| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /*********************************************************************************/ | ||
| 2 | /* Copyright 2009-2024 Barcelona Supercomputing Center */ | ||
| 3 | /* */ | ||
| 4 | /* This file is part of the DLB library. */ | ||
| 5 | /* */ | ||
| 6 | /* DLB is free software: you can redistribute it and/or modify */ | ||
| 7 | /* it under the terms of the GNU Lesser General Public License as published by */ | ||
| 8 | /* the Free Software Foundation, either version 3 of the License, or */ | ||
| 9 | /* (at your option) any later version. */ | ||
| 10 | /* */ | ||
| 11 | /* DLB is distributed in the hope that it will be useful, */ | ||
| 12 | /* but WITHOUT ANY WARRANTY; without even the implied warranty of */ | ||
| 13 | /* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the */ | ||
| 14 | /* GNU Lesser General Public License for more details. */ | ||
| 15 | /* */ | ||
| 16 | /* You should have received a copy of the GNU Lesser General Public License */ | ||
| 17 | /* along with DLB. If not, see <https://www.gnu.org/licenses/>. */ | ||
| 18 | /*********************************************************************************/ | ||
| 19 | |||
| 20 | #include "LB_numThreads/numThreads.h" | ||
| 21 | |||
| 22 | #include "LB_core/spd.h" | ||
| 23 | #include "apis/dlb_errors.h" | ||
| 24 | #include "support/tracing.h" | ||
| 25 | #include "support/debug.h" | ||
| 26 | #include "support/mask_utils.h" | ||
| 27 | |||
| 28 | #include <sched.h> | ||
| 29 | #include <pthread.h> | ||
| 30 | |||
| 31 | #define OMP_SYMBOLS_DEFINED ( \ | ||
| 32 | omp_set_num_threads \ | ||
| 33 | ) | ||
| 34 | |||
| 35 | /* Weak symbols */ | ||
| 36 | void omp_set_num_threads(int nthreads) __attribute__((weak)); | ||
| 37 | int omp_get_max_threads(void) __attribute__((weak)); | ||
| 38 | |||
| 39 | ✗ | static void packed_omp_set_num_threads(int nthreads, void *arg) { | |
| 40 | ✗ | omp_set_num_threads(nthreads); | |
| 41 | } | ||
| 42 | |||
| 43 | |||
| 44 | 100 | void pm_init(pm_interface_t *pm) { | |
| 45 | |||
| 46 | 100 | *pm = (const pm_interface_t) {}; | |
| 47 | |||
| 48 | /* OpenMP */ | ||
| 49 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 96 times.
|
100 | if (OMP_SYMBOLS_DEFINED) { |
| 50 | 4 | pm->dlb_callback_set_num_threads_ptr = packed_omp_set_num_threads; | |
| 51 | } | ||
| 52 | /* Undefined */ | ||
| 53 | else {} | ||
| 54 | 100 | } | |
| 55 | |||
| 56 | 90 | void pm_finalize(pm_interface_t *pm) { | |
| 57 | /* Reset all fields */ | ||
| 58 | 90 | *pm = (const pm_interface_t) {}; | |
| 59 | 90 | } | |
| 60 | |||
| 61 | 28 | int pm_get_num_threads(void) { | |
| 62 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 28 times.
|
28 | return omp_get_max_threads ? omp_get_max_threads() : 1; |
| 63 | } | ||
| 64 | |||
| 65 | 139 | int pm_callback_set(pm_interface_t *pm, dlb_callbacks_t which, | |
| 66 | dlb_callback_t callback, void *arg) { | ||
| 67 |
10/10✓ Branch 0 taken 39 times.
✓ Branch 1 taken 1 times.
✓ Branch 2 taken 12 times.
✓ Branch 3 taken 1 times.
✓ Branch 4 taken 1 times.
✓ Branch 5 taken 42 times.
✓ Branch 6 taken 34 times.
✓ Branch 7 taken 4 times.
✓ Branch 8 taken 4 times.
