Commit | Line | Data |
---|---|---|
c906108c | 1 | /* Implement a cached obstack. |
c2d11a7d JM |
2 | Written by Fred Fish <fnf@cygnus.com> |
3 | Rewritten by Jim Blandy <jimb@cygnus.com> | |
2c7ef074 | 4 | |
42a4f53d | 5 | Copyright (C) 1999-2019 Free Software Foundation, Inc. |
c906108c | 6 | |
c5aa993b | 7 | This file is part of GDB. |
c906108c | 8 | |
c5aa993b JM |
9 | This program is free software; you can redistribute it and/or modify |
10 | it under the terms of the GNU General Public License as published by | |
a9762ec7 | 11 | the Free Software Foundation; either version 3 of the License, or |
c5aa993b | 12 | (at your option) any later version. |
c906108c | 13 | |
c5aa993b JM |
14 | This program is distributed in the hope that it will be useful, |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | GNU General Public License for more details. | |
c906108c | 18 | |
c5aa993b | 19 | You should have received a copy of the GNU General Public License |
a9762ec7 | 20 | along with this program. If not, see <http://www.gnu.org/licenses/>. */ |
c906108c SS |
21 | |
22 | #include "defs.h" | |
d55e5aa6 | 23 | #include "gdb_obstack.h" |
4de283e4 | 24 | #include "bcache.h" |
c906108c | 25 | |
39ef2f62 CB |
26 | #include <algorithm> |
27 | ||
af5f3db6 AC |
28 | /* The type used to hold a single bcache string. The user data is |
29 | stored in d.data. Since it can be any type, it needs to have the | |
30 | same alignment as the most strict alignment of any type on the host | |
31 | machine. I don't know of any really correct way to do this in | |
32 | stock ANSI C, so just do it the same way obstack.h does. */ | |
33 | ||
34 | struct bstring | |
35 | { | |
49df298f | 36 | /* Hash chain. */ |
af5f3db6 | 37 | struct bstring *next; |
49df298f AC |
38 | /* Assume the data length is no more than 64k. */ |
39 | unsigned short length; | |
40 | /* The half hash hack. This contains the upper 16 bits of the hash | |
41 | value and is used as a pre-check when comparing two strings and | |
42 | avoids the need to do length or memcmp calls. It proves to be | |
43 | roughly 100% effective. */ | |
44 | unsigned short half_hash; | |
af5f3db6 AC |
45 | |
46 | union | |
47 | { | |
48 | char data[1]; | |
49 | double dummy; | |
50 | } | |
51 | d; | |
52 | }; | |
53 | ||
c2d11a7d JM |
54 | \f |
55 | /* Growing the bcache's hash table. */ | |
56 | ||
57 | /* If the average chain length grows beyond this, then we want to | |
58 | resize our hash table. */ | |
59 | #define CHAIN_LENGTH_THRESHOLD (5) | |
60 | ||
25629dfd TT |
61 | void |
62 | bcache::expand_hash_table () | |
c906108c | 63 | { |
c2d11a7d JM |
64 | /* A table of good hash table sizes. Whenever we grow, we pick the |
65 | next larger size from this table. sizes[i] is close to 1 << (i+10), | |
66 | so we roughly double the table size each time. After we fall off | |
67 | the end of this table, we just double. Don't laugh --- there have | |
68 | been executables sighted with a gigabyte of debug info. */ | |
69 | static unsigned long sizes[] = { | |
70 | 1021, 2053, 4099, 8191, 16381, 32771, | |
71 | 65537, 131071, 262144, 524287, 1048573, 2097143, | |
