diff options
Diffstat (limited to 'Carpet/LoopControl')
| -rw-r--r-- | Carpet/LoopControl/src/loopcontrol.c | 62 | ||||
| -rw-r--r-- | Carpet/LoopControl/src/loopcontrol.h | 31 |
2 files changed, 58 insertions, 35 deletions
diff --git a/Carpet/LoopControl/src/loopcontrol.c b/Carpet/LoopControl/src/loopcontrol.c index e928e91e2..a0bbe564b 100644 --- a/Carpet/LoopControl/src/loopcontrol.c +++ b/Carpet/LoopControl/src/loopcontrol.c @@ -71,11 +71,11 @@ lc_statmap_t * lc_statmap_list = NULL; /* Find all possible thread topologies */ /* This finds all possible thread topologies which can be expressed as NIxNJxNK x NIIxNJJxNKK. More complex topologies, e.g. based on a - recursive subdiviston, are not considered (and cannot be expressed - with the data structures used in LoopControl). I expect that more - complex topologies are not necessary, since the number of treads is - usually quite small and contains many small factors in its prime - decomposition. */ + recursive subdivision, are not considered (and cannot be expressed + with the data structures currently used in LoopControl). I expect + that more complex topologies are not necessary, since the number of + threads is usually quite small and contains many small factors in + its prime decomposition. */ static void find_thread_topologies (lc_topology_t * restrict const topologies, @@ -122,6 +122,7 @@ find_thread_topologies (lc_topology_t * restrict const topologies, } +#if 0 /* Find "good" tiling specifications */ /* This calculates a subset of all possible thread specifications. @@ -130,11 +131,10 @@ find_thread_topologies (lc_topology_t * restrict const topologies, "equally", so that one does not have to spend much effort investigating tiling specifications with very similar properties. For example, if there are 200 grid points, then half of the - possible tiling specifications consists of splitting the domain - into two subdomains with [100+N, 100-N] points. This is avoided by + possible tiling specifications consist of splitting the domain into + two subdomains with [100+N, 100-N] points. This is avoided by covering all possible tiling specifications in exponentially growing step sizes. */ -#if 0 static int tiling_compare (const void * const a, const void * const b) { @@ -372,7 +372,7 @@ lc_stattime_find_create (lc_statset_t * restrict const ls, if (! lt) { lt = malloc (sizeof * lt); - assert(lt); + assert (lt); lc_stattime_init (lt, ls, state); } @@ -406,23 +406,33 @@ lc_statset_init (lc_statset_t * restrict const ls, /*** Threads ****************************************************************/ static int saved_maxthreads = -1; - static lc_topology_t * restrict * saved_topologies = NULL; + static lc_topology_t * * saved_topologies = NULL; static int * restrict saved_ntopologies = NULL; - + + // Allocate memory the first time this function is called if (saved_maxthreads < 0) { saved_maxthreads = omp_get_max_threads(); saved_topologies = malloc (saved_maxthreads * sizeof * saved_topologies ); - assert(saved_topologies); saved_ntopologies = malloc (saved_maxthreads * sizeof * saved_ntopologies); - assert(saved_ntopologies); + assert (saved_topologies ); + assert (saved_ntopologies); for (int n=0; n<saved_maxthreads; ++n) { saved_topologies [n] = NULL; saved_ntopologies[n] = -1; } } - + // Reallocate memory in case we need more if (num_threads > saved_maxthreads) { - CCTK_WARN (CCTK_WARN_ABORT, "Thread count inconsistency"); + int old_saved_maxthreads = saved_maxthreads; + saved_maxthreads = num_threads; + saved_topologies = realloc (saved_topologies, saved_maxthreads * sizeof * saved_topologies ); + saved_ntopologies = realloc (saved_ntopologies, saved_maxthreads * sizeof * saved_ntopologies); + assert (saved_topologies ); + assert (saved_ntopologies); + for (int n=old_saved_maxthreads; n<saved_maxthreads; ++n) { + saved_topologies [n] = NULL; + saved_ntopologies[n] = -1; + } } if (! saved_topologies[num_threads-1]) { @@ -436,7 +446,7 @@ lc_statset_init (lc_statset_t * restrict const ls, saved_topologies[num_threads-1] = malloc (maxntopologies * sizeof * saved_topologies[num_threads-1]); - assert(saved_topologies[num_threads-1]); + assert (saved_topologies[num_threads-1]); find_thread_topologies (saved_topologies[num_threads-1], maxntopologies, & saved_ntopologies[num_threads-1], @@ -445,6 +455,7 @@ lc_statset_init (lc_statset_t * restrict const ls, realloc (saved_topologies[num_threads-1], (saved_ntopologies[num_threads-1] * sizeof * saved_topologies[num_threads-1])); + assert (saved_topologies[num_threads-1]); if (debug) { printf ("Found %d possible thread topologies\n", @@ -481,12 +492,12 @@ lc_statset_init (lc_statset_t * restrict const ls, printf ("Dimension %d: %d points\n", d, ls->npoints[d]); } ls->tilings[d] = malloc (maxntilings * sizeof * ls->tilings[d]); - assert(ls->tilings[d]); + assert (ls->tilings[d]); find_tiling_specifications (ls->tilings[d], maxntilings, & ls->ntilings[d], ls->npoints[d]); ls->topology_ntilings[d] = malloc (ls->ntopologies * sizeof * ls->topology_ntilings[d]); - assert(ls->topology_ntilings[d]); + assert (ls->topology_ntilings[d]); for (int n = 0; n < ls->ntopologies; ++n) { int tiling; for (tiling = 1; tiling < ls->ntilings[d]; ++tiling) { @@ -582,7 +593,7 @@ lc_statset_find_create (lc_statmap_t * restrict const lm, if (! ls) { ls = malloc (sizeof * ls); - assert(ls); + assert (ls); lc_statset_init (ls, lm, num_threads, npoints); } @@ -631,7 +642,8 @@ lc_control_init (lc_control_t * restrict const lc, lc_statmap_t * restrict const lm, int const imin, int const jmin, int const kmin, int const imax, int const jmax, int const kmax, - int const ilsh, int const jlsh, int const klsh) + int const ilsh, int const jlsh, int const klsh, + int const di) { DECLARE_CCTK_PARAMETERS; @@ -645,6 +657,7 @@ lc_control_init (lc_control_t * restrict const lc, assert (imin >= 0 && imax <= ilsh && ilsh >= 0); assert (jmin >= 0 && jmax <= jlsh && jlsh >= 0); assert (kmin >= 0 && kmax <= klsh && klsh >= 0); + assert (di > 0); /* Copy arguments */ lc->imin = imin; @@ -656,6 +669,7 @@ lc_control_init (lc_control_t * restrict const lc, lc->ilsh = ilsh; lc->jlsh = jlsh; lc->klsh = klsh; + lc->di = di; /* vector size */ @@ -663,11 +677,8 @@ lc_control_init (lc_control_t * restrict const lc, lc_statset_t * restrict ls; #pragma omp single copyprivate (ls) { - /* Get number of threads */ - /* int const num_threads = omp_get_num_threads(); */ - /* int const num_threads = omp_get_max_threads(); */ - /* Calculate