ceph/src/dpdk/app/test/test_memcpy_perf.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdint.h>
  35 #include <stdio.h>
  36 #include <string.h>
  37 #include <stdlib.h>
  38
  39 #include <rte_common.h>
  40 #include <rte_cycles.h>
  41 #include <rte_random.h>
  42 #include <rte_malloc.h>
  43
  44 #include <rte_memcpy.h>
  45
  46 #include "test.h"
  47
  48 /*
  49  * Set this to the maximum buffer size you want to test. If it is 0, then the
  50  * values in the buf_sizes[] array below will be used.
  51  */
  52 #define TEST_VALUE_RANGE        0
  53
  54 /* List of buffer sizes to test */
  55 #if TEST_VALUE_RANGE == 0
  56 static size_t buf_sizes[] = {
  57         1, 2, 3, 4, 5, 6, 7, 8, 9, 12, 15, 16, 17, 31, 32, 33, 63, 64, 65, 127, 128,
  58         129, 191, 192, 193, 255, 256, 257, 319, 320, 321, 383, 384, 385, 447, 448,
  59         449, 511, 512, 513, 767, 768, 769, 1023, 1024, 1025, 1518, 1522, 1536, 1600,
  60         2048, 2560, 3072, 3584, 4096, 4608, 5120, 5632, 6144, 6656, 7168, 7680, 8192
  61 };
  62 /* MUST be as large as largest packet size above */
  63 #define SMALL_BUFFER_SIZE       8192
  64 #else /* TEST_VALUE_RANGE != 0 */
  65 static size_t buf_sizes[TEST_VALUE_RANGE];
  66 #define SMALL_BUFFER_SIZE       TEST_VALUE_RANGE
  67 #endif /* TEST_VALUE_RANGE == 0 */
  68
  69
  70 /*
  71  * Arrays of this size are used for measuring uncached memory accesses by
  72  * picking a random location within the buffer. Make this smaller if there are
  73  * memory allocation errors.
  74  */
  75 #define LARGE_BUFFER_SIZE       (100 * 1024 * 1024)
  76
  77 /* How many times to run timing loop for performance tests */
  78 #define TEST_ITERATIONS         1000000
  79 #define TEST_BATCH_SIZE         100
  80
  81 /* Data is aligned on this many bytes (power of 2) */
  82 #ifdef RTE_MACHINE_CPUFLAG_AVX512F
  83 #define ALIGNMENT_UNIT          64
  84 #elif defined RTE_MACHINE_CPUFLAG_AVX2
  85 #define ALIGNMENT_UNIT          32
  86 #else /* RTE_MACHINE_CPUFLAG */
  87 #define ALIGNMENT_UNIT          16
  88 #endif /* RTE_MACHINE_CPUFLAG */
  89
  90 /*
  91  * Pointers used in performance tests. The two large buffers are for uncached
  92  * access where random addresses within the buffer are used for each
  93  * memcpy. The two small buffers are for cached access.