72 | 4194301, 8388617, 16777213, 33554467, 67108859, 134217757, | |
73 | 268435459, 536870923, 1073741827, 2147483659UL | |
74 | }; | |
745b8ca0 | 75 | unsigned int new_num_buckets; |
c2d11a7d | 76 | struct bstring **new_buckets; |
745b8ca0 | 77 | unsigned int i; |
c2d11a7d | 78 | |
2160782c AC |
79 | /* Count the stats. Every unique item needs to be re-hashed and |
80 | re-entered. */ | |
25629dfd TT |
81 | m_expand_count++; |
82 | m_expand_hash_count += m_unique_count; | |
2160782c | 83 | |
c2d11a7d | 84 | /* Find the next size. */ |
25629dfd | 85 | new_num_buckets = m_num_buckets * 2; |
c2d11a7d | 86 | for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++) |
25629dfd | 87 | if (sizes[i] > m_num_buckets) |
c2d11a7d JM |
88 | { |
89 | new_num_buckets = sizes[i]; | |
90 | break; | |
91 | } | |
c2d11a7d JM |
92 | |
93 | /* Allocate the new table. */ | |
94 | { | |
95 | size_t new_size = new_num_buckets * sizeof (new_buckets[0]); | |
b6de9da4 | 96 | |
c2d11a7d JM |
97 | new_buckets = (struct bstring **) xmalloc (new_size); |
98 | memset (new_buckets, 0, new_size); | |
99 | ||
25629dfd TT |
100 | m_structure_size -= m_num_buckets * sizeof (m_bucket[0]); |
101 | m_structure_size += new_size; | |
c2d11a7d | 102 | } |
c906108c | 103 | |
c2d11a7d | 104 | /* Rehash all existing strings. */ |
25629dfd | 105 | for (i = 0; i < m_num_buckets; i++) |
c906108c | 106 | { |
c2d11a7d JM |
107 | struct bstring *s, *next; |
108 | ||
25629dfd | 109 | for (s = m_bucket[i]; s; s = next) |
c906108c | 110 | { |
c2d11a7d JM |
111 | struct bstring **new_bucket; |
112 | next = s->next; | |
113 | ||
25629dfd | 114 | new_bucket = &new_buckets[(m_hash_function (&s->d.data, s->length) |
c2d11a7d JM |
115 | % new_num_buckets)]; |
116 | s->next = *new_bucket; | |
117 | *new_bucket = s; | |
c906108c SS |
118 | } |
119 | } | |
c2d11a7d JM |
120 | |
121 | /* Plug in the new table. */ | |
25629dfd TT |
122 | xfree (m_bucket); |
123 | m_bucket = new_buckets; | |
124 | m_num_buckets = new_num_buckets; | |
c906108c SS |
125 | } |
126 | ||
c2d11a7d JM |
127 | \f |
128 | /* Looking up things in the bcache. */ | |
129 | ||
130 | /* The number of bytes needed to allocate a struct bstring whose data | |
131 | is N bytes long. */ | |
132 | #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n)) | |
133 | ||
2e618c13 AR |
134 | /* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has |
135 | never seen those bytes before, add a copy of them to BCACHE. In | |
136 | either case, return a pointer to BCACHE's copy of that string. If | |
137 | optional ADDED is not NULL, return 1 in case of new entry or 0 if | |
138 | returning an old entry. */ | |
139 | ||
11d31d94 | 140 | const void * |
25629dfd | 141 | bcache::insert (const void *addr, int length, int *added) |
c906108c | 142 | { |
49df298f AC |
143 | unsigned long full_hash; |
144 | unsigned short half_hash; | |
c2d11a7d JM |
145 | int hash_index; |
146 | struct bstring *s; | |
c906108c | 147 | |
2e618c13 AR |
148 | if (added) |
149 | *added = 0; | |
150 | ||
e6ad058a TT |
151 | /* Lazily initialize the obstack. This can save quite a bit of |
152 | memory in some cases. */ | |
