]>
Commit | Line | Data |
---|---|---|
818ab71a | 1 | /* Copyright (C) 2005-2016 Free Software Foundation, Inc. |
12aac30b JJ |
2 | Contributed by Jakub Jelinek <jakub@redhat.com>. |
3 | ||
f1f3453e TS |
4 | This file is part of the GNU Offloading and Multi Processing Library |
5 | (libgomp). | |
12aac30b JJ |
6 | |
7 | Libgomp is free software; you can redistribute it and/or modify it | |
748086b7 JJ |
8 | under the terms of the GNU General Public License as published by |
9 | the Free Software Foundation; either version 3, or (at your option) | |
10 | any later version. | |
12aac30b JJ |
11 | |
12 | Libgomp is distributed in the hope that it will be useful, but WITHOUT ANY | |
13 | WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS | |
748086b7 | 14 | FOR A PARTICULAR PURPOSE. See the GNU General Public License for |
12aac30b JJ |
15 | more details. |
16 | ||
748086b7 JJ |
17 | Under Section 7 of GPL version 3, you are granted additional |
18 | permissions described in the GCC Runtime Library Exception, version | |
19 | 3.1, as published by the Free Software Foundation. | |
20 | ||
21 | You should have received a copy of the GNU General Public License and | |
22 | a copy of the GCC Runtime Library Exception along with this program; | |
23 | see the files COPYING3 and COPYING.RUNTIME respectively. If not, see | |
24 | <http://www.gnu.org/licenses/>. */ | |
12aac30b JJ |
25 | |
26 | /* This file contains system specific routines related to counting | |
27 | online processors and dynamic load balancing. */ | |
28 | ||
29 | #ifndef _GNU_SOURCE | |
30 | #define _GNU_SOURCE 1 | |
31 | #endif | |
32 | #include "libgomp.h" | |
e0b23d9f | 33 | #include "proc.h" |
acf0174b | 34 | #include <errno.h> |
12aac30b JJ |
35 | #include <stdlib.h> |
36 | #include <unistd.h> | |
37 | #ifdef HAVE_GETLOADAVG | |
38 | # ifdef HAVE_SYS_LOADAVG_H | |
39 | # include <sys/loadavg.h> | |
40 | # endif | |
41 | #endif | |
42 | ||
43 | #ifdef HAVE_PTHREAD_AFFINITY_NP | |
acf0174b JJ |
44 | unsigned long gomp_cpuset_size; |
45 | static unsigned long gomp_get_cpuset_size; | |
46 | cpu_set_t *gomp_cpusetp; | |
47 | ||
e0b23d9f | 48 | unsigned long |
acf0174b | 49 | gomp_cpuset_popcount (unsigned long cpusetsize, cpu_set_t *cpusetp) |
12aac30b | 50 | { |
acf0174b JJ |
51 | #ifdef CPU_COUNT_S |
52 | /* glibc 2.7 and above provide a macro for this. */ | |
53 | return CPU_COUNT_S (cpusetsize, cpusetp); | |
12aac30b | 54 | #else |
acf0174b JJ |
55 | #ifdef CPU_COUNT |
56 | if (cpusetsize == sizeof (cpu_set_t)) | |
57 | /* glibc 2.6 and above provide a macro for this. */ | |
58 | return CPU_COUNT (cpusetp); | |
59 | #endif | |
12aac30b JJ |
60 | size_t i; |
61 | unsigned long ret = 0; | |
acf0174b | 62 | extern int check[sizeof (cpusetp->__bits[0]) == sizeof (unsigned long int) |
f89163fd | 63 | ? 1 : -1] __attribute__((unused)); |
12aac30b | 64 | |
acf0174b | 65 | for (i = 0; i < cpusetsize / sizeof (cpusetp->__bits[0]); i++) |
12aac30b JJ |
66 | { |
67 | unsigned long int mask = cpusetp->__bits[i]; | |
68 | if (mask == 0) | |
69 | continue; | |
70 | ret += __builtin_popcountl (mask); | |
71 | } | |
72 | return ret; | |
73 | #endif | |
74 | } | |
75 | #endif | |
76 | ||
77 | /* At startup, determine the default number of threads. It would seem | |
78 | this should be related to the number of cpus online. */ | |
79 | ||
80 | void | |
81 | gomp_init_num_threads (void) | |
82 | { | |
83 | #ifdef HAVE_PTHREAD_AFFINITY_NP | |
acf0174b JJ |
84 | #if defined (_SC_NPROCESSORS_CONF) && defined (CPU_ALLOC_SIZE) |
85 | gomp_cpuset_size = sysconf (_SC_NPROCESSORS_CONF); | |
86 | gomp_cpuset_size = CPU_ALLOC_SIZE (gomp_cpuset_size); | |
87 | #else | |
88 | gomp_cpuset_size = sizeof (cpu_set_t); | |
89 | #endif | |
12aac30b | 90 | |
acf0174b JJ |
91 | gomp_cpusetp = (cpu_set_t *) gomp_malloc (gomp_cpuset_size); |
92 | do | |
12aac30b | 93 | { |
acf0174b JJ |
94 | int ret = pthread_getaffinity_np (pthread_self (), gomp_cpuset_size, |
95 | gomp_cpusetp); | |
96 | if (ret == 0) | |
97 | { | |
acf0174b JJ |
98 | /* Count only the CPUs this process can use. */ |
99 | gomp_global_icv.nthreads_var | |
100 | = gomp_cpuset_popcount (gomp_cpuset_size, gomp_cpusetp); | |
