/* * Copyright (c) 2018 Jiri Svoboda * Copyright (c) 2018 Vojtech Horky * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * - Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * - The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /** @addtogroup hbench * @{ */ /** * @file */ #include #include #include #include #include #include #include #include #include #include #include #include #include "hbench.h" #define MAX_ERROR_STR_LENGTH 1024 static void short_report(bench_run_t *info, int run_index, benchmark_t *bench, uint64_t workload_size) { csv_report_add_entry(info, run_index, bench, workload_size); usec_t duration_usec = NSEC2USEC(stopwatch_get_nanos(&info->stopwatch)); printf("Completed %" PRIu64 " operations in %llu us", workload_size, duration_usec); if (duration_usec > 0) { double nanos = stopwatch_get_nanos(&info->stopwatch); double thruput = (double) workload_size / (nanos / 1000000000.0l); printf(", %.0f ops/s.\n", thruput); } else { printf(".\n"); } } /** Estimate square root value. * * @param value The value to compute square root of. * @param precision Required precision (e.g. 0.00001). * * @details * * This is a temporary solution until we have proper sqrt() implementation * in libmath. * * The algorithm uses Babylonian method [1]. * * [1] https://en.wikipedia.org/wiki/Methods_of_computing_square_roots#Babylonian_method */ static double estimate_square_root(double value, double precision) { double estimate = 1.; double prev_estimate = estimate + 10 * precision; while (fabs(estimate - prev_estimate) > precision) { prev_estimate = estimate; estimate = (prev_estimate + value / prev_estimate) / 2.; } return estimate; } /** Compute available statistics from given stopwatches. * * We compute normal mean for average duration of the workload and geometric * mean for average thruput. Note that geometric mean is necessary to compute * average throughput correctly - consider the following example: * - we run always 60 operations, * - first run executes in 30 s (i.e. 2 ops/s) * - and second one in 10 s (6 ops/s). * Then, naively, average throughput would be (2+6)/2 = 4 [ops/s]. However, we * actually executed 60 + 60 ops in 30 + 10 seconds. So the actual average * throughput is 3 ops/s (which is exactly what geometric mean means). * */ static void compute_stats(bench_run_t *runs, size_t run_count, uint64_t workload_size, double precision, double *out_duration_avg, double *out_duration_sigma, double *out_thruput_avg) { double inv_thruput_sum = 0.0; double nanos_sum = 0.0; double nanos_sum2 = 0.0; for (size_t i = 0; i < run_count; i++) { double nanos = stopwatch_get_nanos(&runs[i].stopwatch); double thruput = (double) workload_size / nanos; inv_thruput_sum += 1.0 / thruput; nanos_sum += nanos; nanos_sum2 += nanos * nanos; } *out_duration_avg = nanos_sum / run_count; double sigma2 = (nanos_sum2 - nanos_sum * (*out_duration_avg)) / ((double) run_count - 1); // FIXME: implement sqrt properly if (run_count > 1) { *out_duration_sigma = estimate_square_root(sigma2, precision); } else { *out_duration_sigma = NAN; } *out_thruput_avg = 1.0 / (inv_thruput_sum / run_count); } static void summary_stats(bench_run_t *runs, size_t run_count, benchmark_t *bench, uint64_t workload_size) { double duration_avg, duration_sigma, thruput_avg; compute_stats(runs, run_count, workload_size, 0.001, &duration_avg, &duration_sigma, &thruput_avg); printf("Average: %" PRIu64 " ops in %.0f us (sd %.0f us); " "%.0f ops/s; Samples: %zu\n", workload_size, duration_avg / 1000.0, duration_sigma / 1000.0, thruput_avg * 1000000000.0, run_count); } static bool run_benchmark(bench_env_t *env, benchmark_t *bench) { printf("Warm up and determine workload size...\n"); /* * We share this buffer across all runs as we know that it is * used only on failure (and we abort after first error). */ char *error_msg = malloc(MAX_ERROR_STR_LENGTH + 1); if (error_msg == NULL) { printf("Out of memory!\n"); return false; } str_cpy(error_msg, MAX_ERROR_STR_LENGTH, ""); bench_run_t helper_run; bench_run_init(&helper_run, error_msg, MAX_ERROR_STR_LENGTH); bool ret = true; if (bench->setup != NULL) { ret = bench->setup(env, &helper_run); if (!ret) { goto leave_error; } } /* * Find workload size that is big enough to last few seconds. * We also check that uint64_t is big enough. */ uint64_t workload_size = 0; for (size_t bits = 0; bits <= 64; bits++) { if (bits == 64) { str_cpy(error_msg, MAX_ERROR_STR_LENGTH, "Workload too small even for 1 << 63"); goto leave_error; } workload_size = ((uint64_t) 1) << bits; bench_run_t run; bench_run_init(&run, error_msg, MAX_ERROR_STR_LENGTH); bool ok = bench->entry(env, &run, workload_size); if (!ok) { goto leave_error; } short_report(&run, -1, bench, workload_size); nsec_t duration = stopwatch_get_nanos(&run.stopwatch); if (duration > env->minimal_run_duration_nanos) { break; } } printf("Workload size set to %" PRIu64 ", measuring %zu samples.