✓ Branch 9 taken 1 times.
|
139 | switch(which) { |
| 68 | 39 | case dlb_callback_set_num_threads: | |
| 69 | 39 | pm->dlb_callback_set_num_threads_ptr = (dlb_callback_set_num_threads_t)callback; | |
| 70 | 39 | pm->dlb_callback_set_num_threads_arg = arg; | |
| 71 | 39 | break; | |
| 72 | 1 | case dlb_callback_set_active_mask: | |
| 73 | 1 | pm->dlb_callback_set_active_mask_ptr = (dlb_callback_set_active_mask_t)callback; | |
| 74 | 1 | pm->dlb_callback_set_active_mask_arg = arg; | |
| 75 | 1 | break; | |
| 76 | 12 | case dlb_callback_set_process_mask: | |
| 77 | 12 | pm->dlb_callback_set_process_mask_ptr = (dlb_callback_set_process_mask_t)callback; | |
| 78 | 12 | pm->dlb_callback_set_process_mask_arg = arg; | |
| 79 | 12 | break; | |
| 80 | 1 | case dlb_callback_add_active_mask: | |
| 81 | 1 | pm->dlb_callback_add_active_mask_ptr = (dlb_callback_add_active_mask_t)callback; | |
| 82 | 1 | pm->dlb_callback_add_active_mask_arg = arg; | |
| 83 | 1 | break; | |
| 84 | 1 | case dlb_callback_add_process_mask: | |
| 85 | 1 | pm->dlb_callback_add_process_mask_ptr = (dlb_callback_add_process_mask_t)callback; | |
| 86 | 1 | pm->dlb_callback_add_process_mask_arg = arg; | |
| 87 | 1 | break; | |
| 88 | 42 | case dlb_callback_enable_cpu: | |
| 89 | 42 | pm->dlb_callback_enable_cpu_ptr = (dlb_callback_enable_cpu_t)callback; | |
| 90 | 42 | pm->dlb_callback_enable_cpu_arg = arg; | |
| 91 | 42 | break; | |
| 92 | 34 | case dlb_callback_disable_cpu: | |
| 93 | 34 | pm->dlb_callback_disable_cpu_ptr = (dlb_callback_disable_cpu_t)callback; | |
| 94 | 34 | pm->dlb_callback_disable_cpu_arg = arg; | |
| 95 | 34 | break; | |
| 96 | 4 | case dlb_callback_enable_cpu_set: | |
| 97 | 4 | pm->dlb_callback_enable_cpu_set_ptr = (dlb_callback_enable_cpu_set_t)callback; | |
| 98 | 4 | pm->dlb_callback_enable_cpu_set_arg = arg; | |
| 99 | 4 | break; | |
| 100 | 4 | case dlb_callback_disable_cpu_set: | |
| 101 | 4 | pm->dlb_callback_disable_cpu_set_ptr = (dlb_callback_disable_cpu_set_t)callback; | |
| 102 | 4 | pm->dlb_callback_disable_cpu_set_arg = arg; | |
| 103 | 4 | break; | |
| 104 | 1 | default: | |
| 105 | 1 | return DLB_ERR_NOCBK; | |
| 106 | } | ||
| 107 | 138 | return DLB_SUCCESS; | |
| 108 | } | ||
| 109 | |||
| 110 | 12 | int pm_callback_get(const pm_interface_t *pm, dlb_callbacks_t which, | |
| 111 | dlb_callback_t *callback, void **arg) { | ||
| 112 |
10/10✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
✓ Branch 2 taken 1 times.
✓ Branch 3 taken 1 times.
✓ Branch 4 taken 1 times.
✓ Branch 5 taken 3 times.
✓ Branch 6 taken 1 times.
✓ Branch 7 taken 1 times.
✓ Branch 8 taken 1 times.
✓ Branch 9 taken 1 times.
