Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 /* Copyright (c) 2020 Facebook */
0003 #define _GNU_SOURCE
0004 #include <argp.h>
0005 #include <linux/compiler.h>
0006 #include <sys/time.h>
0007 #include <sched.h>
0008 #include <fcntl.h>
0009 #include <pthread.h>
0010 #include <sys/sysinfo.h>
0011 #include <signal.h>
0012 #include "bench.h"
0013 #include "testing_helpers.h"
0014 
0015 struct env env = {
0016     .warmup_sec = 1,
0017     .duration_sec = 5,
0018     .affinity = false,
0019     .consumer_cnt = 1,
0020     .producer_cnt = 1,
0021 };
0022 
0023 static int libbpf_print_fn(enum libbpf_print_level level,
0024             const char *format, va_list args)
0025 {
0026     if (level == LIBBPF_DEBUG && !env.verbose)
0027         return 0;
0028     return vfprintf(stderr, format, args);
0029 }
0030 
0031 void setup_libbpf(void)
0032 {
0033     libbpf_set_strict_mode(LIBBPF_STRICT_ALL);
0034     libbpf_set_print(libbpf_print_fn);
0035 }
0036 
0037 void false_hits_report_progress(int iter, struct bench_res *res, long delta_ns)
0038 {
0039     long total = res->false_hits  + res->hits + res->drops;
0040 
0041     printf("Iter %3d (%7.3lfus): ",
0042            iter, (delta_ns - 1000000000) / 1000.0);
0043 
0044     printf("%ld false hits of %ld total operations. Percentage = %2.2f %%\n",
0045            res->false_hits, total, ((float)res->false_hits / total) * 100);
0046 }
0047 
0048 void false_hits_report_final(struct bench_res res[], int res_cnt)
0049 {
0050     long total_hits = 0, total_drops = 0, total_false_hits = 0, total_ops = 0;
0051     int i;
0052 
0053     for (i = 0; i < res_cnt; i++) {
0054         total_hits += res[i].hits;
0055         total_false_hits += res[i].false_hits;
0056         total_drops += res[i].drops;
0057     }
0058     total_ops = total_hits + total_false_hits + total_drops;
0059 
0060     printf("Summary: %ld false hits of %ld total operations. ",
0061            total_false_hits, total_ops);
0062     printf("Percentage =  %2.2f %%\n",
0063            ((float)total_false_hits / total_ops) * 100);
0064 }
0065 
0066 void hits_drops_report_progress(int iter, struct bench_res *res, long delta_ns)
0067 {
0068     double hits_per_sec, drops_per_sec;
0069     double hits_per_prod;
0070 
0071     hits_per_sec = res->hits / 1000000.0 / (delta_ns / 1000000000.0);
0072     hits_per_prod = hits_per_sec / env.producer_cnt;
0073     drops_per_sec = res->drops / 1000000.0 / (delta_ns / 1000000000.0);
0074 
0075     printf("Iter %3d (%7.3lfus): ",
0076            iter, (delta_ns - 1000000000) / 1000.0);
0077 
0078     printf("hits %8.3lfM/s (%7.3lfM/prod), drops %8.3lfM/s, total operations %8.3lfM/s\n",
0079            hits_per_sec, hits_per_prod, drops_per_sec, hits_per_sec + drops_per_sec);
0080 }
0081 
0082 void
0083 grace_period_latency_basic_stats(struct bench_res res[], int res_cnt, struct basic_stats *gp_stat)
0084 {
0085     int i;
0086 
0087     memset(gp_stat, 0, sizeof(struct basic_stats));
0088 
0089     for (i = 0; i < res_cnt; i++)
0090         gp_stat->mean += res[i].gp_ns / 1000.0 / (double)res[i].gp_ct / (0.0 + res_cnt);
0091 
0092 #define IT_MEAN_DIFF (res[i].gp_ns / 1000.0 / (double)res[i].gp_ct - gp_stat->mean)
0093     if (res_cnt > 1) {
0094         for (i = 0; i < res_cnt; i++)
0095             gp_stat->stddev += (IT_MEAN_DIFF * IT_MEAN_DIFF) / (res_cnt - 1.0);
0096     }
0097     gp_stat->stddev = sqrt(gp_stat->stddev);
0098 #undef IT_MEAN_DIFF
