Further improve string bench timing
[glibc.git] / support / blob_repeat.c
blob974582215f39ad37d6cc73d362bfbda95877ca32
1 /* Repeating a memory blob, with alias mapping optimization.
2 Copyright (C) 2018-2019 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 <http://www.gnu.org/licenses/>. */
19 #include <errno.h>
20 #include <fcntl.h>
21 #include <stdbool.h>
22 #include <stdlib.h>
23 #include <string.h>
24 #include <support/blob_repeat.h>
25 #include <support/check.h>
26 #include <support/test-driver.h>
27 #include <support/support.h>
28 #include <support/xunistd.h>
29 #include <sys/mman.h>
30 #include <unistd.h>
31 #include <wchar.h>
33 /* Small allocations should use malloc directly instead of the mmap
34 optimization because mappings carry a lot of overhead. */
35 static const size_t maximum_small_size = 4 * 1024 * 1024;
37 /* Internal helper for fill. */
38 static void
39 fill0 (char *target, const char *element, size_t element_size,
40 size_t count)
42 while (count > 0)
44 memcpy (target, element, element_size);
45 target += element_size;
46 --count;
50 /* Fill the buffer at TARGET with COUNT copies of the ELEMENT_SIZE
51 bytes starting at ELEMENT. */
52 static void
53 fill (char *target, const char *element, size_t element_size,
54 size_t count)
56 if (element_size == 0 || count == 0)
57 return;
58 else if (element_size == 1)
59 memset (target, element[0], count);
60 else if (element_size == sizeof (wchar_t))
62 wchar_t wc;
63 memcpy (&wc, element, sizeof (wc));
64 wmemset ((wchar_t *) target, wc, count);
66 else if (element_size < 1024 && count > 4096)
68 /* Use larger copies for really small element sizes. */
69 char buffer[8192];
70 size_t buffer_count = sizeof (buffer) / element_size;
71 fill0 (buffer, element, element_size, buffer_count);
72 while (count > 0)
74 size_t copy_count = buffer_count;
75 if (copy_count > count)
76 copy_count = count;
77 size_t copy_bytes = copy_count * element_size;
78 memcpy (target, buffer, copy_bytes);
79 target += copy_bytes;
80 count -= copy_count;
83 else
84 fill0 (target, element, element_size, count);
87 /* Use malloc instead of mmap for small allocations and unusual size
88 combinations. */
89 static struct support_blob_repeat
90 allocate_malloc (size_t total_size, const void *element, size_t element_size,
91 size_t count)
93 void *buffer = malloc (total_size);
94 if (buffer == NULL)
95 return (struct support_blob_repeat) { 0 };
96 fill (buffer, element, element_size, count);
97 return (struct support_blob_repeat)
99 .start = buffer,
100 .size = total_size,
101 .use_malloc = true
105 /* Return the least common multiple of PAGE_SIZE and ELEMENT_SIZE,
106 avoiding overflow. This assumes that PAGE_SIZE is a power of
107 two. */
108 static size_t
109 minimum_stride_size (size_t page_size, size_t element_size)
111 TEST_VERIFY_EXIT (page_size > 0);
112 TEST_VERIFY_EXIT (element_size > 0);
114 /* Compute the number of trailing zeros common to both sizes. */
115 unsigned int common_zeros = __builtin_ctzll (page_size | element_size);
117 /* In the product, this power of two appears twice, but in the least
118 common multiple, it appears only once. Therefore, shift one
119 factor. */
120 size_t multiple;
121 if (__builtin_mul_overflow (page_size >> common_zeros, element_size,
122 &multiple))
123 return 0;
124 return multiple;
127 /* Allocations larger than maximum_small_size potentially use mmap
128 with alias mappings. */
129 static struct support_blob_repeat
130 allocate_big (size_t total_size, const void *element, size_t element_size,
131 size_t count)
133 unsigned long page_size = xsysconf (_SC_PAGESIZE);
134 size_t stride_size = minimum_stride_size (page_size, element_size);
135 if (stride_size == 0)
137 errno = EOVERFLOW;
138 return (struct support_blob_repeat) { 0 };
141 /* Ensure that the stride size is at least maximum_small_size. This
142 is necessary to reduce the number of distinct mappings. */
143 if (stride_size < maximum_small_size)
144 stride_size
145 = ((maximum_small_size + stride_size - 1) / stride_size) * stride_size;
