1 | /* A type-safe hash table template. |
2 | Copyright (C) 2012-2022 Free Software Foundation, Inc. |
3 | Contributed by Lawrence Crowl <crowl@google.com> |
4 | |
5 | This file is part of GCC. |
6 | |
7 | GCC is free software; you can redistribute it and/or modify it under |
8 | the terms of the GNU General Public License as published by the Free |
9 | Software Foundation; either version 3, or (at your option) any later |
10 | version. |
11 | |
12 | GCC is distributed in the hope that it will be useful, but WITHOUT ANY |
13 | WARRANTY; without even the implied warranty of MERCHANTABILITY or |
14 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
15 | for more details. |
16 | |
17 | You should have received a copy of the GNU General Public License |
18 | along with GCC; see the file COPYING3. If not see |
19 | <http://www.gnu.org/licenses/>. */ |
20 | |
21 | |
22 | /* This file implements a typed hash table. |
23 | The implementation borrows from libiberty's htab_t in hashtab.h. |
24 | |
25 | |
26 | INTRODUCTION TO TYPES |
27 | |
28 | Users of the hash table generally need to be aware of three types. |
29 | |
30 | 1. The type being placed into the hash table. This type is called |
31 | the value type. |
32 | |
33 | 2. The type used to describe how to handle the value type within |
34 | the hash table. This descriptor type provides the hash table with |
35 | several things. |
36 | |
37 | - A typedef named 'value_type' to the value type (from above). |
38 | Provided a suitable Descriptor class it may be a user-defined, |
39 | non-POD type. |
40 | |
41 | - A static member function named 'hash' that takes a value_type |
42 | (or 'const value_type &') and returns a hashval_t value. |
43 | |
44 | - A typedef named 'compare_type' that is used to test when a value |
45 | is found. This type is the comparison type. Usually, it will be |
46 | the same as value_type and may be a user-defined, non-POD type. |
47 | If it is not the same type, you must generally explicitly compute |
48 | hash values and pass them to the hash table. |
49 | |
50 | - A static member function named 'equal' that takes a value_type |
51 | and a compare_type, and returns a bool. Both arguments can be |
52 | const references. |
53 | |
54 | - A static function named 'remove' that takes an value_type pointer |
55 | and frees the memory allocated by it. This function is used when |
56 | individual elements of the table need to be disposed of (e.g., |
57 | when deleting a hash table, removing elements from the table, etc). |
58 | |
59 | - An optional static function named 'keep_cache_entry'. This |
60 | function is provided only for garbage-collected elements that |
61 | are not marked by the normal gc mark pass. It describes what |
62 | what should happen to the element at the end of the gc mark phase. |
63 | The return value should be: |
64 | - 0 if the element should be deleted |
65 | - 1 if the element should be kept and needs to be marked |
66 | - -1 if the element should be kept and is already marked. |
67 | Returning -1 rather than 1 is purely an optimization. |
68 | |
69 | 3. The type of the hash table itself. (More later.) |
70 | |
71 | In very special circumstances, users may need to know about a fourth type. |
72 | |
73 | 4. The template type used to describe how hash table memory |
74 | is allocated. This type is called the allocator type. It is |
75 | parameterized on the value type. It provides two functions: |
76 | |
77 | - A static member function named 'data_alloc'. This function |
78 | allocates the data elements in the table. |
79 | |
80 | - A static member function named 'data_free'. This function |
81 | deallocates the data elements in the table. |
82 | |
83 | Hash table are instantiated with two type arguments. |
84 | |
85 | * The descriptor type, (2) above. |
86 | |
87 | * The allocator type, (4) above. In general, you will not need to |
88 | provide your own allocator type. By default, hash tables will use |
89 | the class template xcallocator, which uses malloc/free for allocation. |
90 | |
91 | |
92 | DEFINING A DESCRIPTOR TYPE |
93 | |
94 | The first task in using the hash table is to describe the element type. |
95 | We compose this into a few steps. |
96 | |
97 | 1. Decide on a removal policy for values stored in the table. |
98 | hash-traits.h provides class templates for the four most common |
99 | policies: |
100 | |
101 | * typed_free_remove implements the static 'remove' member function |
102 | by calling free(). |
103 | |
104 | * typed_noop_remove implements the static 'remove' member function |
105 | by doing nothing. |
106 | |
107 | * ggc_remove implements the static 'remove' member by doing nothing, |
108 | but instead provides routines for gc marking and for PCH streaming. |
109 | Use this for garbage-collected data that needs to be preserved across |
110 | collections. |
111 | |
112 | * ggc_cache_remove is like ggc_remove, except that it does not |
113 | mark the entries during the normal gc mark phase. Instead it |
114 | uses 'keep_cache_entry' (described above) to keep elements that |
115 | were not collected and delete those that were. Use this for |
116 | garbage-collected caches that should not in themselves stop |
117 | the data from being collected. |
118 | |
119 | You can use these policies by simply deriving the descriptor type |
