1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
|
/******************************************************************************
* Copyright (c) Intel Corporation - All rights reserved. *
* This file is part of the LIBXSMM library. *
* *
* For information on the license, see the LICENSE file. *
* Further information: https://github.com/hfp/libxsmm/ *
* SPDX-License-Identifier: BSD-3-Clause *
******************************************************************************/
/* Hans Pabst (Intel Corp.)
******************************************************************************/
#include <libxsmm_sync.h>
#include <libxsmm.h>
#if defined(_OPENMP)
# include <omp.h>
#endif
/* measure non-contended latency of RO-lock */
#define MEASURE_LATENCY_RO(LOCK_KIND, LOCKPTR, NREPEAT, NR) do { \
libxsmm_timer_tickint latency = 0; \
double duration; \
int i; \
for (i = 0; i < (NREPEAT) / 4; ++i) { \
const libxsmm_timer_tickint tick = libxsmm_timer_tick(); \
LIBXSMM_LOCK_ACQREAD(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_RELREAD(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_ACQREAD(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_RELREAD(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_ACQREAD(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_RELREAD(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_ACQREAD(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_RELREAD(LOCK_KIND, LOCKPTR); \
latency += libxsmm_timer_ncycles(tick, libxsmm_timer_tick()); \
} \
duration = libxsmm_timer_duration(0, latency); \
if (0 < duration) { \
printf("\tro-latency: %.0f ns (call/s %.0f MHz, %.0f cycles)\n", \
duration * (NR) * 1e9, (NREPEAT) / (1e6 * duration), latency * (NR)); \
} \
} while(0)
/* measure non-contended latency of RW-lock */
#define MEASURE_LATENCY_RW(LOCK_KIND, LOCKPTR, NREPEAT, NR) do { \
libxsmm_timer_tickint latency = 0; \
double duration; \
int i; \
for (i = 0; i < (NREPEAT) / 4; ++i) { \
const libxsmm_timer_tickint tick = libxsmm_timer_tick(); \
LIBXSMM_LOCK_ACQUIRE(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_RELEASE(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_ACQUIRE(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_RELEASE(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_ACQUIRE(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_RELEASE(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_ACQUIRE(LOCK_KIND, LOCKPTR); \
LIBXSMM_LOCK_RELEASE(LOCK_KIND, LOCKPTR); \
latency += libxsmm_timer_ncycles(tick, libxsmm_timer_tick()); \
} \
duration = libxsmm_timer_duration(0, latency); \
if (0 < duration) { \
printf("\trw-latency: %.0f ns (call/s %.0f MHz, %.0f cycles)\n", \
duration * (NR) * 1e9, (NREPEAT) / (1e6 * duration), latency * (NR)); \
} \
} while(0)
#if defined(_OPENMP)
# define MEASURE_THROUGHPUT_PARALLEL(NTHREADS) LIBXSMM_PRAGMA(omp parallel num_threads(NTHREADS))
# define MEASURE_THROUGHPUT_ATOMIC LIBXSMM_PRAGMA(omp atomic)
#else
# define MEASURE_THROUGHPUT_PARALLEL(NTHREADS)
# define MEASURE_THROUGHPUT_ATOMIC
#endif
#define MEASURE_THROUGHPUT(LOCK_KIND, LOCKPTR, NREPEAT, NTHREADS, WORK_R, WORK_W, NW, NT) do { \
libxsmm_timer_tickint throughput = 0; \
double duration; \
MEASURE_THROUGHPUT_PARALLEL(NTHREADS) \
{ \
int n, nn; \
libxsmm_timer_tickint t1, t2, d = 0; \
const libxsmm_timer_tickint t0 = libxsmm_timer_tick(); \
for (n = 0; n < (NREPEAT); n = nn) { \
nn = n + 1; \
if (0 != (nn % (NW))) { /* read */ \
LIBXSMM_LOCK_ACQREAD(LOCK_KIND, LOCKPTR); \
t1 = libxsmm_timer_tick(); \
t2 = work(t1, WORK_R); \
