.stack[qid].min = min;
- space_sort_struct.stack[qid].max = pivot;
- if (atomic_inc(&space_sort_struct.waiting) >=
- space_sort_struct.stack_size)
+ sort_struct->stack[qid].i = i;
+ sort_struct->stack[qid].j = jj;
+ sort_struct->stack[qid].min = min;
+ sort_struct->stack[qid].max = pivot;
+ if (atomic_inc(&sort_struct->waiting) >=
+ sort_struct->stack_size)
error("Qstack overflow.");
- space_sort_struct.stack[qid].ready = 1;
+ sort_struct->stack[qid].ready = 1;
}
/* Recurse on the right? */
@@ -817,18 +820,18 @@ void space_do_parts_sort() {
/* Recurse on the right? */
if (pivot + 1 < max) {
- qid = atomic_inc(&space_sort_struct.last) %
- space_sort_struct.stack_size;
- while (space_sort_struct.stack[qid].ready)
+ qid = atomic_inc(&sort_struct->last) %
+ sort_struct->stack_size;
+ while (sort_struct->stack[qid].ready)
;
- space_sort_struct.stack[qid].i = jj + 1;
- space_sort_struct.stack[qid].j = j;
- space_sort_struct.stack[qid].min = pivot + 1;
- space_sort_struct.stack[qid].max = max;
- if (atomic_inc(&space_sort_struct.waiting) >=
- space_sort_struct.stack_size)
+ sort_struct->stack[qid].i = jj + 1;
+ sort_struct->stack[qid].j = j;
+ sort_struct->stack[qid].min = pivot + 1;
+ sort_struct->stack[qid].max = max;
+ if (atomic_inc(&sort_struct->waiting) >=
+ sort_struct->stack_size)
error("Qstack overflow.");
- space_sort_struct.stack[qid].ready = 1;
+ sort_struct->stack[qid].ready = 1;
}
/* Recurse on the left? */
@@ -841,7 +844,7 @@ void space_do_parts_sort() {
} /* loop over sub-intervals. */
- atomic_dec(&space_sort_struct.waiting);
+ atomic_dec(&sort_struct->waiting);
} /* main loop. */
}
@@ -862,30 +865,33 @@ void space_gparts_sort(struct space *s, int *ind, size_t N, int min, int max,
const ticks tic = getticks();
- /*Populate the global parallel_sort structure with the input data */
- space_sort_struct.gparts = s->gparts;
- space_sort_struct.ind = ind;
- space_sort_struct.stack_size = 2 * (max - min + 1) + 10 + s->e->nr_threads;
- if ((space_sort_struct.stack = malloc(sizeof(struct qstack) *
- space_sort_struct.stack_size)) == NULL)
+ /*Populate a global parallel_sort structure with the input data */
+ struct parallel_sort sort_struct;
+ sort_struct.gparts = s->gparts;
+ sort_struct.ind = ind;
+ sort_struct.stack_size = 2 * (max - min + 1) + 10 + s->e->nr_threads;
+ if ((sort_struct.stack = malloc(sizeof(struct qstack) *
+ sort_struct.stack_size)) == NULL)
error("Failed to allocate sorting stack.");
- for (int i = 0; i < space_sort_struct.stack_size; i++)
- space_sort_struct.stack[i].ready = 0;
+ for (int i = 0; i < sort_struct.stack_size; i++)
+ sort_struct.stack[i].ready = 0;
/* Add the first interval. */
- space_sort_struct.stack[0].i = 0;
- space_sort_struct.stack[0].j = N - 1;
- space_sort_struct.stack[0].min = min;
- space_sort_struct.stack[0].max = max;
- space_sort_struct.stack[0].ready = 1;
- space_sort_struct.first = 0;
- space_sort_struct.last = 1;
- space_sort_struct.waiting = 1;
-
- /* Launch the sorting tasks. */
- engine_launch(s->e, s->e->nr_threads, (1 << task_type_gpart_sort), 0);
-
- /* Verify space_sort_struct. */
+ sort_struct.stack[0].i = 0;
+ sort_struct.stack[0].j = N - 1;
+ sort_struct.stack[0].min = min;
+ sort_struct.stack[0].max = max;
+ sort_struct.stack[0].ready = 1;
+ sort_struct.first = 0;
+ sort_struct.last = 1;
+ sort_struct.waiting = 1;
+
+ /* Launch the sorting tasks with a stride of zero such that the same
+ map data is passed to each thread. */
+ threadpool_map(&s->e->threadpool, space_gparts_sort_mapper,
+ &sort_struct, s->e->threadpool.num_threads, 0, NULL);
+