25629dfd | 153 | if (m_total_count == 0) |
e6ad058a TT |
154 | { |
155 | /* We could use obstack_specify_allocation here instead, but | |
156 | gdb_obstack.h specifies the allocation/deallocation | |
157 | functions. */ | |
25629dfd | 158 | obstack_init (&m_cache); |
e6ad058a TT |
159 | } |
160 | ||
c2d11a7d | 161 | /* If our average chain length is too high, expand the hash table. */ |
25629dfd TT |
162 | if (m_unique_count >= m_num_buckets * CHAIN_LENGTH_THRESHOLD) |
163 | expand_hash_table (); | |
c2d11a7d | 164 | |
25629dfd TT |
165 | m_total_count++; |
166 | m_total_size += length; | |
c2d11a7d | 167 | |
25629dfd | 168 | full_hash = m_hash_function (addr, length); |
cbd70537 | 169 | |
49df298f | 170 | half_hash = (full_hash >> 16); |
25629dfd | 171 | hash_index = full_hash % m_num_buckets; |
c2d11a7d | 172 | |
25629dfd | 173 | /* Search the hash m_bucket for a string identical to the caller's. |
49df298f AC |
174 | As a short-circuit first compare the upper part of each hash |
175 | values. */ | |
25629dfd | 176 | for (s = m_bucket[hash_index]; s; s = s->next) |
49df298f AC |
177 | { |
178 | if (s->half_hash == half_hash) | |
179 | { | |
180 | if (s->length == length | |
25629dfd | 181 | && m_compare_function (&s->d.data, addr, length)) |
49df298f AC |
182 | return &s->d.data; |
183 | else | |
25629dfd | 184 | m_half_hash_miss_count++; |
49df298f AC |
185 | } |
186 | } | |
c2d11a7d JM |
187 | |
188 | /* The user's string isn't in the list. Insert it after *ps. */ | |
189 | { | |
fe978cb0 | 190 | struct bstring *newobj |
25629dfd | 191 | = (struct bstring *) obstack_alloc (&m_cache, |
224c3ddb | 192 | BSTRING_SIZE (length)); |
b6de9da4 | 193 | |
fe978cb0 PA |
194 | memcpy (&newobj->d.data, addr, length); |
195 | newobj->length = length; | |
25629dfd | 196 | newobj->next = m_bucket[hash_index]; |
fe978cb0 | 197 | newobj->half_hash = half_hash; |
25629dfd | 198 | m_bucket[hash_index] = newobj; |
c2d11a7d | 199 | |
25629dfd TT |
200 | m_unique_count++; |
201 | m_unique_size += length; | |
202 | m_structure_size += BSTRING_SIZE (length); | |
c2d11a7d | 203 | |
2e618c13 AR |
204 | if (added) |
205 | *added = 1; | |
206 | ||
fe978cb0 | 207 | return &newobj->d.data; |
c2d11a7d | 208 | } |
c906108c | 209 | } |
c2d11a7d | 210 | \f |
cbd70537 SW |
211 | |
212 | /* Compare the byte string at ADDR1 of lenght LENGHT to the | |
213 | string at ADDR2. Return 1 if they are equal. */ | |
214 | ||
25629dfd TT |
215 | int |
216 | bcache::compare (const void *addr1, const void *addr2, int length) | |
cbd70537 SW |
217 | { |
218 | return memcmp (addr1, addr2, length) == 0; | |
219 | } | |
220 | ||
c2d11a7d | 221 | /* Free all the storage associated with BCACHE. */ |
25629dfd | 222 | bcache::~bcache () |
c906108c | 223 | { |
e6ad058a | 224 | /* Only free the obstack if we actually initialized it. */ |
25629dfd TT |
225 | if (m_total_count > 0) |
226 | obstack_free (&m_cache, 0); | |
227 | xfree (m_bucket); | |
c2d11a7d JM |
228 | } |
229 | ||
230 | ||
231 | \f | |
232 | /* Printing statistics. */ | |
233 | ||
c2d11a7d JM |
234 | static void |
235 | print_percentage (int portion, int total) | |
