diff --git a/src/cell.c b/src/cell.c
index 356e4769e120bbf395964ad7a3d0ba8cea6a923e..86ff7b2ee78f001986b375dcccb5e4920a4e89cf 100644
--- a/src/cell.c
+++ b/src/cell.c
@@ -47,7 +47,9 @@ ticks cell_timer[ cell_timer_count ];
/* Define the timer macros. */
#ifdef TIMER_VERBOSE
- #define TIMER
+ #ifndef TIMER
+ #define TIMER
+ #endif
#endif
#ifdef TIMER
#define TIMER_TIC ticks tic = getticks();
@@ -61,7 +63,7 @@ ticks cell_timer[ cell_timer_count ];
# define INLINE inline
# endif
#endif
- INLINE ticks timer_toc ( int t , ticks tic ) {
+ INLINE static ticks timer_toc ( int t , ticks tic ) {
ticks d = (getticks() - tic);
__sync_add_and_fetch( &cell_timer[t] , d );
return d;
diff --git a/src/cell.h b/src/cell.h
index 27c4242ffdeac5312863d561f58daa8ead16d93b..f9baf21d11d8dd8211af78fc66ec9c6f6ead2a9e 100644
--- a/src/cell.h
+++ b/src/cell.h
@@ -58,7 +58,7 @@ struct cell {
struct entry *sort;
/* Number of pairs associated with this cell. */
- int nr_pairs;
+ // int nr_pairs;
/* Pointers to the next level of cells. */
struct cell *progeny[8];
diff --git a/src/cycle.h b/src/cycle.h
index e357a017c79a6e9befb73e2b988bd23918b66f37..16f57e7e1ef942d2736f4328be9117b2deab6d6e 100644
--- a/src/cycle.h
+++ b/src/cycle.h
@@ -178,7 +178,7 @@ typedef unsigned long long ticks;
# define INLINE inline
# endif
#endif
-INLINE ticks getticks(void)
+INLINE static ticks getticks(void)
{
ticks ret;
@@ -234,7 +234,7 @@ typedef unsigned long long ticks;
# define INLINE inline
# endif
#endif
-INLINE ticks getticks(void)
+INLINE static ticks getticks(void)
{
unsigned a, d;
asm volatile("rdtsc" : "=a" (a), "=d" (d));
diff --git a/src/engine.c b/src/engine.c
index 53b62c81e16d953f144e7806c72c8c625e776452..1f93b1ab9884fd6834cacb60f00fcdce04f5a422 100644
--- a/src/engine.c
+++ b/src/engine.c
@@ -129,15 +129,16 @@ void engine_ranktasks ( struct engine *e ) {
struct task *t;
struct space *s = e->s;
int *tid = s->tasks_ind;
-
+
/* Run throught the tasks and get all the waits right. */
for ( k = 0 ; k < s->nr_tasks ; k++ ) {
+ tid[k] = k;
for ( j = 0 ; j < s->tasks[k].nr_unlock_tasks ; j++ )
s->tasks[k].unlock_tasks[j]->wait += 1;
}
/* Main loop. */
- for ( rank = 0 ; left < s->nr_tasks ; rank++ ) {
+ for ( j = 0 , rank = 0 ; left < s->nr_tasks ; rank++ ) {
/* Load the tids of tasks with no waits. */
for ( k = left ; k < s->nr_tasks ; k++ )
@@ -308,8 +309,8 @@ void engine_init ( struct engine *e , struct space *s , int nr_threads , int nr_
}
/* Sort the queues topologically. */
- for ( k = 0 ; k < nr_queues ; k++ )
- queue_sort( &e->queues[k] );
+ // for ( k = 0 ; k < nr_queues ; k++ )
+ // queue_sort( &e->queues[k] );
/* Allocate and init the threads. */
if ( ( e->runners = (struct runner *)malloc( sizeof(struct runner) * nr_threads ) ) == NULL )
diff --git a/src/lock.h b/src/lock.h
index dc364177e1057583724e2268ee6d8c1093b77ee2..9ff386119e720b95f1da4a36b7bab5ffc2eb909f 100644
--- a/src/lock.h
+++ b/src/lock.h
@@ -40,7 +40,7 @@
#define lock_type volatile int
#define lock_init( l ) ( *l = 0 )
#define lock_destroy( l ) 0
- INLINE int lock_lock ( volatile int *l ) {
+ INLINE static int lock_lock ( volatile int *l ) {
while ( __sync_val_compare_and_swap( l , 0 , 1 ) != 0 )
while( *l );
return 0;
diff --git a/src/queue.c b/src/queue.c
index 272582519a7e08dc21ba45887c25c8d98cca8bdd..ca248da6f74872f27905a2f01023c954c04241f6 100644
--- a/src/queue.c
+++ b/src/queue.c
@@ -43,7 +43,9 @@
/* Define the timer macros. */
#ifdef TIMER_VERBOSE
- #define TIMER
+ #ifndef TIMER
+ #define TIMER
+ #endif
#endif
#ifdef TIMER
#define TIMER_TIC ticks tic = getticks();
@@ -57,7 +59,7 @@
# define INLINE inline
# endif
#endif
- INLINE ticks timer_toc ( int t , ticks tic ) {
+ INLINE static ticks timer_toc ( int t , ticks tic ) {
ticks d = (getticks() - tic);
__sync_add_and_fetch( &queue_timer[t] , d );
return d;
@@ -445,6 +447,8 @@ void queue_sort ( struct queue *q ) {
int *weight, *wait;
int *data = q->tid;
struct task *t;
+
+ printf( "queue_sort: sorting queue with %i tasks.\n" , q->count );
/* Allocate and pre-compute each task's weight. */
if ( ( weight = (int *)alloca( sizeof(int) * q->count ) ) == NULL ||
diff --git a/src/runner.c b/src/runner.c
index 21c4135f44f2b394f248c8e5aae875c993cc33e7..f79ae2d8d408b9eee017f56763cc831291f35e77 100644
--- a/src/runner.c
+++ b/src/runner.c
@@ -177,14 +177,8 @@ void runner_dosort ( struct runner *r , struct cell *c , int flags ) {
int i, ind, off[8], inds[8], temp_i;
// float shift[3];
float buff[8], px[3];
- struct cell *temp_c;
TIMER_TIC
- /* Does this cell even need to be sorted? */
- for ( temp_c = c ; temp_c != NULL && temp_c->nr_pairs == 0 ; temp_c = temp_c->parent );
- if ( temp_c == NULL )
- return;
-
/* start by allocating the entry arrays. */
if ( lock_lock( &c->lock ) != 0 )
error( "Failed to lock cell." );
@@ -204,6 +198,11 @@ void runner_dosort ( struct runner *r , struct cell *c , int flags ) {
if ( !( flags & (1 << j) ) )
continue;
+ /* Sort any un-sorted progeny. */
+ for ( k = 0 ; k < 8 ; k++ )
+ if ( c->progeny[k] != NULL && ( c->progeny[k]->sorts[0] == NULL || !(c->progeny[k]->sorts[0]->flags & (1 << j)) ) )
+ runner_dosort( r , c->progeny[k] , 1 << j );
+
/* Init the particle index offsets. */
for ( off[0] = 0 , k = 1 ; k < 8 ; k++ )
if ( c->progeny[k-1] != NULL )
@@ -298,7 +297,7 @@ void runner_dosort ( struct runner *r , struct cell *c , int flags ) {
error( "Sorting failed, indices borked." );
}
} */
-
+
#ifdef TIMER_VERBOSE
printf( "runner_dosort[%02i]: %i parts at depth %i (flags = %i%i%i%i%i%i%i%i%i%i%i%i%i) took %.3f ms.\n" ,
r->id , c->count , c->depth ,
@@ -335,7 +334,7 @@ void runner_doghost ( struct runner *r , struct cell *c ) {
runner_doghost( r , c->progeny[k] );
return;
}
-
+
/* Init the IDs that have to be updated. */
if ( ( pid = (int *)alloca( sizeof(int) * count ) ) == NULL )
error( "Call to alloca failed." );
diff --git a/src/runner.h b/src/runner.h
index 9763a2ae50351427aabe8c01c28ae37d47a6b074..4bb83a460fe58c1aed945f22758c30bd8e5a516c 100644
--- a/src/runner.h
+++ b/src/runner.h
@@ -39,9 +39,15 @@ enum {
extern ticks runner_timer[ runner_timer_count ];
+/* SID stuff. */
+extern const char runner_flip[];
+
+
/* Define the timer macros. */
#ifdef TIMER_VERBOSE
- #define TIMER
+ #ifndef TIMER
+ #define TIMER
+ #endif
#endif
#ifdef TIMER
#define TIMER_TIC ticks tic = getticks();
@@ -55,7 +61,7 @@ extern ticks runner_timer[ runner_timer_count ];
# define INLINE inline
# endif
#endif
- INLINE ticks timer_toc ( int t , ticks tic ) {
+ INLINE static ticks timer_toc ( int t , ticks tic ) {
ticks d = (getticks() - tic);
__sync_add_and_fetch( &runner_timer[t] , d );
return d;
diff --git a/src/runner_doiact.h b/src/runner_doiact.h
index fd4dd880116c06a225311df865775f4b765dcbc5..7372c23277c5df7d021dd11ee5c22592b139365d 100644
--- a/src/runner_doiact.h
+++ b/src/runner_doiact.h
@@ -498,7 +498,7 @@ void DOPAIR_SUBSET ( struct runner *r , struct cell *restrict ci , struct part *
#endif
#ifdef TIMER_VERBOSE
- printf( "runner_dopair_subset[%02i]: %i/%i parts at depth %i (r_max=%.3f/%.3f) took %.3f ms.\n" , r->id , count_i , count_j , ci->depth , ci->h_max , cj->h_max , ((double)TIMER_TOC(TIMER_DOPAIR)) / CPU_TPS * 1000 );
+ printf( "runner_dopair_subset[%02i]: %i/%i parts at depth %i (r_max=%.3f/%.3f) took %.3f ms.\n" , r->id , count , count_j , ci->depth , ci->h_max , cj->h_max , ((double)TIMER_TOC(TIMER_DOPAIR)) / CPU_TPS * 1000 );
#else
TIMER_TOC(runner_timer_dopair_subset);
#endif
@@ -607,7 +607,7 @@ void DOSELF_SUBSET ( struct runner *r , struct cell *restrict ci , struct part *
#endif
#ifdef TIMER_VERBOSE
- printf( "runner_doself_subset[%02i]: %i/%i parts at depth %i (r_max=%.3f/%.3f) took %.3f ms.\n" , r->id , count_i , count_j , ci->depth , ci->h_max , cj->h_max , ((double)TIMER_TOC(TIMER_DOPAIR)) / CPU_TPS * 1000 );
+ printf( "runner_doself_subset[%02i]: %i/%i parts at depth %i took %.3f ms.\n" , r->id , count , ci->count , ci->depth , ((double)TIMER_TOC(TIMER_DOSELF)) / CPU_TPS * 1000 );
#else
TIMER_TOC(runner_timer_dopair_subset);
#endif
@@ -624,12 +624,11 @@ void DOSELF_SUBSET ( struct runner *r , struct cell *restrict ci , struct part *
* @param cj The second #cell.
