static hwloc_obj_t insert_task(hwloc_topology_t topology, hwloc_cpuset_t cpuset, const char * name)
{
hwloc_obj_t group, obj;
hwloc_bitmap_and(cpuset, cpuset, hwloc_topology_get_topology_cpuset(topology));
if (hwloc_bitmap_iszero(cpuset))
return NULL;
/* try to insert a group at exact position */
group = hwloc_topology_alloc_group_object(topology);
if (!group)
return NULL;
group->cpuset = hwloc_bitmap_dup(cpuset);
group = hwloc_topology_insert_group_object(topology, group);
if (!group) {
/* try to insert in a larger parent */
char *s;
hwloc_bitmap_asprintf(&s, cpuset);
group = hwloc_get_obj_covering_cpuset(topology, cpuset);
fprintf(stderr, "Inserting process `%s' below parent larger than cpuset %s\n", name, s);
free(s);
}
obj = hwloc_topology_insert_misc_object(topology, group, name);
if (!obj)
fprintf(stderr, "Failed to insert process `%s'\n", name);
else
obj->subtype = strdup("Process");
return obj;
}
static hwloc_obj_t insert_task(hwloc_topology_t topology, hwloc_cpuset_t cpuset, const char * name)
{
hwloc_obj_t obj;
hwloc_bitmap_and(cpuset, cpuset, hwloc_topology_get_topology_cpuset(topology));
if (hwloc_bitmap_iszero(cpuset))
return NULL;
/* try to insert at exact position */
obj = hwloc_topology_insert_misc_object_by_cpuset(topology, cpuset, name);
if (!obj) {
/* try to insert in a larger parent */
char *s;
hwloc_bitmap_asprintf(&s, cpuset);
obj = hwloc_get_obj_covering_cpuset(topology, cpuset);
if (obj) {
obj = hwloc_topology_insert_misc_object_by_parent(topology, obj, name);
fprintf(stderr, "Inserted process `%s' below parent larger than cpuset %s\n", name, s);
} else {
fprintf(stderr, "Failed to insert process `%s' with cpuset %s\n", name, s);
}
free(s);
} else {
hwloc_obj_add_info(obj, "Type", "Process");
}
return obj;
}
static void
output_compute_pu_min_textwidth(struct lstopo_output *output)
{
unsigned fontsize = output->fontsize;
char text[64];
int n;
hwloc_topology_t topology = output->topology;
hwloc_obj_t lastpu;
if (!output->methods->textsize) {
output->min_pu_textwidth = 0;
return;
}
if (output->logical) {
int depth = hwloc_get_type_depth(topology, HWLOC_OBJ_PU);
lastpu = hwloc_get_obj_by_depth(topology, depth, hwloc_get_nbobjs_by_depth(topology, depth)-1);
} else {
unsigned lastidx = hwloc_bitmap_last(hwloc_topology_get_topology_cpuset(topology));
lastpu = hwloc_get_pu_obj_by_os_index(topology, lastidx);
}
n = lstopo_obj_snprintf(output, text, sizeof(text), lastpu);
output->min_pu_textwidth = get_textwidth(output, text, n, fontsize);
}
static hwloc_const_bitmap_t
hwloc_fix_cpubind(hwloc_topology_t topology, hwloc_const_bitmap_t set)
{
hwloc_const_bitmap_t topology_set = hwloc_topology_get_topology_cpuset(topology);
hwloc_const_bitmap_t complete_set = hwloc_topology_get_complete_cpuset(topology);
if (!topology_set) {
/* The topology is composed of several systems, the cpuset is ambiguous. */
errno = EXDEV;
return NULL;
}
if (hwloc_bitmap_iszero(set)) {
errno = EINVAL;
return NULL;
}
if (!hwloc_bitmap_isincluded(set, complete_set)) {
errno = EINVAL;
return NULL;
}
if (hwloc_bitmap_isincluded(topology_set, set))
set = complete_set;
return set;
}
static hwloc_const_nodeset_t
hwloc_fix_membind(hwloc_topology_t topology, hwloc_const_nodeset_t nodeset)
{
hwloc_const_bitmap_t topology_nodeset = hwloc_topology_get_topology_nodeset(topology);
hwloc_const_bitmap_t complete_nodeset = hwloc_topology_get_complete_nodeset(topology);
if (!hwloc_topology_get_topology_cpuset(topology)) {
/* The topology is composed of several systems, the nodeset is thus
* ambiguous. */
errno = EXDEV;
return NULL;
}
if (!complete_nodeset) {
/* There is no NUMA node */
errno = ENODEV;
return NULL;
}
if (hwloc_bitmap_iszero(nodeset)) {
errno = EINVAL;
return NULL;
}
if (!hwloc_bitmap_isincluded(nodeset, complete_nodeset)) {
errno = EINVAL;
return NULL;
}
if (hwloc_bitmap_isincluded(topology_nodeset, nodeset))
