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 | |
ecd75fc8 | 5 | Copyright (C) 1999-2014 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" | |
04ea0df1 | 23 | #include "gdb_obstack.h" |
c906108c | 24 | #include "bcache.h" |
c906108c | 25 | |
af5f3db6 AC |
26 | /* The type used to hold a single bcache string. The user data is |
27 | stored in d.data. Since it can be any type, it needs to have the | |
28 | same alignment as the most strict alignment of any type on the host | |
29 | machine. I don't know of any really correct way to do this in | |
30 | stock ANSI C, so just do it the same way obstack.h does. */ | |
31 | ||
32 | struct bstring | |
33 | { | |
49df298f | 34 | /* Hash chain. */ |
af5f3db6 | 35 | struct bstring *next; |
49df298f AC |
36 | /* Assume the data length is no more than 64k. */ |
37 | unsigned short length; | |
38 | /* The half hash hack. This contains the upper 16 bits of the hash | |
39 | value and is used as a pre-check when comparing two strings and | |
40 | avoids the need to do length or memcmp calls. It proves to be | |
41 | roughly 100% effective. */ | |
42 | unsigned short half_hash; | |
af5f3db6 AC |
43 | |
44 | union | |
45 | { | |
46 | char data[1]; | |
47 | double dummy; | |
48 | } | |
49 | d; | |
50 | }; | |
51 | ||
52 | ||
53 | /* The structure for a bcache itself. The bcache is initialized, in | |
54 | bcache_xmalloc(), by filling it with zeros and then setting the | |
55 | corresponding obstack's malloc() and free() methods. */ | |
56 | ||
57 | struct bcache | |
58 | { | |
59 | /* All the bstrings are allocated here. */ | |
60 | struct obstack cache; | |
61 | ||
62 | /* How many hash buckets we're using. */ | |
63 | unsigned int num_buckets; | |
64 | ||
65 | /* Hash buckets. This table is allocated using malloc, so when we | |
66 | grow the table we can return the old table to the system. */ | |
67 | struct bstring **bucket; | |
68 | ||
69 | /* Statistics. */ | |
70 | unsigned long unique_count; /* number of unique strings */ | |
71 | long total_count; /* total number of strings cached, including dups */ | |
72 | long unique_size; /* size of unique strings, in bytes */ | |
73 | long total_size; /* total number of bytes cached, including dups */ | |
74 | long structure_size; /* total size of bcache, including infrastructure */ | |
2160782c AC |
75 | /* Number of times that the hash table is expanded and hence |
76 | re-built, and the corresponding number of times that a string is | |
77 | [re]hashed as part of entering it into the expanded table. The | |
78 | total number of hashes can be computed by adding TOTAL_COUNT to | |
79 | expand_hash_count. */ | |
80 | unsigned long expand_count; | |
81 | unsigned long expand_hash_count; | |
49df298f AC |
82 | /* Number of times that the half-hash compare hit (compare the upper |
83 | 16 bits of hash values) hit, but the corresponding combined | |
84 | length/data compare missed. */ | |
85 | unsigned long half_hash_miss_count; | |
cbd70537 SW |
86 | |
87 | /* Hash function to be used for this bcache object. */ | |
88 | unsigned long (*hash_function)(const void *addr, int length); | |
89 | ||
90 | /* Compare function to be used for this bcache object. */ | |
91 | int (*compare_function)(const void *, const void *, int length); | |
af5f3db6 AC |
92 | }; |
93 | ||
4a64f543 MS |
94 | /* The old hash function was stolen from SDBM. This is what DB 3.0 |
95 | uses now, and is better than the old one. */ | |
c2d11a7d | 96 | \f |
c2d11a7d | 97 | unsigned long |
d85a5daf | 98 | hash(const void *addr, int length) |
cbd70537 SW |
99 | { |
100 | return hash_continue (addr, length, 0); | |
101 | } | |
102 | ||
103 | /* Continue the calculation of the hash H at the given address. */ | |
104 | ||
105 | unsigned long | |
106 | hash_continue (const void *addr, int length, unsigned long h) | |