147 if (stride_size > total_size)
148 /* The mmap optimization would not save anything. */
149 return allocate_malloc (total_size, element, element_size, count);
151 /* Reserve the memory region. If we cannot create the mapping,
152 there is no reason to set up the backing file. */
153 void *target = mmap (NULL, total_size, PROT_NONE,
154 MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
155 if (target == MAP_FAILED)
156 return (struct support_blob_repeat) { 0 };
158 /* Create the backing file for the repeated mapping. Call mkstemp
159 directly to remove the resources backing the temporary file
160 immediately, once support_blob_repeat_free is called. Using
161 create_temp_file would result in a warning during post-test
162 cleanup. */
163 int fd;
165 char *temppath = xasprintf ("%s/support_blob_repeat-XXXXXX", test_dir);
166 fd = mkstemp (temppath);
167 if (fd < 0)
168 FAIL_EXIT1 ("mkstemp (\"%s\"): %m", temppath);
169 xunlink (temppath);
170 free (temppath);
173 /* Make sure that there is backing storage, so that the fill
174 operation will not fault. */
175 if (posix_fallocate (fd, 0, stride_size) != 0)
176 FAIL_EXIT1 ("posix_fallocate (%zu): %m", stride_size);
178 /* The stride size must still be a multiple of the page size and
179 element size. */
180 TEST_VERIFY_EXIT ((stride_size % page_size) == 0);
181 TEST_VERIFY_EXIT ((stride_size % element_size) == 0);
183 /* Fill the backing store. */
185 void *ptr = mmap (target, stride_size, PROT_READ | PROT_WRITE,
186 MAP_FIXED | MAP_FILE | MAP_SHARED, fd, 0);
187 if (ptr == MAP_FAILED)
189 int saved_errno = errno;
190 xmunmap (target, total_size);
191 xclose (fd);
192 errno = saved_errno;
193 return (struct support_blob_repeat) { 0 };
195 if (ptr != target)
196 FAIL_EXIT1 ("mapping of %zu bytes moved from %p to %p",
197 stride_size, target, ptr);
199 /* Write the repeating data. */
200 fill (target, element, element_size, stride_size / element_size);
202 /* Return to a PROT_NONE mapping, just to be on the safe side. */
203 ptr = mmap (target, stride_size, PROT_NONE,
204 MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
205 if (ptr == MAP_FAILED)
206 FAIL_EXIT1 ("Failed to reinstate PROT_NONE mapping: %m");
207 if (ptr != target)
208 FAIL_EXIT1 ("PROT_NONE mapping of %zu bytes moved from %p to %p",
209 stride_size, target, ptr);
212 /* Create the alias mappings. */
214 size_t remaining_size = total_size;
215 char *current = target;
216 int flags = MAP_FIXED | MAP_FILE | MAP_PRIVATE;
217 #ifdef MAP_NORESERVE
218 flags |= MAP_NORESERVE;
219 #endif
220 while (remaining_size > 0)
222 size_t to_map = stride_size;
223 if (to_map > remaining_size)
224 to_map = remaining_size;
225 void *ptr = mmap (current, to_map, PROT_READ | PROT_WRITE,
226 flags, fd, 0);
227 if (ptr == MAP_FAILED)
229 int saved_errno = errno;
230 xmunmap (target, total_size);
231 xclose (fd);
232 errno = saved_errno;
233 return (struct support_blob_repeat) { 0 };
235 if (ptr != current)
236 FAIL_EXIT1 ("MAP_PRIVATE mapping of %zu bytes moved from %p to %p",
237 to_map, target, ptr);
238 remaining_size -= to_map;
239 current += to_map;
243 xclose (fd);
245 return (struct support_blob_repeat)
247 .start = target,
248 .size = total_size,
249 .use_malloc = false
253 struct support_blob_repeat
254 support_blob_repeat_allocate (const void *element, size_t element_size,
255 size_t count)
257 size_t total_size;
258 if (__builtin_mul_overflow (element_size, count, &total_size))
260 errno = EOVERFLOW;
261 return (struct support_blob_repeat) { 0 };
263 if (total_size <= maximum_small_size)
264 return allocate_malloc (total_size, element, element_size, count);
265 else
266 return allocate_big (total_size, element, element_size, count);
269 void
270 support_blob_repeat_free (struct support_blob_repeat *blob)
272 if (blob->size > 0)
274 int saved_errno = errno;
275 if (blob->use_malloc)
276 free (blob->start);
277 else
278 xmunmap (blob->start, blob->size);
279 errno = saved_errno;
281 *blob = (struct support_blob_repeat) { 0 };