120 | from one of those class template, with the appropriate argument. |
121 | |
122 | Otherwise, you need to write the static 'remove' member function |
123 | in the descriptor class. |
124 | |
125 | 2. Choose a hash function. Write the static 'hash' member function. |
126 | |
127 | 3. Decide whether the lookup function should take as input an object |
128 | of type value_type or something more restricted. Define compare_type |
129 | accordingly. |
130 | |
131 | 4. Choose an equality testing function 'equal' that compares a value_type |
132 | and a compare_type. |
133 | |
134 | If your elements are pointers, it is usually easiest to start with one |
135 | of the generic pointer descriptors described below and override the bits |
136 | you need to change. |
137 | |
138 | AN EXAMPLE DESCRIPTOR TYPE |
139 | |
140 | Suppose you want to put some_type into the hash table. You could define |
141 | the descriptor type as follows. |
142 | |
143 | struct some_type_hasher : nofree_ptr_hash <some_type> |
144 | // Deriving from nofree_ptr_hash means that we get a 'remove' that does |
145 | // nothing. This choice is good for raw values. |
146 | { |
147 | static inline hashval_t hash (const value_type *); |
148 | static inline bool equal (const value_type *, const compare_type *); |
149 | }; |
150 | |
151 | inline hashval_t |
152 | some_type_hasher::hash (const value_type *e) |
153 | { ... compute and return a hash value for E ... } |
154 | |
155 | inline bool |
156 | some_type_hasher::equal (const value_type *p1, const compare_type *p2) |
157 | { ... compare P1 vs P2. Return true if they are the 'same' ... } |
158 | |
159 | |
160 | AN EXAMPLE HASH_TABLE DECLARATION |
161 | |
162 | To instantiate a hash table for some_type: |
163 | |
164 | hash_table <some_type_hasher> some_type_hash_table; |
165 | |
166 | There is no need to mention some_type directly, as the hash table will |
167 | obtain it using some_type_hasher::value_type. |
168 | |
169 | You can then use any of the functions in hash_table's public interface. |
170 | See hash_table for details. The interface is very similar to libiberty's |
171 | htab_t. |
172 | |
173 | If a hash table is used only in some rare cases, it is possible |
174 | to construct the hash_table lazily before first use. This is done |
175 | through: |
176 | |
177 | hash_table <some_type_hasher, true> some_type_hash_table; |
178 | |
179 | which will cause whatever methods actually need the allocated entries |
180 | array to allocate it later. |
181 | |
182 | |
183 | EASY DESCRIPTORS FOR POINTERS |
184 | |
185 | There are four descriptors for pointer elements, one for each of |
186 | the removal policies above: |
187 | |
188 | * nofree_ptr_hash (based on typed_noop_remove) |
189 | * free_ptr_hash (based on typed_free_remove) |
190 | * ggc_ptr_hash (based on ggc_remove) |
191 | * ggc_cache_ptr_hash (based on ggc_cache_remove) |
192 | |
193 | These descriptors hash and compare elements by their pointer value, |
194 | rather than what they point to. So, to instantiate a hash table over |
195 | pointers to whatever_type, without freeing the whatever_types, use: |
196 | |
197 | hash_table <nofree_ptr_hash <whatever_type> > whatever_type_hash_table; |
198 | |
199 | |
200 | HASH TABLE ITERATORS |
201 | |
202 | The hash table provides standard C++ iterators. For example, consider a |
203 | hash table of some_info. We wish to consume each element of the table: |
204 | |
205 | extern void consume (some_info *); |
206 | |
207 | We define a convenience typedef and the hash table: |
208 | |
209 | typedef hash_table <some_info_hasher> info_table_type; |
210 | info_table_type info_table; |
211 | |
212 | Then we write the loop in typical C++ style: |
213 | |
214 | for (info_table_type::iterator iter = info_table.begin (); |
215 | iter != info_table.end (); |
216 | ++iter) |
217 | if ((*iter).status == INFO_READY) |
218 | consume (&*iter); |
219 | |
220 | Or with common sub-expression elimination: |
221 | |
222 | for (info_table_type::iterator iter = info_table.begin (); |
223 | iter != info_table.end (); |
224 | ++iter) |
225 | { |
226 | some_info &elem = *iter; |
227 | if (elem.status == INFO_READY) |
228 | consume (&elem); |
229 | } |
230 | |
231 | One can also use a more typical GCC style: |
232 | |
233 | typedef some_info *some_info_p; |
234 | some_info *elem_ptr; |
235 | info_table_type::iterator iter; |
236 | FOR_EACH_HASH_TABLE_ELEMENT (info_table, elem_ptr, some_info_p, iter) |
237 | if (elem_ptr->status == INFO_READY) |
238 | consume (elem_ptr); |
239 | |
240 | */ |
241 | |
242 | |
243 | #ifndef TYPED_HASHTAB_H |
244 | #define TYPED_HASHTAB_H |
245 | |
246 | #include "statistics.h" |
247 | #include "ggc.h" |
248 | #include "vec.h" |
249 | #include "hashtab.h" |
250 | #include "inchash.h" |
251 | #include "mem-stats-traits.h" |
252 | #include "hash-traits.h" |
253 | #include "hash-map-traits.h" |
254 | |
255 | template<typename, typename, typename> class hash_map; |
256 | template<typename, bool, typename> class hash_set; |
257 | |
258 | /* The ordinary memory allocator. */ |
259 | /* FIXME (crowl): This allocator may be extracted for wider sharing later. */ |
260 | |
261 | template <typename Type> |
262 | struct xcallocator |
263 | { |
264 | static Type *data_alloc (size_t count); |
