odp_lock_perf.c

/* SPDX-License-Identifier: BSD-3-Clause
 * Copyright (c) 2021-2024 Nokia
 */
 
#include <stdio.h>
#include <string.h>
#include <stdint.h>
#include <inttypes.h>
#include <stdlib.h>
#include <getopt.h>
 
#include <odp_api.h>
#include <odp/helper/odph_api.h>
 
#include <export_results.h>
 
/* Max number of workers if num_cpu=0 */
#define DEFAULT_MAX_WORKERS 10
 
/* Max number of counters */
#define MAX_COUNTERS 8
 
#define TEST_INFO(name, test, validate) { name, test, validate }
 
/* Maximum number of results to be held */
#define TEST_MAX_BENCH 50
 
typedef enum repeat_t {
    REPEAT_NO,
    REPEAT_UNTIL_FAIL,
    REPEAT_FOREVER,
} repeat_t;
 
typedef enum place_t {
    PLACE_PACK,
    PLACE_SEPARATE,
    PLACE_ALL_SEPARATE,
} place_t;
 
/* Command line options */
typedef struct test_options_t {
    uint32_t num_cpu;
    uint32_t type;
    uint64_t num_round;
    repeat_t repeat;
    uint32_t num_counter;
    place_t place;
} test_options_t;
 
/* command line options default values */
static test_options_t test_options_def = {
    .num_cpu = 0,
    .type = 0,
    .num_round = 100000,
    .repeat = REPEAT_NO,
    .num_counter = 2,
    .place = 2,
};
 
typedef struct test_global_t test_global_t;
 
/* Test function template */
typedef void (*test_fn_t)(test_global_t *g, uint64_t **counter,
              uint32_t num_counter);
/* Test result validation function template */
typedef int (*validate_fn_t)(test_global_t *g, uint64_t **counter,
                 uint32_t num_counter);
 
/* Worker thread context */
typedef struct test_thread_ctx_t {
    test_global_t *global;
    test_fn_t func;
    uint64_t nsec;
    uint64_t cycles;
    uint32_t idx;
} test_thread_ctx_t;
 
typedef struct results_t {
    const char *test_name;
    double cycles_per_round;
    double nsec_per_op;
    double rounds_per_cpu;
    double total_rounds;
} results_t;
 
/* Global data */
struct test_global_t {
    test_options_t test_options;
    uint32_t cur_type;
    odp_barrier_t barrier;
    odp_cpumask_t cpumask;
    odph_thread_t thread_tbl[ODP_THREAD_COUNT_MAX];
    test_thread_ctx_t thread_ctx[ODP_THREAD_COUNT_MAX];
    test_common_options_t common_options;
    results_t results[TEST_MAX_BENCH];
    struct {
        struct ODP_ALIGNED_CACHE {
            odp_spinlock_t lock;
            uint64_t counter[MAX_COUNTERS];
        } spinlock;
        struct ODP_ALIGNED_CACHE {
            odp_spinlock_recursive_t lock;
            uint64_t counter[MAX_COUNTERS];
        } spinlock_recursive;
        struct ODP_ALIGNED_CACHE {
            odp_rwlock_t lock;
            uint64_t counter[MAX_COUNTERS];
        } rwlock;
        struct ODP_ALIGNED_CACHE {
            odp_rwlock_recursive_t lock;
            uint64_t counter[MAX_COUNTERS];
        } rwlock_recursive;
        struct ODP_ALIGNED_CACHE {
            odp_ticketlock_t lock;
            uint64_t counter[MAX_COUNTERS];
        } ticketlock;
        struct ODP_ALIGNED_CACHE {
            uint64_t counter[MAX_COUNTERS];
        } separate;
        struct {
            uint64_t ODP_ALIGNED_CACHE counter;
        } all_separate[MAX_COUNTERS];
    } item;
};
 
typedef struct {
    const char *name;
    test_fn_t test_fn;
    validate_fn_t validate_fn;
} test_case_t;
 
static test_global_t *test_global;
 
static inline void test_spinlock(test_global_t *g, uint64_t **counter,
                 uint32_t num_counter)
{
    odp_spinlock_t *lock = &g->item.spinlock.lock;
 
    for (uint64_t i = 0; i < g->test_options.num_round; i++) {
        odp_spinlock_lock(lock);
        for (uint32_t j = 0; j < num_counter; j++)
            (*counter[j])++;
        odp_spinlock_unlock(lock);
    }
}
 