return complete_nodeset;
return nodeset;
}
void initialize_hwloc(int nb_workers) {
#ifdef HAVE_HWLOC
hwloc_topology_init(&topology);
hwloc_topology_load(topology);
bool numa_alloc_interleaved = (nb_workers == 0) ? false : true;
numa_alloc_interleaved = cmdline::parse_or_default("numa_alloc_interleaved", numa_alloc_interleaved);
if (numa_alloc_interleaved) {
hwloc_cpuset_t all_cpus =
hwloc_bitmap_dup(hwloc_topology_get_topology_cpuset(topology));
int err = hwloc_set_membind(topology, all_cpus, HWLOC_MEMBIND_INTERLEAVE, 0);
if (err < 0) {
printf("Warning: failed to set NUMA round-robin allocation policy\n");
}
}
#endif
}
static int
hwloc_fix_membind_cpuset(hwloc_topology_t topology, hwloc_nodeset_t nodeset, hwloc_const_cpuset_t cpuset)
{
hwloc_const_bitmap_t topology_set = hwloc_topology_get_topology_cpuset(topology);
hwloc_const_bitmap_t complete_set = hwloc_topology_get_complete_cpuset(topology);
hwloc_const_bitmap_t complete_nodeset = hwloc_topology_get_complete_nodeset(topology);
if (!topology_set) {
/* The topology is composed of several systems, the cpuset is thus
* ambiguous. */
errno = EXDEV;
return -1;
}
if (!complete_nodeset) {
/* There is no NUMA node */
errno = ENODEV;
return -1;
}
if (hwloc_bitmap_iszero(cpuset)) {
errno = EINVAL;
return -1;
}
if (!hwloc_bitmap_isincluded(cpuset, complete_set)) {
errno = EINVAL;
return -1;
}
if (hwloc_bitmap_isincluded(topology_set, cpuset)) {
hwloc_bitmap_copy(nodeset, complete_nodeset);
return 0;
}
hwloc_cpuset_to_nodeset(topology, cpuset, nodeset);
return 0;
}
int main(int argc, char *argv[])
{
const struct hwloc_topology_support *support;
hwloc_topology_t topology;
hwloc_const_bitmap_t topocpuset;
hwloc_bitmap_t cpuset;
unsigned long flags = 0;
DIR *dir;
struct dirent *dirent;
int show_all = 0;
int show_threads = 0;
int get_last_cpu_location = 0;
char *callname;
char *pidcmd = NULL;
int err;
int opt;
callname = strrchr(argv[0], '/');
if (!callname)
callname = argv[0];
else
callname++;
/* skip argv[0], handle options */
argc--;
argv++;
hwloc_utils_check_api_version(callname);
while (argc >= 1) {
opt = 0;
if (!strcmp(argv[0], "-a"))
show_all = 1;
else if (!strcmp(argv[0], "-l") || !strcmp(argv[0], "--logical")) {
logical = 1;
} else if (!strcmp(argv[0], "-p") || !strcmp(argv[0], "--physical")) {
logical = 0;
} else if (!strcmp(argv[0], "-c") || !strcmp(argv[0], "--cpuset")) {
show_cpuset = 1;
} else if (!strcmp(argv[0], "-e") || !strncmp(argv[0], "--get-last-cpu-location", 10)) {
get_last_cpu_location = 1;
} else if (!strcmp(argv[0], "-t") || !strcmp(argv[0], "--threads")) {
#ifdef HWLOC_LINUX_SYS
show_threads = 1;
#else
fprintf (stderr, "Listing threads is currently only supported on Linux\n");
#endif
} else if (!strcmp (argv[0], "--whole-system")) {
flags |= HWLOC_TOPOLOGY_FLAG_WHOLE_SYSTEM;
} else if (!strcmp (argv[0], "--pid-cmd")) {
if (argc < 2) {
usage(callname, stdout);
exit(EXIT_FAILURE);
}
pidcmd = argv[1];
opt = 1;
} else {
fprintf (stderr, "Unrecognized option: %s\n", argv[0]);
usage (callname, stderr);
exit(EXIT_FAILURE);
}
argc -= opt+1;
argv += opt+1;
}
err = hwloc_topology_init(&topology);
if (err)
goto out;
hwloc_topology_set_flags(topology, flags);
err = hwloc_topology_load(topology);
if (err)
goto out_with_topology;
support = hwloc_topology_get_support(topology);
if (get_last_cpu_location) {
if (!support->cpubind->get_proc_last_cpu_location)
goto out_with_topology;
} else {
if (!support->cpubind->get_proc_cpubind)
goto out_with_topology;
}
topocpuset = hwloc_topology_get_topology_cpuset(topology);
dir = opendir("/proc");
if (!dir)
goto out_with_topology;
cpuset = hwloc_bitmap_alloc();
if (!cpuset)
goto out_with_dir;
while ((dirent = readdir(dir))) {
long pid_number;
hwloc_pid_t pid;
char pidoutput[1024];
char *end;
char name[64] = "";
/* management of threads */
unsigned boundthreads = 0, i;
long *tids = NULL; /* NULL if process is not threaded */