*/
-void DOPAIR1 ( struct runner *r , struct cell *restrict ci , struct cell *restrict cj ) {
+void DOPAIR1 ( struct runner *r , struct cell *ci , struct cell *cj ) {
struct engine *restrict e = r->e;
int pid, pjd, k, sid;
double rshift, shift[3] = { 0.0 , 0.0 , 0.0 };
- struct cell *temp;
struct entry *restrict sort_i, *restrict sort_j;
struct part *restrict pi, *restrict pj, *restrict parts_i, *restrict parts_j;
struct cpart *restrict cpi, *restrict cparts_i;
@@ -653,26 +652,9 @@ void DOPAIR1 ( struct runner *r , struct cell *restrict ci , struct cell *restri
/* Anything to do here? */
if ( ci->dt_min > dt_max && cj->dt_min > dt_max )
return;
-
- /* Get the relative distance between the pairs, wrapping. */
- for ( k = 0 ; k < 3 ; k++ ) {
- if ( cj->loc[k] - ci->loc[k] < -e->s->dim[k]/2 )
- shift[k] = e->s->dim[k];
- else if ( cj->loc[k] - ci->loc[k] > e->s->dim[k]/2 )
- shift[k] = -e->s->dim[k];
- }
- /* Get the sorting index. */
- for ( sid = 0 , k = 0 ; k < 3 ; k++ )
- sid = 3*sid + ( (cj->loc[k] - ci->loc[k] + shift[k] < 0) ? 0 : (cj->loc[k] - ci->loc[k] + shift[k] > 0) ? 2 : 1 );
-
- /* Switch the cells around? */
- if ( runner_flip[sid] ) {
- temp = ci; ci = cj; cj = temp;
- for ( k = 0 ; k < 3 ; k++ )
- shift[k] = -shift[k];
- }
- sid = sortlistID[sid];
+ /* Get the sort ID. */
+ sid = space_getsid( e->s , &ci , &cj , shift );
/* Get the cutoff shift. */
for ( rshift = 0.0 , k = 0 ; k < 3 ; k++ )
@@ -709,7 +691,8 @@ void DOPAIR1 ( struct runner *r , struct cell *restrict ci , struct cell *restri
/* if ( ci->split && cj->split && sid == 4 )
printf( "boing!\n" ); */
-
+
+
/* Loop over the parts in ci. */
for ( pid = count_i-1 ; pid >= 0 && sort_i[pid].d + hi_max > dj_min ; pid-- ) {
@@ -857,12 +840,11 @@ void DOPAIR1 ( struct runner *r , struct cell *restrict ci , struct cell *restri
}
-void DOPAIR2 ( struct runner *r , struct cell *restrict ci , struct cell *restrict cj ) {
+void DOPAIR2 ( struct runner *r , struct cell *ci , struct cell *cj ) {
struct engine *restrict e = r->e;
int pid, pjd, k, sid;
double rshift, shift[3] = { 0.0 , 0.0 , 0.0 };
- struct cell *temp;
struct entry *restrict sort_i, *restrict sort_j;
struct entry *restrict sortdt_i = NULL, *restrict sortdt_j = NULL;
int countdt_i = 0, countdt_j = 0;
@@ -888,26 +870,9 @@ void DOPAIR2 ( struct runner *r , struct cell *restrict ci , struct cell *restri
/* Anything to do here? */
if ( ci->dt_min > dt_max && cj->dt_min > dt_max )
return;
-
- /* Get the relative distance between the pairs, wrapping. */
- for ( k = 0 ; k < 3 ; k++ ) {
- if ( cj->loc[k] - ci->loc[k] < -e->s->dim[k]/2 )
- shift[k] = e->s->dim[k];
- else if ( cj->loc[k] - ci->loc[k] > e->s->dim[k]/2 )
- shift[k] = -e->s->dim[k];
- }
- /* Get the sorting index. */
- for ( sid = 0 , k = 0 ; k < 3 ; k++ )
- sid = 3*sid + ( (cj->loc[k] - ci->loc[k] + shift[k] < 0) ? 0 : (cj->loc[k] - ci->loc[k] + shift[k] > 0) ? 2 : 1 );
-
- /* Switch the cells around? */
- if ( runner_flip[sid] ) {
- temp = ci; ci = cj; cj = temp;
- for ( k = 0 ; k < 3 ; k++ )
- shift[k] = -shift[k];
- }
- sid = sortlistID[sid];
+ /* Get the shift ID. */
+ sid = space_getsid( e->s , &ci , &cj , shift );
/* Get the cutoff shift. */
for ( rshift = 0.0 , k = 0 ; k < 3 ; k++ )
@@ -1262,7 +1227,7 @@ void DOSELF1 ( struct runner *r , struct cell *restrict c ) {
pi = &parts[pid];
if ( cpi->dt > dt_max )
continue;
-
+
/* Get the particle position and radius. */
for ( k = 0 ; k < 3 ; k++ )
pix[k] = cpi->x[k];
@@ -1512,9 +1477,9 @@ void DOSELF2 ( struct runner *r , struct cell *restrict c ) {
* redundant computations to find the sid on-the-fly.