265 | static void data_free (Type *memory); |
266 | }; |
267 | |
268 | |
269 | /* Allocate memory for COUNT data blocks. */ |
270 | |
271 | template <typename Type> |
272 | inline Type * |
273 | xcallocator <Type>::data_alloc (size_t count) |
274 | { |
275 | return static_cast <Type *> (xcalloc (count, sizeof (Type))); |
276 | } |
277 | |
278 | |
279 | /* Free memory for data blocks. */ |
280 | |
281 | template <typename Type> |
282 | inline void |
283 | xcallocator <Type>::data_free (Type *memory) |
284 | { |
285 | return ::free (memory); |
286 | } |
287 | |
288 | |
289 | /* Table of primes and their inversion information. */ |
290 | |
291 | struct prime_ent |
292 | { |
293 | hashval_t prime; |
294 | hashval_t inv; |
295 | hashval_t inv_m2; /* inverse of prime-2 */ |
296 | hashval_t shift; |
297 | }; |
298 | |
299 | extern struct prime_ent const prime_tab[]; |
300 | |
301 | /* Limit number of comparisons when calling hash_table<>::verify. */ |
302 | extern unsigned int hash_table_sanitize_eq_limit; |
303 | |
304 | /* Functions for computing hash table indexes. */ |
305 | |
306 | extern unsigned int hash_table_higher_prime_index (unsigned long n) |
307 | ATTRIBUTE_PURE; |
308 | |
309 | extern ATTRIBUTE_NORETURN ATTRIBUTE_COLD void hashtab_chk_error (); |
310 | |
311 | /* Return X % Y using multiplicative inverse values INV and SHIFT. |
312 | |
313 | The multiplicative inverses computed above are for 32-bit types, |
314 | and requires that we be able to compute a highpart multiply. |
315 | |
316 | FIX: I am not at all convinced that |
317 | 3 loads, 2 multiplications, 3 shifts, and 3 additions |
318 | will be faster than |
319 | 1 load and 1 modulus |
320 | on modern systems running a compiler. */ |
321 | |
322 | inline hashval_t |
323 | mul_mod (hashval_t x, hashval_t y, hashval_t inv, int shift) |
324 | { |
325 | hashval_t t1, t2, t3, t4, q, r; |
326 | |
327 | t1 = ((uint64_t)x * inv) >> 32; |
328 | t2 = x - t1; |
329 | t3 = t2 >> 1; |
330 | t4 = t1 + t3; |
331 | q = t4 >> shift; |
332 | r = x - (q * y); |
333 | |
334 | return r; |
335 | } |
336 | |
337 | /* Compute the primary table index for HASH given current prime index. */ |
338 | |
339 | inline hashval_t |
340 | hash_table_mod1 (hashval_t hash, unsigned int index) |
341 | { |
342 | const struct prime_ent *p = &prime_tab[index]; |
343 | gcc_checking_assert (sizeof (hashval_t) * CHAR_BIT <= 32); |
344 | return mul_mod (hash, p->prime, p->inv, p->shift); |
345 | } |
346 | |
347 | /* Compute the secondary table index for HASH given current prime index. */ |
348 | |
349 | inline hashval_t |
350 | hash_table_mod2 (hashval_t hash, unsigned int index) |
351 | { |
352 | const struct prime_ent *p = &prime_tab[index]; |
353 | gcc_checking_assert (sizeof (hashval_t) * CHAR_BIT <= 32); |
354 | return 1 + mul_mod (hash, p->prime - 2, p->inv_m2, p->shift); |
355 | } |
356 | |
357 | class mem_usage; |
358 | |
359 | /* User-facing hash table type. |
360 | |
361 | The table stores elements of type Descriptor::value_type and uses |
362 | the static descriptor functions described at the top of the file |
363 | to hash, compare and remove elements. |
364 | |
365 | Specify the template Allocator to allocate and free memory. |
366 | The default is xcallocator. |
367 | |
368 | Storage is an implementation detail and should not be used outside the |
369 | hash table code. |
370 | |
371 | */ |
372 | template <typename Descriptor, bool Lazy = false, |
373 | template<typename Type> class Allocator = xcallocator> |
374 | class hash_table |
375 | { |
376 | typedef typename Descriptor::value_type value_type; |
377 | typedef typename Descriptor::compare_type compare_type; |
378 | |
379 | public: |
380 | explicit hash_table (size_t, bool ggc = false, |
381 | bool sanitize_eq_and_hash = true, |
382 | bool gather_mem_stats = GATHER_STATISTICS, |
383 | mem_alloc_origin origin = HASH_TABLE_ORIGIN |
384 | CXX_MEM_STAT_INFO); |
385 | explicit hash_table (const hash_table &, bool ggc = false, |
386 | bool sanitize_eq_and_hash = true, |
387 | bool gather_mem_stats = GATHER_STATISTICS, |
388 | mem_alloc_origin origin = HASH_TABLE_ORIGIN |
389 | CXX_MEM_STAT_INFO); |
390 | ~hash_table (); |
391 | |
392 | /* Create a hash_table in gc memory. */ |
393 | static hash_table * |
394 | create_ggc (size_t n, bool sanitize_eq_and_hash = true CXX_MEM_STAT_INFO) |
395 | { |
396 | hash_table *table = ggc_alloc<hash_table> (); |
397 | new (table) hash_table (n, true, sanitize_eq_and_hash, GATHER_STATISTICS, |
398 | HASH_TABLE_ORIGIN PASS_MEM_STAT); |
399 | return table; |
400 | } |
401 | |
402 | /* Current size (in entries) of the hash table. */ |
403 | size_t size () const { return m_size; } |
404 | |
405 | /* Return the current number of elements in this hash table. */ |
406 | size_t elements () const { return m_n_elements - m_n_deleted; } |
407 | |
408 | /* Return the current number of elements in this hash table. */ |
409 | size_t elements_with_deleted () const { return m_n_elements; } |
410 | |
411 | /* This function clears all entries in this hash table. */ |
412 | void empty () { if (elements ()) empty_slow (); } |
413 | |
414 | /* Return true when there are no elements in this hash table. */ |
415 | bool is_empty () const { return elements () == 0; } |
416 | |