static inline void test_spinlock_recursive(test_global_t *g, uint64_t **counter,
                       uint32_t num_counter)
{
    odp_spinlock_recursive_t *lock = &g->item.spinlock_recursive.lock;
 
    for (uint64_t i = 0; i < g->test_options.num_round; i++) {
        odp_spinlock_recursive_lock(lock);
        odp_spinlock_recursive_lock(lock);
        for (uint32_t j = 0; j < num_counter; j++)
            (*counter[j])++;
        odp_spinlock_recursive_unlock(lock);
        odp_spinlock_recursive_unlock(lock);
    }
}
 
static inline void test_rwlock(test_global_t *g, uint64_t **counter,
                   uint32_t num_counter)
{
    odp_rwlock_t *lock = &g->item.rwlock.lock;
 
    for (uint64_t i = 0; i < g->test_options.num_round; i++) {
        odp_rwlock_write_lock(lock);
        for (uint32_t j = 0; j < num_counter; j++)
            (*counter[j])++;
        odp_rwlock_write_unlock(lock);
        odp_rwlock_read_lock(lock);
        for (uint32_t j = 1; j < num_counter; j++)
            if (*counter[0] != *counter[j]) {
                odp_rwlock_read_unlock(lock);
                ODPH_ERR("Error: Counter mismatch\n");
                return;
            }
        odp_rwlock_read_unlock(lock);
    }
}
 
static inline void test_rwlock_recursive(test_global_t *g, uint64_t **counter,
                     uint32_t num_counter)
{
    odp_rwlock_recursive_t *lock = &g->item.rwlock_recursive.lock;
 
    for (uint64_t i = 0; i < g->test_options.num_round; i++) {
        odp_rwlock_recursive_write_lock(lock);
        odp_rwlock_recursive_write_lock(lock);
        for (uint32_t j = 0; j < num_counter; j++)
            (*counter[j])++;
        odp_rwlock_recursive_write_unlock(lock);
        odp_rwlock_recursive_write_unlock(lock);
        odp_rwlock_recursive_read_lock(lock);
        odp_rwlock_recursive_read_lock(lock);
        for (uint32_t j = 1; j < num_counter; j++)
            if (*counter[0] != *counter[j]) {
                odp_rwlock_recursive_read_unlock(lock);
                odp_rwlock_recursive_read_unlock(lock);
                ODPH_ERR("Error: Counter mismatch\n");
                return;
            }
        odp_rwlock_recursive_read_unlock(lock);
        odp_rwlock_recursive_read_unlock(lock);
    }
}
 
static inline void test_ticketlock(test_global_t *g, uint64_t **counter,
                   uint32_t num_counter)
{
    odp_ticketlock_t *lock = &g->item.ticketlock.lock;
 
    for (uint64_t i = 0; i < g->test_options.num_round; i++) {
        odp_ticketlock_lock(lock);
        for (uint32_t j = 0; j < num_counter; j++)
            (*counter[j])++;
        odp_ticketlock_unlock(lock);
    }
}
 
static inline int validate_generic(test_global_t *g, uint64_t **counter,
                   uint32_t num_counter)
{
    int status = 0;
    uint64_t total = (uint64_t)g->test_options.num_cpu * g->test_options.num_round;
 
    for (uint32_t i = 0; i < num_counter; i++) {
        if (*counter[i] != total) {
            status = 1;
            ODPH_ERR("Error: Counter %d value %" PRIu64 " expected %" PRIu64 "\n",
                 i, *counter[i], total);
        }
    }
 
    return status;
}
 
static void print_usage(void)
{
    printf("\n"
           "Lock performance test\n"
           "\n"
           "Usage: odp_lock_perf [options]\n"
           "\n"
           "  -c, --num_cpu          Number of CPUs (worker threads). 0: all available CPUs (or max %d) (default)\n"
           "  -t, --type             Lock type to test. 0: all (default %u)\n"
           "                             1: odp_spinlock_t\n"
           "                             2: odp_spinlock_recursive_t\n"
           "                             3: odp_rwlock_t\n"
           "                             4: odp_rwlock_recursive_t\n"
           "                             5: odp_ticketlock_t\n"
           "  -r, --num_round        Number of rounds (default %" PRIu64 ")\n"
           "  -e, --repeat           Repeat the tests (default %u)\n"
           "                             0: no repeat, run the tests once\n"
           "                             1: repeat until failure\n"
           "                             2: repeat forever\n"
           "  -o, --num_counter      Number of counters (default %u)\n"
           "  -p, --place            Counter placement (default %d)\n"
           "                             0: pack to same cache line with lock\n"
           "                             1: pack to separate cache line\n"
           "                             2: place each counter to separate cache line\n"
           "  -h, --help             This help\n"
           "\n",
           DEFAULT_MAX_WORKERS, test_options_def.type,
           test_options_def.num_round, test_options_def.repeat,
           test_options_def.num_counter, test_options_def.place);
}
 
