4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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.
39 #include <rte_common.h>
40 #include <rte_cycles.h>
41 #include <rte_random.h>
42 #include <rte_malloc.h>
44 #include <rte_memcpy.h>
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.
52 #define TEST_VALUE_RANGE 0
54 /* List of buffer sizes to test */
55 #if TEST_VALUE_RANGE == 0
56 static size_t buf_sizes[] = {
57 0, 1, 7, 8, 9, 15, 16, 17, 31, 32, 33, 63, 64, 65, 127, 128, 129, 255,
58 256, 257, 320, 384, 511, 512, 513, 1023, 1024, 1025, 1518, 1522, 1600,
59 2048, 3072, 4096, 5120, 6144, 7168, 8192
61 /* MUST be as large as largest packet size above */
62 #define SMALL_BUFFER_SIZE 8192
63 #else /* TEST_VALUE_RANGE != 0 */
64 static size_t buf_sizes[TEST_VALUE_RANGE];
65 #define SMALL_BUFFER_SIZE TEST_VALUE_RANGE
66 #endif /* TEST_VALUE_RANGE == 0 */
70 * Arrays of this size are used for measuring uncached memory accesses by
71 * picking a random location within the buffer. Make this smaller if there are
72 * memory allocation errors.
74 #define LARGE_BUFFER_SIZE (100 * 1024 * 1024)
76 /* How many times to run timing loop for performance tests */
77 #define TEST_ITERATIONS 1000000
78 #define TEST_BATCH_SIZE 100
80 /* Data is aligned on this many bytes (power of 2) */
81 #define ALIGNMENT_UNIT 16
84 * Pointers used in performance tests. The two large buffers are for uncached
85 * access where random addresses within the buffer are used for each
86 * memcpy. The two small buffers are for cached access.
88 static uint8_t *large_buf_read, *large_buf_write;
89 static uint8_t *small_buf_read, *small_buf_write;
91 /* Initialise data buffers. */
97 large_buf_read = rte_malloc("memcpy", LARGE_BUFFER_SIZE, ALIGNMENT_UNIT);
98 if (large_buf_read == NULL)
99 goto error_large_buf_read;
101 large_buf_write = rte_malloc("memcpy", LARGE_BUFFER_SIZE, ALIGNMENT_UNIT);
102 if (large_buf_write == NULL)
103 goto error_large_buf_write;
105 small_buf_read = rte_malloc("memcpy", SMALL_BUFFER_SIZE, ALIGNMENT_UNIT);
106 if (small_buf_read == NULL)
107 goto error_small_buf_read;
109 small_buf_write = rte_malloc("memcpy", SMALL_BUFFER_SIZE, ALIGNMENT_UNIT);
110 if (small_buf_write == NULL)
111 goto error_small_buf_write;
113 for (i = 0; i < LARGE_BUFFER_SIZE; i++)
114 large_buf_read[i] = rte_rand();
115 for (i = 0; i < SMALL_BUFFER_SIZE; i++)
116 small_buf_read[i] = rte_rand();
120 error_small_buf_write:
121 rte_free(small_buf_read);
122 error_small_buf_read:
123 rte_free(large_buf_write);
124 error_large_buf_write:
125 rte_free(large_buf_read);
126 error_large_buf_read:
127 printf("ERROR: not enough memory\n");
131 /* Cleanup data buffers */
135 rte_free(large_buf_read);
136 rte_free(large_buf_write);
137 rte_free(small_buf_read);
138 rte_free(small_buf_write);
142 * Get a random offset into large array, with enough space needed to perform
143 * max copy size. Offset is aligned.
146 get_rand_offset(void)
148 return ((rte_rand() % (LARGE_BUFFER_SIZE - SMALL_BUFFER_SIZE)) &
149 ~(ALIGNMENT_UNIT - 1));
152 /* Fill in source and destination addresses. */
154 fill_addr_arrays(size_t *dst_addr, int is_dst_cached,
155 size_t *src_addr, int is_src_cached)
159 for (i = 0; i < TEST_BATCH_SIZE; i++) {
160 dst_addr[i] = (is_dst_cached) ? 0 : get_rand_offset();
161 src_addr[i] = (is_src_cached) ? 0 : get_rand_offset();
166 * WORKAROUND: For some reason the first test doing an uncached write
167 * takes a very long time (~25 times longer than is expected). So we do
168 * it once without timing.
171 do_uncached_write(uint8_t *dst, int is_dst_cached,
172 const uint8_t *src, int is_src_cached, size_t size)
175 size_t dst_addrs[TEST_BATCH_SIZE], src_addrs[TEST_BATCH_SIZE];
177 for (i = 0; i < (TEST_ITERATIONS / TEST_BATCH_SIZE); i++) {
178 fill_addr_arrays(dst_addrs, is_dst_cached,
179 src_addrs, is_src_cached);
180 for (j = 0; j < TEST_BATCH_SIZE; j++)
181 rte_memcpy(dst+dst_addrs[j], src+src_addrs[j], size);
186 * Run a single memcpy performance test. This is a macro to ensure that if
187 * the "size" parameter is a constant it won't be converted to a variable.