\n", workload_size, env->run_count); bench_run_t *runs = calloc(env->run_count, sizeof(bench_run_t)); if (runs == NULL) { snprintf(error_msg, MAX_ERROR_STR_LENGTH, "failed allocating memory"); goto leave_error; } for (size_t i = 0; i < env->run_count; i++) { bench_run_init(&runs[i], error_msg, MAX_ERROR_STR_LENGTH); bool ok = bench->entry(env, &runs[i], workload_size); if (!ok) { free(runs); goto leave_error; } short_report(&runs[i], i, bench, workload_size); } summary_stats(runs, env->run_count, bench, workload_size); printf("\nBenchmark completed\n"); free(runs); goto leave; leave_error: printf("Error: %s\n", error_msg); ret = false; leave: if (bench->teardown != NULL) { bool ok = bench->teardown(env, &helper_run); if (!ok) { printf("Error: %s\n", error_msg); ret = false; } } free(error_msg); return ret; } static int run_benchmarks(bench_env_t *env) { unsigned int count_ok = 0; unsigned int count_fail = 0; char *failed_names = NULL; printf("\n*** Running all benchmarks ***\n\n"); for (size_t it = 0; it < benchmark_count; it++) { printf("%s (%s)\n", benchmarks[it]->name, benchmarks[it]->desc); if (run_benchmark(env, benchmarks[it])) { count_ok++; continue; } if (!failed_names) { failed_names = str_dup(benchmarks[it]->name); } else { char *f = NULL; asprintf(&f, "%s, %s", failed_names, benchmarks[it]->name); if (!f) { printf("Out of memory.\n"); abort(); } free(failed_names); failed_names = f; } count_fail++; } printf("\nCompleted, %u benchmarks run, %u succeeded.\n", count_ok + count_fail, count_ok); if (failed_names) printf("Failed benchmarks: %s\n", failed_names); return count_fail; } static void list_benchmarks(void) { size_t len = 0; for (size_t i = 0; i < benchmark_count; i++) { size_t len_now = str_length(benchmarks[i]->name); if (len_now > len) len = len_now; } assert(can_cast_size_t_to_int(len) && "benchmark name length overflow"); for (size_t i = 0; i < benchmark_count; i++) printf(" %-*s %s\n", (int) len, benchmarks[i]->name, benchmarks[i]->desc); printf(" %-*s Run all benchmarks\n", (int) len, "*"); } static void print_usage(const char *progname) { printf("Usage: %s [options] \n", progname); printf("-h, --help " "Print this help and exit\n"); printf("-d, --duration MILLIS " "Set minimal run duration (milliseconds)\n"); printf("-n, --count N " "Set number of measured runs\n"); printf("-o, --output filename.csv " "Store machine-readable data in filename.csv\n"); printf("-p, --param KEY=VALUE " "Additional parameters for the benchmark\n"); printf(" is one of the following:\n"); list_benchmarks(); } static void handle_param_arg(bench_env_t *env, char *arg) { char *value = NULL; char *key = str_tok(arg, "=", &value); bench_env_param_set(env, key, value); } int main(int argc, char *argv[]) { bench_env_t bench_env; errno_t rc = bench_env_init(&bench_env); if (rc != EOK) { fprintf(stderr, "Failed to initialize internal params structure: %s\n", str_error(rc)); return -5; } const char *short_options = "ho:p:n:d:"; struct option long_options[] = { { "duration", required_argument, NULL, 'd' }, { "help", optional_argument, NULL, 'h' }, { "count", required_argument, NULL, 'n' }, { "output", required_argument, NULL, 'o' }, { "param", required_argument, NULL, 'p' }, { 0, 0, NULL, 0 } }; char *csv_output_filename = NULL; int opt = 0; while ((opt = getopt_long(argc, argv, short_options, long_options, NULL)) > 0) { switch (opt) { case 'd': errno = EOK; bench_env.minimal_run_duration_nanos = MSEC2NSEC(atoll(optarg)); if ((errno != EOK) || (bench_env.minimal_run_duration_nanos <= 0)) { fprintf(stderr, "Invalid -d argument.\n"); return -3; } break; case 'h': print_usage(*argv); return 0; case 'n': errno = EOK; bench_env.run_count = (nsec_t) atoll(optarg); if ((errno != EOK) || (bench_env.run_count <= 0)) { fprintf(stderr, "Invalid -n argument.\n"); return -3; } break; case 'o': csv_output_filename = optarg; break; case 'p': handle_param_arg(&bench_env, optarg); break; case -1: default: break; } } if (optind + 1 != argc) { print_usage(*argv); fprintf(stderr, "Error: specify one benchmark to run or * for all.\n"); return -3; } const char *benchmark = argv[optind]; if (csv_output_filename != NULL) { errno_t rc = csv_report_open(csv_output_filename); if (rc != EOK) { fprintf(stderr, "Failed to open CSV report '%s': %s\n", csv_output_filename, str_error(rc)); return -4; } } int exit_code = 0; if (str_cmp(benchmark, "*") == 0) { exit_code = run_benchmarks(&bench_env); } else { bool benchmark_exists = false; for (size_t i = 0; i < benchmark_count; i++) { if (str_cmp(benchmark, benchmarks[i]->name) == 0) { benchmark_exists = true; exit_code = run_benchmark(&bench_env, benchmarks[i]) ? 0 : -1; break; } } if (!benchmark_exists) { printf("Unknown benchmark \"%s\"\n", benchmark); exit_code = -2; } } csv_report_close(); bench_env_cleanup(&bench_env); return exit_code; } /** @} */