*/
-void DOSUB1 ( struct runner *r , struct cell *restrict ci , struct cell *restrict cj , int sid ) {
+void DOSUB1 ( struct runner *r , struct cell *ci , struct cell *cj , int sid ) {
- int j, k;
+ int j = 0, k;
double shift[3];
float h;
struct space *s = r->e->s;
@@ -1551,31 +1516,9 @@ void DOSUB1 ( struct runner *r , struct cell *restrict ci , struct cell *restric
h = fmin( ci->h[0] , fmin( ci->h[1] , ci->h[2] ) );
/* Get the type of pair if not specified explicitly. */
- if ( sid < 0 ) {
-
- /* Get the relative distance between the pairs, wrapping. */
- for ( k = 0 ; k < 3 ; k++ ) {
- if ( cj->loc[k] - ci->loc[k] < -s->dim[k]/2 )
- shift[k] = s->dim[k];
- else if ( cj->loc[k] - ci->loc[k] > s->dim[k]/2 )
- shift[k] = -s->dim[k];
- else
- shift[k] = 0.0;
- }
-
- /* Get the sorting index. */
- for ( sid = 0 , k = 0 ; k < 3 ; k++ )
- sid = 3*sid + ( (cj->loc[k] - ci->loc[k] + shift[k] < 0) ? 0 : (cj->loc[k] - ci->loc[k] + shift[k] > 0) ? 2 : 1 );
-
- /* Flip? */
- if ( sid < 13 ) {
- struct cell *temp = cj; cj = ci; ci = temp;
- }
- else
- sid = 26 - sid;
-
- }
-
+ // if ( sid < 0 )
+ sid = space_getsid( s , &ci , &cj , shift );
+
/* Recurse? */
if ( ci->split && cj->split &&
ci->h_max*2 < h && cj->h_max*2 < h ) {
@@ -1787,7 +1730,7 @@ void DOSUB1 ( struct runner *r , struct cell *restrict ci , struct cell *restric
#ifdef TIMER_VERBOSE
- printf( "runner_DOSUB[%02i]: flags=%i at depth %i took %.3f ms.\n" , r->id , flags , ci->depth , ((double)TIMER_TOC(TIMER_DOSUB)) / CPU_TPS * 1000 );
+ printf( "runner_dosub1[%02i]: flags=%i at depth %i took %.3f ms.\n" , r->id , sid , ci->depth , ((double)TIMER_TOC(TIMER_DOSUB)) / CPU_TPS * 1000 );
#else
TIMER_TOC(TIMER_DOSUB);
#endif
@@ -1795,7 +1738,7 @@ void DOSUB1 ( struct runner *r , struct cell *restrict ci , struct cell *restric
}
-void DOSUB2 ( struct runner *r , struct cell *restrict ci , struct cell *restrict cj , int sid ) {
+void DOSUB2 ( struct runner *r , struct cell *ci , struct cell *cj , int sid ) {
int j, k;
double shift[3];
@@ -1834,30 +1777,8 @@ void DOSUB2 ( struct runner *r , struct cell *restrict ci , struct cell *restric
h = fmin( ci->h[0] , fmin( ci->h[1] , ci->h[2] ) );
/* Get the type of pair if not specified explicitly. */
- if ( sid < 0 ) {
-
- /* Get the relative distance between the pairs, wrapping. */
- for ( k = 0 ; k < 3 ; k++ ) {
- if ( cj->loc[k] - ci->loc[k] < -s->dim[k]/2 )
- shift[k] = s->dim[k];
- else if ( cj->loc[k] - ci->loc[k] > s->dim[k]/2 )
- shift[k] = -s->dim[k];
- else
- shift[k] = 0.0;
- }
-
- /* Get the sorting index. */
- for ( sid = 0 , k = 0 ; k < 3 ; k++ )
- sid = 3*sid + ( (cj->loc[k] - ci->loc[k] + shift[k] < 0) ? 0 : (cj->loc[k] - ci->loc[k] + shift[k] > 0) ? 2 : 1 );
-
- /* Flip? */
- if ( sid < 13 ) {
- struct cell *temp = cj; cj = ci; ci = temp;
- }
- else
- sid = 26 - sid;
-
- }
+ if ( sid < 0 )
+ sid = space_getsid( s , &ci , &cj , shift );
/* Recurse? */
if ( ci->split && cj->split &&
@@ -2070,7 +1991,7 @@ void DOSUB2 ( struct runner *r , struct cell *restrict ci , struct cell *restric
#ifdef TIMER_VERBOSE
- printf( "runner_dosub[%02i]: flags=%i at depth %i took %.3f ms.\n" , r->id , flags , ci->depth , ((double)TIMER_TOC(TIMER_DOSUB)) / CPU_TPS * 1000 );
+ printf( "runner_dosub2[%02i]: flags=%i at depth %i took %.3f ms.\n" , r->id , sid , ci->depth , ((double)TIMER_TOC(TIMER_DOSUB)) / CPU_TPS * 1000 );
#else
TIMER_TOC(TIMER_DOSUB);
#endif
@@ -2518,7 +2439,7 @@ void DOSUB_SUBSET ( struct runner *r , struct cell *restrict ci , struct part *p
#ifdef TIMER_VERBOSE
- printf( "runner_dosub[%02i]: flags=%i at depth %i took %.3f ms.\n" , r->id , flags , ci->depth , ((double)TIMER_TOC(TIMER_DOSUB)) / CPU_TPS * 1000 );
+ printf( "runner_dosub[%02i]: flags=%i at depth %i took %.3f ms.\n" , r->id , sid , ci->depth , ((double)TIMER_TOC(TIMER_DOSUB)) / CPU_TPS * 1000 );
#else
TIMER_TOC(TIMER_DOSUB);
#endif
diff --git a/src/space.c b/src/space.c
index 51eec54bf80d70041423e317ad6de5d2530e2968..858fc7313f4ce15189b6f49390b61949d4e261ac 100644
--- a/src/space.c
+++ b/src/space.c
@@ -77,6 +77,54 @@ const int sortlistID[27] = {
};
+/**
+ * @brief Get the shift-id of the given pair of cells, swapping them
+ * if need be.
+ *
+ * @param s The space
+ * @param ci Pointer to first #cell.
+ * @param cj Pointer second #cell.
+ * @param shift Vector from ci to cj.
+ *
+ * @return The shift ID and set shift, may or may not swap ci and cj.
+ */
+
+int space_getsid ( struct space *s , struct cell **ci , struct cell **cj , double *shift ) {
+
+ int k, sid = 0;
+ struct cell *temp;
+ double dx[3];
+
+ /* Get the relative distance between the pairs, wrapping. */
+ for ( k = 0 ; k < 3 ; k++ ) {
+ dx[k] = (*cj)->loc[k] - (*ci)->loc[k];
+ if ( dx[k] < -s->dim[k]/2 )
+ shift[k] = s->dim[k];
+ else if ( dx[k] > s->dim[k]/2 )
+ shift[k] = -s->dim[k];
+ else
+ shift[k] = 0.0;
+ dx[k] += shift[k];
+ }
+
+ /* Get the sorting index. */
+ for ( k = 0 ; k < 3 ; k++ )
+ sid = 3*sid + ( (dx[k] < 0.0) ? 0 : ( (dx[k] > 0.0) ? 2 : 1 ) );
+
+ /* Switch the cells around? */
+ if ( runner_flip[sid] ) {
+ temp = *ci; *ci = *cj; *cj = temp;
+ for ( k = 0 ; k < 3 ; k++ )
+ shift[k] = -shift[k];
+ }
+ sid = sortlistID[sid];
+
+ /* Return the sort ID. */
+ return sid;
+
+ }
+
+
/**
* @breif Recursively dismantle a cell tree.
*
@@ -613,6 +661,7 @@ void space_splittasks ( struct space *s ) {
struct cell *ci, *cj;
double hi, hj, shift[3];
struct task *t;
+ float dt_max = s->dt_max;
int pts[7][8] = { { -1 , 12 , 10 , 9 , 4 , 3 , 1 , 0 } ,
{ -1 , -1 , 11 , 10 , 5 , 4 , 2 , 1 } ,
{ -1 , -1 , -1 , 12 , 7 , 6 , 4 , 3 } ,
@@ -633,6 +682,12 @@ void space_splittasks ( struct space *s ) {
/* Get a handle on the cell involved. */
ci = t->ci;
+ /* Ingore this task? */
+ if ( ci->dt_min > dt_max ) {
+ t->type = task_type_none;
+ continue;
+ }
+
/* Is this cell even split? */
if ( !ci->split )
continue;
@@ -645,9 +700,7 @@ void space_splittasks ( struct space *s ) {
/* Wait for this tasks sorts, as we will now have pairwise
components in this sub. */
- for ( k = 0 ; k < 14 ; k++ )
- if ( k == 0 || ci->sorts[k] != ci->sorts[k-1] )
- task_addunlock( ci->sorts[k] , t );
+ space_addsorts( s , t , ci , NULL , -1 );
}
@@ -662,19 +715,14 @@ void space_splittasks ( struct space *s ) {
t->ci = ci->progeny[k];
for ( k += 1 ; k < 8 ; k++ )
if ( ci->progeny[k] != NULL )
- t = space_addtask( s , task_type_self , task_subtype_density , 0 , 0 , ci->progeny[k] , NULL , NULL , 0 , NULL , 0 );
+ space_addtask( s , task_type_self , task_subtype_density , 0 , 0 , ci->progeny[k] , NULL , NULL , 0 , NULL , 0 );
/* Make a task for each pair of progeny. */
for ( j = 0 ; j < 8 ; j++ )
- if ( ci->progeny[j] != NULL && ci->progeny[j]->count > 0 )
+ if ( ci->progeny[j] != NULL )
for ( k = j + 1 ; k < 8 ; k++ )
- if ( ci->progeny[k] != NULL && ci->progeny[k]->count > 0 ) {
- t = space_addtask( s , task_type_pair , task_subtype_density , 0 , 0 , ci->progeny[j] , ci->progeny[k] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[j]->sorts[ pts[j][k] ] , t );
- task_addunlock( ci->progeny[k]->sorts[ pts[j][k] ] , t );
- ci->progeny[k]->nr_pairs += 1;
- ci->progeny[j]->nr_pairs += 1;
- }
+ if ( ci->progeny[k] != NULL )
+ space_addtask( s , task_type_pair , task_subtype_density , pts[j][k] , 0 , ci->progeny[j] , ci->progeny[k] , NULL , 0 , NULL , 0 );
}
}
@@ -688,33 +736,20 @@ void space_splittasks ( struct space *s ) {
hi = fmax( ci->h[0] , fmax( ci->h[1] , ci->h[2] ) );
hj = fmax( cj->h[0] , fmax( cj->h[1] , cj->h[2] ) );
+ /* Ingore this task? */
+ if ( ci->dt_min > dt_max && cj->dt_min > dt_max ) {
+ t->type = task_type_none;
+ continue;
+ }
+
+ /* Get the sort ID, use space_getsid and not t->flags
+ to make sure we get ci and cj swapped if needed. */
+ sid = space_getsid( s , &ci , &cj , shift );
+
/* Should this task be split-up? */
if ( ci->split && cj->split &&
ci->h_max*space_stretch < hi/2 && cj->h_max*space_stretch < hj/2 ) {
- /* Get the relative distance between the pairs, wrapping. */
- for ( k = 0 ; k < 3 ; k++ ) {
- if ( cj->loc[k] - ci->loc[k] < -s->dim[k]/2 )