0099 }
0100 
0101 void
0102 grace_period_ticks_basic_stats(struct bench_res res[], int res_cnt, struct basic_stats *gp_stat)
0103 {
0104     int i;
0105 
0106     memset(gp_stat, 0, sizeof(struct basic_stats));
0107     for (i = 0; i < res_cnt; i++)
0108         gp_stat->mean += res[i].stime / (double)res[i].gp_ct / (0.0 + res_cnt);
0109 
0110 #define IT_MEAN_DIFF (res[i].stime / (double)res[i].gp_ct - gp_stat->mean)
0111     if (res_cnt > 1) {
0112         for (i = 0; i < res_cnt; i++)
0113             gp_stat->stddev += (IT_MEAN_DIFF * IT_MEAN_DIFF) / (res_cnt - 1.0);
0114     }
0115     gp_stat->stddev = sqrt(gp_stat->stddev);
0116 #undef IT_MEAN_DIFF
0117 }
0118 
0119 void hits_drops_report_final(struct bench_res res[], int res_cnt)
0120 {
0121     int i;
0122     double hits_mean = 0.0, drops_mean = 0.0, total_ops_mean = 0.0;
0123     double hits_stddev = 0.0, drops_stddev = 0.0, total_ops_stddev = 0.0;
0124     double total_ops;
0125 
0126     for (i = 0; i < res_cnt; i++) {
0127         hits_mean += res[i].hits / 1000000.0 / (0.0 + res_cnt);
0128         drops_mean += res[i].drops / 1000000.0 / (0.0 + res_cnt);
0129     }
0130     total_ops_mean = hits_mean + drops_mean;
0131 
0132     if (res_cnt > 1)  {
0133         for (i = 0; i < res_cnt; i++) {
0134             hits_stddev += (hits_mean - res[i].hits / 1000000.0) *
0135                        (hits_mean - res[i].hits / 1000000.0) /
0136                        (res_cnt - 1.0);
0137             drops_stddev += (drops_mean - res[i].drops / 1000000.0) *
0138                     (drops_mean - res[i].drops / 1000000.0) /
0139                     (res_cnt - 1.0);
0140             total_ops = res[i].hits + res[i].drops;
0141             total_ops_stddev += (total_ops_mean - total_ops / 1000000.0) *
0142                     (total_ops_mean - total_ops / 1000000.0) /
0143                     (res_cnt - 1.0);
0144         }
0145         hits_stddev = sqrt(hits_stddev);
0146         drops_stddev = sqrt(drops_stddev);
0147         total_ops_stddev = sqrt(total_ops_stddev);
0148     }
0149     printf("Summary: hits %8.3lf \u00B1 %5.3lfM/s (%7.3lfM/prod), ",
0150            hits_mean, hits_stddev, hits_mean / env.producer_cnt);
0151     printf("drops %8.3lf \u00B1 %5.3lfM/s, ",
0152            drops_mean, drops_stddev);
0153     printf("total operations %8.3lf \u00B1 %5.3lfM/s\n",
0154            total_ops_mean, total_ops_stddev);
0155 }
0156 
0157 void ops_report_progress(int iter, struct bench_res *res, long delta_ns)
0158 {
0159     double hits_per_sec, hits_per_prod;
0160 
0161     hits_per_sec = res->hits / 1000000.0 / (delta_ns / 1000000000.0);
0162     hits_per_prod = hits_per_sec / env.producer_cnt;
0163 
0164     printf("Iter %3d (%7.3lfus): ", iter, (delta_ns - 1000000000) / 1000.0);
0165 
0166     printf("hits %8.3lfM/s (%7.3lfM/prod)\n", hits_per_sec, hits_per_prod);
0167 }
0168 
0169 void ops_report_final(struct bench_res res[], int res_cnt)
0170 {
0171     double hits_mean = 0.0, hits_stddev = 0.0;
0172     int i;
0173 
0174     for (i = 0; i < res_cnt; i++)
0175         hits_mean += res[i].hits / 1000000.0 / (0.0 + res_cnt);
0176 
0177     if (res_cnt > 1)  {
0178         for (i = 0; i < res_cnt; i++)
0179             hits_stddev += (hits_mean - res[i].hits / 1000000.0) *
0180                        (hits_mean - res[i].hits / 1000000.0) /
0181                        (res_cnt - 1.0);
0182 
0183         hits_stddev = sqrt(hits_stddev);