hwloc_bitmap_t *tidcpusets = NULL;
pid_number = strtol(dirent->d_name, &end, 10);
if (*end)
/* Not a number */
continue;
pid = hwloc_pid_from_number(pid_number, 0);
#ifdef HWLOC_LINUX_SYS
{
unsigned pathlen = 6 + strlen(dirent->d_name) + 1 + 7 + 1;
char *path;
int file;
ssize_t n;
path = malloc(pathlen);
snprintf(path, pathlen, "/proc/%s/cmdline", dirent->d_name);
file = open(path, O_RDONLY);
free(path);
if (file >= 0) {
n = read(file, name, sizeof(name) - 1);
close(file);
if (n <= 0)
/* Ignore kernel threads and errors */
continue;
name[n] = 0;
}
}
#endif /* HWLOC_LINUX_SYS */
if (show_threads) {
#ifdef HWLOC_LINUX_SYS
/* check if some threads must be displayed */
unsigned pathlen = 6 + strlen(dirent->d_name) + 1 + 4 + 1;
char *path;
DIR *taskdir;
path = malloc(pathlen);
snprintf(path, pathlen, "/proc/%s/task", dirent->d_name);
taskdir = opendir(path);
if (taskdir) {
struct dirent *taskdirent;
long tid;
unsigned n = 0;
/* count threads */
while ((taskdirent = readdir(taskdir))) {
tid = strtol(taskdirent->d_name, &end, 10);
if (*end)
/* Not a number */
continue;
n++;
}
if (n > 1) {
/* if there's more than one thread, see if some are bound */
tids = malloc(n * sizeof(*tids));
tidcpusets = calloc(n+1, sizeof(*tidcpusets));
if (tids && tidcpusets) {
/* reread the directory but gather info now */
rewinddir(taskdir);
i = 0;
while ((taskdirent = readdir(taskdir))) {
tid = strtol(taskdirent->d_name, &end, 10);
if (*end)
/* Not a number */
continue;
if (get_last_cpu_location) {
if (hwloc_linux_get_tid_last_cpu_location(topology, tid, cpuset))
continue;
} else {
if (hwloc_linux_get_tid_cpubind(topology, tid, cpuset))
continue;
}
hwloc_bitmap_and(cpuset, cpuset, topocpuset);
tids[i] = tid;
tidcpusets[i] = hwloc_bitmap_dup(cpuset);
i++;
if (hwloc_bitmap_iszero(cpuset))
continue;
if (hwloc_bitmap_isequal(cpuset, topocpuset) && !show_all)
continue;
boundthreads++;
}
} else {
/* failed to alloc, behave as if there were no threads */
free(tids);
tids = NULL;
free(tidcpusets);
tidcpusets = NULL;
}
}
closedir(taskdir);
}
#endif /* HWLOC_LINUX_SYS */
}
if (get_last_cpu_location) {
if (hwloc_get_proc_last_cpu_location(topology, pid, cpuset, 0))
continue;
} else {
if (hwloc_get_proc_cpubind(topology, pid, cpuset, 0))
continue;
}
hwloc_bitmap_and(cpuset, cpuset, topocpuset);
if (hwloc_bitmap_iszero(cpuset))
continue;
/* don't print anything if the process isn't bound and if no threads are bound and if not showing all */
if (hwloc_bitmap_isequal(cpuset, topocpuset) && (!tids || !boundthreads) && !show_all)
continue;
pidoutput[0] = '\0';
if (pidcmd) {
char *cmd;
FILE *file;
cmd = malloc(strlen(pidcmd)+1+5+2+1);
sprintf(cmd, "%s %u", pidcmd, pid);
file = popen(cmd, "r");
if (file) {
if (fgets(pidoutput, sizeof(pidoutput), file)) {
end = strchr(pidoutput, '\n');
if (end)
*end = '\0';
}
pclose(file);
}
free(cmd);
}
/* print the process */
print_task(topology, pid_number, name, cpuset, pidoutput[0] == '\0' ? NULL : pidoutput, 0);
if (tids)
/* print each tid we found (it's tidcpuset isn't NULL anymore) */
for(i=0; tidcpusets[i] != NULL; i++) {
print_task(topology, tids[i], "", tidcpusets[i], NULL, 1);
hwloc_bitmap_free(tidcpusets[i]);
}
/* free threads stuff */
free(tidcpusets);
free(tids);
}
err = 0;
hwloc_bitmap_free(cpuset);
out_with_dir:
closedir(dir);
out_with_topology:
hwloc_topology_destroy(topology);
out:
return err;
}
void output_console(struct lstopo_output *loutput, const char *filename)
{
hwloc_topology_t topology = loutput->topology;
unsigned topodepth;
int verbose_mode = loutput->verbose_mode;
int logical = loutput->logical;
FILE *output;
output = open_output(filename, loutput->overwrite);
if (!output) {
fprintf(stderr, "Failed to open %s for writing (%s)\n", filename, strerror(errno));
return;
}
topodepth = hwloc_topology_get_depth(topology);
/*
* if verbose_mode == 0, only print the summary.