c2d11a7d | 107 | { |
b6de9da4 | 108 | const unsigned char *k, *e; |
b6de9da4 MS |
109 | |
110 | k = (const unsigned char *)addr; | |
111 | e = k+length; | |
cbd70537 | 112 | for (; k< e;++k) |
b6de9da4 MS |
113 | { |
114 | h *=16777619; | |
115 | h ^= *k; | |
116 | } | |
117 | return (h); | |
c906108c | 118 | } |
c2d11a7d JM |
119 | \f |
120 | /* Growing the bcache's hash table. */ | |
121 | ||
122 | /* If the average chain length grows beyond this, then we want to | |
123 | resize our hash table. */ | |
124 | #define CHAIN_LENGTH_THRESHOLD (5) | |
125 | ||
126 | static void | |
127 | expand_hash_table (struct bcache *bcache) | |
c906108c | 128 | { |
c2d11a7d JM |
129 | /* A table of good hash table sizes. Whenever we grow, we pick the |
130 | next larger size from this table. sizes[i] is close to 1 << (i+10), | |
131 | so we roughly double the table size each time. After we fall off | |
132 | the end of this table, we just double. Don't laugh --- there have | |
133 | been executables sighted with a gigabyte of debug info. */ | |
134 | static unsigned long sizes[] = { | |
135 | 1021, 2053, 4099, 8191, 16381, 32771, | |
136 | 65537, 131071, 262144, 524287, 1048573, 2097143, | |
137 | 4194301, 8388617, 16777213, 33554467, 67108859, 134217757, | |
138 | 268435459, 536870923, 1073741827, 2147483659UL | |
139 | }; | |
745b8ca0 | 140 | unsigned int new_num_buckets; |
c2d11a7d | 141 | struct bstring **new_buckets; |
745b8ca0 | 142 | unsigned int i; |
c2d11a7d | 143 | |
2160782c AC |
144 | /* Count the stats. Every unique item needs to be re-hashed and |
145 | re-entered. */ | |
146 | bcache->expand_count++; | |
147 | bcache->expand_hash_count += bcache->unique_count; | |
148 | ||
c2d11a7d | 149 | /* Find the next size. */ |
745b8ca0 | 150 | new_num_buckets = bcache->num_buckets * 2; |
c2d11a7d JM |
151 | for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++) |
152 | if (sizes[i] > bcache->num_buckets) | |
153 | { | |
154 | new_num_buckets = sizes[i]; | |
155 | break; | |
156 | } | |
c2d11a7d JM |
157 | |
158 | /* Allocate the new table. */ | |
159 | { | |
160 | size_t new_size = new_num_buckets * sizeof (new_buckets[0]); | |
b6de9da4 | 161 | |
c2d11a7d JM |
162 | new_buckets = (struct bstring **) xmalloc (new_size); |
163 | memset (new_buckets, 0, new_size); | |
164 | ||
165 | bcache->structure_size -= (bcache->num_buckets | |
166 | * sizeof (bcache->bucket[0])); | |
167 | bcache->structure_size += new_size; | |
168 | } | |
c906108c | 169 | |
c2d11a7d JM |
170 | /* Rehash all existing strings. */ |
171 | for (i = 0; i < bcache->num_buckets; i++) | |
c906108c | 172 | { |
c2d11a7d JM |
173 | struct bstring *s, *next; |
174 | ||
175 | for (s = bcache->bucket[i]; s; s = next) | |
c906108c | 176 | { |
c2d11a7d JM |
177 | struct bstring **new_bucket; |
178 | next = s->next; | |
179 | ||
d3aff959 TT |
180 | new_bucket = &new_buckets[(bcache->hash_function (&s->d.data, |
181 | s->length) | |
c2d11a7d JM |
182 | % new_num_buckets)]; |
183 | s->next = *new_bucket; | |
184 | *new_bucket = s; | |
c906108c SS |
185 | } |
186 | } | |
c2d11a7d JM |
187 | |
188 | /* Plug in the new table. */ | |
189 | if (bcache->bucket) | |
b8c9b27d | 190 | xfree (bcache->bucket); |
c2d11a7d JM |
191 | bcache->bucket = new_buckets; |
192 | bcache->num_buckets = new_num_buckets; | |
c906108c SS |
193 | } |
194 | ||
c2d11a7d JM |
195 | \f |
196 | /* Looking up things in the bcache. */ | |
197 | ||
198 | /* The number of bytes needed to allocate a struct bstring whose data | |
199 | is N bytes long. */ | |
200 | #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n)) | |
201 | ||
202 | /* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has | |
203 | never seen those bytes before, add a copy of them to BCACHE. In | |
204 | either case, return a pointer to BCACHE's copy of that string. */ | |
2e618c13 | 205 | const void * |