|
12 | switch(which) { |
| 113 | 1 | case dlb_callback_set_num_threads: | |
| 114 | 1 | *callback = (dlb_callback_t)pm->dlb_callback_set_num_threads_ptr; | |
| 115 | 1 | *arg = pm->dlb_callback_set_num_threads_arg; | |
| 116 | 1 | break; | |
| 117 | 1 | case dlb_callback_set_active_mask: | |
| 118 | 1 | *callback = (dlb_callback_t)pm->dlb_callback_set_active_mask_ptr; | |
| 119 | 1 | *arg = pm->dlb_callback_set_active_mask_arg; | |
| 120 | 1 | break; | |
| 121 | 1 | case dlb_callback_set_process_mask: | |
| 122 | 1 | *callback = (dlb_callback_t)pm->dlb_callback_set_process_mask_ptr; | |
| 123 | 1 | *arg = pm->dlb_callback_set_process_mask_arg; | |
| 124 | 1 | break; | |
| 125 | 1 | case dlb_callback_add_active_mask: | |
| 126 | 1 | *callback = (dlb_callback_t)pm->dlb_callback_add_active_mask_ptr; | |
| 127 | 1 | *arg = pm->dlb_callback_add_active_mask_arg; | |
| 128 | 1 | break; | |
| 129 | 1 | case dlb_callback_add_process_mask: | |
| 130 | 1 | *callback = (dlb_callback_t)pm->dlb_callback_add_process_mask_ptr; | |
| 131 | 1 | *arg = pm->dlb_callback_add_process_mask_arg; | |
| 132 | 1 | break; | |
| 133 | 3 | case dlb_callback_enable_cpu: | |
| 134 | 3 | *callback = (dlb_callback_t)pm->dlb_callback_enable_cpu_ptr; | |
| 135 | 3 | *arg = pm->dlb_callback_enable_cpu_arg; | |
| 136 | 3 | break; | |
| 137 | 1 | case dlb_callback_disable_cpu: | |
| 138 | 1 | *callback = (dlb_callback_t)pm->dlb_callback_disable_cpu_ptr; | |
| 139 | 1 | *arg = pm->dlb_callback_disable_cpu_arg; | |
| 140 | 1 | break; | |
| 141 | 1 | case dlb_callback_enable_cpu_set: | |
| 142 | 1 | *callback = (dlb_callback_t)pm->dlb_callback_enable_cpu_set_ptr; | |
| 143 | 1 | *arg = pm->dlb_callback_enable_cpu_set_arg; | |
| 144 | 1 | break; | |
| 145 | 1 | case dlb_callback_disable_cpu_set: | |
| 146 | 1 | *callback = (dlb_callback_t)pm->dlb_callback_disable_cpu_set_ptr; | |
| 147 | 1 | *arg = pm->dlb_callback_disable_cpu_set_arg; | |
| 148 | 1 | break; | |
| 149 | 1 | default: | |
| 150 | 1 | return DLB_ERR_NOCBK; | |
| 151 | } | ||
| 152 | 11 | return DLB_SUCCESS; | |
| 153 | } | ||
| 154 | |||
| 155 | 157 | int update_threads(const pm_interface_t *pm, int threads) { | |
| 156 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 145 times.
|
157 | if (pm->dlb_callback_set_num_threads_ptr == NULL) { |
| 157 | 12 | return DLB_ERR_NOCBK; | |
| 158 | } | ||
| 159 | |||
| 160 | 145 | int cpus_node = mu_get_system_size(); | |
| 161 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 144 times.
|
145 | if (threads > cpus_node) { |
| 162 | 1 | warning("Trying to use more CPUS (%d) than available (%d)", threads, cpus_node); | |
| 163 | 1 | threads = cpus_node; | |
| 164 |
2/2✓ Branch 0 taken 26 times.
✓ Branch 1 taken 118 times.
|
144 | } else if (threads < 1) { |
| 165 | 26 | threads = 1; | |
| 166 | } | ||
| 167 | |||
| 168 | add_event(THREADS_USED_EVENT, threads); | ||
| 169 | |||
| 170 | instrument_event(CALLBACK_EVENT, dlb_callback_set_num_threads, EVENT_BEGIN); | ||
| 171 | 145 | pm->dlb_callback_set_num_threads_ptr(threads, pm->dlb_callback_set_num_threads_arg); | |
| 172 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 173 | 145 | return DLB_SUCCESS; | |
| 174 | } | ||
| 175 | |||
| 176 | 2 | int set_mask(const pm_interface_t *pm, const cpu_set_t *cpu_set) { | |
| 177 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
|
2 | if (pm->dlb_callback_set_active_mask_ptr == NULL) { |
| 178 | 1 | return DLB_ERR_NOCBK; | |
| 179 | } | ||
| 180 | instrument_event(CALLBACK_EVENT, dlb_callback_set_active_mask, EVENT_BEGIN); | ||
| 181 | 1 | pm->dlb_callback_set_active_mask_ptr(cpu_set, pm->dlb_callback_set_active_mask_arg); | |
| 182 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 183 | 1 | return DLB_SUCCESS; | |
| 184 | } | ||
| 185 | |||
| 186 | 13 | int set_process_mask(const pm_interface_t *pm, const cpu_set_t *cpu_set) { | |
| 187 |
2/2✓ Branch 0 taken 11 times.
