util: Cygwin has linux-style pthread_setname_np
[mesa.git] / src / util / u_thread.h
1 /**************************************************************************
2 *
3 * Copyright 1999-2006 Brian Paul
4 * Copyright 2008 VMware, Inc.
5 * All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
21 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
22 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
23 * OTHER DEALINGS IN THE SOFTWARE.
24 *
25 **************************************************************************/
26
27 #ifndef U_THREAD_H_
28 #define U_THREAD_H_
29
30 #include <stdint.h>
31 #include <stdbool.h>
32
33 #include "c11/threads.h"
34 #include "detect_os.h"
35
36 #ifdef HAVE_PTHREAD
37 #include <signal.h>
38 #ifdef PTHREAD_SETAFFINITY_IN_NP_HEADER
39 #include <pthread_np.h>
40 #endif
41 #endif
42
43 #ifdef __FreeBSD__
44 #define cpu_set_t cpuset_t
45 #endif
46
47 static inline thrd_t u_thread_create(int (*routine)(void *), void *param)
48 {
49 thrd_t thread;
50 #ifdef HAVE_PTHREAD
51 sigset_t saved_set, new_set;
52 int ret;
53
54 sigfillset(&new_set);
55 sigdelset(&new_set, SIGSYS);
56 pthread_sigmask(SIG_BLOCK, &new_set, &saved_set);
57 ret = thrd_create( &thread, routine, param );
58 pthread_sigmask(SIG_SETMASK, &saved_set, NULL);
59 #else
60 int ret;
61 ret = thrd_create( &thread, routine, param );
62 #endif
63 if (ret)
64 return 0;
65
66 return thread;
67 }
68
69 static inline void u_thread_setname( const char *name )
70 {
71 #if defined(HAVE_PTHREAD)
72 #if DETECT_OS_LINUX || DETECT_OS_CYGWIN
73 pthread_setname_np(pthread_self(), name);
74 #elif DETECT_OS_FREEBSD || DETECT_OS_OPENBSD
75 pthread_set_name_np(pthread_self(), name);
76 #elif DETECT_OS_NETBSD
77 pthread_setname_np(pthread_self(), "%s", (void *)name);
78 #elif DETECT_OS_APPLE
79 pthread_setname_np(name);
80 #else
81 #error Not sure how to call pthread_setname_np
82 #endif
83 #endif
84 (void)name;
85 }
86
87 /**
88 * An AMD Zen CPU consists of multiple modules where each module has its own L3
89 * cache. Inter-thread communication such as locks and atomics between modules
90 * is very expensive. It's desirable to pin a group of closely cooperating
91 * threads to one group of cores sharing L3.
92 *
93 * \param thread thread
94 * \param L3_index index of the L3 cache
95 * \param cores_per_L3 number of CPU cores shared by one L3
96 */
97 static inline void
98 util_pin_thread_to_L3(thrd_t thread, unsigned L3_index, unsigned cores_per_L3)
99 {
100 #if defined(HAVE_PTHREAD_SETAFFINITY)
101 cpu_set_t cpuset;
102
103 CPU_ZERO(&cpuset);
104 for (unsigned i = 0; i < cores_per_L3; i++)
105 CPU_SET(L3_index * cores_per_L3 + i, &cpuset);
106 pthread_setaffinity_np(thread, sizeof(cpuset), &cpuset);
107 #endif
108 }
109
110 /**
111 * Return the index of L3 that the thread is pinned to. If the thread is
112 * pinned to multiple L3 caches, return -1.
113 *
114 * \param thread thread
115 * \param cores_per_L3 number of CPU cores shared by one L3
116 */
117 static inline int
118 util_get_L3_for_pinned_thread(thrd_t thread, unsigned cores_per_L3)
119 {
120 #if defined(HAVE_PTHREAD_SETAFFINITY)
121 cpu_set_t cpuset;
122
123 if (pthread_getaffinity_np(thread, sizeof(cpuset), &cpuset) == 0) {
124 int L3_index = -1;
125
126 for (unsigned i = 0; i < CPU_SETSIZE; i++) {
127 if (CPU_ISSET(i, &cpuset)) {
128 int x = i / cores_per_L3;
129
130 if (L3_index != x) {
131 if (L3_index == -1)
132 L3_index = x;
133 else
134 return -1; /* multiple L3s are set */
135 }
136 }
137 }
138 return L3_index;
139 }
140 #endif
141 return -1;
142 }
143
144 /*
145 * Thread statistics.
146 */
147
148 /* Return the time of a thread's CPU time clock. */
149 static inline int64_t
150 u_thread_get_time_nano(thrd_t thread)
151 {
152 #if defined(__linux__) && defined(HAVE_PTHREAD)
153 struct timespec ts;
154 clockid_t cid;
155
156 pthread_getcpuclockid(thread, &cid);
157 clock_gettime(cid, &ts);
158 return (int64_t)ts.tv_sec * 1000000000 + ts.tv_nsec;
159 #else
160 return 0;
161 #endif
162 }
163
164 static inline bool u_thread_is_self(thrd_t thread)
165 {
166 #if defined(HAVE_PTHREAD)
167 return pthread_equal(pthread_self(), thread);
168 #endif
169 return false;
170 }
171
172 /*
173 * util_barrier
174 */
175
176 #if defined(HAVE_PTHREAD) && !defined(__APPLE__)
177
178 typedef pthread_barrier_t util_barrier;
179
180 static inline void util_barrier_init(util_barrier *barrier, unsigned count)
181 {
182 pthread_barrier_init(barrier, NULL, count);
183 }
184
185 static inline void util_barrier_destroy(util_barrier *barrier)
186 {
187 pthread_barrier_destroy(barrier);
188 }
189
190 static inline void util_barrier_wait(util_barrier *barrier)
191 {
192 pthread_barrier_wait(barrier);
193 }
194
195
196 #else /* If the OS doesn't have its own, implement barriers using a mutex and a condvar */
197
198 typedef struct {
199 unsigned count;
200 unsigned waiters;
201 uint64_t sequence;
202 mtx_t mutex;
203 cnd_t condvar;
204 } util_barrier;
205
206 static inline void util_barrier_init(util_barrier *barrier, unsigned count)
207 {
208 barrier->count = count;
209 barrier->waiters = 0;
210 barrier->sequence = 0;
211 (void) mtx_init(&barrier->mutex, mtx_plain);
212 cnd_init(&barrier->condvar);
213 }
214
215 static inline void util_barrier_destroy(util_barrier *barrier)
216 {
217 assert(barrier->waiters == 0);
218 mtx_destroy(&barrier->mutex);
219 cnd_destroy(&barrier->condvar);
220 }
221
222 static inline void util_barrier_wait(util_barrier *barrier)
223 {
224 mtx_lock(&barrier->mutex);
225
226 assert(barrier->waiters < barrier->count);
227 barrier->waiters++;
228
229 if (barrier->waiters < barrier->count) {
230 uint64_t sequence = barrier->sequence;
231
232 do {
233 cnd_wait(&barrier->condvar, &barrier->mutex);
234 } while (sequence == barrier->sequence);
235 } else {
236 barrier->waiters = 0;
237 barrier->sequence++;
238 cnd_broadcast(&barrier->condvar);
239 }
240
241 mtx_unlock(&barrier->mutex);
242 }
243
244 #endif
245
246 #endif /* U_THREAD_H_ */