1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
|
/* Detect the number of processors.
Copyright (C) 2009-2020 Free Software Foundation, Inc.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3, or (at your option)
any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, see <https://www.gnu.org/licenses/>. */
/* Written by Glen Lenker and Bruno Haible. */
#include <config.h>
#include "nproc.h"
#include <limits.h>
#include <stdlib.h>
#include <unistd.h>
#if HAVE_PTHREAD_GETAFFINITY_NP && 0
# include <pthread.h>
# include <sched.h>
#endif
#if HAVE_SCHED_GETAFFINITY_LIKE_GLIBC || HAVE_SCHED_GETAFFINITY_NP
# include <sched.h>
#endif
#include <sys/types.h>
#if HAVE_SYS_PSTAT_H
# include <sys/pstat.h>
#endif
#if HAVE_SYS_SYSMP_H
# include <sys/sysmp.h>
#endif
#if HAVE_SYS_PARAM_H
# include <sys/param.h>
#endif
#if HAVE_SYS_SYSCTL_H && ! defined __GLIBC__
# include <sys/sysctl.h>
#endif
#if defined _WIN32 && ! defined __CYGWIN__
# define WIN32_LEAN_AND_MEAN
# include <windows.h>
#endif
#include "c-ctype.h"
#include "minmax.h"
#define ARRAY_SIZE(a) (sizeof (a) / sizeof ((a)[0]))
/* Return the number of processors available to the current process, based
on a modern system call that returns the "affinity" between the current
process and each CPU. Return 0 if unknown or if such a system call does
not exist. */
static unsigned long
num_processors_via_affinity_mask (void)
{
/* glibc >= 2.3.3 with NPTL and NetBSD 5 have pthread_getaffinity_np,
but with different APIs. Also it requires linking with -lpthread.
Therefore this code is not enabled.
glibc >= 2.3.4 has sched_getaffinity whereas NetBSD 5 has
sched_getaffinity_np. */
#if HAVE_PTHREAD_GETAFFINITY_NP && defined __GLIBC__ && 0
{
cpu_set_t set;
if (pthread_getaffinity_np (pthread_self (), sizeof (set), &set) == 0)
{
unsigned long count;
# ifdef CPU_COUNT
/* glibc >= 2.6 has the CPU_COUNT macro. */
count = CPU_COUNT (&set);
# else
size_t i;
count = 0;
for (i = 0; i < CPU_SETSIZE; i++)
if (CPU_ISSET (i, &set))
count++;
# endif
if (count > 0)
return count;
}
}
#elif HAVE_PTHREAD_GETAFFINITY_NP && defined __NetBSD__ && 0
{
cpuset_t *set;
set = cpuset_create ();
if (set != NULL)
{
unsigned long count = 0;
if (pthread_getaffinity_np (pthread_self (), cpuset_size (set), set)
== 0)
{
cpuid_t i;
for (i = 0;; i++)
{
int ret = cpuset_isset (i, set);
if (ret < 0)
break;
if (ret > 0)
count++;
}
}
cpuset_destroy (set);
if (count > 0)
return count;
}
}
#elif HAVE_SCHED_GETAFFINITY_LIKE_GLIBC /* glibc >= 2.3.4 */
{
cpu_set_t set;
if (sched_getaffinity (0, sizeof (set), &set) == 0)
{
unsigned long count;
# ifdef CPU_COUNT
/* glibc >= 2.6 has the CPU_COUNT macro. */
count = CPU_COUNT (&set);
# else
size_t i;
count = 0;
for (i = 0; i < CPU_SETSIZE; i++)
if (CPU_ISSET (i, &set))
count++;
# endif
if (count > 0)
return count;
}
}
#elif HAVE_SCHED_GETAFFINITY_NP /* NetBSD >= 5 */
{
cpuset_t *set;
set = cpuset_create ();
if (set != NULL)
{
unsigned long count = 0;
if (sched_getaffinity_np (getpid (), cpuset_size (set), set) == 0)
{
cpuid_t i;
for (i = 0;; i++)
{
int ret = cpuset_isset (i, set);
if (ret < 0)
break;
if (ret > 0)
count++;
}
}
cpuset_destroy (set);
if (count > 0)
return count;
}
}
#endif
#if defined _WIN32 && ! defined __CYGWIN__
{ /* This works on native Windows platforms. */
DWORD_PTR process_mask;
DWORD_PTR system_mask;
if (GetProcessAffinityMask (GetCurrentProcess (),
&process_mask, &system_mask))
{
DWORD_PTR mask = process_mask;
unsigned long count = 0;
for (; mask != 0; mask = mask >> 1)
if (mask & 1)
count++;
if (count > 0)
return count;
}
}
#endif
return 0;
}
/* Return the total number of processors. Here QUERY must be one of
NPROC_ALL, NPROC_CURRENT. The result is guaranteed to be at least 1. */
static unsigned long int
num_processors_ignoring_omp (enum nproc_query query)
{
/* On systems with a modern affinity mask system call, we have
sysconf (_SC_NPROCESSORS_CONF)
>= sysconf (_SC_NPROCESSORS_ONLN)
>= num_processors_via_affinity_mask ()
The first number is the number of CPUs configured in the system.
The second number is the number of CPUs available to the scheduler.
The third number is the number of CPUs available to the current process.
Note! On Linux systems with glibc, the first and second number come from
the /sys and /proc file systems (see
glibc/sysdeps/unix/sysv/linux/getsysstats.c).