417 | /* This function clears a specified SLOT in a hash table. It is |
418 | useful when you've already done the lookup and don't want to do it |
419 | again. */ |
420 | void clear_slot (value_type *); |
421 | |
422 | /* This function searches for a hash table entry equal to the given |
423 | COMPARABLE element starting with the given HASH value. It cannot |
424 | be used to insert or delete an element. */ |
425 | value_type &find_with_hash (const compare_type &, hashval_t); |
426 | |
427 | /* Like find_slot_with_hash, but compute the hash value from the element. */ |
428 | value_type &find (const value_type &value) |
429 | { |
430 | return find_with_hash (value, Descriptor::hash (value)); |
431 | } |
432 | |
433 | value_type *find_slot (const value_type &value, insert_option insert) |
434 | { |
435 | return find_slot_with_hash (value, Descriptor::hash (value), insert); |
436 | } |
437 | |
438 | /* This function searches for a hash table slot containing an entry |
439 | equal to the given COMPARABLE element and starting with the given |
440 | HASH. To delete an entry, call this with insert=NO_INSERT, then |
441 | call clear_slot on the slot returned (possibly after doing some |
442 | checks). To insert an entry, call this with insert=INSERT, then |
443 | write the value you want into the returned slot. When inserting an |
444 | entry, NULL may be returned if memory allocation fails. */ |
445 | value_type *find_slot_with_hash (const compare_type &comparable, |
446 | hashval_t hash, enum insert_option insert); |
447 | |
448 | /* This function deletes an element with the given COMPARABLE value |
449 | from hash table starting with the given HASH. If there is no |
450 | matching element in the hash table, this function does nothing. */ |
451 | void remove_elt_with_hash (const compare_type &, hashval_t); |
452 | |
453 | /* Like remove_elt_with_hash, but compute the hash value from the |
454 | element. */ |
455 | void remove_elt (const value_type &value) |
456 | { |
457 | remove_elt_with_hash (value, Descriptor::hash (value)); |
458 | } |
459 | |
460 | /* This function scans over the entire hash table calling CALLBACK for |
461 | each live entry. If CALLBACK returns false, the iteration stops. |
462 | ARGUMENT is passed as CALLBACK's second argument. */ |
463 | template <typename Argument, |
464 | int (*Callback) (value_type *slot, Argument argument)> |
465 | void traverse_noresize (Argument argument); |
466 | |
467 | /* Like traverse_noresize, but does resize the table when it is too empty |
468 | to improve effectivity of subsequent calls. */ |
469 | template <typename Argument, |
470 | int (*Callback) (value_type *slot, Argument argument)> |
471 | void traverse (Argument argument); |
472 | |
473 | class iterator |
474 | { |
475 | public: |
476 | iterator () : m_slot (NULL), m_limit (NULL) {} |
477 | |
478 | iterator (value_type *slot, value_type *limit) : |
479 | m_slot (slot), m_limit (limit) {} |
480 | |
481 | inline value_type &operator * () { return *m_slot; } |
482 | void slide (); |
483 | inline iterator &operator ++ (); |
484 | bool operator != (const iterator &other) const |
485 | { |
486 | return m_slot != other.m_slot || m_limit != other.m_limit; |
487 | } |
488 | |
489 | private: |
490 | value_type *m_slot; |
491 | value_type *m_limit; |
492 | }; |
493 | |
494 | iterator begin () const |
495 | { |
496 | if (Lazy && m_entries == NULL) |
497 | return iterator (); |
498 | iterator iter (m_entries, m_entries + m_size); |
499 | iter.slide (); |
500 | return iter; |
501 | } |
502 | |
503 | iterator end () const { return iterator (); } |
504 | |
505 | double collisions () const |
506 | { |
507 | return m_searches ? static_cast <double> (m_collisions) / m_searches : 0; |
508 | } |
509 | |
510 | private: |
511 | /* FIXME: Make the class assignable. See pr90959. */ |
512 | void operator= (hash_table&); |
513 | |
514 | template<typename T> friend void gt_ggc_mx (hash_table<T> *); |
515 | template<typename T> friend void gt_pch_nx (hash_table<T> *); |
516 | template<typename T> friend void |
517 | hashtab_entry_note_pointers (void *, void *, gt_pointer_operator, void *); |
518 | template<typename T, typename U, typename V> friend void |
519 | gt_pch_nx (hash_map<T, U, V> *, gt_pointer_operator, void *); |
520 | template<typename T, typename U> |
521 | friend void gt_pch_nx (hash_set<T, false, U> *, gt_pointer_operator, void *); |
522 | template<typename T> friend void gt_pch_nx (hash_table<T> *, |
523 | gt_pointer_operator, void *); |
524 | |
525 | template<typename T> friend void gt_cleare_cache (hash_table<T> *); |
526 | |
527 | void empty_slow (); |
528 | |
529 | value_type *alloc_entries (size_t n CXX_MEM_STAT_INFO) const; |
530 | value_type *find_empty_slot_for_expand (hashval_t); |
531 | void verify (const compare_type &comparable, hashval_t hash); |
532 | bool too_empty_p (unsigned int); |
533 | void expand (); |
534 | static bool is_deleted (value_type &v) |
535 | { |
536 | return Descriptor::is_deleted (v); |
537 | } |
538 | |
539 | static bool is_empty (value_type &v) |
540 | { |
541 | return Descriptor::is_empty (v); |
542 | } |
543 | |
544 | static void mark_deleted (value_type &v) |
545 | { |
546 | Descriptor::mark_deleted (v); |
547 | } |
548 | |
549 | static void mark_empty (value_type &v) |
550 | { |