236 | { | |
237 | if (total == 0) | |
4a64f543 | 238 | /* i18n: Like "Percentage of duplicates, by count: (not applicable)". */ |
3d263c1d | 239 | printf_filtered (_("(not applicable)\n")); |
c906108c | 240 | else |
b2ba182e | 241 | printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total)); |
c2d11a7d JM |
242 | } |
243 | ||
244 | ||
245 | /* Print statistics on BCACHE's memory usage and efficacity at | |
246 | eliminating duplication. NAME should describe the kind of data | |
247 | BCACHE holds. Statistics are printed using `printf_filtered' and | |
248 | its ilk. */ | |
249 | void | |
25629dfd | 250 | bcache::print_statistics (const char *type) |
c2d11a7d JM |
251 | { |
252 | int occupied_buckets; | |
253 | int max_chain_length; | |
254 | int median_chain_length; | |
2160782c AC |
255 | int max_entry_size; |
256 | int median_entry_size; | |
c2d11a7d | 257 | |
2160782c AC |
258 | /* Count the number of occupied buckets, tally the various string |
259 | lengths, and measure chain lengths. */ | |
c2d11a7d | 260 | { |
745b8ca0 | 261 | unsigned int b; |
25629dfd TT |
262 | int *chain_length = XCNEWVEC (int, m_num_buckets + 1); |
263 | int *entry_size = XCNEWVEC (int, m_unique_count + 1); | |
2160782c | 264 | int stringi = 0; |
c2d11a7d JM |
265 | |
266 | occupied_buckets = 0; | |
267 | ||
25629dfd | 268 | for (b = 0; b < m_num_buckets; b++) |
c2d11a7d | 269 | { |
25629dfd | 270 | struct bstring *s = m_bucket[b]; |
c2d11a7d JM |
271 | |
272 | chain_length[b] = 0; | |
273 | ||
274 | if (s) | |
275 | { | |
276 | occupied_buckets++; | |
277 | ||
278 | while (s) | |
279 | { | |
25629dfd | 280 | gdb_assert (b < m_num_buckets); |
c2d11a7d | 281 | chain_length[b]++; |
25629dfd | 282 | gdb_assert (stringi < m_unique_count); |
2160782c | 283 | entry_size[stringi++] = s->length; |
c2d11a7d JM |
284 | s = s->next; |
285 | } | |
286 | } | |
287 | } | |
288 | ||
4a64f543 MS |
289 | /* To compute the median, we need the set of chain lengths |
290 | sorted. */ | |
39ef2f62 CB |
291 | std::sort (chain_length, chain_length + m_num_buckets); |
292 | std::sort (entry_size, entry_size + m_unique_count); | |
c2d11a7d | 293 | |
25629dfd | 294 | if (m_num_buckets > 0) |
c2d11a7d | 295 | { |
25629dfd TT |
296 | max_chain_length = chain_length[m_num_buckets - 1]; |
297 | median_chain_length = chain_length[m_num_buckets / 2]; | |
c2d11a7d JM |
298 | } |
299 | else | |
300 | { | |
301 | max_chain_length = 0; | |
302 | median_chain_length = 0; | |
303 | } | |
25629dfd | 304 | if (m_unique_count > 0) |
2160782c | 305 | { |
25629dfd TT |
306 | max_entry_size = entry_size[m_unique_count - 1]; |
307 | median_entry_size = entry_size[m_unique_count / 2]; | |
2160782c AC |
308 | } |
309 | else | |
310 | { | |
311 | max_entry_size = 0; | |
312 | median_entry_size = 0; | |
313 | } | |
314 | ||
315 | xfree (chain_length); | |
316 | xfree (entry_size); | |
c2d11a7d JM |
317 | } |
318 | ||
25629dfd TT |
319 | printf_filtered (_(" M_Cached '%s' statistics:\n"), type); |
320 | printf_filtered (_(" Total object count: %ld\n"), m_total_count); | |
321 | printf_filtered (_(" Unique object count: %lu\n"), m_unique_count); | |