In some situations these file systems are not mounted, and the sysconf call
returns 1 or 2 (<https://sourceware.org/bugzilla/show_bug.cgi?id=21542>),
which does not reflect the reality. */
if (query == NPROC_CURRENT)
{
/* Try the modern affinity mask system call. */
{
unsigned long nprocs = num_processors_via_affinity_mask ();
if (nprocs > 0)
return nprocs;
}
#if defined _SC_NPROCESSORS_ONLN
{ /* This works on glibc, Mac OS X 10.5, FreeBSD, AIX, OSF/1, Solaris,
Cygwin, Haiku. */
long int nprocs = sysconf (_SC_NPROCESSORS_ONLN);
if (nprocs > 0)
return nprocs;
}
#endif
}
else /* query == NPROC_ALL */
{
#if defined _SC_NPROCESSORS_CONF
{ /* This works on glibc, Mac OS X 10.5, FreeBSD, AIX, OSF/1, Solaris,
Cygwin, Haiku. */
long int nprocs = sysconf (_SC_NPROCESSORS_CONF);
# if __GLIBC__ >= 2 && defined __linux__
/* On Linux systems with glibc, this information comes from the /sys and
/proc file systems (see glibc/sysdeps/unix/sysv/linux/getsysstats.c).
In some situations these file systems are not mounted, and the
sysconf call returns 1 or 2. But we wish to guarantee that
num_processors (NPROC_ALL) >= num_processors (NPROC_CURRENT). */
if (nprocs == 1 || nprocs == 2)
{
unsigned long nprocs_current = num_processors_via_affinity_mask ();
if (/* nprocs_current > 0 && */ nprocs_current > nprocs)
nprocs = nprocs_current;
}
# endif
if (nprocs > 0)
return nprocs;
}
#endif
}
#if HAVE_PSTAT_GETDYNAMIC
{ /* This works on HP-UX. */
struct pst_dynamic psd;
if (pstat_getdynamic (&psd, sizeof psd, 1, 0) >= 0)
{
/* The field psd_proc_cnt contains the number of active processors.
In newer releases of HP-UX 11, the field psd_max_proc_cnt includes
deactivated processors. */
if (query == NPROC_CURRENT)
{
if (psd.psd_proc_cnt > 0)
return psd.psd_proc_cnt;
}
else
{
if (psd.psd_max_proc_cnt > 0)
return psd.psd_max_proc_cnt;
}
}
}
#endif
#if HAVE_SYSMP && defined MP_NAPROCS && defined MP_NPROCS
{ /* This works on IRIX. */
/* MP_NPROCS yields the number of installed processors.
MP_NAPROCS yields the number of processors available to unprivileged
processes. */
int nprocs =
sysmp (query == NPROC_CURRENT && getuid () != 0
? MP_NAPROCS
: MP_NPROCS);
if (nprocs > 0)
return nprocs;
}
#endif
/* Finally, as fallback, use the APIs that don't distinguish between
NPROC_CURRENT and NPROC_ALL. */
#if HAVE_SYSCTL && ! defined __GLIBC__ && defined HW_NCPU
{ /* This works on Mac OS X, FreeBSD, NetBSD, OpenBSD. */
int nprocs;
size_t len = sizeof (nprocs);
static int mib[2] = { CTL_HW, HW_NCPU };
if (sysctl (mib, ARRAY_SIZE (mib), &nprocs, &len, NULL, 0) == 0
&& len == sizeof (nprocs)
&& 0 < nprocs)
return nprocs;
}
#endif
#if defined _WIN32 && ! defined __CYGWIN__
{ /* This works on native Windows platforms. */
SYSTEM_INFO system_info;
GetSystemInfo (&system_info);
if (0 < system_info.dwNumberOfProcessors)
return system_info.dwNumberOfProcessors;
}
#endif
return 1;
}
/* Parse OMP environment variables without dependence on OMP.
Return 0 for invalid values. */
static unsigned long int
parse_omp_threads (char const* threads)
{
unsigned long int ret = 0;
if (threads == NULL)
return ret;
/* The OpenMP spec says that the value assigned to the environment variables
"may have leading and trailing white space". */
while (*threads != '\0' && c_isspace (*threads))
threads++;
/* Convert it from positive decimal to 'unsigned long'. */
if (c_isdigit (*threads))
{
char *endptr = NULL;
unsigned long int value = strtoul (threads, &endptr, 10);
if (endptr != NULL)
{
while (*endptr != '\0' && c_isspace (*endptr))
endptr++;
if (*endptr == '\0')
return value;
/* Also accept the first value in a nesting level,
since we can't determine the nesting level from env vars. */
else if (*endptr == ',')
return value;
}
}
return ret;
}
unsigned long int
num_processors (enum nproc_query query)
{
unsigned long int omp_env_limit = ULONG_MAX;
if (query == NPROC_CURRENT_OVERRIDABLE)
{
unsigned long int omp_env_threads;
/* Honor the OpenMP environment variables, recognized also by all
programs that are based on OpenMP. */
omp_env_threads = parse_omp_threads (getenv ("OMP_NUM_THREADS"));
omp_env_limit = parse_omp_threads (getenv ("OMP_THREAD_LIMIT"));
if (! omp_env_limit)
omp_env_limit = ULONG_MAX;
if (omp_env_threads)
return MIN (omp_env_threads, omp_env_limit);
query = NPROC_CURRENT;
}
/* Here query is one of NPROC_ALL, NPROC_CURRENT. */
{
unsigned long nprocs = num_processors_ignoring_omp (query);
return MIN (nprocs, omp_env_limit);
}
}
|