0184     }
0185     printf("Summary: throughput %8.3lf \u00B1 %5.3lf M ops/s (%7.3lfM ops/prod), ",
0186            hits_mean, hits_stddev, hits_mean / env.producer_cnt);
0187     printf("latency %8.3lf ns/op\n", 1000.0 / hits_mean * env.producer_cnt);
0188 }
0189 
0190 void local_storage_report_progress(int iter, struct bench_res *res,
0191                    long delta_ns)
0192 {
0193     double important_hits_per_sec, hits_per_sec;
0194     double delta_sec = delta_ns / 1000000000.0;
0195 
0196     hits_per_sec = res->hits / 1000000.0 / delta_sec;
0197     important_hits_per_sec = res->important_hits / 1000000.0 / delta_sec;
0198 
0199     printf("Iter %3d (%7.3lfus): ", iter, (delta_ns - 1000000000) / 1000.0);
0200 
0201     printf("hits %8.3lfM/s ", hits_per_sec);
0202     printf("important_hits %8.3lfM/s\n", important_hits_per_sec);
0203 }
0204 
0205 void local_storage_report_final(struct bench_res res[], int res_cnt)
0206 {
0207     double important_hits_mean = 0.0, important_hits_stddev = 0.0;
0208     double hits_mean = 0.0, hits_stddev = 0.0;
0209     int i;
0210 
0211     for (i = 0; i < res_cnt; i++) {
0212         hits_mean += res[i].hits / 1000000.0 / (0.0 + res_cnt);
0213         important_hits_mean += res[i].important_hits / 1000000.0 / (0.0 + res_cnt);
0214     }
0215 
0216     if (res_cnt > 1)  {
0217         for (i = 0; i < res_cnt; i++) {
0218             hits_stddev += (hits_mean - res[i].hits / 1000000.0) *
0219                        (hits_mean - res[i].hits / 1000000.0) /
0220                        (res_cnt - 1.0);
0221             important_hits_stddev +=
0222                        (important_hits_mean - res[i].important_hits / 1000000.0) *
0223                        (important_hits_mean - res[i].important_hits / 1000000.0) /
0224                        (res_cnt - 1.0);
0225         }
0226 
0227         hits_stddev = sqrt(hits_stddev);
0228         important_hits_stddev = sqrt(important_hits_stddev);
0229     }
0230     printf("Summary: hits throughput %8.3lf \u00B1 %5.3lf M ops/s, ",
0231            hits_mean, hits_stddev);
0232     printf("hits latency %8.3lf ns/op, ", 1000.0 / hits_mean);
0233     printf("important_hits throughput %8.3lf \u00B1 %5.3lf M ops/s\n",
0234            important_hits_mean, important_hits_stddev);
0235 }
0236 
0237 const char *argp_program_version = "benchmark";
0238 const char *argp_program_bug_address = "<bpf@vger.kernel.org>";
0239 const char argp_program_doc[] =
0240 "benchmark    Generic benchmarking framework.\n"
0241 "\n"
0242 "This tool runs benchmarks.\n"
0243 "\n"
0244 "USAGE: benchmark <bench-name>\n"
0245 "\n"
0246 "EXAMPLES:\n"
0247 "    # run 'count-local' benchmark with 1 producer and 1 consumer\n"
0248 "    benchmark count-local\n"
0249 "    # run 'count-local' with 16 producer and 8 consumer thread, pinned to CPUs\n"
0250 "    benchmark -p16 -c8 -a count-local\n";
0251 
0252 enum {
0253     ARG_PROD_AFFINITY_SET = 1000,
0254     ARG_CONS_AFFINITY_SET = 1001,
0255 };
0256 
0257 static const struct argp_option opts[] = {
0258     { "list", 'l', NULL, 0, "List available benchmarks"},
0259     { "duration", 'd', "SEC", 0, "Duration of benchmark, seconds"},
0260     { "warmup", 'w', "SEC", 0, "Warm-up period, seconds"},
0261     { "producers", 'p', "NUM", 0, "Number of producer threads"},
0262     { "consumers", 'c', "NUM", 0, "Number of consumer threads"},
0263     { "verbose", 'v', NULL, 0, "Verbose debug output"},