3d263c1d | 322 | printf_filtered (_(" Percentage of duplicates, by count: ")); |
25629dfd | 323 | print_percentage (m_total_count - m_unique_count, m_total_count); |
c2d11a7d JM |
324 | printf_filtered ("\n"); |
325 | ||
25629dfd TT |
326 | printf_filtered (_(" Total object size: %ld\n"), m_total_size); |
327 | printf_filtered (_(" Unique object size: %ld\n"), m_unique_size); | |
3d263c1d | 328 | printf_filtered (_(" Percentage of duplicates, by size: ")); |
25629dfd | 329 | print_percentage (m_total_size - m_unique_size, m_total_size); |
c2d11a7d JM |
330 | printf_filtered ("\n"); |
331 | ||
3d263c1d BI |
332 | printf_filtered (_(" Max entry size: %d\n"), max_entry_size); |
333 | printf_filtered (_(" Average entry size: ")); | |
25629dfd TT |
334 | if (m_unique_count > 0) |
335 | printf_filtered ("%ld\n", m_unique_size / m_unique_count); | |
2160782c | 336 | else |
4a64f543 | 337 | /* i18n: "Average entry size: (not applicable)". */ |
3d263c1d BI |
338 | printf_filtered (_("(not applicable)\n")); |
339 | printf_filtered (_(" Median entry size: %d\n"), median_entry_size); | |
2160782c AC |
340 | printf_filtered ("\n"); |
341 | ||
3e43a32a MS |
342 | printf_filtered (_(" \ |
343 | Total memory used by bcache, including overhead: %ld\n"), | |
25629dfd | 344 | m_structure_size); |
3d263c1d | 345 | printf_filtered (_(" Percentage memory overhead: ")); |
25629dfd | 346 | print_percentage (m_structure_size - m_unique_size, m_unique_size); |
3d263c1d | 347 | printf_filtered (_(" Net memory savings: ")); |
25629dfd | 348 | print_percentage (m_total_size - m_structure_size, m_total_size); |
c2d11a7d JM |
349 | printf_filtered ("\n"); |
350 | ||
4a64f543 | 351 | printf_filtered (_(" Hash table size: %3d\n"), |
25629dfd | 352 | m_num_buckets); |
3d263c1d | 353 | printf_filtered (_(" Hash table expands: %lu\n"), |
25629dfd | 354 | m_expand_count); |
3d263c1d | 355 | printf_filtered (_(" Hash table hashes: %lu\n"), |
25629dfd | 356 | m_total_count + m_expand_hash_count); |
3d263c1d | 357 | printf_filtered (_(" Half hash misses: %lu\n"), |
25629dfd | 358 | m_half_hash_miss_count); |
3d263c1d | 359 | printf_filtered (_(" Hash table population: ")); |
25629dfd | 360 | print_percentage (occupied_buckets, m_num_buckets); |
3d263c1d | 361 | printf_filtered (_(" Median hash chain length: %3d\n"), |
c2d11a7d | 362 | median_chain_length); |
3d263c1d | 363 | printf_filtered (_(" Average hash chain length: ")); |
25629dfd TT |
364 | if (m_num_buckets > 0) |
365 | printf_filtered ("%3lu\n", m_unique_count / m_num_buckets); | |
c906108c | 366 | else |
4a64f543 | 367 | /* i18n: "Average hash chain length: (not applicable)". */ |
3d263c1d | 368 | printf_filtered (_("(not applicable)\n")); |
4a64f543 MS |
369 | printf_filtered (_(" Maximum hash chain length: %3d\n"), |
370 | max_chain_length); | |
c2d11a7d | 371 | printf_filtered ("\n"); |
c906108c | 372 | } |
af5f3db6 AC |
373 | |
374 | int | |
25629dfd | 375 | bcache::memory_used () |
af5f3db6 | 376 | { |
25629dfd | 377 | if (m_total_count == 0) |
e6ad058a | 378 | return 0; |
25629dfd | 379 | return obstack_memory_used (&m_cache); |
af5f3db6 | 380 | } |