12747160 | 206 | bcache (const void *addr, int length, struct bcache *cache) |
2e618c13 | 207 | { |
12747160 | 208 | return bcache_full (addr, length, cache, NULL); |
2e618c13 AR |
209 | } |
210 | ||
211 | /* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has | |
212 | never seen those bytes before, add a copy of them to BCACHE. In | |
213 | either case, return a pointer to BCACHE's copy of that string. If | |
214 | optional ADDED is not NULL, return 1 in case of new entry or 0 if | |
215 | returning an old entry. */ | |
216 | ||
11d31d94 TT |
217 | const void * |
218 | bcache_full (const void *addr, int length, struct bcache *bcache, int *added) | |
c906108c | 219 | { |
49df298f AC |
220 | unsigned long full_hash; |
221 | unsigned short half_hash; | |
c2d11a7d JM |
222 | int hash_index; |
223 | struct bstring *s; | |
c906108c | 224 | |
2e618c13 AR |
225 | if (added) |
226 | *added = 0; | |
227 | ||
e6ad058a TT |
228 | /* Lazily initialize the obstack. This can save quite a bit of |
229 | memory in some cases. */ | |
230 | if (bcache->total_count == 0) | |
231 | { | |
232 | /* We could use obstack_specify_allocation here instead, but | |
233 | gdb_obstack.h specifies the allocation/deallocation | |
234 | functions. */ | |
235 | obstack_init (&bcache->cache); | |
236 | } | |
237 | ||
c2d11a7d JM |
238 | /* If our average chain length is too high, expand the hash table. */ |
239 | if (bcache->unique_count >= bcache->num_buckets * CHAIN_LENGTH_THRESHOLD) | |
240 | expand_hash_table (bcache); | |
241 | ||
242 | bcache->total_count++; | |
243 | bcache->total_size += length; | |
244 | ||
cbd70537 SW |
245 | full_hash = bcache->hash_function (addr, length); |
246 | ||
49df298f AC |
247 | half_hash = (full_hash >> 16); |
248 | hash_index = full_hash % bcache->num_buckets; | |
c2d11a7d | 249 | |
49df298f AC |
250 | /* Search the hash bucket for a string identical to the caller's. |
251 | As a short-circuit first compare the upper part of each hash | |
252 | values. */ | |
c2d11a7d | 253 | for (s = bcache->bucket[hash_index]; s; s = s->next) |
49df298f AC |
254 | { |
255 | if (s->half_hash == half_hash) | |
256 | { | |
257 | if (s->length == length | |
cbd70537 | 258 | && bcache->compare_function (&s->d.data, addr, length)) |
49df298f AC |
259 | return &s->d.data; |
260 | else | |
261 | bcache->half_hash_miss_count++; | |
262 | } | |
263 | } | |
c2d11a7d JM |
264 | |
265 | /* The user's string isn't in the list. Insert it after *ps. */ | |
266 | { | |
267 | struct bstring *new | |
268 | = obstack_alloc (&bcache->cache, BSTRING_SIZE (length)); | |
b6de9da4 | 269 | |
c2d11a7d JM |
270 | memcpy (&new->d.data, addr, length); |
271 | new->length = length; | |
272 | new->next = bcache->bucket[hash_index]; | |
49df298f | 273 | new->half_hash = half_hash; |
c2d11a7d JM |
274 | bcache->bucket[hash_index] = new; |
275 | ||
276 | bcache->unique_count++; | |
277 | bcache->unique_size += length; | |
278 | bcache->structure_size += BSTRING_SIZE (length); | |
279 | ||
2e618c13 AR |
280 | if (added) |
281 | *added = 1; | |
282 | ||
c2d11a7d JM |
283 | return &new->d.data; |
284 | } | |
c906108c | 285 | } |
c2d11a7d | 286 | \f |
cbd70537 SW |
287 | |
288 | /* Compare the byte string at ADDR1 of lenght LENGHT to the | |
289 | string at ADDR2. Return 1 if they are equal. */ | |
290 | ||
291 | static int | |
292 | bcache_compare (const void *addr1, const void *addr2, int length) | |
293 | { | |
294 | return memcmp (addr1, addr2, length) == 0; | |
295 | } | |
296 | ||
af5f3db6 AC |
297 | /* Allocating and freeing bcaches. */ |
298 | ||
cbd70537 SW |
299 | /* Allocated a bcache. HASH_FUNCTION and COMPARE_FUNCTION can be used |
300 | to pass in custom hash, and compare functions to be used by this | |
4a64f543 MS |
301 | bcache. If HASH_FUNCTION is NULL hash() is used and if |