✓ Branch 1 taken 2 times.
|
13 | if (pm->dlb_callback_set_process_mask_ptr == NULL) { |
| 188 | 11 | return DLB_ERR_NOCBK; | |
| 189 | } | ||
| 190 | instrument_event(CALLBACK_EVENT, dlb_callback_set_process_mask, EVENT_BEGIN); | ||
| 191 | 2 | pm->dlb_callback_set_process_mask_ptr(cpu_set, pm->dlb_callback_set_process_mask_arg); | |
| 192 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 193 | 2 | return DLB_SUCCESS; | |
| 194 | } | ||
| 195 | |||
| 196 | 2 | int add_mask(const pm_interface_t *pm, const cpu_set_t *cpu_set) { | |
| 197 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
|
2 | if (pm->dlb_callback_add_active_mask_ptr == NULL) { |
| 198 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
|
1 | if (pm->dlb_callback_enable_cpu_ptr != NULL) { |
| 199 | /* fallback to enable_cpu */ | ||
| 200 | ✗ | for (int cpuid = mu_get_first_cpu(cpu_set); | |
| 201 | ✗ | cpuid >= 0; | |
| 202 | ✗ | cpuid = mu_get_next_cpu(cpu_set, cpuid)) { | |
| 203 | instrument_event(CALLBACK_EVENT, dlb_callback_enable_cpu, EVENT_BEGIN); | ||
| 204 | ✗ | pm->dlb_callback_enable_cpu_ptr(cpuid, pm->dlb_callback_enable_cpu_arg); | |
| 205 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 206 | } | ||
| 207 | ✗ | return DLB_SUCCESS; | |
| 208 | } | ||
| 209 | 1 | return DLB_ERR_NOCBK; | |
| 210 | } | ||
| 211 | instrument_event(CALLBACK_EVENT, dlb_callback_add_active_mask, EVENT_BEGIN); | ||
| 212 | 1 | pm->dlb_callback_add_active_mask_ptr(cpu_set, pm->dlb_callback_add_active_mask_arg); | |
| 213 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 214 | 1 | return DLB_SUCCESS; | |
| 215 | } | ||
| 216 | |||
| 217 | 2 | int add_process_mask(const pm_interface_t *pm, const cpu_set_t *cpu_set) { | |
| 218 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
|
2 | if (pm->dlb_callback_add_process_mask_ptr == NULL) { |
| 219 | 1 | return DLB_ERR_NOCBK; | |
| 220 | } | ||
| 221 | instrument_event(CALLBACK_EVENT, dlb_callback_add_process_mask, EVENT_BEGIN); | ||
| 222 | 1 | pm->dlb_callback_add_process_mask_ptr(cpu_set, pm->dlb_callback_add_process_mask_arg); | |
| 223 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 224 | 1 | return DLB_SUCCESS; | |
| 225 | } | ||
| 226 | |||
| 227 | 210 | int enable_cpu(const pm_interface_t *pm, int cpuid) { | |
| 228 | /* fallback case */ | ||
| 229 |
2/2✓ Branch 0 taken 3 times.
✓ Branch 1 taken 207 times.
|
210 | if (pm->dlb_callback_enable_cpu_ptr == NULL |
| 230 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 3 times.
|
3 | && pm->dlb_callback_add_active_mask_ptr) { |
| 231 | cpu_set_t cpu_set; | ||
| 232 | ✗ | CPU_ZERO(&cpu_set); | |
| 233 | ✗ | CPU_SET(cpuid, &cpu_set); | |
| 234 | ✗ | return enable_cpu_set(pm, &cpu_set); | |
| 235 | } | ||
| 236 | |||
| 237 |
2/2✓ Branch 0 taken 3 times.
✓ Branch 1 taken 207 times.
|
210 | if (pm->dlb_callback_enable_cpu_ptr == NULL) { |
| 238 | 3 | return DLB_ERR_NOCBK; | |
| 239 | } | ||
| 240 | instrument_event(CALLBACK_EVENT, dlb_callback_enable_cpu, EVENT_BEGIN); | ||
| 241 | 207 | pm->dlb_callback_enable_cpu_ptr(cpuid, pm->dlb_callback_enable_cpu_arg); | |
| 242 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 243 | 207 | return DLB_SUCCESS; | |
| 244 | } | ||
| 245 | |||
| 246 | 56 | int enable_cpu_set(const pm_interface_t *pm, const cpu_set_t *cpu_set) { | |
| 247 |
2/2✓ Branch 0 taken 52 times.
✓ Branch 1 taken 4 times.
|
56 | if (pm->dlb_callback_enable_cpu_set_ptr == NULL) { |
| 248 |
1/2✓ Branch 0 taken 52 times.
