1 /* Measure memcpy performance.
2 Copyright (C) 2016-2021 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
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.
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.
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/>. */
19 #define MIN_PAGE_SIZE (512*1024+4096)
21 #define TEST_NAME "memcpy"
22 #include "bench-string.h"
26 #define MAX_COPIES 8192
30 typedef struct { uint16_t size
; uint16_t freq
; } freq_data_t
;
31 typedef struct { uint8_t align
; uint16_t freq
; } align_data_t
;
33 #define SIZE_NUM 65536
34 #define SIZE_MASK (SIZE_NUM-1)
35 static uint8_t size_arr
[SIZE_NUM
];
37 /* Frequency data for memcpy of less than 4096 bytes based on SPEC2017. */
38 static freq_data_t size_freq
[] =
40 { 32, 22320}, { 16, 9554}, { 8, 8915}, {152, 5327}, { 4, 2159}, {292, 2035},
41 { 12, 1608}, { 24, 1343}, {1152, 895}, {144, 813}, {884, 733}, {284, 721},
42 {120, 661}, { 2, 649}, {882, 550}, { 5, 475}, { 7, 461}, {108, 460},
43 { 10, 361}, { 9, 361}, { 6, 334}, { 3, 326}, {464, 308}, {2048, 303},
44 { 1, 298}, { 64, 250}, { 11, 197}, {296, 194}, { 68, 187}, { 15, 185},
45 {192, 184}, {1764, 183}, { 13, 173}, {560, 126}, {160, 115}, {288, 96},
46 {104, 96}, {1144, 83}, { 18, 80}, { 23, 78}, { 40, 77}, { 19, 68},
47 { 48, 63}, { 17, 57}, { 72, 54}, {1280, 51}, { 20, 49}, { 28, 47},
48 { 22, 46}, {640, 45}, { 25, 41}, { 14, 40}, { 56, 37}, { 27, 35},
49 { 35, 33}, {384, 33}, { 29, 32}, { 80, 30}, {4095, 22}, {232, 22},
50 { 36, 19}, {184, 17}, { 21, 17}, {256, 16}, { 44, 15}, { 26, 15},
51 { 31, 14}, { 88, 14}, {176, 13}, { 33, 12}, {1024, 12}, {208, 11},
52 { 62, 11}, {128, 10}, {704, 10}, {324, 10}, { 96, 10}, { 60, 9},
53 {136, 9}, {124, 9}, { 34, 8}, { 30, 8}, {480, 8}, {1344, 8},
54 {273, 7}, {520, 7}, {112, 6}, { 52, 6}, {344, 6}, {336, 6},
55 {504, 5}, {168, 5}, {424, 5}, { 0, 4}, { 76, 3}, {200, 3},
56 {512, 3}, {312, 3}, {240, 3}, {960, 3}, {264, 2}, {672, 2},
57 { 38, 2}, {328, 2}, { 84, 2}, { 39, 2}, {216, 2}, { 42, 2},
58 { 37, 2}, {1608, 2}, { 70, 2}, { 46, 2}, {536, 2}, {280, 1},
59 {248, 1}, { 47, 1}, {1088, 1}, {1288, 1}, {224, 1}, { 41, 1},
60 { 50, 1}, { 49, 1}, {808, 1}, {360, 1}, {440, 1}, { 43, 1},
61 { 45, 1}, { 78, 1}, {968, 1}, {392, 1}, { 54, 1}, { 53, 1},
62 { 59, 1}, {376, 1}, {664, 1}, { 58, 1}, {272, 1}, { 66, 1},
63 {2688, 1}, {472, 1}, {568, 1}, {720, 1}, { 51, 1}, { 63, 1},
64 { 86, 1}, {496, 1}, {776, 1}, { 57, 1}, {680, 1}, {792, 1},
65 {122, 1}, {760, 1}, {824, 1}, {552, 1}, { 67, 1}, {456, 1},
66 {984, 1}, { 74, 1}, {408, 1}, { 75, 1}, { 92, 1}, {576, 1},
67 {116, 1}, { 65, 1}, {117, 1}, { 82, 1}, {352, 1}, { 55, 1},
68 {100, 1}, { 90, 1}, {696, 1}, {111, 1}, {880, 1}, { 79, 1},
69 {488, 1}, { 61, 1}, {114, 1}, { 94, 1}, {1032, 1}, { 98, 1},
70 { 87, 1}, {584, 1}, { 85, 1}, {648, 1}, {0, 0}
73 #define ALIGN_NUM 1024
74 #define ALIGN_MASK (ALIGN_NUM-1)
75 static uint8_t src_align_arr
[ALIGN_NUM
];
76 static uint8_t dst_align_arr
[ALIGN_NUM
];
78 /* Source alignment frequency for memcpy based on SPEC2017. */
79 static align_data_t src_align_freq
[] =
81 {8, 300}, {16, 292}, {32, 168}, {64, 153}, {4, 79}, {2, 14}, {1, 18}, {0, 0}