LIBXSMM_LOCK_RELREAD(LOCK_KIND, LOCKPTR); \
d += libxsmm_timer_ncycles(t1, t2); \
} \
else { /* write */ \
LIBXSMM_LOCK_ACQUIRE(LOCK_KIND, LOCKPTR); \
t1 = libxsmm_timer_tick(); \
t2 = work(t1, WORK_W); \
LIBXSMM_LOCK_RELEASE(LOCK_KIND, LOCKPTR); \
d += libxsmm_timer_ncycles(t1, t2); \
} \
} \
t1 = libxsmm_timer_ncycles(t0, libxsmm_timer_tick()); \
MEASURE_THROUGHPUT_ATOMIC \
throughput += t1 - d; \
} \
duration = libxsmm_timer_duration(0, throughput); \
if (0 < duration) { \
const double r = 1.0 / (NT); \
printf("\tthroughput: %.0f us (call/s %.0f kHz, %.0f cycles)\n", \
duration * r * 1e6, (NT) / (1e3 * duration), throughput * r); \
} \
} while(0)
#define BENCHMARK(LOCK_KIND, IMPL, NTHREADS, WORK_R, WORK_W, WRATIOPERC, NREPEAT_LAT, NREPEAT_TPT) do { \
const int nw = 0 < (WRATIOPERC) ? (100 / (WRATIOPERC)) : ((NREPEAT_TPT) + 1); \
const int nt = (NREPEAT_TPT) * (NTHREADS); \
const double nr = 1.0 / (NREPEAT_LAT); \
LIBXSMM_LOCK_ATTR_TYPE(LOCK_KIND) attr; \
LIBXSMM_LOCK_TYPE(LOCK_KIND) lock; \
LIBXSMM_ASSERT(0 < nt); \
printf("Latency and throughput of \"%s\" (%s) for nthreads=%i wratio=%i%% work_r=%i work_w=%i nlat=%i ntpt=%i\n", \
LIBXSMM_STRINGIFY(LOCK_KIND), IMPL, NTHREADS, WRATIOPERC, WORK_R, WORK_W, NREPEAT_LAT, NREPEAT_TPT); \
LIBXSMM_LOCK_ATTR_INIT(LOCK_KIND, &attr); \
LIBXSMM_LOCK_INIT(LOCK_KIND, &lock, &attr); \
LIBXSMM_LOCK_ATTR_DESTROY(LOCK_KIND, &attr); \
MEASURE_LATENCY_RO(LOCK_KIND, &lock, NREPEAT_LAT, nr); \
MEASURE_LATENCY_RW(LOCK_KIND, &lock, NREPEAT_LAT, nr); \
MEASURE_THROUGHPUT(LOCK_KIND, &lock, NREPEAT_TPT, NTHREADS, WORK_R, WORK_W, nw, nt); \
LIBXSMM_LOCK_DESTROY(LOCK_KIND, &lock); \
} while(0)
libxsmm_timer_tickint work(libxsmm_timer_tickint start, libxsmm_timer_tickint duration);
libxsmm_timer_tickint work(libxsmm_timer_tickint start, libxsmm_timer_tickint duration)
{
const libxsmm_timer_tickint end = start + duration;
libxsmm_timer_tickint tick = start;
do {
libxsmm_timer_tickint i, s = 0;
for (i = 0; i < ((end - tick) / 4); ++i) s += i;
tick = libxsmm_timer_tick();
}
while(tick < end);
return tick;
}
int main(int argc, char* argv[])
{
#if defined(_OPENMP)
const int max_nthreads = omp_get_max_threads();
#else
const int max_nthreads = 1;
#endif
const int nthreads = LIBXSMM_MAX(1 < argc ? atoi(argv[1]) : max_nthreads, 1);
const int wratioperc = LIBXSMM_CLMP(2 < argc ? atoi(argv[2]) : 5, 0, 100);
const int work_r = LIBXSMM_MAX(3 < argc ? atoi(argv[3]) : 100, 1);
const int work_w = LIBXSMM_MAX(4 < argc ? atoi(argv[4]) : (10 * work_r), 1);
const int nlat = LIBXSMM_MAX(5 < argc ? atoi(argv[5]) : 2000000, 1);
const int ntpt = LIBXSMM_MAX(6 < argc ? atoi(argv[6]) : 10000, 1);
libxsmm_init();
printf("LIBXSMM: default lock-kind \"%s\" (%s)\n\n", LIBXSMM_STRINGIFY(LIBXSMM_LOCK_DEFAULT),
#if defined(LIBXSMM_LOCK_SYSTEM_SPINLOCK)
"OS-native");
#else
"Other");
#endif
#if defined(LIBXSMM_LOCK_SYSTEM_SPINLOCK)
BENCHMARK(LIBXSMM_LOCK_SPINLOCK, "OS-native", nthreads, work_r, work_w, wratioperc, nlat, ntpt);
#else
BENCHMARK(LIBXSMM_LOCK_SPINLOCK, "Other", nthreads, work_r, work_w, wratioperc, nlat, ntpt);
#endif
#if defined(LIBXSMM_LOCK_SYSTEM_MUTEX)
BENCHMARK(LIBXSMM_LOCK_MUTEX, "OS-native", nthreads, work_r, work_w, wratioperc, nlat, ntpt);
#else
BENCHMARK(LIBXSMM_LOCK_MUTEX, "Other", nthreads, work_r, work_w, wratioperc, nlat, ntpt);
#endif
#if defined(LIBXSMM_LOCK_SYSTEM_RWLOCK)
BENCHMARK(LIBXSMM_LOCK_RWLOCK, "OS-native", nthreads, work_r, work_w, wratioperc, nlat, ntpt);
#else
BENCHMARK(LIBXSMM_LOCK_RWLOCK, "Other", nthreads, work_r, work_w, wratioperc, nlat, ntpt);
#endif
return EXIT_SUCCESS;
}
|