* if verbose_mode == 1, only print the topology tree.
* if verbose_mode > 1, print both.
*/
if (lstopo_show_only != (hwloc_obj_type_t)-1) {
if (verbose_mode > 1)
fprintf(output, "Only showing %s objects\n", hwloc_obj_type_string(lstopo_show_only));
output_only (topology, hwloc_get_root_obj(topology), output, logical, verbose_mode);
} else if (verbose_mode >= 1) {
output_topology (topology, hwloc_get_root_obj(topology), NULL, output, 0, logical, verbose_mode);
fprintf(output, "\n");
}
if ((verbose_mode > 1 || !verbose_mode) && lstopo_show_only == (hwloc_obj_type_t)-1) {
hwloc_lstopo_show_summary(output, topology);
}
if (verbose_mode > 1 && lstopo_show_only == (hwloc_obj_type_t)-1) {
const struct hwloc_distances_s * distances;
unsigned depth;
for (depth = 0; depth < topodepth; depth++) {
distances = hwloc_get_whole_distance_matrix_by_depth(topology, depth);
if (!distances || !distances->latency)
continue;
fprintf(output, "relative latency matrix between %ss (depth %u) by %s indexes:\n",
hwloc_obj_type_string(hwloc_get_depth_type(topology, depth)),
depth,
logical ? "logical" : "physical");
hwloc_utils_print_distance_matrix(output, topology, hwloc_get_root_obj(topology), distances->nbobjs, depth, distances->latency, logical);
}
}
if (verbose_mode > 1 && lstopo_show_only == (hwloc_obj_type_t)-1) {
hwloc_const_bitmap_t complete = hwloc_topology_get_complete_cpuset(topology);
hwloc_const_bitmap_t topo = hwloc_topology_get_topology_cpuset(topology);
hwloc_const_bitmap_t allowed = hwloc_topology_get_allowed_cpuset(topology);
if (!hwloc_bitmap_isequal(topo, complete)) {
hwloc_bitmap_t unknown = hwloc_bitmap_alloc();
char *unknownstr;
hwloc_bitmap_copy(unknown, complete);
hwloc_bitmap_andnot(unknown, unknown, topo);
hwloc_bitmap_asprintf(&unknownstr, unknown);
fprintf (output, "%d processors not represented in topology: %s\n", hwloc_bitmap_weight(unknown), unknownstr);
free(unknownstr);
hwloc_bitmap_free(unknown);
}
if (!hwloc_bitmap_isequal(topo, allowed)) {
hwloc_bitmap_t disallowed = hwloc_bitmap_alloc();
char *disallowedstr;
hwloc_bitmap_copy(disallowed, topo);
hwloc_bitmap_andnot(disallowed, disallowed, allowed);
hwloc_bitmap_asprintf(&disallowedstr, disallowed);
fprintf(output, "%d processors represented but not allowed: %s\n", hwloc_bitmap_weight(disallowed), disallowedstr);
free(disallowedstr);
hwloc_bitmap_free(disallowed);
}
if (!hwloc_topology_is_thissystem(topology))
fprintf (output, "Topology not from this system\n");
}
if (output != stdout)
fclose(output);
}
void output_console(hwloc_topology_t topology, const char *filename, int logical, int legend __hwloc_attribute_unused, int verbose_mode)
{
unsigned topodepth;
FILE *output;
if (!filename || !strcmp(filename, "-"))
output = stdout;
else {
output = open_file(filename, "w");
if (!output) {
fprintf(stderr, "Failed to open %s for writing (%s)\n", filename, strerror(errno));
return;
}
}
topodepth = hwloc_topology_get_depth(topology);
/*
* if verbose_mode == 0, only print the summary.
* if verbose_mode == 1, only print the topology tree.
* if verbose_mode > 1, print both.