+ /* Verify sort_struct. */
/* for (int i = 1; i < N; i++)
if (ind[i - 1] > ind[i])
error("Sorting failed (ind[%i]=%i,ind[%i]=%i), min=%i, max=%i.", i - 1,
@@ -894,36 +900,39 @@ void space_gparts_sort(struct space *s, int *ind, size_t N, int min, int max,
message("Sorting succeeded."); */
/* Clean up. */
- free(space_sort_struct.stack);
+ free(sort_struct.stack);
if (verbose)
message("took %.3f %s.", clocks_from_ticks(getticks() - tic),
clocks_getunit());
}
-void space_do_gparts_sort() {
+void space_gparts_sort_mapper(void *map_data, void *extra_data) {
+
+ /* Unpack the mapping data. */
+ struct parallel_sort *sort_struct = (struct parallel_sort *)map_data;
/* Pointers to the sorting data. */
- int *ind = space_sort_struct.ind;
- struct gpart *gparts = space_sort_struct.gparts;
+ int *ind = sort_struct->ind;
+ struct gpart *gparts = sort_struct->gparts;
/* Main loop. */
- while (space_sort_struct.waiting) {
+ while (sort_struct->waiting) {
/* Grab an interval off the queue. */
int qid =
- atomic_inc(&space_sort_struct.first) % space_sort_struct.stack_size;
+ atomic_inc(&sort_struct->first) % sort_struct->stack_size;
/* Wait for the entry to be ready, or for the sorting do be done. */
- while (!space_sort_struct.stack[qid].ready)
- if (!space_sort_struct.waiting) return;
+ while (!sort_struct->stack[qid].ready)
+ if (!sort_struct->waiting) return;
/* Get the stack entry. */
- ptrdiff_t i = space_sort_struct.stack[qid].i;
- ptrdiff_t j = space_sort_struct.stack[qid].j;
- int min = space_sort_struct.stack[qid].min;
- int max = space_sort_struct.stack[qid].max;
- space_sort_struct.stack[qid].ready = 0;
+ ptrdiff_t i = sort_struct->stack[qid].i;
+ ptrdiff_t j = sort_struct->stack[qid].j;
+ int min = sort_struct->stack[qid].min;
+ int max = sort_struct->stack[qid].max;
+ sort_struct->stack[qid].ready = 0;
/* Loop over sub-intervals. */
while (1) {
@@ -949,7 +958,7 @@ void space_do_gparts_sort() {
}
}
- /* Verify space_sort_struct. */
+ /* Verify sort_struct. */
/* for (int k = i; k <= jj; k++)
if (ind[k] > pivot) {
message("sorting failed at k=%i, ind[k]=%i, pivot=%i, i=%i, j=%i.", k,
@@ -968,18 +977,18 @@ void space_do_gparts_sort() {
/* Recurse on the left? */
if (jj > i && pivot > min) {
- qid = atomic_inc(&space_sort_struct.last) %
- space_sort_struct.stack_size;
- while (space_sort_struct.stack[qid].ready)
+ qid = atomic_inc(&sort_struct->last) %
+ sort_struct->stack_size;
+ while (sort_struct->stack[qid].ready)
;
- space_sort_struct.stack[qid].i = i;
- space_sort_struct.stack[qid].j = jj;
- space_sort_struct.stack[qid].min = min;
- space_sort_struct.stack[qid].max = pivot;
- if (atomic_inc(&space_sort_struct.waiting) >=
- space_sort_struct.stack_size)
+ sort_struct->stack[qid].i = i;
+ sort_struct->stack[qid].j = jj;
+ sort_struct->stack[qid].min = min;
+ sort_struct->stack[qid].max = pivot;
+ if (atomic_inc(&sort_struct->waiting) >=
+ sort_struct->stack_size)
error("Qstack overflow.");
- space_sort_struct.stack[qid].ready = 1;
+ sort_struct->stack[qid].ready = 1;
}
/* Recurse on the right? */
@@ -993,18 +1002,18 @@ void space_do_gparts_sort() {
/* Recurse on the right? */
if (pivot + 1 < max) {
- qid = atomic_inc(&space_sort_struct.last) %
- space_sort_struct.stack_size;
- while (space_sort_struct.stack[qid].ready)
+ qid = atomic_inc(&sort_struct->last) %
+ sort_struct->stack_size;
+ while (sort_struct->stack[qid].ready)
;
- space_sort_struct.stack[qid].i = jj + 1;
- space_sort_struct.stack[qid].j = j;
- space_sort_struct.stack[qid].min = pivot + 1;
- space_sort_struct.stack[qid].max = max;