  94  */
  95 static uint8_t *large_buf_read, *large_buf_write;
  96 static uint8_t *small_buf_read, *small_buf_write;
  97
  98 /* Initialise data buffers. */
  99 static int
 100 init_buffers(void)
 101 {
 102         unsigned i;
 103
 104         large_buf_read = rte_malloc("memcpy", LARGE_BUFFER_SIZE + ALIGNMENT_UNIT, ALIGNMENT_UNIT);
 105         if (large_buf_read == NULL)
 106                 goto error_large_buf_read;
 107
 108         large_buf_write = rte_malloc("memcpy", LARGE_BUFFER_SIZE + ALIGNMENT_UNIT, ALIGNMENT_UNIT);
 109         if (large_buf_write == NULL)
 110                 goto error_large_buf_write;
 111
 112         small_buf_read = rte_malloc("memcpy", SMALL_BUFFER_SIZE + ALIGNMENT_UNIT, ALIGNMENT_UNIT);
 113         if (small_buf_read == NULL)
 114                 goto error_small_buf_read;
 115
 116         small_buf_write = rte_malloc("memcpy", SMALL_BUFFER_SIZE + ALIGNMENT_UNIT, ALIGNMENT_UNIT);
 117         if (small_buf_write == NULL)
 118                 goto error_small_buf_write;
 119
 120         for (i = 0; i < LARGE_BUFFER_SIZE; i++)
 121                 large_buf_read[i] = rte_rand();
 122         for (i = 0; i < SMALL_BUFFER_SIZE; i++)
 123                 small_buf_read[i] = rte_rand();
 124
 125         return 0;
 126
 127 error_small_buf_write:
 128         rte_free(small_buf_read);
 129 error_small_buf_read:
 130         rte_free(large_buf_write);
 131 error_large_buf_write:
 132         rte_free(large_buf_read);
 133 error_large_buf_read:
 134         printf("ERROR: not enough memory\n");
 135         return -1;
 136 }
 137
 138 /* Cleanup data buffers */
 139 static void
 140 free_buffers(void)
 141 {
 142         rte_free(large_buf_read);
 143         rte_free(large_buf_write);
 144         rte_free(small_buf_read);
 145         rte_free(small_buf_write);
 146 }
 147
 148 /*
 149  * Get a random offset into large array, with enough space needed to perform
 150  * max copy size. Offset is aligned, uoffset is used for unalignment setting.
 151  */
 152 static inline size_t
 153 get_rand_offset(size_t uoffset)
 154 {
 155         return ((rte_rand() % (LARGE_BUFFER_SIZE - SMALL_BUFFER_SIZE)) &
 156                         ~(ALIGNMENT_UNIT - 1)) + uoffset;
 157 }
 158
 159 /* Fill in source and destination addresses. */
 160 static inline void
 161 fill_addr_arrays(size_t *dst_addr, int is_dst_cached, size_t dst_uoffset,
 162                                  size_t *src_addr, int is_src_cached, size_t src_uoffset)
 163 {
 164         unsigned int i;
 165
 166         for (i = 0; i < TEST_BATCH_SIZE; i++) {
 167                 dst_addr[i] = (is_dst_cached) ? dst_uoffset : get_rand_offset(dst_uoffset);
 168                 src_addr[i] = (is_src_cached) ? src_uoffset : get_rand_offset(src_uoffset);
 169         }
 170 }
 171
 172 /*
 173  * WORKAROUND: For some reason the first test doing an uncached write
 174  * takes a very long time (~25 times longer than is expected). So we do
 175  * it once without timing.
 176  */
 177 static void
 178 do_uncached_write(uint8_t *dst, int is_dst_cached,
 179                                   const uint8_t *src, int is_src_cached, size_t size)
 180 {
 181         unsigned i, j;
 182         size_t dst_addrs[TEST_BATCH_SIZE], src_addrs[TEST_BATCH_SIZE];
 183
 184         for (i = 0; i < (TEST_ITERATIONS / TEST_BATCH_SIZE); i++) {
 185                 fill_addr_arrays(dst_addrs, is_dst_cached, 0,
 186                                                  src_addrs, is_src_cached, 0);
 187                 for (j = 0; j < TEST_BATCH_SIZE; j++) {
 188                         rte_memcpy(dst+dst_addrs[j], src+src_addrs[j], size);
 189                 }
 190         }
 191 }
 192
 193 /*
 194  * Run a single memcpy performance test. This is a macro to ensure that if
 195  * the "size" parameter is a constant it won't be converted to a variable.