*/
if (lstopo_show_only != (hwloc_obj_type_t)-1) {
if (verbose_mode > 1)
fprintf(output, "Only showing %s objects\n", hwloc_obj_type_string(lstopo_show_only));
output_only (topology, hwloc_get_root_obj(topology), output, logical, verbose_mode);
} else if (verbose_mode >= 1) {
output_topology (topology, hwloc_get_root_obj(topology), NULL, output, 0, logical, verbose_mode);
fprintf(output, "\n");
}
if ((verbose_mode > 1 || !verbose_mode) && lstopo_show_only == (hwloc_obj_type_t)-1) {
hwloc_lstopo_show_summary(output, topology);
}
if (verbose_mode > 1 && lstopo_show_only == (hwloc_obj_type_t)-1) {
const struct hwloc_distances_s * distances;
unsigned depth;
for (depth = 0; depth < topodepth; depth++) {
distances = hwloc_get_whole_distance_matrix_by_depth(topology, depth);
if (!distances || !distances->latency)
continue;
printf("latency matrix between %ss (depth %u) by %s indexes:\n",
hwloc_obj_type_string(hwloc_get_depth_type(topology, depth)),
depth,
logical ? "logical" : "physical");
hwloc_utils_print_distance_matrix(topology, hwloc_get_root_obj(topology), distances->nbobjs, depth, distances->latency, logical);
}
}
if (verbose_mode > 1 && lstopo_show_only == (hwloc_obj_type_t)-1) {
hwloc_const_bitmap_t complete = hwloc_topology_get_complete_cpuset(topology);
hwloc_const_bitmap_t topo = hwloc_topology_get_topology_cpuset(topology);
hwloc_const_bitmap_t online = hwloc_topology_get_online_cpuset(topology);
hwloc_const_bitmap_t allowed = hwloc_topology_get_allowed_cpuset(topology);
if (complete && !hwloc_bitmap_isequal(topo, complete)) {
hwloc_bitmap_t unknown = hwloc_bitmap_alloc();
char *unknownstr;
hwloc_bitmap_copy(unknown, complete);
hwloc_bitmap_andnot(unknown, unknown, topo);
hwloc_bitmap_asprintf(&unknownstr, unknown);
fprintf (output, "%d processors not represented in topology: %s\n", hwloc_bitmap_weight(unknown), unknownstr);
free(unknownstr);
hwloc_bitmap_free(unknown);
}
if (complete && !hwloc_bitmap_isequal(online, complete)) {
hwloc_bitmap_t offline = hwloc_bitmap_alloc();
char *offlinestr;
hwloc_bitmap_copy(offline, complete);
hwloc_bitmap_andnot(offline, offline, online);
hwloc_bitmap_asprintf(&offlinestr, offline);
fprintf (output, "%d processors offline: %s\n", hwloc_bitmap_weight(offline), offlinestr);
free(offlinestr);
hwloc_bitmap_free(offline);
}
if (complete && !hwloc_bitmap_isequal(allowed, online)) {
if (!hwloc_bitmap_isincluded(online, allowed)) {
hwloc_bitmap_t forbidden = hwloc_bitmap_alloc();
char *forbiddenstr;
hwloc_bitmap_copy(forbidden, online);
hwloc_bitmap_andnot(forbidden, forbidden, allowed);
hwloc_bitmap_asprintf(&forbiddenstr, forbidden);
fprintf(output, "%d processors online but not allowed: %s\n", hwloc_bitmap_weight(forbidden), forbiddenstr);
free(forbiddenstr);
hwloc_bitmap_free(forbidden);
}
if (!hwloc_bitmap_isincluded(allowed, online)) {
hwloc_bitmap_t potential = hwloc_bitmap_alloc();
char *potentialstr;
hwloc_bitmap_copy(potential, allowed);
hwloc_bitmap_andnot(potential, potential, online);
hwloc_bitmap_asprintf(&potentialstr, potential);
fprintf(output, "%d processors allowed but not online: %s\n", hwloc_bitmap_weight(potential), potentialstr);
free(potentialstr);
hwloc_bitmap_free(potential);
}
}
if (!hwloc_topology_is_thissystem(topology))
fprintf (output, "Topology not from this system\n");
}
if (output != stdout)
fclose(output);
}
int main(void)
{
hwloc_topology_t topology;
hwloc_bitmap_t set, set2;
hwloc_const_bitmap_t cset_available, cset_all;
hwloc_obj_t obj;
char *buffer;
char type[64];
unsigned i;
int err;
/* create a topology */
err = hwloc_topology_init(&topology);
if (err < 0) {
fprintf(stderr, "failed to initialize the topology\n");
return EXIT_FAILURE;
}
err = hwloc_topology_load(topology);
if (err < 0) {
fprintf(stderr, "failed to load the topology\n");
hwloc_topology_destroy(topology);
return EXIT_FAILURE;
}
/* retrieve the entire set of available PUs */
cset_available = hwloc_topology_get_topology_cpuset(topology);
/* retrieve the CPU binding of the current entire process */
set = hwloc_bitmap_alloc();
if (!set) {
fprintf(stderr, "failed to allocate a bitmap\n");
hwloc_topology_destroy(topology);
return EXIT_FAILURE;
}
err = hwloc_get_cpubind(topology, set, HWLOC_CPUBIND_PROCESS);
if (err < 0) {
fprintf(stderr, "failed to get cpu binding\n");
hwloc_bitmap_free(set);
hwloc_topology_destroy(topology);
}
/* display the processing units that cannot be used by this process */
if (hwloc_bitmap_isequal(set, cset_available)) {
printf("this process can use all available processing units in the system\n");
} else {
/* compute the set where we currently cannot run.