- shift[k] = s->dim[k];
- else if ( cj->loc[k] - ci->loc[k] > s->dim[k]/2 )
- shift[k] = -s->dim[k];
- else
- shift[k] = 0.0;
- }
-
- /* Get the sorting index. */
- for ( sid = 0 , k = 0 ; k < 3 ; k++ )
- sid = 3*sid + ( (cj->loc[k] - ci->loc[k] + shift[k] < 0) ? 0 : (cj->loc[k] - ci->loc[k] + shift[k] > 0) ? 2 : 1 );
-
- /* Flip? */
- if ( sid < 13 ) {
- cj = t->ci;
- ci = t->cj;
- t->ci = ci; t->cj = cj;
- }
- else
- sid = 26 - sid;
-
/* Replace by a single sub-task? */
if ( space_dosub &&
ci->count < space_subsize && cj->count < space_subsize &&
@@ -723,12 +758,10 @@ void space_splittasks ( struct space *s ) {
/* Make this task a sub task. */
t->type = task_type_sub;
t->flags = sid;
+ t->ci = ci; t->cj = cj;
- /* Make it depend on all the sorts of its two cells. */
- for ( k = 0 ; k < 14 ; k++ )
- task_addunlock( ci->sorts[k] , t );
- for ( k = 0 ; k < 14 ; k++ )
- task_addunlock( cj->sorts[k] , t );
+ /* Create the sorts recursively. */
+ space_addsorts( s , t , ci , cj , sid );
/* Don't go any further. */
continue;
@@ -738,267 +771,145 @@ void space_splittasks ( struct space *s ) {
/* Take a step back (we're going to recycle the current task)... */
tid -= 1;
- /* Remove the dependency of this task on the sorts of ci and cj. */
- task_rmunlock( ci->sorts[sid] , t );
- task_rmunlock( cj->sorts[sid] , t );
- ci->nr_pairs -= 1;
- cj->nr_pairs -= 1;
- t->nr_unlock_cells = 0;
-
/* For each different sorting type... */
switch ( sid ) {
case 0: /* ( 1 , 1 , 1 ) */
- t->ci = ci->progeny[7]; t->cj = cj->progeny[0];
- task_addunlock( ci->progeny[7]->sorts[0] , t ); task_addunlock( cj->progeny[0]->sorts[0] , t );
- ci->progeny[7]->nr_pairs += 1;
- cj->progeny[0]->nr_pairs += 1;
+ t->ci = ci->progeny[7]; t->cj = cj->progeny[0]; t->flags = 0;
break;
case 1: /* ( 1 , 1 , 0 ) */
- t->ci = ci->progeny[6]; t->cj = cj->progeny[0];
- task_addunlock( ci->progeny[6]->sorts[1] , t ); task_addunlock( cj->progeny[0]->sorts[1] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[1] , t ); task_addunlock( cj->progeny[1]->sorts[1] , t );
+ t->ci = ci->progeny[6]; t->cj = cj->progeny[0]; t->flags = 1;
+ t = space_addtask( s , task_type_pair , t->subtype , 1 , 0 , ci->progeny[7] , cj->progeny[1] , NULL , 0 , NULL , 0 );
t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[0] , t ); task_addunlock( cj->progeny[1]->sorts[0] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[2] , t ); task_addunlock( cj->progeny[0]->sorts[2] , t );
- ci->progeny[6]->nr_pairs += 2;
- ci->progeny[7]->nr_pairs += 2;
- cj->progeny[0]->nr_pairs += 2;
- cj->progeny[1]->nr_pairs += 2;
+ t = space_addtask( s , task_type_pair , t->subtype , 2 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
break;
case 2: /* ( 1 , 1 , -1 ) */
- t->ci = ci->progeny[6]; t->cj = cj->progeny[1];
- task_addunlock( ci->progeny[6]->sorts[2] , t ); task_addunlock( cj->progeny[1]->sorts[2] , t );
- ci->progeny[6]->nr_pairs += 1;
- cj->progeny[1]->nr_pairs += 1;
+ t->ci = ci->progeny[6]; t->cj = cj->progeny[1]; t->flags = 2;
break;
case 3: /* ( 1 , 0 , 1 ) */
- t->ci = ci->progeny[5]; t->cj = cj->progeny[0];
- task_addunlock( ci->progeny[5]->sorts[3] , t ); task_addunlock( cj->progeny[0]->sorts[3] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[3] , t ); task_addunlock( cj->progeny[2]->sorts[3] , t );
+ t->ci = ci->progeny[5]; t->cj = cj->progeny[0]; t->flags = 3;
+ t = space_addtask( s , task_type_pair , t->subtype , 3 , 0 , ci->progeny[7] , cj->progeny[2] , NULL , 0 , NULL , 0 );
t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[0] , t ); task_addunlock( cj->progeny[2]->sorts[0] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[6] , t ); task_addunlock( cj->progeny[0]->sorts[6] , t );
- ci->progeny[5]->nr_pairs += 2;
- ci->progeny[7]->nr_pairs += 2;
- cj->progeny[0]->nr_pairs += 2;
- cj->progeny[2]->nr_pairs += 2;
+ t = space_addtask( s , task_type_pair , t->subtype , 6 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
break;
case 4: /* ( 1 , 0 , 0 ) */
- t->ci = ci->progeny[4]; t->cj = cj->progeny[0];
- task_addunlock( ci->progeny[4]->sorts[4] , t ); task_addunlock( cj->progeny[0]->sorts[4] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[5] , t ); task_addunlock( cj->progeny[0]->sorts[5] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[7] , t ); task_addunlock( cj->progeny[0]->sorts[7] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[8] , t ); task_addunlock( cj->progeny[0]->sorts[8] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[4] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[4]->sorts[3] , t ); task_addunlock( cj->progeny[1]->sorts[3] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[4] , t ); task_addunlock( cj->progeny[1]->sorts[4] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[6] , t ); task_addunlock( cj->progeny[1]->sorts[6] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[7] , t ); task_addunlock( cj->progeny[1]->sorts[7] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[4] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[4]->sorts[1] , t ); task_addunlock( cj->progeny[2]->sorts[1] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[2] , t ); task_addunlock( cj->progeny[2]->sorts[2] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[4] , t ); task_addunlock( cj->progeny[2]->sorts[4] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[5] , t ); task_addunlock( cj->progeny[2]->sorts[5] , t );
+ t->ci = ci->progeny[4]; t->cj = cj->progeny[0]; t->flags = 4;
+ t = space_addtask( s , task_type_pair , t->subtype , 5 , 0 , ci->progeny[5] , cj->progeny[0] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 7 , 0 , ci->progeny[6] , cj->progeny[0] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 8 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 3 , 0 , ci->progeny[4] , cj->progeny[1] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 4 , 0 , ci->progeny[5] , cj->progeny[1] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 6 , 0 , ci->progeny[6] , cj->progeny[1] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 7 , 0 , ci->progeny[7] , cj->progeny[1] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 1 , 0 , ci->progeny[4] , cj->progeny[2] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 2 , 0 , ci->progeny[5] , cj->progeny[2] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 4 , 0 , ci->progeny[6] , cj->progeny[2] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 5 , 0 , ci->progeny[7] , cj->progeny[2] , NULL , 0 , NULL , 0 );
t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[4] , cj->progeny[3] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[4]->sorts[0] , t ); task_addunlock( cj->progeny[3]->sorts[0] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[3] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[1] , t ); task_addunlock( cj->progeny[3]->sorts[1] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[3] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[3] , t ); task_addunlock( cj->progeny[3]->sorts[3] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[3] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[4] , t ); task_addunlock( cj->progeny[3]->sorts[4] , t );
- ci->progeny[4]->nr_pairs += 4;
- ci->progeny[5]->nr_pairs += 4;
- ci->progeny[6]->nr_pairs += 4;
- ci->progeny[7]->nr_pairs += 4;
- cj->progeny[0]->nr_pairs += 4;
- cj->progeny[1]->nr_pairs += 4;
- cj->progeny[2]->nr_pairs += 4;
- cj->progeny[3]->nr_pairs += 4;
+ t = space_addtask( s , task_type_pair , t->subtype , 1 , 0 , ci->progeny[5] , cj->progeny[3] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 3 , 0 , ci->progeny[6] , cj->progeny[3] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 4 , 0 , ci->progeny[7] , cj->progeny[3] , NULL , 0 , NULL , 0 );
break;
case 5: /* ( 1 , 0 , -1 ) */
- t->ci = ci->progeny[4]; t->cj = cj->progeny[1];