0264     { "affinity", 'a', NULL, 0, "Set consumer/producer thread affinity"},
0265     { "prod-affinity", ARG_PROD_AFFINITY_SET, "CPUSET", 0,
0266       "Set of CPUs for producer threads; implies --affinity"},
0267     { "cons-affinity", ARG_CONS_AFFINITY_SET, "CPUSET", 0,
0268       "Set of CPUs for consumer threads; implies --affinity"},
0269     {},
0270 };
0271 
0272 extern struct argp bench_ringbufs_argp;
0273 extern struct argp bench_bloom_map_argp;
0274 extern struct argp bench_bpf_loop_argp;
0275 extern struct argp bench_local_storage_argp;
0276 extern struct argp bench_local_storage_rcu_tasks_trace_argp;
0277 extern struct argp bench_strncmp_argp;
0278 
0279 static const struct argp_child bench_parsers[] = {
0280     { &bench_ringbufs_argp, 0, "Ring buffers benchmark", 0 },
0281     { &bench_bloom_map_argp, 0, "Bloom filter map benchmark", 0 },
0282     { &bench_bpf_loop_argp, 0, "bpf_loop helper benchmark", 0 },
0283     { &bench_local_storage_argp, 0, "local_storage benchmark", 0 },
0284     { &bench_strncmp_argp, 0, "bpf_strncmp helper benchmark", 0 },
0285     { &bench_local_storage_rcu_tasks_trace_argp, 0,
0286         "local_storage RCU Tasks Trace slowdown benchmark", 0 },
0287     {},
0288 };
0289 
0290 static error_t parse_arg(int key, char *arg, struct argp_state *state)
0291 {
0292     static int pos_args;
0293 
0294     switch (key) {
0295     case 'v':
0296         env.verbose = true;
0297         break;
0298     case 'l':
0299         env.list = true;
0300         break;
0301     case 'd':
0302         env.duration_sec = strtol(arg, NULL, 10);
0303         if (env.duration_sec <= 0) {
0304             fprintf(stderr, "Invalid duration: %s\n", arg);
0305             argp_usage(state);
0306         }
0307         break;
0308     case 'w':
0309         env.warmup_sec = strtol(arg, NULL, 10);
0310         if (env.warmup_sec <= 0) {
0311             fprintf(stderr, "Invalid warm-up duration: %s\n", arg);
0312             argp_usage(state);
0313         }
0314         break;
0315     case 'p':
0316         env.producer_cnt = strtol(arg, NULL, 10);
0317         if (env.producer_cnt <= 0) {
0318             fprintf(stderr, "Invalid producer count: %s\n", arg);
0319             argp_usage(state);
0320         }
0321         break;
0322     case 'c':
0323         env.consumer_cnt = strtol(arg, NULL, 10);
0324         if (env.consumer_cnt <= 0) {
0325             fprintf(stderr, "Invalid consumer count: %s\n", arg);
0326             argp_usage(state);
0327         }
0328         break;
0329     case 'a':
0330         env.affinity = true;
0331         break;
0332     case ARG_PROD_AFFINITY_SET:
0333         env.affinity = true;
0334         if (parse_num_list(arg, &env.prod_cpus.cpus,
0335                    &env.prod_cpus.cpus_len)) {
0336             fprintf(stderr, "Invalid format of CPU set for producers.");
0337             argp_usage(state);
0338         }
0339         break;
0340     case ARG_CONS_AFFINITY_SET:
0341         env.affinity = true;
0342         if (parse_num_list(arg, &env.cons_cpus.cpus,
0343                    &env.cons_cpus.cpus_len)) {
0344             fprintf(stderr, "Invalid format of CPU set for consumers.");
0345             argp_usage(state);
0346         }
0347         break;
0348     case ARGP_KEY_ARG:
0349         if (pos_args++) {
0350             fprintf(stderr,
0351                 "Unrecognized positional argument: %s\n", arg);