101 | if (gomp_global_icv.nthreads_var == 0) | |
102 | break; | |
103 | gomp_get_cpuset_size = gomp_cpuset_size; | |
104 | #ifdef CPU_ALLOC_SIZE | |
f89163fd | 105 | unsigned long i; |
acf0174b JJ |
106 | for (i = gomp_cpuset_size * 8; i; i--) |
107 | if (CPU_ISSET_S (i - 1, gomp_cpuset_size, gomp_cpusetp)) | |
108 | break; | |
109 | gomp_cpuset_size = CPU_ALLOC_SIZE (i); | |
110 | #endif | |
111 | return; | |
112 | } | |
113 | if (ret != EINVAL) | |
114 | break; | |
115 | #ifdef CPU_ALLOC_SIZE | |
116 | if (gomp_cpuset_size < sizeof (cpu_set_t)) | |
117 | gomp_cpuset_size = sizeof (cpu_set_t); | |
118 | else | |
119 | gomp_cpuset_size = gomp_cpuset_size * 2; | |
120 | if (gomp_cpuset_size < 8 * sizeof (cpu_set_t)) | |
121 | gomp_cpusetp | |
122 | = (cpu_set_t *) gomp_realloc (gomp_cpusetp, gomp_cpuset_size); | |
123 | else | |
124 | { | |
125 | /* Avoid gomp_fatal if too large memory allocation would be | |
126 | requested, e.g. kernel returning EINVAL all the time. */ | |
127 | void *p = realloc (gomp_cpusetp, gomp_cpuset_size); | |
128 | if (p == NULL) | |
129 | break; | |
130 | gomp_cpusetp = (cpu_set_t *) p; | |
131 | } | |
132 | #else | |
133 | break; | |
134 | #endif | |
12aac30b | 135 | } |
acf0174b JJ |
136 | while (1); |
137 | gomp_cpuset_size = 0; | |
138 | gomp_global_icv.nthreads_var = 1; | |
139 | free (gomp_cpusetp); | |
140 | gomp_cpusetp = NULL; | |
12aac30b JJ |
141 | #endif |
142 | #ifdef _SC_NPROCESSORS_ONLN | |
a68ab351 | 143 | gomp_global_icv.nthreads_var = sysconf (_SC_NPROCESSORS_ONLN); |
12aac30b JJ |
144 | #endif |
145 | } | |
146 | ||
147 | static int | |
148 | get_num_procs (void) | |
149 | { | |
150 | #ifdef HAVE_PTHREAD_AFFINITY_NP | |
acf0174b | 151 | if (gomp_places_list == NULL) |
12aac30b JJ |
152 | { |
153 | /* Count only the CPUs this process can use. */ | |
acf0174b JJ |
154 | if (gomp_cpusetp |
155 | && pthread_getaffinity_np (pthread_self (), gomp_get_cpuset_size, | |
156 | gomp_cpusetp) == 0) | |
12aac30b | 157 | { |
acf0174b | 158 | int ret = gomp_cpuset_popcount (gomp_get_cpuset_size, gomp_cpusetp); |
12aac30b JJ |
159 | return ret != 0 ? ret : 1; |
160 | } | |
161 | } | |
162 | else | |
163 | { | |
12aac30b JJ |
164 | /* We can't use pthread_getaffinity_np in this case |
165 | (we have changed it ourselves, it binds to just one CPU). | |
166 | Count instead the number of different CPUs we are | |
e4ba7a60 JJ |
167 | using. gomp_init_affinity updated gomp_available_cpus to |
168 | the number of CPUs in the GOMP_AFFINITY mask that we are | |
169 | allowed to use though. */ | |
170 | return gomp_available_cpus; | |
12aac30b JJ |
171 | } |
172 | #endif | |
173 | #ifdef _SC_NPROCESSORS_ONLN | |
174 | return sysconf (_SC_NPROCESSORS_ONLN); | |
175 | #else | |
a68ab351 | 176 | return gomp_icv (false)->nthreads_var; |
12aac30b JJ |
177 | #endif |
178 | } | |
179 | ||
180 | /* When OMP_DYNAMIC is set, at thread launch determine the number of | |
181 | threads we should spawn for this team. */ | |
182 | /* ??? I have no idea what best practice for this is. Surely some | |
183 | function of the number of processors that are *still* online and | |
184 | the load average. Here I use the number of processors online | |
185 | minus the 15 minute load average. */ | |
186 | ||
187 | unsigned | |
188 | gomp_dynamic_max_threads (void) | |
189 | { | |
a68ab351 | 190 | unsigned n_onln, loadavg, nthreads_var = gomp_icv (false)->nthreads_var; |
12aac30b JJ |
191 | |
192 | n_onln = get_num_procs (); | |
a68ab351 JJ |
193 | if (n_onln > nthreads_var) |
194 | n_onln = nthreads_var; | |
12aac30b JJ |
195 | |
196 | loadavg = 0; | |
197 | #ifdef HAVE_GETLOADAVG | |
198 | { | |
199 | double dloadavg[3]; | |
200 | if (getloadavg (dloadavg, 3) == 3) | |
201 | { | |
202 | /* Add 0.1 to get a kind of biased rounding. */ | |
203 | loadavg = dloadavg[2] + 0.1; | |
204 | } | |
205 | } | |
206 | #endif | |
207 | ||
208 | if (loadavg >= n_onln) | |
209 | return 1; | |
210 | else | |
211 | return n_onln - loadavg; | |
212 | } | |
213 | ||
214 | int | |
215 | omp_get_num_procs (void) | |
216 | { | |
217 | return get_num_procs (); | |
218 | } | |
219 | ||
220 | ialias (omp_get_num_procs) |