- task_addunlock( ci->progeny[4]->sorts[5] , t ); task_addunlock( cj->progeny[1]->sorts[5] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[3] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[5] , t ); task_addunlock( cj->progeny[3]->sorts[5] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[4] , cj->progeny[3] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[4]->sorts[2] , t ); task_addunlock( cj->progeny[3]->sorts[2] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[8] , t ); task_addunlock( cj->progeny[1]->sorts[8] , t );
- ci->progeny[4]->nr_pairs += 2;
- ci->progeny[6]->nr_pairs += 2;
- cj->progeny[1]->nr_pairs += 2;
- cj->progeny[3]->nr_pairs += 2;
+ t->ci = ci->progeny[4]; t->cj = cj->progeny[1]; t->flags = 5;
+ t = space_addtask( s , task_type_pair , t->subtype , 5 , 0 , ci->progeny[6] , cj->progeny[3] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 2 , 0 , ci->progeny[4] , cj->progeny[3] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 8 , 0 , ci->progeny[6] , cj->progeny[1] , NULL , 0 , NULL , 0 );
break;
case 6: /* ( 1 , -1 , 1 ) */
- t->ci = ci->progeny[5]; t->cj = cj->progeny[2];
- task_addunlock( ci->progeny[5]->sorts[6] , t ); task_addunlock( cj->progeny[2]->sorts[6] , t );
- ci->progeny[5]->nr_pairs += 1;
- cj->progeny[2]->nr_pairs += 1;
+ t->ci = ci->progeny[5]; t->cj = cj->progeny[2]; t->flags = 6;
break;
case 7: /* ( 1 , -1 , 0 ) */
- t->ci = ci->progeny[4]; t->cj = cj->progeny[3];
- task_addunlock( ci->progeny[4]->sorts[6] , t ); task_addunlock( cj->progeny[3]->sorts[6] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[8] , t ); task_addunlock( cj->progeny[2]->sorts[8] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[4] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[4]->sorts[7] , t ); task_addunlock( cj->progeny[2]->sorts[7] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[3] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[7] , t ); task_addunlock( cj->progeny[3]->sorts[7] , t );
- ci->progeny[4]->nr_pairs += 2;
- ci->progeny[5]->nr_pairs += 2;
- cj->progeny[2]->nr_pairs += 2;
- cj->progeny[3]->nr_pairs += 2;
+ t->ci = ci->progeny[4]; t->cj = cj->progeny[3]; t->flags = 6;
+ t = space_addtask( s , task_type_pair , t->subtype , 8 , 0 , ci->progeny[5] , cj->progeny[2] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 7 , 0 , ci->progeny[4] , cj->progeny[2] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 7 , 0 , ci->progeny[5] , cj->progeny[3] , NULL , 0 , NULL , 0 );
break;
case 8: /* ( 1 , -1 , -1 ) */
- t->ci = ci->progeny[4]; t->cj = cj->progeny[3];
- task_addunlock( ci->progeny[4]->sorts[8] , t ); task_addunlock( cj->progeny[3]->sorts[8] , t );
- ci->progeny[4]->nr_pairs += 1;
- cj->progeny[3]->nr_pairs += 1;
+ t->ci = ci->progeny[4]; t->cj = cj->progeny[3]; t->flags = 8;
break;
case 9: /* ( 0 , 1 , 1 ) */
- t->ci = ci->progeny[3]; t->cj = cj->progeny[0];
- task_addunlock( ci->progeny[3]->sorts[9] , t ); task_addunlock( cj->progeny[0]->sorts[9] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[4] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[9] , t ); task_addunlock( cj->progeny[4]->sorts[9] , t );
+ t->ci = ci->progeny[3]; t->cj = cj->progeny[0]; t->flags = 9;
+ t = space_addtask( s , task_type_pair , t->subtype , 9 , 0 , ci->progeny[7] , cj->progeny[4] , NULL , 0 , NULL , 0 );
t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[3] , cj->progeny[4] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[3]->sorts[0] , t ); task_addunlock( cj->progeny[4]->sorts[0] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[8] , t ); task_addunlock( cj->progeny[0]->sorts[8] , t );
- ci->progeny[3]->nr_pairs += 2;
- ci->progeny[7]->nr_pairs += 2;
- cj->progeny[0]->nr_pairs += 2;
- cj->progeny[4]->nr_pairs += 2;
+ t = space_addtask( s , task_type_pair , t->subtype , 8 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
break;
case 10: /* ( 0 , 1 , 0 ) */
- t->ci = ci->progeny[2]; t->cj = cj->progeny[0];
- task_addunlock( ci->progeny[2]->sorts[10] , t ); task_addunlock( cj->progeny[0]->sorts[10] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[3] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[3]->sorts[11] , t ); task_addunlock( cj->progeny[0]->sorts[11] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[7] , t ); task_addunlock( cj->progeny[0]->sorts[7] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[6] , t ); task_addunlock( cj->progeny[0]->sorts[6] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[2] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[2]->sorts[9] , t ); task_addunlock( cj->progeny[1]->sorts[9] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[3] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[3]->sorts[10] , t ); task_addunlock( cj->progeny[1]->sorts[10] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[8] , t ); task_addunlock( cj->progeny[1]->sorts[8] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[7] , t ); task_addunlock( cj->progeny[1]->sorts[7] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[2] , cj->progeny[4] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[2]->sorts[1] , t ); task_addunlock( cj->progeny[4]->sorts[1] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[3] , cj->progeny[4] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[3]->sorts[2] , t ); task_addunlock( cj->progeny[4]->sorts[2] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[4] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[10] , t ); task_addunlock( cj->progeny[4]->sorts[10] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[4] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[11] , t ); task_addunlock( cj->progeny[4]->sorts[11] , t );
+ t->ci = ci->progeny[2]; t->cj = cj->progeny[0]; t->flags = 10;
+ t = space_addtask( s , task_type_pair , t->subtype , 11 , 0 , ci->progeny[3] , cj->progeny[0] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 7 , 0 , ci->progeny[6] , cj->progeny[0] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 6 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 9 , 0 , ci->progeny[2] , cj->progeny[1] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 10 , 0 , ci->progeny[3] , cj->progeny[1] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 8 , 0 , ci->progeny[6] , cj->progeny[1] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 7 , 0 , ci->progeny[7] , cj->progeny[1] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 1 , 0 , ci->progeny[2] , cj->progeny[4] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 2 , 0 , ci->progeny[3] , cj->progeny[4] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 10 , 0 , ci->progeny[6] , cj->progeny[4] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 11 , 0 , ci->progeny[7] , cj->progeny[4] , NULL , 0 , NULL , 0 );
t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[2] , cj->progeny[5] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[2]->sorts[0] , t ); task_addunlock( cj->progeny[5]->sorts[0] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[3] , cj->progeny[5] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[3]->sorts[1] , t ); task_addunlock( cj->progeny[5]->sorts[1] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[5] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[9] , t ); task_addunlock( cj->progeny[5]->sorts[9] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[5] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[10] , t ); task_addunlock( cj->progeny[5]->sorts[10] , t );
- ci->progeny[2]->nr_pairs += 4;
- ci->progeny[3]->nr_pairs += 4;
- ci->progeny[6]->nr_pairs += 4;
- ci->progeny[7]->nr_pairs += 4;
- cj->progeny[0]->nr_pairs += 4;
- cj->progeny[1]->nr_pairs += 4;
- cj->progeny[4]->nr_pairs += 4;
- cj->progeny[5]->nr_pairs += 4;
+ t = space_addtask( s , task_type_pair , t->subtype , 1 , 0 , ci->progeny[3] , cj->progeny[5] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 9 , 0 , ci->progeny[6] , cj->progeny[5] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 10 , 0 , ci->progeny[7] , cj->progeny[5] , NULL , 0 , NULL , 0 );
break;
case 11: /* ( 0 , 1 , -1 ) */
- t->ci = ci->progeny[2]; t->cj = cj->progeny[1];