0352             argp_usage(state);
0353         }
0354         env.bench_name = strdup(arg);
0355         break;
0356     default:
0357         return ARGP_ERR_UNKNOWN;
0358     }
0359     return 0;
0360 }
0361 
0362 static void parse_cmdline_args(int argc, char **argv)
0363 {
0364     static const struct argp argp = {
0365         .options = opts,
0366         .parser = parse_arg,
0367         .doc = argp_program_doc,
0368         .children = bench_parsers,
0369     };
0370     if (argp_parse(&argp, argc, argv, 0, NULL, NULL))
0371         exit(1);
0372     if (!env.list && !env.bench_name) {
0373         argp_help(&argp, stderr, ARGP_HELP_DOC, "bench");
0374         exit(1);
0375     }
0376 }
0377 
0378 static void collect_measurements(long delta_ns);
0379 
0380 static __u64 last_time_ns;
0381 static void sigalarm_handler(int signo)
0382 {
0383     long new_time_ns = get_time_ns();
0384     long delta_ns = new_time_ns - last_time_ns;
0385 
0386     collect_measurements(delta_ns);
0387 
0388     last_time_ns = new_time_ns;
0389 }
0390 
0391 /* set up periodic 1-second timer */
0392 static void setup_timer()
0393 {
0394     static struct sigaction sigalarm_action = {
0395         .sa_handler = sigalarm_handler,
0396     };
0397     struct itimerval timer_settings = {};
0398     int err;
0399 
0400     last_time_ns = get_time_ns();
0401     err = sigaction(SIGALRM, &sigalarm_action, NULL);
0402     if (err < 0) {
0403         fprintf(stderr, "failed to install SIGALRM handler: %d\n", -errno);
0404         exit(1);
0405     }
0406     timer_settings.it_interval.tv_sec = 1;
0407     timer_settings.it_value.tv_sec = 1;
0408     err = setitimer(ITIMER_REAL, &timer_settings, NULL);
0409     if (err < 0) {
0410         fprintf(stderr, "failed to arm interval timer: %d\n", -errno);
0411         exit(1);
0412     }
0413 }
0414 
0415 static void set_thread_affinity(pthread_t thread, int cpu)
0416 {
0417     cpu_set_t cpuset;
0418 
0419     CPU_ZERO(&cpuset);
0420     CPU_SET(cpu, &cpuset);
0421     if (pthread_setaffinity_np(thread, sizeof(cpuset), &cpuset)) {
0422         fprintf(stderr, "setting affinity to CPU #%d failed: %d\n",
0423             cpu, errno);
0424         exit(1);
0425     }
0426 }
0427 
0428 static int next_cpu(struct cpu_set *cpu_set)
0429 {
0430     if (cpu_set->cpus) {
0431         int i;
0432 
0433         /* find next available CPU */
0434         for (i = cpu_set->next_cpu; i < cpu_set->cpus_len; i++) {
0435             if (cpu_set->cpus[i]) {
0436                 cpu_set->next_cpu = i + 1;
0437                 return i;
0438             }
0439         }
0440         fprintf(stderr, "Not enough CPUs specified, need CPU #%d or higher.\n", i);
0441         exit(1);
0442     }
0443 
0444     return cpu_set->next_cpu++;
0445 }
0446 
0447 static struct bench_state {
0448     int res_cnt;
0449     struct bench_res *results;
0450     pthread_t *consumers;
0451     pthread_t *producers;
0452 } state;
0453 
0454 const struct bench *bench = NULL;
0455 
0456 extern const struct bench bench_count_global;
0457 extern const struct bench bench_count_local;
0458 extern const struct bench bench_rename_base;
0459 extern const struct bench bench_rename_kprobe;
0460 extern const struct bench bench_rename_kretprobe;
0461 extern const struct bench bench_rename_rawtp;
0462 extern const struct bench bench_rename_fentry;
0463 extern const struct bench bench_rename_fexit;