✗ Branch 1 not taken.
|
52 | if (pm->dlb_callback_enable_cpu_ptr != NULL) { |
| 249 | /* fallback to enable_cpu */ | ||
| 250 | 52 | for (int cpuid = mu_get_first_cpu(cpu_set); | |
| 251 |
2/2✓ Branch 0 taken 139 times.
✓ Branch 1 taken 52 times.
|
191 | cpuid >= 0; |
| 252 | 139 | cpuid = mu_get_next_cpu(cpu_set, cpuid)) { | |
| 253 | instrument_event(CALLBACK_EVENT, dlb_callback_enable_cpu, EVENT_BEGIN); | ||
| 254 | 139 | pm->dlb_callback_enable_cpu_ptr(cpuid, pm->dlb_callback_enable_cpu_arg); | |
| 255 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 256 | } | ||
| 257 | 52 | return DLB_SUCCESS; | |
| 258 | } | ||
| 259 | ✗ | return DLB_ERR_NOCBK; | |
| 260 | } | ||
| 261 | instrument_event(CALLBACK_EVENT, dlb_callback_enable_cpu_set, EVENT_BEGIN); | ||
| 262 | 4 | pm->dlb_callback_enable_cpu_set_ptr(cpu_set, pm->dlb_callback_enable_cpu_set_arg); | |
| 263 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 264 | 4 | return DLB_SUCCESS; | |
| 265 | } | ||
| 266 | |||
| 267 | 31 | int disable_cpu(const pm_interface_t *pm, int cpuid) { | |
| 268 | /* fallback case */ | ||
| 269 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 30 times.
|
31 | if (pm->dlb_callback_disable_cpu_ptr == NULL |
| 270 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
|
1 | && pm->dlb_callback_set_active_mask_ptr) { |
| 271 | cpu_set_t cpu_set; | ||
| 272 | ✗ | sched_getaffinity(0, sizeof(cpu_set_t), &cpu_set); | |
| 273 | ✗ | CPU_CLR(cpuid, &cpu_set); | |
| 274 | ✗ | return set_mask(pm, &cpu_set); | |
| 275 | } | ||
| 276 | |||
| 277 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 30 times.
|
31 | if (pm->dlb_callback_disable_cpu_ptr == NULL) { |
| 278 | 1 | return DLB_ERR_NOCBK; | |
| 279 | } | ||
| 280 | instrument_event(CALLBACK_EVENT, dlb_callback_disable_cpu, EVENT_BEGIN); | ||
| 281 | 30 | pm->dlb_callback_disable_cpu_ptr(cpuid, pm->dlb_callback_disable_cpu_arg); | |
| 282 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 283 | 30 | return DLB_SUCCESS; | |
| 284 | } | ||
| 285 | |||
| 286 | 15 | int disable_cpu_set(const pm_interface_t *pm, const cpu_set_t *cpu_set) { | |
| 287 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 3 times.
|
15 | if (pm->dlb_callback_disable_cpu_set_ptr == NULL) { |
| 288 |
1/2✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
|
12 | if (pm->dlb_callback_disable_cpu_ptr != NULL) { |
| 289 | /* fallback to disable_cpu */ | ||
| 290 | 12 | for (int cpuid = mu_get_first_cpu(cpu_set); | |
| 291 |
2/2✓ Branch 0 taken 29 times.
✓ Branch 1 taken 12 times.
|
41 | cpuid >= 0; |
| 292 | 29 | cpuid = mu_get_next_cpu(cpu_set, cpuid)) { | |
| 293 | instrument_event(CALLBACK_EVENT, dlb_callback_disable_cpu, EVENT_BEGIN); | ||
| 294 | 29 | pm->dlb_callback_disable_cpu_ptr(cpuid, pm->dlb_callback_disable_cpu_arg); | |
| 295 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 296 | } | ||
| 297 | 12 | return DLB_SUCCESS; | |
| 298 | } | ||
| 299 | ✗ | return DLB_ERR_NOCBK; | |
| 300 | } | ||
| 301 | instrument_event(CALLBACK_EVENT, dlb_callback_disable_cpu_set, EVENT_BEGIN); | ||
| 302 | 3 | pm->dlb_callback_disable_cpu_set_ptr(cpu_set, pm->dlb_callback_disable_cpu_set_arg); | |
| 303 | instrument_event(CALLBACK_EVENT, 0, EVENT_END); | ||
| 304 | 3 | return DLB_SUCCESS; | |
| 305 | } | ||
| 306 |