84 /* Destination alignment frequency for memcpy based on SPEC2017. */
85 static align_data_t dst_align_freq
[] =
87 {8, 265}, {16, 263}, {64, 209}, {32, 174}, {4, 90}, {2, 10}, {1, 13}, {0, 0}
97 static copy_t copy
[MAX_COPIES
];
99 typedef char *(*proto_t
) (char *, const char *, size_t);
102 init_copy_distribution (void)
104 int i
, j
, freq
, size
, n
;
106 for (n
= i
= 0; (freq
= size_freq
[i
].freq
) != 0; i
++)
107 for (j
= 0, size
= size_freq
[i
].size
; j
< freq
; j
++)
108 size_arr
[n
++] = size
;
109 assert (n
== SIZE_NUM
);
111 for (n
= i
= 0; (freq
= src_align_freq
[i
].freq
) != 0; i
++)
112 for (j
= 0, size
= src_align_freq
[i
].align
; j
< freq
; j
++)
113 src_align_arr
[n
++] = size
- 1;
114 assert (n
== ALIGN_NUM
);
116 for (n
= i
= 0; (freq
= dst_align_freq
[i
].freq
) != 0; i
++)
117 for (j
= 0, size
= dst_align_freq
[i
].align
; j
< freq
; j
++)
118 dst_align_arr
[n
++] = size
- 1;
119 assert (n
== ALIGN_NUM
);
124 do_one_test (json_ctx_t
*json_ctx
, impl_t
*impl
, char *dst
, char *src
,
125 copy_t
*copy
, size_t n
)
127 timing_t start
, stop
, cur
;
128 size_t iters
= INNER_LOOP_ITERS_MEDIUM
;
130 for (int j
= 0; j
< n
; j
++)
131 CALL (impl
, dst
+ copy
[j
].dst
, src
+ copy
[j
].src
, copy
[j
].len
);
134 for (int i
= 0; i
< iters
; ++i
)
135 for (int j
= 0; j
< n
; j
++)
136 CALL (impl
, dst
+ copy
[j
].dst
, src
+ copy
[j
].src
, copy
[j
].len
);
139 TIMING_DIFF (cur
, start
, stop
);
141 json_element_double (json_ctx
, (double) cur
/ (double) iters
);
145 do_test (json_ctx_t
*json_ctx
, size_t max_size
)
149 memset (buf1
, 1, max_size
);
151 /* Create a random set of copies with the given size and alignment
153 for (i
= 0; i
< MAX_COPIES
; i
++)
155 copy
[i
].dst
= (rand () & (max_size
- 1));
156 copy
[i
].dst
&= ~dst_align_arr
[rand () & ALIGN_MASK
];
157 copy
[i
].src
= (rand () & (max_size
- 1));
158 copy
[i
].src
&= ~src_align_arr
[rand () & ALIGN_MASK
];
159 copy
[i
].len
= size_arr
[rand () & SIZE_MASK
];
162 json_element_object_begin (json_ctx
);
163 json_attr_uint (json_ctx
, "max-size", (double) max_size
);
164 json_array_begin (json_ctx
, "timings");
166 FOR_EACH_IMPL (impl
, 0)
167 do_one_test (json_ctx
, impl
, (char *) buf2
, (char *) buf1
, copy
, i
);
169 json_array_end (json_ctx
);
170 json_element_object_end (json_ctx
);
179 init_copy_distribution ();
181 json_init (&json_ctx
, 0, stdout
);
183 json_document_begin (&json_ctx
);
184 json_attr_string (&json_ctx
, "timing_type", TIMING_TYPE
);
186 json_attr_object_begin (&json_ctx
, "functions");
187 json_attr_object_begin (&json_ctx
, TEST_NAME
);
188 json_attr_string (&json_ctx
, "bench-variant", "random");
190 json_array_begin (&json_ctx
, "ifuncs");
191 FOR_EACH_IMPL (impl
, 0)
192 json_element_string (&json_ctx
, impl
->name
);
193 json_array_end (&json_ctx
);
195 json_array_begin (&json_ctx
, "results");
196 for (int i
= 4; i
<= 512; i
= i
* 2)
197 do_test (&json_ctx
, i
* 1024);
199 json_array_end (&json_ctx
);
200 json_attr_object_end (&json_ctx
);
201 json_attr_object_end (&json_ctx
);
202 json_document_end (&json_ctx
);
207 #include <support/test-driver.c>