0464 extern const struct bench bench_trig_base;
0465 extern const struct bench bench_trig_tp;
0466 extern const struct bench bench_trig_rawtp;
0467 extern const struct bench bench_trig_kprobe;
0468 extern const struct bench bench_trig_fentry;
0469 extern const struct bench bench_trig_fentry_sleep;
0470 extern const struct bench bench_trig_fmodret;
0471 extern const struct bench bench_trig_uprobe_base;
0472 extern const struct bench bench_trig_uprobe_with_nop;
0473 extern const struct bench bench_trig_uretprobe_with_nop;
0474 extern const struct bench bench_trig_uprobe_without_nop;
0475 extern const struct bench bench_trig_uretprobe_without_nop;
0476 extern const struct bench bench_rb_libbpf;
0477 extern const struct bench bench_rb_custom;
0478 extern const struct bench bench_pb_libbpf;
0479 extern const struct bench bench_pb_custom;
0480 extern const struct bench bench_bloom_lookup;
0481 extern const struct bench bench_bloom_update;
0482 extern const struct bench bench_bloom_false_positive;
0483 extern const struct bench bench_hashmap_without_bloom;
0484 extern const struct bench bench_hashmap_with_bloom;
0485 extern const struct bench bench_bpf_loop;
0486 extern const struct bench bench_strncmp_no_helper;
0487 extern const struct bench bench_strncmp_helper;
0488 extern const struct bench bench_bpf_hashmap_full_update;
0489 extern const struct bench bench_local_storage_cache_seq_get;
0490 extern const struct bench bench_local_storage_cache_interleaved_get;
0491 extern const struct bench bench_local_storage_cache_hashmap_control;
0492 extern const struct bench bench_local_storage_tasks_trace;
0493 
0494 static const struct bench *benchs[] = {
0495     &bench_count_global,
0496     &bench_count_local,
0497     &bench_rename_base,
0498     &bench_rename_kprobe,
0499     &bench_rename_kretprobe,
0500     &bench_rename_rawtp,
0501     &bench_rename_fentry,
0502     &bench_rename_fexit,
0503     &bench_trig_base,
0504     &bench_trig_tp,
0505     &bench_trig_rawtp,
0506     &bench_trig_kprobe,
0507     &bench_trig_fentry,
0508     &bench_trig_fentry_sleep,
0509     &bench_trig_fmodret,
0510     &bench_trig_uprobe_base,
0511     &bench_trig_uprobe_with_nop,
0512     &bench_trig_uretprobe_with_nop,
0513     &bench_trig_uprobe_without_nop,
0514     &bench_trig_uretprobe_without_nop,
0515     &bench_rb_libbpf,
0516     &bench_rb_custom,
0517     &bench_pb_libbpf,
0518     &bench_pb_custom,
0519     &bench_bloom_lookup,
0520     &bench_bloom_update,
0521     &bench_bloom_false_positive,
0522     &bench_hashmap_without_bloom,
0523     &bench_hashmap_with_bloom,
0524     &bench_bpf_loop,
0525     &bench_strncmp_no_helper,
0526     &bench_strncmp_helper,
0527     &bench_bpf_hashmap_full_update,
0528     &bench_local_storage_cache_seq_get,
0529     &bench_local_storage_cache_interleaved_get,
0530     &bench_local_storage_cache_hashmap_control,
0531     &bench_local_storage_tasks_trace,
0532 };
0533 
0534 static void setup_benchmark()
0535 {
0536     int i, err;
0537 
0538     if (!env.bench_name) {
0539         fprintf(stderr, "benchmark name is not specified\n");
0540         exit(1);
0541     }
0542 
0543     for (i = 0; i < ARRAY_SIZE(benchs); i++) {
0544         if (strcmp(benchs[i]->name, env.bench_name) == 0) {
0545             bench = benchs[i];
0546             break;
0547         }
0548     }
0549     if (!bench) {
0550         fprintf(stderr, "benchmark '%s' not found\n", env.bench_name);