551 | Descriptor::mark_empty (v); |
552 | } |
553 | |
554 | /* Table itself. */ |
555 | typename Descriptor::value_type *m_entries; |
556 | |
557 | size_t m_size; |
558 | |
559 | /* Current number of elements including also deleted elements. */ |
560 | size_t m_n_elements; |
561 | |
562 | /* Current number of deleted elements in the table. */ |
563 | size_t m_n_deleted; |
564 | |
565 | /* The following member is used for debugging. Its value is number |
566 | of all calls of `htab_find_slot' for the hash table. */ |
567 | unsigned int m_searches; |
568 | |
569 | /* The following member is used for debugging. Its value is number |
570 | of collisions fixed for time of work with the hash table. */ |
571 | unsigned int m_collisions; |
572 | |
573 | /* Current size (in entries) of the hash table, as an index into the |
574 | table of primes. */ |
575 | unsigned int m_size_prime_index; |
576 | |
577 | /* if m_entries is stored in ggc memory. */ |
578 | bool m_ggc; |
579 | |
580 | /* True if the table should be sanitized for equal and hash functions. */ |
581 | bool m_sanitize_eq_and_hash; |
582 | |
583 | /* If we should gather memory statistics for the table. */ |
584 | #if GATHER_STATISTICS |
585 | bool m_gather_mem_stats; |
586 | #else |
587 | static const bool m_gather_mem_stats = false; |
588 | #endif |
589 | }; |
590 | |
591 | /* As mem-stats.h heavily utilizes hash maps (hash tables), we have to include |
592 | mem-stats.h after hash_table declaration. */ |
593 | |
594 | #include "mem-stats.h" |
595 | #include "hash-map.h" |
596 | |
597 | extern mem_alloc_description<mem_usage>& hash_table_usage (void); |
598 | |
599 | /* Support function for statistics. */ |
600 | extern void dump_hash_table_loc_statistics (void); |
601 | |
602 | template<typename Descriptor, bool Lazy, |
603 | template<typename Type> class Allocator> |
604 | hash_table<Descriptor, Lazy, Allocator>::hash_table (size_t size, bool ggc, |
605 | bool sanitize_eq_and_hash, |
606 | bool gather_mem_stats |
607 | ATTRIBUTE_UNUSED, |
608 | mem_alloc_origin origin |
609 | MEM_STAT_DECL) : |
610 | m_n_elements (0), m_n_deleted (0), m_searches (0), m_collisions (0), |
611 | m_ggc (ggc), m_sanitize_eq_and_hash (sanitize_eq_and_hash) |
612 | #if GATHER_STATISTICS |
613 | , m_gather_mem_stats (gather_mem_stats) |
614 | #endif |
615 | { |
616 | unsigned int size_prime_index; |
617 | |
618 | size_prime_index = hash_table_higher_prime_index (size); |
619 | size = prime_tab[size_prime_index].prime; |
620 | |
621 | if (m_gather_mem_stats) |
622 | hash_table_usage ().register_descriptor (this, origin, ggc |
623 | FINAL_PASS_MEM_STAT); |
624 | |
625 | if (Lazy) |
626 | m_entries = NULL; |
627 | else |
628 | m_entries = alloc_entries (size PASS_MEM_STAT); |
629 | m_size = size; |
630 | m_size_prime_index = size_prime_index; |
631 | } |
632 | |
633 | template<typename Descriptor, bool Lazy, |
634 | template<typename Type> class Allocator> |
635 | hash_table<Descriptor, Lazy, Allocator>::hash_table (const hash_table &h, |
636 | bool ggc, |
637 | bool sanitize_eq_and_hash, |
638 | bool gather_mem_stats |
639 | ATTRIBUTE_UNUSED, |
640 | mem_alloc_origin origin |
641 | MEM_STAT_DECL) : |
642 | m_n_elements (h.m_n_elements), m_n_deleted (h.m_n_deleted), |
643 | m_searches (0), m_collisions (0), m_ggc (ggc), |
644 | m_sanitize_eq_and_hash (sanitize_eq_and_hash) |
645 | #if GATHER_STATISTICS |
646 | , m_gather_mem_stats (gather_mem_stats) |
647 | #endif |
648 | { |
649 | size_t size = h.m_size; |
650 | |
651 | if (m_gather_mem_stats) |
652 | hash_table_usage ().register_descriptor (this, origin, ggc |
653 | FINAL_PASS_MEM_STAT); |
654 | |
655 | if (Lazy && h.m_entries == NULL) |
656 | m_entries = NULL; |
657 | else |
658 | { |
659 | value_type *nentries = alloc_entries (size PASS_MEM_STAT); |
660 | for (size_t i = 0; i < size; ++i) |
661 | { |
662 | value_type &entry = h.m_entries[i]; |
663 | if (is_deleted (entry)) |
664 | mark_deleted (nentries[i]); |
665 | else if (!is_empty (entry)) |
666 | new ((void*) (nentries + i)) value_type (entry); |
667 | } |
668 | m_entries = nentries; |
669 | } |
670 | m_size = size; |
671 | m_size_prime_index = h.m_size_prime_index; |
672 | } |
673 | |
674 | template<typename Descriptor, bool Lazy, |
675 | template<typename Type> class Allocator> |
676 | hash_table<Descriptor, Lazy, Allocator>::~hash_table () |
677 | { |
678 | if (!Lazy || m_entries) |
679 | { |
680 | for (size_t i = m_size - 1; i < m_size; i--) |
681 | if (!is_empty (m_entries[i]) && !is_deleted (m_entries[i])) |
682 | Descriptor::remove (m_entries[i]); |
683 | |
684 | if (!m_ggc) |
685 | Allocator <value_type> ::data_free (m_entries); |
686 | else |
687 | ggc_free (m_entries); |
688 | if (m_gather_mem_stats) |
689 | hash_table_usage ().release_instance_overhead (this, |
690 | sizeof (value_type) |
691 | * m_size, true); |
692 | } |
693 | else if (m_gather_mem_stats) |
694 | hash_table_usage ().unregister_descriptor (this); |
695 | } |
696 | |
697 | /* This function returns an array of empty hash table elements. */ |
698 | |
699 | template<typename Descriptor, bool Lazy, |
700 | template<typename Type> class Allocator> |
701 | inline typename hash_table<Descriptor, Lazy, Allocator>::value_type * |
702 | hash_table<Descriptor, Lazy, |