* we can't modify cset_available because it's a system read-only one,
* so we do set = available &~ set
*/
hwloc_bitmap_andnot(set, cset_available, set);
hwloc_bitmap_asprintf(&buffer, set);
printf("process cannot use %d process units (%s) among %u in the system\n",
hwloc_bitmap_weight(set), buffer, hwloc_bitmap_weight(cset_available));
free(buffer);
/* restore set where it was before the &~ operation above */
hwloc_bitmap_andnot(set, cset_available, set);
}
/* print the smallest object covering the current process binding */
obj = hwloc_get_obj_covering_cpuset(topology, set);
hwloc_obj_type_snprintf(type, sizeof(type), obj, 0);
printf("process is bound within object %s logical index %u\n", type, obj->logical_index);
/* retrieve the single PU where the current thread actually runs within this process binding */
set2 = hwloc_bitmap_alloc();
if (!set2) {
fprintf(stderr, "failed to allocate a bitmap\n");
hwloc_bitmap_free(set);
hwloc_topology_destroy(topology);
return EXIT_FAILURE;
}
err = hwloc_get_last_cpu_location(topology, set2, HWLOC_CPUBIND_THREAD);
if (err < 0) {
fprintf(stderr, "failed to get last cpu location\n");
hwloc_bitmap_free(set);
hwloc_bitmap_free(set2);
hwloc_topology_destroy(topology);
}
/* sanity checks that are not actually needed but help the reader */
/* this thread runs within the process binding */
assert(hwloc_bitmap_isincluded(set2, set));
/* this thread runs on a single PU at a time */
assert(hwloc_bitmap_weight(set2) == 1);
/* print the logical number of the PU where that thread runs */
/* extract the PU OS index from the bitmap */
i = hwloc_bitmap_first(set2);
obj = hwloc_get_pu_obj_by_os_index(topology, i);
printf("thread is now running on PU logical index %u (OS/physical index %u)\n",
obj->logical_index, i);
/* migrate this single thread to where other PUs within the current binding */
hwloc_bitmap_andnot(set2, set, set2);
err = hwloc_set_cpubind(topology, set2, HWLOC_CPUBIND_THREAD);
if (err < 0) {
fprintf(stderr, "failed to set thread binding\n");
hwloc_bitmap_free(set);
hwloc_bitmap_free(set2);
hwloc_topology_destroy(topology);
}
/* reprint the PU where that thread runs */
err = hwloc_get_last_cpu_location(topology, set2, HWLOC_CPUBIND_THREAD);
if (err < 0) {
fprintf(stderr, "failed to get last cpu location\n");
hwloc_bitmap_free(set);
hwloc_bitmap_free(set2);
hwloc_topology_destroy(topology);
}
/* print the logical number of the PU where that thread runs */
/* extract the PU OS index from the bitmap */
i = hwloc_bitmap_first(set2);
obj = hwloc_get_pu_obj_by_os_index(topology, i);
printf("thread is running on PU logical index %u (OS/physical index %u)\n",
obj->logical_index, i);
hwloc_bitmap_free(set);
hwloc_bitmap_free(set2);
/* retrieve the entire set of all PUs */
cset_all = hwloc_topology_get_complete_cpuset(topology);
if (hwloc_bitmap_isequal(cset_all, cset_available)) {
printf("all hardware PUs are available\n");
} else {
printf("only %d hardware PUs are available in the machine among %d\n",
hwloc_bitmap_weight(cset_available), hwloc_bitmap_weight(cset_all));
}
hwloc_topology_destroy(topology);
return EXIT_SUCCESS;
}
int main(void)
{
hwloc_topology_t topology;
cl_int clret;
cl_platform_id *platform_ids;
unsigned nrp, nrd, count, i, j;
int err;
hwloc_topology_init(&topology);
hwloc_topology_set_type_filter(topology, HWLOC_OBJ_PCI_DEVICE, HWLOC_TYPE_FILTER_KEEP_IMPORTANT);
hwloc_topology_set_type_filter(topology, HWLOC_OBJ_OS_DEVICE, HWLOC_TYPE_FILTER_KEEP_IMPORTANT);