static void print_info(test_options_t *test_options)
{
    printf("\nLock performance test configuration:\n");
    printf("  num cpu          %u\n", test_options->num_cpu);
    printf("  type             %u\n", test_options->type);
    printf("  num rounds       %" PRIu64 "\n", test_options->num_round);
    printf("  repeat           %u\n", test_options->repeat);
    printf("  num counters     %u\n", test_options->num_counter);
    printf("  place            %u\n", test_options->place);
    printf("\n\n");
}
 
static int output_summary(test_global_t *global)
{
    int results_size = ODPH_ARRAY_SIZE(global->results);
    results_t res;
 
    if (global->common_options.is_export) {
        if (test_common_write("function name,rounds/cpu (M/s),"
                      "total rounds (M/s),cycles/round,nsec/round\n")) {
            test_common_write_term();
            return -1;
        }
    }
 
    printf("Average results over %i threads:\n", global->test_options.num_cpu);
    printf("%-33s %-18s %-20s %-14s %-12s\n", "function name", "rounds/cpu (M/s)",
           "total rounds (M/s)", "cycles/round", "nsec/round");
    printf("----------------------------------------------------------------------------"
        "---------------------\n");
    for (int i = 0; i < results_size && global->results[i].test_name; i++) {
        res = global->results[i];
        printf("[%02d] %-28s %-18.2f %-20.2f %-14.2f %-12.2f\n", i + 1,
               res.test_name, res.rounds_per_cpu, res.total_rounds,
               res.cycles_per_round, res.nsec_per_op);
        if (global->common_options.is_export) {
            if (test_common_write("%s,%f,%f,%f,%f\n", res.test_name, res.rounds_per_cpu,
                          res.total_rounds, res.cycles_per_round,
                          res.nsec_per_op)){
                test_common_write_term();
                return -1;
            }
        }
    }
 
    if (global->common_options.is_export)
        test_common_write_term();
 
    return 0;
}
 
static int parse_options(int argc, char *argv[], test_options_t *test_options)
{
    int opt;
    int ret = 0;
 
    static const struct option longopts[] = {
        { "num_cpu", required_argument, NULL, 'c' },
        { "type", required_argument, NULL, 't' },
        { "num_round", required_argument, NULL, 'r' },
        { "repeat", required_argument, NULL, 'e' },
        { "num_counter", required_argument, NULL, 'o' },
        { "place", required_argument, NULL, 'p' },
        { "help", no_argument, NULL, 'h' },
        { NULL, 0, NULL, 0 }
    };
 
    static const char *shortopts = "+c:t:r:e:o:p:h";
 
    *test_options = test_options_def;
 
    while (1) {
        opt = getopt_long(argc, argv, shortopts, longopts, NULL);
 
        if (opt == -1)
            break;
 
        switch (opt) {
        case 'c':
            test_options->num_cpu = atoi(optarg);
            break;
        case 't':
            test_options->type = atoi(optarg);
            break;
        case 'r':
            test_options->num_round = atoll(optarg);
            break;
        case 'e':
            test_options->repeat = atoi(optarg);
            break;
        case 'o':
            test_options->num_counter = atoi(optarg);
            break;
        case 'p':
            test_options->place = atoi(optarg);
            break;
        case 'h':
            /* fall through */
        default:
            print_usage();
            ret = -1;
            break;
        }
    }
 
    if (test_options->num_round < 1) {
        ODPH_ERR("Invalid number of test rounds: %" PRIu64 "\n",
             test_options->num_round);
        return -1;
    }
 
    if (test_options->num_counter < 1 ||
        test_options->num_counter > MAX_COUNTERS) {
        ODPH_ERR("Invalid number of counters: %" PRIu32 "\n",
             test_options->num_counter);
        return -1;
    }
 
    return ret;
}
 
static int set_num_cpu(test_global_t *global)
{
    int ret, max_num;
    test_options_t *test_options = &global->test_options;
    int num_cpu = test_options->num_cpu;
 