703 | Allocator>::alloc_entries (size_t n MEM_STAT_DECL) const |
704 | { |
705 | value_type *nentries; |
706 | |
707 | if (m_gather_mem_stats) |
708 | hash_table_usage ().register_instance_overhead (sizeof (value_type) * n, this); |
709 | |
710 | if (!m_ggc) |
711 | nentries = Allocator <value_type> ::data_alloc (n); |
712 | else |
713 | nentries = ::ggc_cleared_vec_alloc<value_type> (n PASS_MEM_STAT); |
714 | |
715 | gcc_assert (nentries != NULL); |
716 | if (!Descriptor::empty_zero_p) |
717 | for (size_t i = 0; i < n; i++) |
718 | mark_empty (nentries[i]); |
719 | |
720 | return nentries; |
721 | } |
722 | |
723 | /* Similar to find_slot, but without several unwanted side effects: |
724 | - Does not call equal when it finds an existing entry. |
725 | - Does not change the count of elements/searches/collisions in the |
726 | hash table. |
727 | This function also assumes there are no deleted entries in the table. |
728 | HASH is the hash value for the element to be inserted. */ |
729 | |
730 | template<typename Descriptor, bool Lazy, |
731 | template<typename Type> class Allocator> |
732 | typename hash_table<Descriptor, Lazy, Allocator>::value_type * |
733 | hash_table<Descriptor, Lazy, |
734 | Allocator>::find_empty_slot_for_expand (hashval_t hash) |
735 | { |
736 | hashval_t index = hash_table_mod1 (hash, m_size_prime_index); |
737 | size_t size = m_size; |
738 | value_type *slot = m_entries + index; |
739 | hashval_t hash2; |
740 | |
741 | if (is_empty (*slot)) |
742 | return slot; |
743 | gcc_checking_assert (!is_deleted (*slot)); |
744 | |
745 | hash2 = hash_table_mod2 (hash, m_size_prime_index); |
746 | for (;;) |
747 | { |
748 | index += hash2; |
749 | if (index >= size) |
750 | index -= size; |
751 | |
752 | slot = m_entries + index; |
753 | if (is_empty (*slot)) |
754 | return slot; |
755 | gcc_checking_assert (!is_deleted (*slot)); |
756 | } |
757 | } |
758 | |
759 | /* Return true if the current table is excessively big for ELTS elements. */ |
760 | |
761 | template<typename Descriptor, bool Lazy, |
762 | template<typename Type> class Allocator> |
763 | inline bool |
764 | hash_table<Descriptor, Lazy, Allocator>::too_empty_p (unsigned int elts) |
765 | { |
766 | return elts * 8 < m_size && m_size > 32; |
767 | } |
768 | |
769 | /* The following function changes size of memory allocated for the |
770 | entries and repeatedly inserts the table elements. The occupancy |
771 | of the table after the call will be about 50%. Naturally the hash |
772 | table must already exist. Remember also that the place of the |
773 | table entries is changed. If memory allocation fails, this function |
774 | will abort. */ |
775 | |
776 | template<typename Descriptor, bool Lazy, |
777 | template<typename Type> class Allocator> |
778 | void |
779 | hash_table<Descriptor, Lazy, Allocator>::expand () |
780 | { |
781 | value_type *oentries = m_entries; |
782 | unsigned int oindex = m_size_prime_index; |
783 | size_t osize = size (); |
784 | value_type *olimit = oentries + osize; |
785 | size_t elts = elements (); |
786 | |
787 | /* Resize only when table after removal of unused elements is either |
788 | too full or too empty. */ |
789 | unsigned int nindex; |
790 | size_t nsize; |
791 | if (elts * 2 > osize || too_empty_p (elts)) |
792 | { |
793 | nindex = hash_table_higher_prime_index (elts * 2); |
794 | nsize = prime_tab[nindex].prime; |
795 | } |
796 | else |
797 | { |
798 | nindex = oindex; |
799 | nsize = osize; |
800 | } |
801 | |
802 | value_type *nentries = alloc_entries (nsize); |
803 | |
804 | if (m_gather_mem_stats) |
805 | hash_table_usage ().release_instance_overhead (this, sizeof (value_type) |
806 | * osize); |
807 | |
808 | m_entries = nentries; |
809 | m_size = nsize; |
810 | m_size_prime_index = nindex; |
811 | m_n_elements -= m_n_deleted; |
812 | m_n_deleted = 0; |
813 | |
814 | value_type *p = oentries; |
815 | do |
816 | { |
817 | value_type &x = *p; |
818 | |
819 | if (!is_empty (x) && !is_deleted (x)) |
820 | { |
821 | value_type *q = find_empty_slot_for_expand (Descriptor::hash (x)); |
822 | new ((void*) q) value_type (std::move (x)); |
823 | /* After the resources of 'x' have been moved to a new object at 'q', |
824 | we now have to destroy the 'x' object, to end its lifetime. */ |
825 | x.~value_type (); |
826 | } |
827 | |
828 | p++; |
829 | } |
830 | while (p < olimit); |
831 | |
832 | if (!m_ggc) |
833 | Allocator <value_type> ::data_free (oentries); |
834 | else |
835 | ggc_free (oentries); |
836 | } |
837 | |
838 | /* Implements empty() in cases where it isn't a no-op. */ |
839 | |
840 | template<typename Descriptor, bool Lazy, |
841 | template<typename Type> class Allocator> |
842 | void |
843 | hash_table<Descriptor, Lazy, Allocator>::empty_slow () |
844 | { |
845 | size_t size = m_size; |
846 | size_t nsize = size; |
847 | value_type *entries = m_entries; |
848 | |
849 | for (size_t i = size - 1; i < size; i--) |
850 | if (!is_empty (entries[i]) && !is_deleted (entries[i])) |
851 | Descriptor::remove (entries[i]); |
852 | |
853 | /* Instead of clearing megabyte, downsize the table. */ |
854 | if (size > 1024*1024 / sizeof (value_type)) |
855 | nsize = 1024 / sizeof (value_type); |
856 | else if (too_empty_p (m_n_elements)) |
857 | nsize = m_n_elements * 2; |
858 | |
859 | if (nsize != size) |
860 | { |