hwloc_topology_load(topology);
clret = clGetPlatformIDs(0, NULL, &nrp);
if (CL_SUCCESS != clret || !nrp)
return 0;
platform_ids = malloc(nrp * sizeof(*platform_ids));
if (!platform_ids)
return 0;
clret = clGetPlatformIDs(nrp, platform_ids, &nrp);
if (CL_SUCCESS != clret || !nrp)
return 0;
count = 0;
for(i=0; i<nrp; i++) {
cl_device_id *device_ids;
clret = clGetDeviceIDs(platform_ids[i], CL_DEVICE_TYPE_ALL, 0, NULL, &nrd);
if (CL_SUCCESS != clret || !nrd)
continue;
device_ids = malloc(nrd * sizeof(*device_ids));
if (!device_ids)
continue;
clret = clGetDeviceIDs(platform_ids[i], CL_DEVICE_TYPE_ALL, nrd, device_ids, &nrd);
if (CL_SUCCESS != clret || !nrd)
continue;
for(j=0; j<nrd; j++) {
hwloc_bitmap_t set;
hwloc_obj_t osdev, osdev2, ancestor;
const char *value;
unsigned p, d;
osdev = hwloc_opencl_get_device_osdev_by_index(topology, i, j);
/* we currently insert all OpenCL devices, except CPU devices */
if (!osdev) {
cl_device_type type;
clGetDeviceInfo(device_ids[j], CL_DEVICE_TYPE, sizeof(type), &type, NULL);
assert(type == CL_DEVICE_TYPE_CPU);
continue;
}
/* try to get it from PCI locality (only works with AMD extensions) */
osdev2 = hwloc_opencl_get_device_osdev(topology, device_ids[j]);
if (osdev2) {
assert(osdev == osdev2);
}
ancestor = hwloc_get_non_io_ancestor_obj(topology, osdev);
set = hwloc_bitmap_alloc();
err = hwloc_opencl_get_device_cpuset(topology, device_ids[j], set);
if (err < 0) {
printf("no cpuset for platform %u device %u\n", i, j);
} else {
char *cpuset_string = NULL;
hwloc_bitmap_asprintf(&cpuset_string, set);
printf("got cpuset %s for platform %u device %u\n", cpuset_string, i, j);
free(cpuset_string);
if (hwloc_bitmap_isequal(hwloc_topology_get_complete_cpuset(topology), hwloc_topology_get_topology_cpuset(topology)))
/* only compare if the topology is complete, otherwise things can be significantly different */
assert(hwloc_bitmap_isequal(set, ancestor->cpuset));
}
hwloc_bitmap_free(set);
printf("found OSDev %s\n", osdev->name);
err = sscanf(osdev->name, "opencl%ud%u", &p, &d);
assert(err == 2);
assert(p == i);
assert(d == j);
value = hwloc_obj_get_info_by_name(osdev, "Backend");
err = strcmp(value, "OpenCL");
assert(!err);
assert(osdev->attr->osdev.type == HWLOC_OBJ_OSDEV_COPROC);
value = osdev->subtype;
assert(value);
err = strcmp(value, "OpenCL");
assert(!err);
value = hwloc_obj_get_info_by_name(osdev, "GPUModel");
printf("found OSDev model %s\n", value);
count++;
}
}
hwloc_topology_destroy(topology);
return 0;
}
int main(void)
{
hwloc_topology_t topology;
int i;
int err;
hwloc_topology_init(&topology);
hwloc_topology_set_type_filter(topology, HWLOC_OBJ_PCI_DEVICE, HWLOC_TYPE_FILTER_KEEP_IMPORTANT);
hwloc_topology_set_type_filter(topology, HWLOC_OBJ_OS_DEVICE, HWLOC_TYPE_FILTER_KEEP_IMPORTANT);
hwloc_topology_load(topology);
for(i=0; ; i++) {
hwloc_bitmap_t set;
hwloc_obj_t osdev, ancestor;
const char *value;
osdev = hwloc_intel_mic_get_device_osdev_by_index(topology, i);
if (!osdev)
break;
assert(osdev);
ancestor = hwloc_get_non_io_ancestor_obj(topology, osdev);
printf("found OSDev %s\n", osdev->name);
err = strncmp(osdev->name, "mic", 3);
assert(!err);
assert(atoi(osdev->name+3) == (int) i);
assert(osdev->attr->osdev.type == HWLOC_OBJ_OSDEV_COPROC);
value = hwloc_obj_get_info_by_name(osdev, "CoProcType");
err = strcmp(value, "MIC");
assert(!err);
value = hwloc_obj_get_info_by_name(osdev, "MICFamily");
printf("found MICFamily %s\n", value);