0551         exit(1);
0552     }
0553 
0554     printf("Setting up benchmark '%s'...\n", bench->name);
0555 
0556     state.producers = calloc(env.producer_cnt, sizeof(*state.producers));
0557     state.consumers = calloc(env.consumer_cnt, sizeof(*state.consumers));
0558     state.results = calloc(env.duration_sec + env.warmup_sec + 2,
0559                    sizeof(*state.results));
0560     if (!state.producers || !state.consumers || !state.results)
0561         exit(1);
0562 
0563     if (bench->validate)
0564         bench->validate();
0565     if (bench->setup)
0566         bench->setup();
0567 
0568     for (i = 0; i < env.consumer_cnt; i++) {
0569         err = pthread_create(&state.consumers[i], NULL,
0570                      bench->consumer_thread, (void *)(long)i);
0571         if (err) {
0572             fprintf(stderr, "failed to create consumer thread #%d: %d\n",
0573                 i, -errno);
0574             exit(1);
0575         }
0576         if (env.affinity)
0577             set_thread_affinity(state.consumers[i],
0578                         next_cpu(&env.cons_cpus));
0579     }
0580 
0581     /* unless explicit producer CPU list is specified, continue after
0582      * last consumer CPU
0583      */
0584     if (!env.prod_cpus.cpus)
0585         env.prod_cpus.next_cpu = env.cons_cpus.next_cpu;
0586 
0587     for (i = 0; i < env.producer_cnt; i++) {
0588         err = pthread_create(&state.producers[i], NULL,
0589                      bench->producer_thread, (void *)(long)i);
0590         if (err) {
0591             fprintf(stderr, "failed to create producer thread #%d: %d\n",
0592                 i, -errno);
0593             exit(1);
0594         }
0595         if (env.affinity)
0596             set_thread_affinity(state.producers[i],
0597                         next_cpu(&env.prod_cpus));
0598     }
0599 
0600     printf("Benchmark '%s' started.\n", bench->name);
0601 }
0602 
0603 static pthread_mutex_t bench_done_mtx = PTHREAD_MUTEX_INITIALIZER;
0604 static pthread_cond_t bench_done = PTHREAD_COND_INITIALIZER;
0605 
0606 static void collect_measurements(long delta_ns) {
0607     int iter = state.res_cnt++;
0608     struct bench_res *res = &state.results[iter];
0609 
0610     bench->measure(res);
0611 
0612     if (bench->report_progress)
0613         bench->report_progress(iter, res, delta_ns);
0614 
0615     if (iter == env.duration_sec + env.warmup_sec) {
0616         pthread_mutex_lock(&bench_done_mtx);
0617         pthread_cond_signal(&bench_done);
0618         pthread_mutex_unlock(&bench_done_mtx);
0619     }
0620 }
0621 
0622 int main(int argc, char **argv)
0623 {
0624     parse_cmdline_args(argc, argv);
0625 
0626     if (env.list) {
0627         int i;
0628 
0629         printf("Available benchmarks:\n");
0630         for (i = 0; i < ARRAY_SIZE(benchs); i++) {
0631             printf("- %s\n", benchs[i]->name);
0632         }
0633         return 0;
0634     }
0635 
0636     setup_benchmark();
0637 
0638     setup_timer();
0639 
0640     pthread_mutex_lock(&bench_done_mtx);
0641     pthread_cond_wait(&bench_done, &bench_done_mtx);
0642     pthread_mutex_unlock(&bench_done_mtx);
0643 
0644     if (bench->report_final)
0645         /* skip first sample */
0646         bench->report_final(state.results + env.warmup_sec,
0647                     state.res_cnt - env.warmup_sec);
0648 
0649     return 0;
0650 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!