 196  */
 197 #define SINGLE_PERF_TEST(dst, is_dst_cached, dst_uoffset,                   \
 198                          src, is_src_cached, src_uoffset, size)             \
 199 do {                                                                        \
 200     unsigned int iter, t;                                                   \
 201     size_t dst_addrs[TEST_BATCH_SIZE], src_addrs[TEST_BATCH_SIZE];          \
 202     uint64_t start_time, total_time = 0;                                    \
 203     uint64_t total_time2 = 0;                                               \
 204     for (iter = 0; iter < (TEST_ITERATIONS / TEST_BATCH_SIZE); iter++) {    \
 205         fill_addr_arrays(dst_addrs, is_dst_cached, dst_uoffset,             \
 206                          src_addrs, is_src_cached, src_uoffset);            \
 207         start_time = rte_rdtsc();                                           \
 208         for (t = 0; t < TEST_BATCH_SIZE; t++)                               \
 209             rte_memcpy(dst+dst_addrs[t], src+src_addrs[t], size);           \
 210         total_time += rte_rdtsc() - start_time;                             \
 211     }                                                                       \
 212     for (iter = 0; iter < (TEST_ITERATIONS / TEST_BATCH_SIZE); iter++) {    \
 213         fill_addr_arrays(dst_addrs, is_dst_cached, dst_uoffset,             \
 214                          src_addrs, is_src_cached, src_uoffset);            \
 215         start_time = rte_rdtsc();                                           \
 216         for (t = 0; t < TEST_BATCH_SIZE; t++)                               \
 217             memcpy(dst+dst_addrs[t], src+src_addrs[t], size);               \
 218         total_time2 += rte_rdtsc() - start_time;                            \
 219     }                                                                       \
 220     printf("%8.0f -",  (double)total_time /TEST_ITERATIONS);                \
 221     printf("%5.0f",  (double)total_time2 / TEST_ITERATIONS);                \
 222 } while (0)
 223
 224 /* Run aligned memcpy tests for each cached/uncached permutation */
 225 #define ALL_PERF_TESTS_FOR_SIZE(n)                                       \
 226 do {                                                                     \
 227     if (__builtin_constant_p(n))                                         \
 228         printf("\nC%6u", (unsigned)n);                                   \
 229     else                                                                 \
 230         printf("\n%7u", (unsigned)n);                                    \
 231     SINGLE_PERF_TEST(small_buf_write, 1, 0, small_buf_read, 1, 0, n);    \
 232     SINGLE_PERF_TEST(large_buf_write, 0, 0, small_buf_read, 1, 0, n);    \
 233     SINGLE_PERF_TEST(small_buf_write, 1, 0, large_buf_read, 0, 0, n);    \
 234     SINGLE_PERF_TEST(large_buf_write, 0, 0, large_buf_read, 0, 0, n);    \
 235 } while (0)
 236
 237 /* Run unaligned memcpy tests for each cached/uncached permutation */
 238 #define ALL_PERF_TESTS_FOR_SIZE_UNALIGNED(n)                             \
 239 do {                                                                     \
 240     if (__builtin_constant_p(n))                                         \
 241         printf("\nC%6u", (unsigned)n);                                   \
 242     else                                                                 \
 243         printf("\n%7u", (unsigned)n);                                    \
 244     SINGLE_PERF_TEST(small_buf_write, 1, 1, small_buf_read, 1, 5, n);    \
 245     SINGLE_PERF_TEST(large_buf_write, 0, 1, small_buf_read, 1, 5, n);    \
 246     SINGLE_PERF_TEST(small_buf_write, 1, 1, large_buf_read, 0, 5, n);    \
 247     SINGLE_PERF_TEST(large_buf_write, 0, 1, large_buf_read, 0, 5, n);    \
 248 } while (0)
 249
 250 /* Run memcpy tests for constant length */