302 | COMPARE_FUNCTION is NULL memcmp() is used. */ | |
cbd70537 | 303 | |
af5f3db6 | 304 | struct bcache * |
cbd70537 | 305 | bcache_xmalloc (unsigned long (*hash_function)(const void *, int length), |
4a64f543 MS |
306 | int (*compare_function)(const void *, |
307 | const void *, | |
308 | int length)) | |
af5f3db6 AC |
309 | { |
310 | /* Allocate the bcache pre-zeroed. */ | |
fc270c35 | 311 | struct bcache *b = XCNEW (struct bcache); |
b6de9da4 | 312 | |
cbd70537 SW |
313 | if (hash_function) |
314 | b->hash_function = hash_function; | |
315 | else | |
316 | b->hash_function = hash; | |
317 | ||
318 | if (compare_function) | |
319 | b->compare_function = compare_function; | |
320 | else | |
321 | b->compare_function = bcache_compare; | |
af5f3db6 AC |
322 | return b; |
323 | } | |
c2d11a7d JM |
324 | |
325 | /* Free all the storage associated with BCACHE. */ | |
c906108c | 326 | void |
af5f3db6 | 327 | bcache_xfree (struct bcache *bcache) |
c906108c | 328 | { |
af5f3db6 AC |
329 | if (bcache == NULL) |
330 | return; | |
e6ad058a TT |
331 | /* Only free the obstack if we actually initialized it. */ |
332 | if (bcache->total_count > 0) | |
333 | obstack_free (&bcache->cache, 0); | |
af5f3db6 AC |
334 | xfree (bcache->bucket); |
335 | xfree (bcache); | |
c2d11a7d JM |
336 | } |
337 | ||
338 | ||
339 | \f | |
340 | /* Printing statistics. */ | |
341 | ||
c2d11a7d JM |
342 | static void |
343 | print_percentage (int portion, int total) | |
344 | { | |
345 | if (total == 0) | |
4a64f543 | 346 | /* i18n: Like "Percentage of duplicates, by count: (not applicable)". */ |
3d263c1d | 347 | printf_filtered (_("(not applicable)\n")); |
c906108c | 348 | else |
b2ba182e | 349 | printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total)); |
c2d11a7d JM |
350 | } |
351 | ||
352 | ||
353 | /* Print statistics on BCACHE's memory usage and efficacity at | |
354 | eliminating duplication. NAME should describe the kind of data | |
355 | BCACHE holds. Statistics are printed using `printf_filtered' and | |
356 | its ilk. */ | |
357 | void | |
358 | print_bcache_statistics (struct bcache *c, char *type) | |
359 | { | |
360 | int occupied_buckets; | |
361 | int max_chain_length; | |
362 | int median_chain_length; | |
2160782c AC |
363 | int max_entry_size; |
364 | int median_entry_size; | |
c2d11a7d | 365 | |
2160782c AC |
366 | /* Count the number of occupied buckets, tally the various string |
367 | lengths, and measure chain lengths. */ | |
c2d11a7d | 368 | { |
745b8ca0 | 369 | unsigned int b; |
fc270c35 TT |
370 | int *chain_length = XCNEWVEC (int, c->num_buckets + 1); |
371 | int *entry_size = XCNEWVEC (int, c->unique_count + 1); | |
2160782c | 372 | int stringi = 0; |
c2d11a7d JM |
373 | |
374 | occupied_buckets = 0; | |
375 | ||
376 | for (b = 0; b < c->num_buckets; b++) | |
377 | { | |
378 | struct bstring *s = c->bucket[b]; | |
379 | ||
380 | chain_length[b] = 0; | |
381 | ||
382 | if (s) | |
383 | { | |
384 | occupied_buckets++; | |
385 | ||
386 | while (s) | |
387 | { | |
2160782c | 388 | gdb_assert (b < c->num_buckets); |
c2d11a7d | 389 | chain_length[b]++; |
2160782c AC |
390 | gdb_assert (stringi < c->unique_count); |
391 | entry_size[stringi++] = s->length; | |
c2d11a7d JM |
392 | s = s->next; |
393 | } | |
394 | } | |
395 | } | |
396 | ||
4a64f543 MS |
397 | /* To compute the median, we need the set of chain lengths |
398 | sorted. */ | |
c2d11a7d | 399 | qsort (chain_length, c->num_buckets, sizeof (chain_length[0]), |
dc146f7c | 400 | compare_positive_ints); |
2160782c | 401 | qsort (entry_size, c->unique_count, sizeof (entry_size[0]), |
dc146f7c | 402 | compare_positive_ints); |
c2d11a7d JM |
403 | |
404 | if (c->num_buckets > 0) | |
405 | { | |
406 | max_chain_length = chain_length[c->num_buckets - 1]; | |
407 | median_chain_length = chain_length[c->num_buckets / 2]; | |
408 | } | |