- if (atomic_inc(&space_sort_struct.waiting) >=
- space_sort_struct.stack_size)
+ sort_struct->stack[qid].i = jj + 1;
+ sort_struct->stack[qid].j = j;
+ sort_struct->stack[qid].min = pivot + 1;
+ sort_struct->stack[qid].max = max;
+ if (atomic_inc(&sort_struct->waiting) >=
+ sort_struct->stack_size)
error("Qstack overflow.");
- space_sort_struct.stack[qid].ready = 1;
+ sort_struct->stack[qid].ready = 1;
}
/* Recurse on the left? */
@@ -1017,7 +1026,7 @@ void space_do_gparts_sort() {
} /* loop over sub-intervals. */
- atomic_dec(&space_sort_struct.waiting);
+ atomic_dec(&sort_struct->waiting);
} /* main loop. */
}
diff --git a/src/space.h b/src/space.h
index ff046562c720b092d89db6672c355672a275adf4..e70e77f15f0b1ea898fb328746b484a6e61f6738 100644
--- a/src/space.h
+++ b/src/space.h
@@ -127,7 +127,6 @@ struct parallel_sort {
unsigned int stack_size;
volatile unsigned int first, last, waiting;
};
-extern struct parallel_sort space_sort_struct;
/* function prototypes. */
void space_parts_sort(struct space *s, int *ind, size_t N, int min, int max,
@@ -151,6 +150,8 @@ void space_map_parts_xparts(struct space *s,
struct cell *c));
void space_map_cells_post(struct space *s, int full,
void (*fun)(struct cell *c, void *data), void *data);
+void space_parts_sort_mapper(void *map_data, void *extra_data);
+void space_gparts_sort_mapper(void *map_data, void *extra_data);
void space_rebuild(struct space *s, double h_max, int verbose);
void space_recycle(struct space *s, struct cell *c);
void space_split(struct space *s, struct cell *cells, int verbose);
diff --git a/src/task.c b/src/task.c
index 964916ce5e0881fa9ad5afb25edac38ab06cf689..f7a04454b70688830f9577999c789cb3be7ee619 100644
--- a/src/task.c
+++ b/src/task.c
@@ -47,10 +47,9 @@
/* Task type names. */
const char *taskID_names[task_type_count] = {
- "none", "sort", "self", "pair", "sub",
- "init", "ghost", "drift", "kick", "send",
- "recv", "grav_pp", "grav_mm", "grav_up", "grav_down",
- "grav_external", "part_sort", "gpart_sort", "comm_root "};
+ "none", "sort", "self", "pair", "sub", "init",
+ "ghost", "drift", "kick", "send", "recv", "grav_pp",
+ "grav_mm", "grav_up", "grav_down", "grav_external", "comm_root "};
const char *subtaskID_names[task_type_count] = {"none", "density",
"force", "grav"};
@@ -83,9 +82,7 @@ float task_overlap(const struct task *ta, const struct task *tb) {
/* First check if any of the two tasks are of a type that don't
use cells. */
if (ta == NULL || tb == NULL || ta->type == task_type_none ||
- ta->type == task_type_part_sort || ta->type == task_type_gpart_sort ||
- tb->type == task_type_none || tb->type == task_type_part_sort ||
- tb->type == task_type_gpart_sort)
+ tb->type == task_type_none)
return 0.0f;
/* Compute the union of the cell data. */
diff --git a/src/task.h b/src/task.h
index 503d2e35e0afcc4598c5ca20487309504398c4c1..8193ce970fd67779aa6c8d62e0be5e0b5ea7cd83 100644
--- a/src/task.h
+++ b/src/task.h
@@ -49,8 +49,6 @@ enum task_types {
task_type_grav_up,
task_type_grav_down,
task_type_grav_external,
- task_type_part_sort,
- task_type_gpart_sort,
task_type_comm_root,
task_type_count
};
diff --git a/src/threadpool.h b/src/threadpool.h
index 729543fd59bc5cf0a8110bd2b54ef78a0d7c7de1..6e97cf73083f04d26c7e53871f4059162190e5bd 100644
--- a/src/threadpool.h
+++ b/src/threadpool.h
@@ -45,7 +45,7 @@ struct threadpool {
void *map_data, *map_extra_data;
volatile size_t map_data_count, map_data_size, map_data_stride;
volatile threadpool_map_function map_function;
-
+
/* Counter for the number of threads that are done. */
volatile int num_threads_done;
};