    /* One thread used for the main thread */
    if (num_cpu > ODP_THREAD_COUNT_MAX - 1) {
        ODPH_ERR("Too many workers. Maximum is %i.\n", ODP_THREAD_COUNT_MAX - 1);
        return -1;
    }
 
    max_num = num_cpu;
    if (num_cpu == 0) {
        max_num = ODP_THREAD_COUNT_MAX - 1;
        if (max_num > DEFAULT_MAX_WORKERS)
            max_num = DEFAULT_MAX_WORKERS;
    }
 
    ret = odp_cpumask_default_worker(&global->cpumask, max_num);
 
    if (num_cpu && ret != num_cpu) {
        ODPH_ERR("Too many workers. Max supported %i.\n", ret);
        return -1;
    }
 
    /* Zero: all available workers */
    if (num_cpu == 0) {
        if (ret > max_num) {
            ODPH_ERR("Too many cpus from odp_cpumask_default_worker(): %i\n", ret);
            return -1;
        }
 
        num_cpu = ret;
        test_options->num_cpu = num_cpu;
    }
 
    odp_barrier_init(&global->barrier, num_cpu);
 
    return 0;
}
 
static int init_test(test_global_t *g, const char *name)
{
    printf("TEST: %s\n", name);
 
    memset(&g->item, 0, sizeof(g->item));
    odp_spinlock_init(&g->item.spinlock.lock);
    odp_spinlock_recursive_init(&g->item.spinlock_recursive.lock);
    odp_rwlock_init(&g->item.rwlock.lock);
    odp_rwlock_recursive_init(&g->item.rwlock_recursive.lock);
    odp_ticketlock_init(&g->item.ticketlock.lock);
 
    return 0;
}
 
static void fill_counter_ptrs(test_global_t *g, uint64_t **counter_out)
{
    test_options_t *test_options = &g->test_options;
 
    memset(counter_out, 0, sizeof(uint64_t *) * MAX_COUNTERS);
 
    switch (test_options->place) {
    case PLACE_PACK:
        for (uint32_t i = 0; i < test_options->num_counter; i++) {
            switch (g->cur_type) {
            case 0:
                counter_out[i] = &g->item.spinlock.counter[i];
                break;
            case 1:
                counter_out[i] = &g->item.spinlock_recursive.counter[i];
                break;
            case 2:
                counter_out[i] = &g->item.rwlock.counter[i];
                break;
            case 3:
                counter_out[i] = &g->item.rwlock_recursive.counter[i];
                break;
            case 4:
                counter_out[i] = &g->item.ticketlock.counter[i];
                break;
            }
        }
        break;
    case PLACE_SEPARATE:
        for (uint32_t i = 0; i < test_options->num_counter; i++)
            counter_out[i] = &g->item.separate.counter[i];
        break;
    case PLACE_ALL_SEPARATE:
        for (uint32_t i = 0; i < test_options->num_counter; i++)
            counter_out[i] = &g->item.all_separate[i].counter;
        break;
    }
}
 
static int run_test(void *arg)
{
    uint64_t nsec;
    odp_time_t t1, t2;
    uint64_t cycles;
    uint64_t c1, c2;
    test_thread_ctx_t *thread_ctx = arg;
    test_global_t *global = thread_ctx->global;
    test_options_t *test_options = &global->test_options;
    test_fn_t test_func = thread_ctx->func;
    uint64_t *counter[MAX_COUNTERS];
 
    fill_counter_ptrs(global, counter);
 
    /* Start all workers at the same time */
    odp_barrier_wait(&global->barrier);
 
    t1 = odp_time_local_strict();
    c1 = odp_cpu_cycles();
 
    test_func(global, counter, test_options->num_counter);
 
    c2 = odp_cpu_cycles();
    t2 = odp_time_local_strict();
 
    nsec = odp_time_diff_ns(t2, t1);
    cycles = odp_cpu_cycles_diff(c2, c1);
 