189 #define SINGLE_PERF_TEST(dst, is_dst_cached, src, is_src_cached, size) do { \
190 unsigned int iter, t; \
191 size_t dst_addrs[TEST_BATCH_SIZE], src_addrs[TEST_BATCH_SIZE]; \
192 uint64_t start_time, total_time = 0; \
193 uint64_t total_time2 = 0; \
194 for (iter = 0; iter < (TEST_ITERATIONS / TEST_BATCH_SIZE); iter++) { \
195 fill_addr_arrays(dst_addrs, is_dst_cached, \
196 src_addrs, is_src_cached); \
197 start_time = rte_rdtsc(); \
198 for (t = 0; t < TEST_BATCH_SIZE; t++) \
199 rte_memcpy(dst+dst_addrs[t], src+src_addrs[t], size); \
200 total_time += rte_rdtsc() - start_time; \
202 for (iter = 0; iter < (TEST_ITERATIONS / TEST_BATCH_SIZE); iter++) { \
203 fill_addr_arrays(dst_addrs, is_dst_cached, \
204 src_addrs, is_src_cached); \
205 start_time = rte_rdtsc(); \
206 for (t = 0; t < TEST_BATCH_SIZE; t++) \
207 memcpy(dst+dst_addrs[t], src+src_addrs[t], size); \
208 total_time2 += rte_rdtsc() - start_time; \
210 printf("%8.0f -", (double)total_time /TEST_ITERATIONS); \
211 printf("%5.0f", (double)total_time2 / TEST_ITERATIONS); \
214 /* Run memcpy() tests for each cached/uncached permutation. */
215 #define ALL_PERF_TESTS_FOR_SIZE(n) do { \
216 if (__builtin_constant_p(n)) \
217 printf("\nC%6u", (unsigned)n); \
219 printf("\n%7u", (unsigned)n); \
220 SINGLE_PERF_TEST(small_buf_write, 1, small_buf_read, 1, n); \
221 SINGLE_PERF_TEST(large_buf_write, 0, small_buf_read, 1, n); \
222 SINGLE_PERF_TEST(small_buf_write, 1, large_buf_read, 0, n); \
223 SINGLE_PERF_TEST(large_buf_write, 0, large_buf_read, 0, n); \
227 * Run performance tests for a number of different sizes and cached/uncached
233 const unsigned num_buf_sizes = sizeof(buf_sizes) / sizeof(buf_sizes[0]);
237 ret = init_buffers();
241 #if TEST_VALUE_RANGE != 0
242 /* Setup buf_sizes array, if required */
243 for (i = 0; i < TEST_VALUE_RANGE; i++)
247 /* See function comment */
248 do_uncached_write(large_buf_write, 0, small_buf_read, 1, SMALL_BUFFER_SIZE);
250 printf("\n** rte_memcpy() - memcpy perf. tests (C = compile-time constant) **\n"
251 "======= ============== ============== ============== ==============\n"
252 " Size Cache to cache Cache to mem Mem to cache Mem to mem\n"
253 "(bytes) (ticks) (ticks) (ticks) (ticks)\n"
254 "------- -------------- -------------- -------------- --------------");
256 /* Do tests where size is a variable */
257 for (i = 0; i < num_buf_sizes; i++) {
258 ALL_PERF_TESTS_FOR_SIZE((size_t)buf_sizes[i]);
260 printf("\n------- -------------- -------------- -------------- --------------");
261 /* Do tests where size is a compile-time constant */
262 ALL_PERF_TESTS_FOR_SIZE(63U);
263 ALL_PERF_TESTS_FOR_SIZE(64U);
264 ALL_PERF_TESTS_FOR_SIZE(65U);
265 ALL_PERF_TESTS_FOR_SIZE(255U);
266 ALL_PERF_TESTS_FOR_SIZE(256U);
267 ALL_PERF_TESTS_FOR_SIZE(257U);
268 ALL_PERF_TESTS_FOR_SIZE(1023U);
269 ALL_PERF_TESTS_FOR_SIZE(1024U);
270 ALL_PERF_TESTS_FOR_SIZE(1025U);
271 ALL_PERF_TESTS_FOR_SIZE(1518U);
273 printf("\n======= ============== ============== ============== ==============\n\n");
282 test_memcpy_perf(void)
292 static struct test_command memcpy_perf_cmd = {
293 .command = "memcpy_perf_autotest",
294 .callback = test_memcpy_perf,
296 REGISTER_TEST_COMMAND(memcpy_perf_cmd);