 251 #define ALL_PERF_TEST_FOR_CONSTANT                                      \
 252 do {                                                                    \
 253     TEST_CONSTANT(6U); TEST_CONSTANT(64U); TEST_CONSTANT(128U);         \
 254     TEST_CONSTANT(192U); TEST_CONSTANT(256U); TEST_CONSTANT(512U);      \
 255     TEST_CONSTANT(768U); TEST_CONSTANT(1024U); TEST_CONSTANT(1536U);    \
 256 } while (0)
 257
 258 /* Run all memcpy tests for aligned constant cases */
 259 static inline void
 260 perf_test_constant_aligned(void)
 261 {
 262 #define TEST_CONSTANT ALL_PERF_TESTS_FOR_SIZE
 263         ALL_PERF_TEST_FOR_CONSTANT;
 264 #undef TEST_CONSTANT
 265 }
 266
 267 /* Run all memcpy tests for unaligned constant cases */
 268 static inline void
 269 perf_test_constant_unaligned(void)
 270 {
 271 #define TEST_CONSTANT ALL_PERF_TESTS_FOR_SIZE_UNALIGNED
 272         ALL_PERF_TEST_FOR_CONSTANT;
 273 #undef TEST_CONSTANT
 274 }
 275
 276 /* Run all memcpy tests for aligned variable cases */
 277 static inline void
 278 perf_test_variable_aligned(void)
 279 {
 280         unsigned n = sizeof(buf_sizes) / sizeof(buf_sizes[0]);
 281         unsigned i;
 282         for (i = 0; i < n; i++) {
 283                 ALL_PERF_TESTS_FOR_SIZE((size_t)buf_sizes[i]);
 284         }
 285 }
 286
 287 /* Run all memcpy tests for unaligned variable cases */
 288 static inline void
 289 perf_test_variable_unaligned(void)
 290 {
 291         unsigned n = sizeof(buf_sizes) / sizeof(buf_sizes[0]);
 292         unsigned i;
 293         for (i = 0; i < n; i++) {
 294                 ALL_PERF_TESTS_FOR_SIZE_UNALIGNED((size_t)buf_sizes[i]);
 295         }
 296 }
 297
 298 /* Run all memcpy tests */
 299 static int
 300 perf_test(void)
 301 {
 302         int ret;
 303
 304         ret = init_buffers();
 305         if (ret != 0)
 306                 return ret;
 307
 308 #if TEST_VALUE_RANGE != 0
 309         /* Set up buf_sizes array, if required */
 310         unsigned i;
 311         for (i = 0; i < TEST_VALUE_RANGE; i++)
 312                 buf_sizes[i] = i;
 313 #endif
 314
 315         /* See function comment */
 316         do_uncached_write(large_buf_write, 0, small_buf_read, 1, SMALL_BUFFER_SIZE);
 317
 318         printf("\n** rte_memcpy() - memcpy perf. tests (C = compile-time constant) **\n"
 319                    "======= ============== ============== ============== ==============\n"
 320                    "   Size Cache to cache   Cache to mem   Mem to cache     Mem to mem\n"
 321                    "(bytes)        (ticks)        (ticks)        (ticks)        (ticks)\n"
 322                    "------- -------------- -------------- -------------- --------------");
 323
 324         printf("\n========================== %2dB aligned ============================", ALIGNMENT_UNIT);
 325         /* Do aligned tests where size is a variable */
 326         perf_test_variable_aligned();
 327         printf("\n------- -------------- -------------- -------------- --------------");
 328         /* Do aligned tests where size is a compile-time constant */
 329         perf_test_constant_aligned();
 330         printf("\n=========================== Unaligned =============================");
 331         /* Do unaligned tests where size is a variable */
 332         perf_test_variable_unaligned();
 333         printf("\n------- -------------- -------------- -------------- --------------");
 334         /* Do unaligned tests where size is a compile-time constant */
 335         perf_test_constant_unaligned();
 336         printf("\n======= ============== ============== ============== ==============\n\n");
 337
 338         free_buffers();
 339
 340         return 0;
 341 }
 342
 343 static int
 344 test_memcpy_perf(void)
 345 {
 346         int ret;
 347
 348         ret = perf_test();
 349         if (ret != 0)
 350                 return -1;
 351         return 0;
 352 }
 353
 354 REGISTER_TEST_COMMAND(memcpy_perf_autotest, test_memcpy_perf);