    /* Update stats */
    thread_ctx->nsec = nsec;
    thread_ctx->cycles = cycles;
 
    return 0;
}
 
static int start_workers(test_global_t *global, odp_instance_t instance,
             test_fn_t func)
{
    odph_thread_common_param_t param;
    int i, ret;
    test_options_t *test_options = &global->test_options;
    int num_cpu = test_options->num_cpu;
    odph_thread_param_t thr_param[num_cpu];
 
    odph_thread_common_param_init(&param);
    param.instance = instance;
    param.cpumask = &global->cpumask;
 
    for (i = 0; i < num_cpu; i++) {
        test_thread_ctx_t *thread_ctx = &global->thread_ctx[i];
 
        thread_ctx->global = global;
        thread_ctx->idx = i;
        thread_ctx->func = func;
 
        odph_thread_param_init(&thr_param[i]);
        thr_param[i].thr_type = ODP_THREAD_WORKER;
        thr_param[i].start = run_test;
        thr_param[i].arg = thread_ctx;
    }
 
    ret = odph_thread_create(global->thread_tbl, &param, thr_param,
                 num_cpu);
    if (ret != num_cpu) {
        ODPH_ERR("Failed to create all threads %i\n", ret);
        return -1;
    }
 
    return 0;
}
 
static int validate_results(test_global_t *global, validate_fn_t validate)
{
    test_options_t *test_options = &global->test_options;
    uint64_t *counter[MAX_COUNTERS];
 
    fill_counter_ptrs(global, counter);
 
    if (validate(global, counter, test_options->num_counter))
        return -1;
 
    return 0;
}
 
static test_case_t test_suite[] = {
    TEST_INFO("odp_spinlock", test_spinlock, validate_generic),
    TEST_INFO("odp_spinlock_recursive", test_spinlock_recursive, validate_generic),
    TEST_INFO("odp_rwlock", test_rwlock, validate_generic),
    TEST_INFO("odp_rwlock_recursive", test_rwlock_recursive, validate_generic),
    TEST_INFO("odp_ticketlock", test_ticketlock, validate_generic),
};
 
ODP_STATIC_ASSERT(ODPH_ARRAY_SIZE(test_suite) < TEST_MAX_BENCH,
          "Result array is too small to hold all the results");
 
static void output_results(test_global_t *global, int idx)
{
    int i, num;
    double cycles_per_round, nsec_ave, nsec_per_round, rounds_per_cpu, total_rounds;
    test_options_t *test_options = &global->test_options;
    int num_cpu = test_options->num_cpu;
    uint64_t num_round = test_options->num_round;
    uint64_t nsec_sum = 0;
    uint64_t cycles_sum = 0;
 
    global->results[idx].test_name = test_suite[idx].name;
 
    for (i = 0; i < ODP_THREAD_COUNT_MAX; i++) {
        nsec_sum += global->thread_ctx[i].nsec;
        cycles_sum += global->thread_ctx[i].cycles;
    }
 
    if (nsec_sum == 0) {
        printf("No results.\n");
        return;
    }
 
    nsec_ave = (double)nsec_sum / num_cpu;
    nsec_per_round = (double)nsec_sum / (num_cpu * num_round);
    cycles_per_round = (double)cycles_sum / (num_cpu * num_round);
    num = 0;
 
    rounds_per_cpu = num_round / (nsec_ave / 1000.0);
    total_rounds = ((uint64_t)num_cpu * num_round) / (nsec_ave / 1000.0);
 
    global->results[idx].cycles_per_round = cycles_per_round;
    global->results[idx].rounds_per_cpu = rounds_per_cpu;
    global->results[idx].nsec_per_op = nsec_per_round;
    global->results[idx].total_rounds = total_rounds;
 
    printf("------------------------------------------------\n");
    printf("Per thread results (Millions of rounds per sec):\n");
    printf("------------------------------------------------\n");
    printf("          1        2        3        4        5        6        7        8        9       10");
 
    for (i = 0; i < ODP_THREAD_COUNT_MAX; i++) {
        if (global->thread_ctx[i].nsec) {
            if ((num % 10) == 0)
                printf("\n   ");
 
            printf("%8.3f ", num_round / (global->thread_ctx[i].nsec / 1000.0));
            num++;
        }
    }
    printf("\n\n");
 
    printf("Average results over %i threads:\n", num_cpu);
    printf("------------------------------------------\n");
    printf("  duration:       %8.4f  sec\n", nsec_ave / ODP_TIME_SEC_IN_NS);
    printf("  cycles per round    %8.2f\n", cycles_per_round);
    printf("  nsec per round:     %8.2f\n", nsec_per_round);
    printf("  rounds per cpu:     %8.3fM rounds/sec\n", rounds_per_cpu);
    printf("  total rounds:       %8.3fM rounds/sec\n", total_rounds);
    printf("\n\n");
}
 
int main(int argc, char **argv)
{
    odph_helper_options_t helper_options;
    odp_instance_t instance;
    odp_init_t init;
    odp_shm_t shm;
    test_options_t test_options;
    int num_tests, i;
    test_common_options_t common_options;
 