861 | unsigned int nindex = hash_table_higher_prime_index (nsize); |
862 | |
863 | nsize = prime_tab[nindex].prime; |
864 | |
865 | if (!m_ggc) |
866 | Allocator <value_type> ::data_free (m_entries); |
867 | else |
868 | ggc_free (m_entries); |
869 | |
870 | m_entries = alloc_entries (nsize); |
871 | m_size = nsize; |
872 | m_size_prime_index = nindex; |
873 | } |
874 | else if (Descriptor::empty_zero_p) |
875 | memset ((void *) entries, 0, size * sizeof (value_type)); |
876 | else |
877 | for (size_t i = 0; i < size; i++) |
878 | mark_empty (entries[i]); |
879 | |
880 | m_n_deleted = 0; |
881 | m_n_elements = 0; |
882 | } |
883 | |
884 | /* This function clears a specified SLOT in a hash table. It is |
885 | useful when you've already done the lookup and don't want to do it |
886 | again. */ |
887 | |
888 | template<typename Descriptor, bool Lazy, |
889 | template<typename Type> class Allocator> |
890 | void |
891 | hash_table<Descriptor, Lazy, Allocator>::clear_slot (value_type *slot) |
892 | { |
893 | gcc_checking_assert (!(slot < m_entries || slot >= m_entries + size () |
894 | || is_empty (*slot) || is_deleted (*slot))); |
895 | |
896 | Descriptor::remove (*slot); |
897 | |
898 | mark_deleted (*slot); |
899 | m_n_deleted++; |
900 | } |
901 | |
902 | /* This function searches for a hash table entry equal to the given |
903 | COMPARABLE element starting with the given HASH value. It cannot |
904 | be used to insert or delete an element. */ |
905 | |
906 | template<typename Descriptor, bool Lazy, |
907 | template<typename Type> class Allocator> |
908 | typename hash_table<Descriptor, Lazy, Allocator>::value_type & |
909 | hash_table<Descriptor, Lazy, Allocator> |
910 | ::find_with_hash (const compare_type &comparable, hashval_t hash) |
911 | { |
912 | m_searches++; |
913 | size_t size = m_size; |
914 | hashval_t index = hash_table_mod1 (hash, m_size_prime_index); |
915 | |
916 | if (Lazy && m_entries == NULL) |
917 | m_entries = alloc_entries (size); |
918 | |
919 | #if CHECKING_P |
920 | if (m_sanitize_eq_and_hash) |
921 | verify (comparable, hash); |
922 | #endif |
923 | |
924 | value_type *entry = &m_entries[index]; |
925 | if (is_empty (*entry) |
926 | || (!is_deleted (*entry) && Descriptor::equal (*entry, comparable))) |
927 | return *entry; |
928 | |
929 | hashval_t hash2 = hash_table_mod2 (hash, m_size_prime_index); |
930 | for (;;) |
931 | { |
932 | m_collisions++; |
933 | index += hash2; |
934 | if (index >= size) |
935 | index -= size; |
936 | |
937 | entry = &m_entries[index]; |
938 | if (is_empty (*entry) |
939 | || (!is_deleted (*entry) && Descriptor::equal (*entry, comparable))) |
940 | return *entry; |
941 | } |
942 | } |
943 | |
944 | /* This function searches for a hash table slot containing an entry |
945 | equal to the given COMPARABLE element and starting with the given |
946 | HASH. To delete an entry, call this with insert=NO_INSERT, then |
947 | call clear_slot on the slot returned (possibly after doing some |
948 | checks). To insert an entry, call this with insert=INSERT, then |
949 | write the value you want into the returned slot. When inserting an |
950 | entry, NULL may be returned if memory allocation fails. */ |
951 | |
952 | template<typename Descriptor, bool Lazy, |
953 | template<typename Type> class Allocator> |
954 | typename hash_table<Descriptor, Lazy, Allocator>::value_type * |
955 | hash_table<Descriptor, Lazy, Allocator> |
956 | ::find_slot_with_hash (const compare_type &comparable, hashval_t hash, |
957 | enum insert_option insert) |
958 | { |
959 | if (Lazy && m_entries == NULL) |
960 | { |
961 | if (insert == INSERT) |
962 | m_entries = alloc_entries (m_size); |
963 | else |
964 | return NULL; |
965 | } |
966 | if (insert == INSERT && m_size * 3 <= m_n_elements * 4) |
967 | expand (); |
968 | |
969 | #if CHECKING_P |
970 | if (m_sanitize_eq_and_hash) |
971 | verify (comparable, hash); |
972 | #endif |
973 | |
974 | m_searches++; |
975 | value_type *first_deleted_slot = NULL; |
976 | hashval_t index = hash_table_mod1 (hash, m_size_prime_index); |
977 | hashval_t hash2 = hash_table_mod2 (hash, m_size_prime_index); |
978 | value_type *entry = &m_entries[index]; |
979 | size_t size = m_size; |
980 | if (is_empty (*entry)) |
981 | goto empty_entry; |
982 | else if (is_deleted (*entry)) |
983 | first_deleted_slot = &m_entries[index]; |
984 | else if (Descriptor::equal (*entry, comparable)) |
985 | return &m_entries[index]; |
986 | |
987 | for (;;) |
988 | { |
989 | m_collisions++; |
990 | index += hash2; |
991 | if (index >= size) |
992 | index -= size; |
993 | |
994 | entry = &m_entries[index]; |
995 | if (is_empty (*entry)) |
996 | goto empty_entry; |
997 | else if (is_deleted (*entry)) |
998 | { |
999 | if (!first_deleted_slot) |
1000 | first_deleted_slot = &m_entries[index]; |
1001 | } |
1002 | else if (Descriptor::equal (*entry, comparable)) |
1003 | return &m_entries[index]; |
1004 | } |
1005 | |
1006 | empty_entry: |
1007 | if (insert == NO_INSERT) |
1008 | return NULL; |
1009 | |
1010 | if (first_deleted_slot) |
1011 | { |
1012 | m_n_deleted--; |
1013 | mark_empty (*first_deleted_slot); |
1014 | return first_deleted_slot; |
1015 | } |
1016 | |
1017 | m_n_elements++; |
1018 | return &m_entries[index]; |
1019 | } |
1020 | |
1021 | /* Verify that all existing elements in th hash table which are |
