1 /* Measure memcpy performance.
2 Copyright (C) 2016-2022 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
9
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
14
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library; if not, see
17 <https://www.gnu.org/licenses/>. */
18
19 #define MIN_SIZE 32768
20 #define MAX_SIZE (1024*1024)
21 #define NUM_TESTS 16384
22
23 #define MIN_PAGE_SIZE (MAX_SIZE + getpagesize())
24 #define TEST_MAIN
25 #define TEST_NAME "memcpy"
26 #include "bench-string.h"
27 #include <assert.h>
28 #include "json-lib.h"
29
30 IMPL (memcpy, 1)
31
32 typedef struct { uint16_t size; uint16_t freq; } freq_data_t;
33 typedef struct { uint8_t align; uint16_t freq; } align_data_t;
34
35 #define SIZE_NUM 65536
36 #define SIZE_MASK (SIZE_NUM-1)
37 static uint8_t size_arr[SIZE_NUM];
38
39 /* Frequency data for memcpy of less than 4096 bytes based on SPEC2017. */
40 static freq_data_t size_freq[] =
41 {
42 {32,22320}, { 16,9554}, { 8,8915}, {152,5327}, { 4,2159}, {292,2035},
43 { 12,1608}, { 24,1343}, {1152,895}, {144, 813}, {884, 733}, {284, 721},
44 {120, 661}, { 2, 649}, {882, 550}, { 5, 475}, { 7, 461}, {108, 460},
45 { 10, 361}, { 9, 361}, { 6, 334}, { 3, 326}, {464, 308}, {2048,303},
46 { 1, 298}, { 64, 250}, { 11, 197}, {296, 194}, { 68, 187}, { 15, 185},
47 {192, 184}, {1764,183}, { 13, 173}, {560, 126}, {160, 115}, {288, 96},
48 {104, 96}, {1144, 83}, { 18, 80}, { 23, 78}, { 40, 77}, { 19, 68},
49 { 48, 63}, { 17, 57}, { 72, 54}, {1280, 51}, { 20, 49}, { 28, 47},
50 { 22, 46}, {640, 45}, { 25, 41}, { 14, 40}, { 56, 37}, { 27, 35},
51 { 35, 33}, {384, 33}, { 29, 32}, { 80, 30}, {4095, 22}, {232, 22},
52 { 36, 19}, {184, 17}, { 21, 17}, {256, 16}, { 44, 15}, { 26, 15},
53 { 31, 14}, { 88, 14}, {176, 13}, { 33, 12}, {1024, 12}, {208, 11},
54 { 62, 11}, {128, 10}, {704, 10}, {324, 10}, { 96, 10}, { 60, 9},
55 {136, 9}, {124, 9}, { 34, 8}, { 30, 8}, {480, 8}, {1344, 8},
56 {273, 7}, {520, 7}, {112, 6}, { 52, 6}, {344, 6}, {336, 6},
57 {504, 5}, {168, 5}, {424, 5}, { 0, 4}, { 76, 3}, {200, 3},
58 {512, 3}, {312, 3}, {240, 3}, {960, 3}, {264, 2}, {672, 2},
59 { 38, 2}, {328, 2}, { 84, 2}, { 39, 2}, {216, 2}, { 42, 2},
60 { 37, 2}, {1608, 2}, { 70, 2}, { 46, 2}, {536, 2}, {280, 1},
61 {248, 1}, { 47, 1}, {1088, 1}, {1288, 1}, {224, 1}, { 41, 1},
62 { 50, 1}, { 49, 1}, {808, 1}, {360, 1}, {440, 1}, { 43, 1},
63 { 45, 1}, { 78, 1}, {968, 1}, {392, 1}, { 54, 1}, { 53, 1},
64 { 59, 1}, {376, 1}, {664, 1}, { 58, 1}, {272, 1}, { 66, 1},
65 {2688, 1}, {472, 1}, {568, 1}, {720, 1}, { 51, 1}, { 63, 1},
66 { 86, 1}, {496, 1}, {776, 1}, { 57, 1}, {680, 1}, {792, 1},
67 {122, 1}, {760, 1}, {824, 1}, {552, 1}, { 67, 1}, {456, 1},
68 {984, 1}, { 74, 1}, {408, 1}, { 75, 1}, { 92, 1}, {576, 1},
69 {116, 1}, { 65, 1}, {117, 1}, { 82, 1}, {352, 1}, { 55, 1},
70 {100, 1}, { 90, 1}, {696, 1}, {111, 1}, {880, 1}, { 79, 1},
71 {488, 1}, { 61, 1}, {114, 1}, { 94, 1}, {1032, 1}, { 98, 1},
72 { 87, 1}, {584, 1}, { 85, 1}, {648, 1}, {0, 0}
73 };
74
75 #define ALIGN_NUM 1024
76 #define ALIGN_MASK (ALIGN_NUM-1)
77 static uint8_t src_align_arr[ALIGN_NUM];
78 static uint8_t dst_align_arr[ALIGN_NUM];
79
80 /* Source alignment frequency for memcpy based on SPEC2017. */
81 static align_data_t src_align_freq[] =
82 {
83 {8, 300}, {16, 292}, {32, 168}, {64, 153}, {4, 79}, {2, 14}, {1, 18}, {0, 0}
84 };
85
86 /* Destination alignment frequency for memcpy based on SPEC2017. */
87 static align_data_t dst_align_freq[] =
88 {
89 {8, 265}, {16, 263}, {64, 209}, {32, 174}, {4, 90}, {2, 10}, {1, 13}, {0, 0}