    /* Let helper collect its own arguments (e.g. --odph_proc) */
    argc = odph_parse_options(argc, argv);
    if (odph_options(&helper_options)) {
        ODPH_ERR("Error: reading ODP helper options failed.\n");
        exit(EXIT_FAILURE);
    }
 
    argc = test_common_parse_options(argc, argv);
    if (test_common_options(&common_options)) {
        ODPH_ERR("Error: reading test common options failed\n");
        exit(EXIT_FAILURE);
    }
 
    if (parse_options(argc, argv, &test_options))
        exit(EXIT_FAILURE);
 
    /* List features not to be used */
    odp_init_param_init(&init);
    init.not_used.feat.cls = 1;
    init.not_used.feat.compress = 1;
    init.not_used.feat.crypto = 1;
    init.not_used.feat.ipsec = 1;
    init.not_used.feat.schedule = 1;
    init.not_used.feat.stash = 1;
    init.not_used.feat.timer = 1;
    init.not_used.feat.tm = 1;
 
    init.mem_model = helper_options.mem_model;
 
    /* Init ODP before calling anything else */
    if (odp_init_global(&instance, &init, NULL)) {
        ODPH_ERR("Global init failed.\n");
        exit(EXIT_FAILURE);
    }
 
    /* Init this thread */
    if (odp_init_local(instance, ODP_THREAD_CONTROL)) {
        ODPH_ERR("Local init failed.\n");
        exit(EXIT_FAILURE);
    }
 
    /* Reserve memory for global data from shared mem */
    shm = odp_shm_reserve("test_global", sizeof(test_global_t),
                  ODP_CACHE_LINE_SIZE, 0);
 
    if (shm == ODP_SHM_INVALID) {
        ODPH_ERR("Shared memory reserve failed.\n");
        exit(EXIT_FAILURE);
    }
 
    test_global = odp_shm_addr(shm);
    if (test_global == NULL) {
        ODPH_ERR("Shared memory alloc failed.\n");
        exit(EXIT_FAILURE);
    }
    memset(test_global, 0, sizeof(test_global_t));
    test_global->test_options = test_options;
    test_global->common_options = common_options;
 
    odp_sys_info_print();
 
    if (set_num_cpu(test_global))
        exit(EXIT_FAILURE);
 
    print_info(&test_global->test_options);
 
    /* Loop all test cases */
    num_tests = ODPH_ARRAY_SIZE(test_suite);
 
    while (1) {
        for (i = 0; i < num_tests; i++) {
            if (test_options.type && test_options.type != (uint32_t)i + 1)
                continue;
 
            test_global->cur_type = i;
 
            /* Initialize test variables */
            if (init_test(test_global, test_suite[i].name)) {
                ODPH_ERR("Failed to initialize test.\n");
                exit(EXIT_FAILURE);
            }
 
            /* Start workers */
            if (start_workers(test_global, instance, test_suite[i].test_fn))
                exit(EXIT_FAILURE);
 
            /* Wait workers to exit */
            odph_thread_join(test_global->thread_tbl,
                     test_global->test_options.num_cpu);
 
            output_results(test_global, i);
 
            /* Validate test results */
            if (validate_results(test_global, test_suite[i].validate_fn)) {
                ODPH_ERR("Test %s result validation failed.\n",
                     test_suite[i].name);
                if (test_options.repeat != REPEAT_FOREVER)
                    exit(EXIT_FAILURE);
            }
        }
 
        if (test_options.repeat == REPEAT_NO)
            break;
    }
 
    if (output_summary(test_global)) {
        ODPH_ERR("Outputting summary failed.\n");
        exit(EXIT_FAILURE);
    }
 
    if (odp_shm_free(shm)) {
        ODPH_ERR("Shm free failed.\n");
        exit(EXIT_FAILURE);
    }
 
    if (odp_term_local()) {
        ODPH_ERR("Local terminate failed.\n");
        exit(EXIT_FAILURE);
    }
 
    if (odp_term_global(instance)) {
        ODPH_ERR("Global terminate failed.\n");
        exit(EXIT_FAILURE);
    }
 
    return 0;
}