emo/xbps
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

lib/plist.c: share workload more efficient between threads

Browse Source 
Before this patch xbps had a constant slice count for every thread.
This resulted in unbalanced thread usage if the workload was
homogenous over the slices.

This patch replaces the naive slice count approach by spinlock based
reservation.
This commit is contained in:
Enno Boland 2018-07-02 13:18:17 +02:00
parent 3f4d6a93be
commit bfd6b2e88e
1 changed files with 54 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

80
lib/plist.c
View File

@ -38,7 +38,10 @@ struct thread_data {
xbps_dictionary_t dict;
struct xbps_handle *xhp;
unsigned int start;
unsigned int end;
unsigned int arraycount;
unsigned int *reserved;
pthread_spinlock_t *reserved_lock;
unsigned int slicecount;
int (*fn)(struct xbps_handle *, xbps_object_t, const char *, void *, bool *);
void *fn_arg;
};
@ -59,23 +62,33 @@ array_foreach_thread(void *arg)
const char *key;
int rv;
bool loop_done = false;
unsigned i = thd->start;
unsigned int end = i + thd->slicecount;
/* process pkgs from start until end */
for (unsigned int i = thd->start; i < thd->end; i++) {
obj = xbps_array_get(thd->array, i);
if (xbps_object_type(thd->dict) == XBPS_TYPE_DICTIONARY) {
pkgd = xbps_dictionary_get_keysym(thd->dict, obj);
key = xbps_dictionary_keysym_cstring_nocopy(obj);
/* ignore internal objs */
if (strncmp(key, "_XBPS_", 6) == 0)
continue;
} else {
pkgd = obj;
key = NULL;
while(i < thd->arraycount) {
/* process pkgs from start until end */
for (; i < end && i < thd->arraycount; i++) {
obj = xbps_array_get(thd->array, i);
if (xbps_object_type(thd->dict) == XBPS_TYPE_DICTIONARY) {
pkgd = xbps_dictionary_get_keysym(thd->dict, obj);
key = xbps_dictionary_keysym_cstring_nocopy(obj);
/* ignore internal objs */
if (strncmp(key, "_XBPS_", 6) == 0)
continue;
} else {
pkgd = obj;
key = NULL;
}
rv = (*thd->fn)(thd->xhp, pkgd, key, thd->fn_arg, &loop_done);
if (rv != 0 || loop_done)
return NULL;
}
rv = (*thd->fn)(thd->xhp, pkgd, key, thd->fn_arg, &loop_done);
if (rv != 0 || loop_done)
break;
/* Reserve more elements to compute */
pthread_spin_lock(thd->reserved_lock);
i = *thd->reserved;
end = i + thd->slicecount;
*thd->reserved = end;
pthread_spin_unlock(thd->reserved_lock);
}
return NULL;
}
@ -88,8 +101,10 @@ xbps_array_foreach_cb_multi(struct xbps_handle *xhp,
void *arg)
{
struct thread_data *thd;
unsigned int arraycount, slicecount, pkgcount;
unsigned int arraycount, slicecount;
int rv = 0, maxthreads;
unsigned int reserved = 0;
pthread_spinlock_t reserved_lock;
assert(fn != NULL);
@ -101,13 +116,25 @@ xbps_array_foreach_cb_multi(struct xbps_handle *xhp,
return 0;
maxthreads = (int)sysconf(_SC_NPROCESSORS_ONLN);
if (maxthreads <= 0 || maxthreads == 1) /* use single threaded routine */
if (maxthreads <= 1 || arraycount <= 1) /* use single threaded routine */
return xbps_array_foreach_cb(xhp, array, dict, fn, arg);
if (pthread_spin_init(&reserved_lock, PTHREAD_PROCESS_PRIVATE) != 0)
return 0;
thd = calloc(maxthreads, sizeof(*thd));
assert(thd);
slicecount = arraycount / maxthreads;
pkgcount = 0;
// maxthread is boundchecked to be > 1
if((unsigned int)maxthreads >= arraycount) {
maxthreads = arraycount;
slicecount = 1;
} else {
slicecount = arraycount / maxthreads;
if (slicecount > 32) {
slicecount = 32;
}
}
for (int i = 0; i < maxthreads; i++) {
thd[i].array = array;
@ -115,20 +142,21 @@ xbps_array_foreach_cb_multi(struct xbps_handle *xhp,
thd[i].xhp = xhp;
thd[i].fn = fn;
thd[i].fn_arg = arg;
thd[i].start = pkgcount;
if (i + 1 >= maxthreads)
thd[i].end = arraycount;
else
thd[i].end = pkgcount + slicecount;
thd[i].start = i * slicecount;
thd[i].reserved = &reserved;
thd[i].reserved_lock = &reserved_lock;
thd[i].slicecount = slicecount;
thd[i].arraycount = arraycount;
pthread_create(&thd[i].thread, NULL,
array_foreach_thread, &thd[i]);
pkgcount += slicecount;
}
/* wait for all threads */
for (int i = 0; i < maxthreads; i++)
rv = pthread_join(thd[i].thread, NULL);
free(thd);
pthread_spin_destroy(&reserved_lock);
return rv;
}