90 };
91
92 typedef struct
93 {
94 uint64_t src : 24;
95 uint64_t dst : 24;
96 uint64_t len : 16;
97 } copy_t;
98
99 static copy_t test_arr[NUM_TESTS];
100
101 typedef char *(*proto_t) (char *, const char *, size_t);
102
103 static void
init_copy_distribution(void)104 init_copy_distribution (void)
105 {
106 int i, j, freq, size, n;
107
108 for (n = i = 0; (freq = size_freq[i].freq) != 0; i++)
109 for (j = 0, size = size_freq[i].size; j < freq; j++)
110 size_arr[n++] = size;
111 assert (n == SIZE_NUM);
112
113 for (n = i = 0; (freq = src_align_freq[i].freq) != 0; i++)
114 for (j = 0, size = src_align_freq[i].align; j < freq; j++)
115 src_align_arr[n++] = size - 1;
116 assert (n == ALIGN_NUM);
117
118 for (n = i = 0; (freq = dst_align_freq[i].freq) != 0; i++)
119 for (j = 0, size = dst_align_freq[i].align; j < freq; j++)
120 dst_align_arr[n++] = size - 1;
121 assert (n == ALIGN_NUM);
122 }
123
124
125 static void
do_one_test(json_ctx_t * json_ctx,impl_t * impl,char * dst,char * src,copy_t * copy,size_t n)126 do_one_test (json_ctx_t *json_ctx, impl_t *impl, char *dst, char *src,
127 copy_t *copy, size_t n)
128 {
129 timing_t start, stop, cur;
130 size_t iters = INNER_LOOP_ITERS_MEDIUM;
131
132 for (int j = 0; j < n; j++)
133 CALL (impl, dst + copy[j].dst, src + copy[j].src, copy[j].len);
134
135 TIMING_NOW (start);
136 for (int i = 0; i < iters; ++i)
137 for (int j = 0; j < n; j++)
138 CALL (impl, dst + copy[j].dst, src + copy[j].src, copy[j].len);
139 TIMING_NOW (stop);
140
141 TIMING_DIFF (cur, start, stop);
142
143 json_element_double (json_ctx, (double) cur / (double) iters);
144 }
145
146 static void
do_test(json_ctx_t * json_ctx,size_t max_size)147 do_test (json_ctx_t *json_ctx, size_t max_size)
148 {
149 int i;
150
151 memset (buf1, 1, max_size);
152
153 /* Create a random set of copies with the given size and alignment
154 distributions. */
155 for (i = 0; i < NUM_TESTS; i++)
156 {
157 test_arr[i].dst = (rand () & (max_size - 1));
158 test_arr[i].dst &= ~dst_align_arr[rand () & ALIGN_MASK];
159 test_arr[i].src = (rand () & (max_size - 1));
160 test_arr[i].src &= ~src_align_arr[rand () & ALIGN_MASK];
161 test_arr[i].len = size_arr[rand () & SIZE_MASK];
162 }
163
164 json_element_object_begin (json_ctx);
165 json_attr_uint (json_ctx, "length", (double) max_size);
166 json_array_begin (json_ctx, "timings");
167
168 FOR_EACH_IMPL (impl, 0)
169 do_one_test (json_ctx, impl, (char *) buf2, (char *) buf1, test_arr, i);
170
171 json_array_end (json_ctx);
172 json_element_object_end (json_ctx);
173 }
174
175 int
test_main(void)176 test_main (void)
177 {
178 json_ctx_t json_ctx;
179
180 test_init ();
181 init_copy_distribution ();
182
183 json_init (&json_ctx, 0, stdout);
184
185 json_document_begin (&json_ctx);
186 json_attr_string (&json_ctx, "timing_type", TIMING_TYPE);
187
188 json_attr_object_begin (&json_ctx, "functions");
189 json_attr_object_begin (&json_ctx, TEST_NAME);
190 json_attr_string (&json_ctx, "bench-variant", "random");
191
192 json_array_begin (&json_ctx, "ifuncs");
193 FOR_EACH_IMPL (impl, 0)
194 json_element_string (&json_ctx, impl->name);
195 json_array_end (&json_ctx);
196
197 json_array_begin (&json_ctx, "results");
198 for (int i = MIN_SIZE; i <= MAX_SIZE; i = i * 2)
199 do_test (&json_ctx, i);
200
201 json_array_end (&json_ctx);
202 json_attr_object_end (&json_ctx);
203 json_attr_object_end (&json_ctx);
204 json_document_end (&json_ctx);
205
206 return ret;
207 }
208
209 #include <support/test-driver.c>
210