value = hwloc_obj_get_info_by_name(osdev, "MICSKU");
printf("found MICSKU %s\n", value);
value = hwloc_obj_get_info_by_name(osdev, "MICActiveCores");
printf("found MICActiveCores %s\n", value);
value = hwloc_obj_get_info_by_name(osdev, "MICMemorySize");
printf("found MICMemorySize %s\n", value);
set = hwloc_bitmap_alloc();
err = hwloc_intel_mic_get_device_cpuset(topology, i, set);
if (err < 0) {
printf("failed to get cpuset for device %d\n", i);
} else {
char *cpuset_string = NULL;
hwloc_bitmap_asprintf(&cpuset_string, set);
printf("got cpuset %s for device %d\n", cpuset_string, i);
if (hwloc_bitmap_isequal(hwloc_topology_get_complete_cpuset(topology), hwloc_topology_get_topology_cpuset(topology)))
/* only compare if the topology is complete, otherwise things can be significantly different */
assert(hwloc_bitmap_isequal(set, ancestor->cpuset));
free(cpuset_string);
}
hwloc_bitmap_free(set);
}
hwloc_topology_destroy(topology);
return 0;
}
int main(void)
{
hwloc_topology_t topology;
cl_int clret;
cl_platform_id *platform_ids;
unsigned nrp, nrd, count, i, j;
int err;
hwloc_topology_init(&topology);
hwloc_topology_set_flags(topology, HWLOC_TOPOLOGY_FLAG_IO_DEVICES);
hwloc_topology_load(topology);
clret = clGetPlatformIDs(0, NULL, &nrp);
if (CL_SUCCESS != clret || !nrp)
return 0;
platform_ids = malloc(nrp * sizeof(*platform_ids));
if (!platform_ids)
return 0;
clret = clGetPlatformIDs(nrp, platform_ids, &nrp);
if (CL_SUCCESS != clret || !nrp)
return 0;
count = 0;
for(i=0; i<nrp; i++) {
cl_device_id *device_ids;
clret = clGetDeviceIDs(platform_ids[i], CL_DEVICE_TYPE_ALL, 0, NULL, &nrd);
if (CL_SUCCESS != clret || !nrd)
continue;
device_ids = malloc(nrd * sizeof(*device_ids));
if (!device_ids)
continue;
clret = clGetDeviceIDs(platform_ids[i], CL_DEVICE_TYPE_ALL, nrd, device_ids, &nrd);
if (CL_SUCCESS != clret || !nrd)
continue;
for(j=0; j<nrd; j++) {
hwloc_bitmap_t set;
hwloc_obj_t osdev, osdev2, ancestor;
const char *value;
unsigned p, d;
osdev = hwloc_opencl_get_device_osdev(topology, device_ids[j]);
osdev2 = hwloc_opencl_get_device_osdev_by_index(topology, i, j);
assert(osdev == osdev2);
if (!osdev) {
printf("no osdev for platform %u device %u\n", i, j);
continue;
}
ancestor = hwloc_get_non_io_ancestor_obj(topology, osdev);
set = hwloc_bitmap_alloc();
err = hwloc_opencl_get_device_cpuset(topology, device_ids[j], set);
if (err < 0) {
printf("no cpuset for platform %u device %u\n", i, j);
} else {
char *cpuset_string = NULL;
hwloc_bitmap_asprintf(&cpuset_string, set);
printf("got cpuset %s for platform %u device %u\n", cpuset_string, i, j);
free(cpuset_string);
if (hwloc_bitmap_isequal(hwloc_topology_get_complete_cpuset(topology), hwloc_topology_get_topology_cpuset(topology)))
/* only compare if the topology is complete, otherwise things can be significantly different */
assert(hwloc_bitmap_isequal(set, ancestor->cpuset));
}
hwloc_bitmap_free(set);
printf("found OSDev %s\n", osdev->name);
err = sscanf(osdev->name, "opencl%ud%u", &p, &d);
assert(err == 2);
assert(p == i);
assert(d == j);
value = hwloc_obj_get_info_by_name(osdev, "Backend");
err = strcmp(value, "OpenCL");
assert(!err);
assert(osdev->attr->osdev.type == HWLOC_OBJ_OSDEV_COPROC);
value = hwloc_obj_get_info_by_name(osdev, "CoProcType");
err = strcmp(value, "OpenCL");
assert(!err);
value = hwloc_obj_get_info_by_name(osdev, "GPUModel");
printf("found OSDev model %s\n", value);
count++;
}
}
hwloc_topology_destroy(topology);
return 0;
}