- task_addunlock( ci->progeny[2]->sorts[11] , t ); task_addunlock( cj->progeny[1]->sorts[11] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[5] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[11] , t ); task_addunlock( cj->progeny[5]->sorts[11] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[2] , cj->progeny[5] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[2]->sorts[2] , t ); task_addunlock( cj->progeny[5]->sorts[2] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[6] , cj->progeny[1] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[6]->sorts[6] , t ); task_addunlock( cj->progeny[1]->sorts[6] , t );
- ci->progeny[2]->nr_pairs += 2;
- ci->progeny[6]->nr_pairs += 2;
- cj->progeny[1]->nr_pairs += 2;
- cj->progeny[5]->nr_pairs += 2;
+ t->ci = ci->progeny[2]; t->cj = cj->progeny[1]; t->flags = 11;
+ t = space_addtask( s , task_type_pair , t->subtype , 11 , 0 , ci->progeny[6] , cj->progeny[5] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 2 , 0 , ci->progeny[2] , cj->progeny[5] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 6 , 0 , ci->progeny[6] , cj->progeny[1] , NULL , 0 , NULL , 0 );
break;
case 12: /* ( 0 , 0 , 1 ) */
- t->ci = ci->progeny[1]; t->cj = cj->progeny[0];
- task_addunlock( ci->progeny[1]->sorts[12] , t ); task_addunlock( cj->progeny[0]->sorts[12] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[3] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[3]->sorts[11] , t ); task_addunlock( cj->progeny[0]->sorts[11] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[5] , t ); task_addunlock( cj->progeny[0]->sorts[5] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[2] , t ); task_addunlock( cj->progeny[0]->sorts[2] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[1] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[1]->sorts[9] , t ); task_addunlock( cj->progeny[2]->sorts[9] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[3] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[3]->sorts[12] , t ); task_addunlock( cj->progeny[2]->sorts[12] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[8] , t ); task_addunlock( cj->progeny[2]->sorts[8] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[2] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[5] , t ); task_addunlock( cj->progeny[2]->sorts[5] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[1] , cj->progeny[4] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[1]->sorts[3] , t ); task_addunlock( cj->progeny[4]->sorts[3] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[3] , cj->progeny[4] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[3]->sorts[6] , t ); task_addunlock( cj->progeny[4]->sorts[6] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[4] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[12] , t ); task_addunlock( cj->progeny[4]->sorts[12] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[4] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[11] , t ); task_addunlock( cj->progeny[4]->sorts[11] , t );
+ t->ci = ci->progeny[1]; t->cj = cj->progeny[0]; t->flags = 12;
+ t = space_addtask( s , task_type_pair , t->subtype , 11 , 0 , ci->progeny[3] , cj->progeny[0] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 5 , 0 , ci->progeny[5] , cj->progeny[0] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 2 , 0 , ci->progeny[7] , cj->progeny[0] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 9 , 0 , ci->progeny[1] , cj->progeny[2] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 12 , 0 , ci->progeny[3] , cj->progeny[2] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 8 , 0 , ci->progeny[5] , cj->progeny[2] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 5 , 0 , ci->progeny[7] , cj->progeny[2] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 3 , 0 , ci->progeny[1] , cj->progeny[4] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 6 , 0 , ci->progeny[3] , cj->progeny[4] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 12 , 0 , ci->progeny[5] , cj->progeny[4] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 11 , 0 , ci->progeny[7] , cj->progeny[4] , NULL , 0 , NULL , 0 );
t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[1] , cj->progeny[6] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[1]->sorts[0] , t ); task_addunlock( cj->progeny[6]->sorts[0] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[3] , cj->progeny[6] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[3]->sorts[3] , t ); task_addunlock( cj->progeny[6]->sorts[3] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[5] , cj->progeny[6] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[5]->sorts[9] , t ); task_addunlock( cj->progeny[6]->sorts[9] , t );
- t = space_addtask( s , task_type_pair , t->subtype , 0 , 0 , ci->progeny[7] , cj->progeny[6] , NULL , 0 , NULL , 0 );
- task_addunlock( ci->progeny[7]->sorts[12] , t ); task_addunlock( cj->progeny[6]->sorts[12] , t );
- ci->progeny[1]->nr_pairs += 4;
- ci->progeny[3]->nr_pairs += 4;
- ci->progeny[5]->nr_pairs += 4;
- ci->progeny[7]->nr_pairs += 4;
- cj->progeny[0]->nr_pairs += 4;
- cj->progeny[2]->nr_pairs += 4;
- cj->progeny[4]->nr_pairs += 4;
- cj->progeny[6]->nr_pairs += 4;
+ t = space_addtask( s , task_type_pair , t->subtype , 3 , 0 , ci->progeny[3] , cj->progeny[6] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 9 , 0 , ci->progeny[5] , cj->progeny[6] , NULL , 0 , NULL , 0 );
+ t = space_addtask( s , task_type_pair , t->subtype , 12 , 0 , ci->progeny[7] , cj->progeny[6] , NULL , 0 , NULL , 0 );
break;
}
} /* split this task? */
+ /* Otherwise, if not spilt, stitch-up the sorting. */
+ else {
+
+ /* Create the sort for ci. */
+ if ( ci->sorts[0] == NULL )
+ ci->sorts[0] = space_addtask( s , task_type_sort , 0 , 1 << sid , 0 , ci , NULL , NULL , 0 , NULL , 0 );
+ ci->sorts[0]->flags |= (1 << sid);
+ task_addunlock( ci->sorts[0] , t );
+
+ /* Create the sort for cj. */
+ if ( cj->sorts[0] == NULL )
+ cj->sorts[0] = space_addtask( s , task_type_sort , 0 , 1 << sid , 0 , cj , NULL , NULL , 0 , NULL , 0 );
+ cj->sorts[0]->flags |= (1 << sid);
+ task_addunlock( cj->sorts[0] , t );
+
+ }
+
} /* pair interaction? */
} /* loop over all tasks. */
@@ -1007,80 +918,213 @@ void space_splittasks ( struct space *s ) {
/**
- * @brief Fill the #space's task list.
+ * @brief Generate the sorts for a sub recursively.
*
* @param s The #space we are working in.
- * @param do_sort Flag to add sorting tasks to the list.
*/
-void space_maketasks ( struct space *s , int do_sort ) {
+void space_addsorts ( struct space *s , struct task *t , struct cell *ci , struct cell *cj , int sid ) {
- int i, j, k, ii, jj, kk, iii, jjj, kkk, cid, cjd;
- int *cdim = s->cdim;
- struct task *t , *t2;
- int counts[task_type_count];
+ float h;
+ double shift[3];
+ int j, k;
- /* Recursive function to generate sorting tasks in the cell tree. */
- void maketasks_sort_rec ( struct cell *c ) {
+ /* Get the cell dimensions. */
+ h = fmin( ci->h[0] , fmin( ci->h[1] , ci->h[2] ) );
+
+ /* Single-cell sub? */
+ if ( cj == NULL ) {
+
+ /* If there is further splitting, add the pairs recursively. */
+ if ( ci->split ) {
+
+ /* Recurse for each progeny. */
+ for ( j = 0 ; j < 8 ; j++ )
+ if ( ci->progeny[j] != NULL )
+ space_addsorts( s , t , ci->progeny[j] , NULL , -1 );
+
+ /* Recurse for each pair of progeny. */
+ for ( j = 0 ; j < 8 ; j++ )
+ if ( ci->progeny[j] != NULL )
+ for ( k = j + 1 ; k < 8 ; k++ )
+ if ( ci->progeny[k] != NULL )
+ space_addsorts( s , t , ci->progeny[j] , ci->progeny[k] , -1 );
- int j, k;
- struct task *t;
+ }
- /* Clear the waits on this cell. */
- c->wait = 0;
+ }
- /* Start by generating the sort task. */
- if ( c->count > 0 ) {
+ /* Otherwise, it's a pair. */
+ else {
- if ( do_sort ) {
- if ( c->count < 1000 ) {
- t = space_addtask( s , task_type_sort , task_subtype_none , 0x1fff , 0 , c , NULL , NULL , 0 , NULL , 0 );
- for ( k = 0 ; k < 13 ; k++ )
- c->sorts[k] = t;
- }
- else if ( c->count < 5000 ) {
- t = space_addtask( s , task_type_sort , task_subtype_none , 0x7f , 0 , c , NULL , NULL , 0 , NULL , 0 );
- for ( k = 0 ; k < 7 ; k++ )