number of points */ + /* TODO: Take vector size into account */ int npoints[3]; npoints[0] = lc_max (imax - imin, 0); npoints[1] = lc_max (jmax - jmin, 0); @@ -1110,7 +1121,8 @@ CCTK_FNAME (lc_control_init) (lc_control_t * restrict const lc, lc_control_init (lc, lm, * imin - 1, * jmin - 1, * kmin - 1, * imax, * jmax, * kmax, - * ilsh, * jlsh, * klsh); + * ilsh, * jlsh, * klsh, + 1); } CCTK_FCALL diff --git a/Carpet/LoopControl/src/loopcontrol.h b/Carpet/LoopControl/src/loopcontrol.h index 0f3b6d80d..e23806fcb 100644 --- a/Carpet/LoopControl/src/loopcontrol.h +++ b/Carpet/LoopControl/src/loopcontrol.h @@ -242,6 +242,7 @@ typedef struct lc_control_t { int imin, jmin, kmin; int imax, jmax, kmax; int ilsh, jlsh, klsh; + int di; /* Control settings for thread parallelism (useful for debugging) */ int iiimin, jjjmin, kkkmin; @@ -303,7 +304,8 @@ lc_control_init (lc_control_t * restrict lc, lc_statmap_t * restrict lm, int imin, int jmin, int kmin, int imax, int jmax, int kmax, - int ilsh, int jlsh, int klsh); + int ilsh, int jlsh, int klsh, + int di); void lc_control_finish (lc_control_t * restrict lc); @@ -311,15 +313,24 @@ lc_control_finish (lc_control_t * restrict lc); #define LC_LOOP3(name, i,j,k, imin,jmin,kmin, imax,jmax,kmax, ilsh,jlsh,klsh) \ + LC_LOOP3VEC(name, i,j,k, imin,jmin,kmin, imax,jmax,kmax, ilsh,jlsh,klsh, 1) +#define LC_ENDLOOP3(name) \ + LC_ENDLOOP3VEC(name) + +#define LC_LOOP3VEC(name, i,j,k, imin_,jmin_,kmin_, imax_,jmax_,kmax_, ilsh_,jlsh_,klsh_, di_) \ do { \ static int lc_initialised = 0; \ static lc_statmap_t lc_lm; \ if (! lc_initialised) { \ lc_statmap_init (& lc_initialised, & lc_lm, #name); \ } \ + int const lc_di = (di_); \ lc_control_t lc_lc; \ lc_control_init (& lc_lc, & lc_lm, \ - imin,jmin,kmin, imax,jmax,kmax, ilsh,jlsh,klsh); \ + (imin_), (jmin_), (kmin_), \ + (imax_), (jmax_), (kmax_), \ + (ilsh_), (jlsh_), (klsh_), \ + lc_di); \ \ /* Coarse loop */ \ for (int lc_kk = lc_lc.kkmin; \ @@ -327,33 +338,33 @@ lc_control_finish (lc_control_t * restrict lc); lc_kk += lc_lc.kkstep) \ { \ int const lc_kmin = lc_kk + lc_lc.kkkkmin; \ - int const lc_kmax = \ - lc_min (lc_kk + lc_lc.kkkkmax, lc_lc.kkmax); \ + int const lc_kmax = lc_min (lc_kk + lc_lc.kkkkmax, lc_lc.kkmax); \ \ for (int lc_jj = lc_lc.jjmin; \ lc_jj < lc_lc.jjmax; \ lc_jj += lc_lc.jjstep) \ { \ int const lc_jmin = lc_jj + lc_lc.jjjjmin; \ - int const lc_jmax = \ - lc_min (lc_jj + lc_lc.jjjjmax, lc_lc.jjmax); \ + int const lc_jmax = lc_min (lc_jj + lc_lc.jjjjmax, lc_lc.jjmax); \ \ for (int lc_ii = lc_lc.iimin; \ lc_ii < lc_lc.iimax; \ lc_ii += lc_lc.iistep) \ { \ int const lc_imin = lc_ii + lc_lc.iiiimin; \ - int const lc_imax = \ - lc_min (lc_ii + lc_lc.iiiimax, lc_lc.iimax); \ + int const lc_imax = lc_min (lc_ii + lc_lc.iiiimax, lc_lc.iimax); \ \ /* Fine loop */ \ for (int k = lc_kmin; k < lc_kmax; ++k) { \ for (int j = lc_jmin; j < lc_jmax; ++j) { \ LC_PRELOOP_STATEMENTS \ { \ - for (int i = lc_imin; i < lc_imax; ++i) { + int const lc_ipos = \ + lc_imin + lc_lc.ilsh * (j + lc_lc.jlsh * k); \ + int const lc_ioffset = (lc_ipos & - lc_di) - lc_ipos; \ + for (int i = lc_imin + lc_ioffset; i < lc_imax; i += lc_di) { -#define LC_ENDLOOP3(name) \ +#define LC_ENDLOOP3VEC(name) \ } \ } \ LC_POSTLOOP_STATEMENTS \ |