409 | else | |
410 | { | |
411 | max_chain_length = 0; | |
412 | median_chain_length = 0; | |
413 | } | |
2160782c AC |
414 | if (c->unique_count > 0) |
415 | { | |
416 | max_entry_size = entry_size[c->unique_count - 1]; | |
417 | median_entry_size = entry_size[c->unique_count / 2]; | |
418 | } | |
419 | else | |
420 | { | |
421 | max_entry_size = 0; | |
422 | median_entry_size = 0; | |
423 | } | |
424 | ||
425 | xfree (chain_length); | |
426 | xfree (entry_size); | |
c2d11a7d JM |
427 | } |
428 | ||
3d263c1d BI |
429 | printf_filtered (_(" Cached '%s' statistics:\n"), type); |
430 | printf_filtered (_(" Total object count: %ld\n"), c->total_count); | |
431 | printf_filtered (_(" Unique object count: %lu\n"), c->unique_count); | |
432 | printf_filtered (_(" Percentage of duplicates, by count: ")); | |
c2d11a7d JM |
433 | print_percentage (c->total_count - c->unique_count, c->total_count); |
434 | printf_filtered ("\n"); | |
435 | ||
3d263c1d BI |
436 | printf_filtered (_(" Total object size: %ld\n"), c->total_size); |
437 | printf_filtered (_(" Unique object size: %ld\n"), c->unique_size); | |
438 | printf_filtered (_(" Percentage of duplicates, by size: ")); | |
c2d11a7d JM |
439 | print_percentage (c->total_size - c->unique_size, c->total_size); |
440 | printf_filtered ("\n"); | |
441 | ||
3d263c1d BI |
442 | printf_filtered (_(" Max entry size: %d\n"), max_entry_size); |
443 | printf_filtered (_(" Average entry size: ")); | |
2160782c AC |
444 | if (c->unique_count > 0) |
445 | printf_filtered ("%ld\n", c->unique_size / c->unique_count); | |
446 | else | |
4a64f543 | 447 | /* i18n: "Average entry size: (not applicable)". */ |
3d263c1d BI |
448 | printf_filtered (_("(not applicable)\n")); |
449 | printf_filtered (_(" Median entry size: %d\n"), median_entry_size); | |
2160782c AC |
450 | printf_filtered ("\n"); |
451 | ||
3e43a32a MS |
452 | printf_filtered (_(" \ |
453 | Total memory used by bcache, including overhead: %ld\n"), | |
c2d11a7d | 454 | c->structure_size); |
3d263c1d | 455 | printf_filtered (_(" Percentage memory overhead: ")); |
c2d11a7d | 456 | print_percentage (c->structure_size - c->unique_size, c->unique_size); |
3d263c1d | 457 | printf_filtered (_(" Net memory savings: ")); |
c2d11a7d JM |
458 | print_percentage (c->total_size - c->structure_size, c->total_size); |
459 | printf_filtered ("\n"); | |
460 | ||
4a64f543 MS |
461 | printf_filtered (_(" Hash table size: %3d\n"), |
462 | c->num_buckets); | |
3d263c1d | 463 | printf_filtered (_(" Hash table expands: %lu\n"), |
2160782c | 464 | c->expand_count); |
3d263c1d | 465 | printf_filtered (_(" Hash table hashes: %lu\n"), |
2160782c | 466 | c->total_count + c->expand_hash_count); |
3d263c1d | 467 | printf_filtered (_(" Half hash misses: %lu\n"), |
49df298f | 468 | c->half_hash_miss_count); |
3d263c1d | 469 | printf_filtered (_(" Hash table population: ")); |
c2d11a7d | 470 | print_percentage (occupied_buckets, c->num_buckets); |
3d263c1d | 471 | printf_filtered (_(" Median hash chain length: %3d\n"), |
c2d11a7d | 472 | median_chain_length); |
3d263c1d | 473 | printf_filtered (_(" Average hash chain length: ")); |
c2d11a7d | 474 | if (c->num_buckets > 0) |
745b8ca0 | 475 | printf_filtered ("%3lu\n", c->unique_count / c->num_buckets); |
c906108c | 476 | else |
4a64f543 | 477 | /* i18n: "Average hash chain length: (not applicable)". */ |
3d263c1d | 478 | printf_filtered (_("(not applicable)\n")); |
4a64f543 MS |
479 | printf_filtered (_(" Maximum hash chain length: %3d\n"), |
480 | max_chain_length); | |
c2d11a7d | 481 | printf_filtered ("\n"); |
c906108c | 482 | } |
af5f3db6 AC |
483 | |
484 | int | |
485 | bcache_memory_used (struct bcache *bcache) | |
486 | { | |
e6ad058a TT |
487 | if (bcache->total_count == 0) |
488 | return 0; | |
af5f3db6 AC |
489 | return obstack_memory_used (&bcache->cache); |
490 | } |