- c->sorts[k] = t;
- t = space_addtask( s , task_type_sort , task_subtype_none , 0x1f80 , 0 , c , NULL , NULL , 0 , NULL , 0 );
- for ( k = 7 ; k < 14 ; k++ )
- c->sorts[k] = t;
- }
- else {
- c->sorts[0] = c->sorts[1] = space_addtask( s , task_type_sort , task_subtype_none , 0x1 + 0x2 , 0 , c , NULL , NULL , 0 , NULL , 0 );
- c->sorts[2] = c->sorts[3] = space_addtask( s , task_type_sort , task_subtype_none , 0x4 + 0x8 , 0 , c , NULL , NULL , 0 , NULL , 0 );
- c->sorts[4] = c->sorts[5] = space_addtask( s , task_type_sort , task_subtype_none , 0x10 + 0x20 , 0 , c , NULL , NULL , 0 , NULL , 0 );
- c->sorts[6] = c->sorts[7] = space_addtask( s , task_type_sort , task_subtype_none , 0x40 + 0x80 , 0 , c , NULL , NULL , 0 , NULL , 0 );
- c->sorts[8] = c->sorts[9] = space_addtask( s , task_type_sort , task_subtype_none , 0x100 + 0x200 , 0 , c , NULL , NULL , 0 , NULL , 0 );
- c->sorts[10] = c->sorts[11] = space_addtask( s , task_type_sort , task_subtype_none , 0x400 + 0x800 , 0 , c , NULL , NULL , 0 , NULL , 0 );
- c->sorts[12] = c->sorts[13] = space_addtask( s , task_type_sort , task_subtype_none , 0x1000 , 0 , c , NULL , NULL , 0 , NULL , 0 );
- }
- }
+ /* Get the sort ID if not specified. */
+ // if ( sid < 0 )
+ sid = space_getsid( s , &ci , &cj , shift );
+
+ /* If there is no further splitting, add the sorts. */
+ if ( !ci->split || !cj->split ||
+ ci->h_max*2 >= h || cj->h_max*2 >= h ) {
+
+ /* Create and add the sort for ci. */
+ if ( ci->sorts[0] == NULL )
+ ci->sorts[0] = space_addtask( s , task_type_sort , 0 , 1 << sid , 0 , ci , NULL , NULL , 0 , NULL , 0 );
+ ci->sorts[0]->flags |= (1 << sid);
+ task_addunlock( ci->sorts[0] , t );
+
+ /* Create and add the sort for cj. */
+ if ( cj->sorts[0] == NULL )
+ cj->sorts[0] = space_addtask( s , task_type_sort , 0 , 1 << sid , 0 , cj , NULL , NULL , 0 , NULL , 0 );
+ cj->sorts[0]->flags |= (1 << sid);
+ task_addunlock( cj->sorts[0] , t );
}
+
+ /* Otherwise, recurse. */
+ else {
+
+ /* For each different sorting type... */
+ switch ( sid ) {
+
+ case 0: /* ( 1 , 1 , 1 ) */
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[0] , 0 );
+ break;
+
+ case 1: /* ( 1 , 1 , 0 ) */
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[0] , 1 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[1] , 1 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[1] , 0 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[0] , 2 );
+ break;
+
+ case 2: /* ( 1 , 1 , -1 ) */
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[1] , 2 );
+ break;
+
+ case 3: /* ( 1 , 0 , 1 ) */
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[0] , 3 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[2] , 3 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[2] , 0 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[0] , 6 );
+ break;
+
+ case 4: /* ( 1 , 0 , 0 ) */
+ space_addsorts( s , t , ci->progeny[4] , cj->progeny[0] , 4 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[0] , 5 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[0] , 7 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[0] , 8 );
+ space_addsorts( s , t , ci->progeny[4] , cj->progeny[1] , 3 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[1] , 4 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[1] , 6 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[1] , 7 );
+ space_addsorts( s , t , ci->progeny[4] , cj->progeny[2] , 1 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[2] , 2 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[2] , 4 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[2] , 5 );
+ space_addsorts( s , t , ci->progeny[4] , cj->progeny[3] , 0 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[3] , 1 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[3] , 3 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[3] , 4 );
+ break;
+
+ case 5: /* ( 1 , 0 , -1 ) */
+ space_addsorts( s , t , ci->progeny[4] , cj->progeny[1] , 5 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[3] , 5 );
+ space_addsorts( s , t , ci->progeny[4] , cj->progeny[3] , 2 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[1] , 8 );
+ break;
+
+ case 6: /* ( 1 , -1 , 1 ) */
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[2] , 6 );
+ break;
+
+ case 7: /* ( 1 , -1 , 0 ) */
+ space_addsorts( s , t , ci->progeny[4] , cj->progeny[3] , 6 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[2] , 8 );
+ space_addsorts( s , t , ci->progeny[4] , cj->progeny[2] , 7 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[3] , 7 );
+ break;
+
+ case 8: /* ( 1 , -1 , -1 ) */
+ space_addsorts( s , t , ci->progeny[4] , cj->progeny[3] , 8 );
+ break;
+
+ case 9: /* ( 0 , 1 , 1 ) */
+ space_addsorts( s , t , ci->progeny[3] , cj->progeny[0] , 9 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[4] , 9 );
+ space_addsorts( s , t , ci->progeny[3] , cj->progeny[4] , 0 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[0] , 8 );
+ break;
+
+ case 10: /* ( 0 , 1 , 0 ) */
+ space_addsorts( s , t , ci->progeny[2] , cj->progeny[0] , 10 );
+ space_addsorts( s , t , ci->progeny[3] , cj->progeny[0] , 11 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[0] , 7 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[0] , 6 );
+ space_addsorts( s , t , ci->progeny[2] , cj->progeny[1] , 9 );
+ space_addsorts( s , t , ci->progeny[3] , cj->progeny[1] , 10 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[1] , 8 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[1] , 7 );
+ space_addsorts( s , t , ci->progeny[2] , cj->progeny[4] , 1 );
+ space_addsorts( s , t , ci->progeny[3] , cj->progeny[4] , 2 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[4] , 10 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[4] , 11 );
+ space_addsorts( s , t , ci->progeny[2] , cj->progeny[5] , 0 );
+ space_addsorts( s , t , ci->progeny[3] , cj->progeny[5] , 1 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[5] , 9 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[5] , 10 );
+ break;
+
+ case 11: /* ( 0 , 1 , -1 ) */
+ space_addsorts( s , t , ci->progeny[2] , cj->progeny[1] , 11 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[5] , 11 );
+ space_addsorts( s , t , ci->progeny[2] , cj->progeny[5] , 2 );
+ space_addsorts( s , t , ci->progeny[6] , cj->progeny[1] , 6 );
+ break;
+
+ case 12: /* ( 0 , 0 , 1 ) */
+ space_addsorts( s , t , ci->progeny[1] , cj->progeny[0] , 12 );
+ space_addsorts( s , t , ci->progeny[3] , cj->progeny[0] , 11 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[0] , 5 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[0] , 2 );
+ space_addsorts( s , t , ci->progeny[1] , cj->progeny[2] , 9 );
+ space_addsorts( s , t , ci->progeny[3] , cj->progeny[2] , 12 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[2] , 8 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[2] , 5 );
+ space_addsorts( s , t , ci->progeny[1] , cj->progeny[4] , 3 );
+ space_addsorts( s , t , ci->progeny[3] , cj->progeny[4] , 6 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[4] , 12 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[4] , 11 );
+ space_addsorts( s , t , ci->progeny[1] , cj->progeny[6] , 0 );
+ space_addsorts( s , t , ci->progeny[3] , cj->progeny[6] , 3 );
+ space_addsorts( s , t , ci->progeny[5] , cj->progeny[6] , 9 );
+ space_addsorts( s , t , ci->progeny[7] , cj->progeny[6] , 12 );
+ break;
+
+ } /* switch. */
+
+ } /* recurse. */
- /* Otherwise, add the interactions between progeny. */
- if ( c->split ) {
-
- /* Loop over the progeny. */
- for ( k = 0 ; k < 8 ; k++ )
- if ( c->progeny[k] != NULL ) {
-
- /* Recurse. */
- maketasks_sort_rec( c->progeny[k] );
-
- /* Add dependencies between the sorts. */
- for ( j = 0 ; j < 14 ; j++ )
- if ( j == 0 || c->sorts[j] != c->sorts[j-1] )
- task_addunlock( c->progeny[k]->sorts[j] , c->sorts[j] );
-
- }
-
- }
+ } /* it's a pair. */
+
+ }
+
+
+/**
+ * @brief Fill the #space's task list.
+ *
+ * @param s The #space we are working in.
+ * @param do_sort Flag to add sorting tasks to the list.
+ */
+
+void space_maketasks ( struct space *s , int do_sort ) {
+
+ int i, j, k, ii, jj, kk, iii, jjj, kkk, cid, cjd, sid;
+ int *cdim = s->cdim;
+ struct task *t, *t2;
+ struct cell *ci, *cj;
+ // float dt_max = s->dt_max;
+ int counts[task_type_count];
- } /* void maketasks_sort_rec. */
-
-
/* Allocate the task-list, if needed. */
if ( s->tasks == NULL || s->tasks_size < s->tot_cells * space_maxtaskspercell ) {
if ( s->tasks != NULL )
@@ -1095,10 +1139,6 @@ void space_maketasks ( struct space *s , int do_sort ) {
}
s->nr_tasks = 0;
- /* Loop over the cells and generate their sorting tasks. */
- for ( k = 0 ; k < s->nr_cells ; k++ )
- maketasks_sort_rec( &s->cells[k] );
-
/* Run through the highest level of cells and add pairs. */
for ( i = 0 ; i < cdim[0] ; i++ )
for ( j = 0 ; j < cdim[1] ; j++ )
@@ -1106,7 +1146,11 @@ void space_maketasks ( struct space *s , int do_sort ) {
cid = cell_getid( cdim , i , j , k );
if ( s->cells[cid].count == 0 )
continue;
- space_addtask( s , task_type_self , task_subtype_density , 0 , 0 , &s->cells[cid] , NULL , NULL , 0 , NULL , 0 );
+ ci = &s->cells[cid];
+ if ( ci->count == 0 )
+ continue;
+ // if ( ci->dt_min <= dt_max )
+ space_addtask( s , task_type_self , task_subtype_density , 0 , 0 , ci , NULL , NULL , 0 , NULL , 0 );
for ( ii = -1 ; ii < 2 ; ii++ ) {
iii = i + ii;
if ( !s->periodic && ( iii < 0 || iii >= cdim[0] ) )
@@ -1123,15 +1167,12 @@ void space_maketasks ( struct space *s , int do_sort ) {
continue;
kkk = ( kkk + cdim[2] ) % cdim[2];
cjd = cell_getid( cdim , iii , jjj , kkk );
- if ( s->cells[cjd].count == 0 )
- continue;
- if ( cid >= cjd )
+ cj = &s->cells[cjd];
+ if ( cid >= cjd || cj->count == 0 /* ||
+ ( ci->dt_min > dt_max && cj->dt_min > dt_max ) */ )
continue;
- t = space_addtask( s , task_type_pair , task_subtype_density , 0 , 0 , &s->cells[cid] , &s->cells[cjd] , NULL , 0 , NULL , 0 );
- task_addunlock( s->cells[cid].sorts[ sortlistID[ (kk+1) + 3*( (jj+1) + 3*(ii+1) ) ] ] , t );
- task_addunlock( s->cells[cjd].sorts[ sortlistID[ (kk+1) + 3*( (jj+1) + 3*(ii+1) ) ] ] , t );
- s->cells[cid].nr_pairs += 1;
- s->cells[cjd].nr_pairs += 1;
+ sid = sortlistID[ (kk+1) + 3*( (jj+1) + 3*(ii+1) ) ];
+ t = space_addtask( s , task_type_pair , task_subtype_density , sid , 0 , ci , cj , NULL , 0 , NULL , 0 );
}
}
}
@@ -1140,50 +1181,15 @@ void space_maketasks ( struct space *s , int do_sort ) {
/* Split the tasks. */
space_splittasks( s );
- /* Remove pairs and self-interactions for cells with no parts in dt. */
+ /* Make each sort depend on the sorts of its progeny. */
for ( k = 0 ; k < s->nr_tasks ; k++ ) {
t = &s->tasks[k];
- if ( t->type == task_type_self ) {
- if ( t->ci->dt_min > s->dt_max )
- t->type = task_type_none;
- }
- else if ( t->type == task_type_pair ) {
- if ( t->ci->dt_min > s->dt_max && t->cj->dt_min > s->dt_max ) {
- t->type = task_type_none;
- for ( j = 0 ; j < 13 ; j++ )
- task_rmunlock_blind( t->ci->sorts[j] , t );
- for ( j = 0 ; j < 13 ; j++ )
- task_rmunlock_blind( t->cj->sorts[j] , t );
- }
- }
- else if ( t->type == task_type_sub ) {
- if ( t->ci->dt_min > s->dt_max && ( t->cj == NULL || t->cj->dt_min > s->dt_max ) ) {
- t->type = task_type_none;
- for ( j = 0 ; j < 13 ; j++ )
- task_rmunlock_blind( t->ci->sorts[j] , t );
- if ( t->cj != NULL )
- for ( j = 0 ; j < 13 ; j++ )
- task_rmunlock_blind( t->cj->sorts[j] , t );
- }
- }
+ if ( t->type == task_type_sort && t->ci->split )
+ for ( j = 0 ; j < 8 ; j++ )
+ if ( t->ci->progeny[j] != NULL && t->ci->progeny[j]->sorts[0] != NULL )
+ task_addunlock( t->ci->progeny[j]->sorts[0] , t );
}
- /* Remove sort tasks with no dependencies. */
- for ( k = 0 ; k < s->nr_tasks ; k++ ) {
- t = &s->tasks[k];
- if ( t->type == task_type_sort && t->nr_unlock_tasks == 0 ) {
- if ( t->ci->split )
- for ( i = 0 ; i < 13 ; i++ )
- if ( t->flags & ( 1 << i ) ) {
- for ( j = 0 ; j < 8 ; j++ )
- if ( t->ci->progeny[j] != NULL )
- task_rmunlock_blind( t->ci->progeny[j]->sorts[i] , t );
- t->ci->sorts[i] = NULL;
- }
- t->type = task_type_none;
- }
- }
-
/* Count the number of tasks associated with each cell and
store the density tasks in each cell. */
space_map_cells( s , 1 , &space_map_clearnrtasks , NULL );
@@ -1245,7 +1251,7 @@ void space_maketasks ( struct space *s , int do_sort ) {
task_addunlock( t->ci->ghost , t2 );
task_addunlock( t->cj->ghost , t2 );
}
-
+
/* Otherwise, sub interaction? */
else if ( t->type == task_type_sub && t->subtype == task_subtype_density ) {
task_addunlock( t , t->ci->super->ghost );
@@ -1272,7 +1278,7 @@ void space_maketasks ( struct space *s , int do_sort ) {
for ( k = 1 ; k < task_type_count ; k++ )
printf( " %s=%i" , taskID_names[k] , counts[k] );
printf( " ]\n" );
-
+
}
diff --git a/src/space.h b/src/space.h
index 0c52849363cae498a2d58930d4906e013bb3dc28..2dc0f8878f7404942bf1bd975826183284893c58 100644
--- a/src/space.h
+++ b/src/space.h
@@ -26,7 +26,7 @@
#define space_splitratio 0.875
#define space_splitsize_default 400
#define space_subsize_default 1000
-#define space_dosub 0
+#define space_dosub 1
#define space_stretch 1.0
#define space_maxtaskspercell 43
@@ -98,10 +98,12 @@ struct space {
/* function prototypes. */
+void space_addsorts ( struct space *s , struct task *t , struct cell *ci , struct cell *cj , int sid );
void parts_sort ( struct part *parts , int *ind , int N , int min , int max );
struct cell *space_getcell ( struct space *s );
struct task *space_gettask ( struct space *s );
struct task *space_addtask ( struct space *s , int type , int subtype , int flags , int wait , struct cell *ci , struct cell *cj , struct task *unlock_tasks[] , int nr_unlock_tasks , struct cell *unlock_cells[] , int nr_unlock_cells );
+int space_getsid ( struct space *s , struct cell **ci , struct cell **cj , double *shift );
void space_init ( struct space *s , double dim[3] , struct part *parts , int N , int periodic , double h_max );
void space_maketasks ( struct space *s , int do_sort );
void space_map_cells ( struct space *s , int full , void (*fun)( struct cell *c , void *data ) , void *data );
diff --git a/src/task.c b/src/task.c
index 81b9babfcac4df407aa31756a2bc3526d93d8aca..a4c3eb5fba51f863c8614745a343c8c84f08fc14 100644
--- a/src/task.c
+++ b/src/task.c
@@ -44,6 +44,30 @@ const char *taskID_names[task_type_count] = { "none" , "sort" , "self" , "pair"
#define error(s) { fprintf( stderr , "%s:%s:%i: %s\n" , __FILE__ , __FUNCTION__ , __LINE__ , s ); abort(); }
+/**
+ * @breif Remove all unlocks to tasks that are of the given type.
+ *
+ * @param t The #task.
+ * @param type The task type ID to remove.
+ */
+
+void task_cleanunlock ( struct task *t , int type ) {
+
+ int k;
+
+ lock_lock( &t->lock );
+
+ for ( k = 0 ; k < t->nr_unlock_tasks ; k++ )
+ if ( t->unlock_tasks[k]->type == type ) {
+ t->nr_unlock_tasks -= 1;
+ t->unlock_tasks[k] = t->unlock_tasks[ t->nr_unlock_tasks ];
+ }
+
+ lock_unlock_blind( &t->lock );
+
+ }
+
+
/**
* @brief Remove an unlock_task from the given task.
*
@@ -108,10 +132,6 @@ void task_addunlock( struct task *ta , struct task *tb ) {
int k;
- /* Bogus? */
- if ( ta == NULL || tb == NULL )
- return;
-
lock_lock( &ta->lock );
/* Check if ta already unlocks tb. */
diff --git a/src/task.h b/src/task.h
index cd1c692f8147e9c5a354c5da02addedf3eb85148..662d838decbb6d98e67cab4df418bf53acf6da50 100644
--- a/src/task.h
+++ b/src/task.h
@@ -65,4 +65,5 @@ struct task {
/* Function prototypes. */
void task_rmunlock( struct task *ta , struct task *tb );
void task_rmunlock_blind( struct task *ta , struct task *tb );
+void task_cleanunlock ( struct task *t , int type );
void task_addunlock( struct task *ta , struct task *tb );