1022 | equal to COMPARABLE have an equal HASH value provided as argument. */ |
1023 | |
1024 | template<typename Descriptor, bool Lazy, |
1025 | template<typename Type> class Allocator> |
1026 | void |
1027 | hash_table<Descriptor, Lazy, Allocator> |
1028 | ::verify (const compare_type &comparable, hashval_t hash) |
1029 | { |
1030 | for (size_t i = 0; i < MIN (hash_table_sanitize_eq_limit, m_size); i++) |
1031 | { |
1032 | value_type *entry = &m_entries[i]; |
1033 | if (!is_empty (*entry) && !is_deleted (*entry) |
1034 | && hash != Descriptor::hash (*entry) |
1035 | && Descriptor::equal (*entry, comparable)) |
1036 | hashtab_chk_error (); |
1037 | } |
1038 | } |
1039 | |
1040 | /* This function deletes an element with the given COMPARABLE value |
1041 | from hash table starting with the given HASH. If there is no |
1042 | matching element in the hash table, this function does nothing. */ |
1043 | |
1044 | template<typename Descriptor, bool Lazy, |
1045 | template<typename Type> class Allocator> |
1046 | void |
1047 | hash_table<Descriptor, Lazy, Allocator> |
1048 | ::remove_elt_with_hash (const compare_type &comparable, hashval_t hash) |
1049 | { |
1050 | value_type *slot = find_slot_with_hash (comparable, hash, NO_INSERT); |
1051 | if (slot == NULL) |
1052 | return; |
1053 | |
1054 | Descriptor::remove (*slot); |
1055 | |
1056 | mark_deleted (*slot); |
1057 | m_n_deleted++; |
1058 | } |
1059 | |
1060 | /* This function scans over the entire hash table calling CALLBACK for |
1061 | each live entry. If CALLBACK returns false, the iteration stops. |
1062 | ARGUMENT is passed as CALLBACK's second argument. */ |
1063 | |
1064 | template<typename Descriptor, bool Lazy, |
1065 | template<typename Type> class Allocator> |
1066 | template<typename Argument, |
1067 | int (*Callback) |
1068 | (typename hash_table<Descriptor, Lazy, Allocator>::value_type *slot, |
1069 | Argument argument)> |
1070 | void |
1071 | hash_table<Descriptor, Lazy, Allocator>::traverse_noresize (Argument argument) |
1072 | { |
1073 | if (Lazy && m_entries == NULL) |
1074 | return; |
1075 | |
1076 | value_type *slot = m_entries; |
1077 | value_type *limit = slot + size (); |
1078 | |
1079 | do |
1080 | { |
1081 | value_type &x = *slot; |
1082 | |
1083 | if (!is_empty (x) && !is_deleted (x)) |
1084 | if (! Callback (slot, argument)) |
1085 | break; |
1086 | } |
1087 | while (++slot < limit); |
1088 | } |
1089 | |
1090 | /* Like traverse_noresize, but does resize the table when it is too empty |
1091 | to improve effectivity of subsequent calls. */ |
1092 | |
1093 | template <typename Descriptor, bool Lazy, |
1094 | template <typename Type> class Allocator> |
1095 | template <typename Argument, |
1096 | int (*Callback) |
1097 | (typename hash_table<Descriptor, Lazy, Allocator>::value_type *slot, |
1098 | Argument argument)> |
1099 | void |
1100 | hash_table<Descriptor, Lazy, Allocator>::traverse (Argument argument) |
1101 | { |
1102 | if (too_empty_p (elements ()) && (!Lazy || m_entries)) |
1103 | expand (); |
1104 | |
1105 | traverse_noresize <Argument, Callback> (argument); |
1106 | } |
1107 | |
1108 | /* Slide down the iterator slots until an active entry is found. */ |
1109 | |
1110 | template<typename Descriptor, bool Lazy, |
1111 | template<typename Type> class Allocator> |
1112 | void |
1113 | hash_table<Descriptor, Lazy, Allocator>::iterator::slide () |
1114 | { |
1115 | for ( ; m_slot < m_limit; ++m_slot ) |
1116 | { |
1117 | value_type &x = *m_slot; |
1118 | if (!is_empty (x) && !is_deleted (x)) |
1119 | return; |
1120 | } |
1121 | m_slot = NULL; |
1122 | m_limit = NULL; |
1123 | } |
1124 | |
1125 | /* Bump the iterator. */ |
1126 | |
1127 | template<typename Descriptor, bool Lazy, |
1128 | template<typename Type> class Allocator> |
1129 | inline typename hash_table<Descriptor, Lazy, Allocator>::iterator & |
1130 | hash_table<Descriptor, Lazy, Allocator>::iterator::operator ++ () |
1131 | { |
1132 | ++m_slot; |
1133 | slide (); |
1134 | return *this; |
1135 | } |
1136 | |
1137 | |
1138 | /* Iterate through the elements of hash_table HTAB, |
1139 | using hash_table <....>::iterator ITER, |
1140 | storing each element in RESULT, which is of type TYPE. */ |
1141 | |
1142 | #define FOR_EACH_HASH_TABLE_ELEMENT(HTAB, RESULT, TYPE, ITER) \ |
1143 | for ((ITER) = (HTAB).begin (); \ |
1144 | (ITER) != (HTAB).end () ? (RESULT = *(ITER) , true) : false; \ |
1145 | ++(ITER)) |
1146 | |
1147 | /* ggc walking routines. */ |
1148 | |
1149 | template<typename E> |
1150 | static inline void |
1151 | gt_ggc_mx (hash_table<E> *h) |
1152 | { |
1153 | typedef hash_table<E> table; |
1154 | |
1155 | if (!ggc_test_and_set_mark (h->m_entries)) |
1156 | return; |
1157 | |
1158 | for (size_t i = 0; i < h->m_size; i++) |
1159 | { |
1160 | if (table::is_empty (h->m_entries[i]) |
1161 | || table::is_deleted (h->m_entries[i])) |
1162 | continue; |
1163 | |
1164 | /* Use ggc_maxbe_mx so we don't mark right away for cache tables; we'll |
1165 | mark in gt_cleare_cache if appropriate. */ |
1166 | E::ggc_maybe_mx (h->m_entries[i]); |
1167 | } |
1168 | } |
1169 | |
1170 | template<typename D> |
1171 | static inline void |
1172 | hashtab_entry_note_pointers (void *obj, void *h, gt_pointer_operator op, |
1173 | void *cookie) |
1174 | { |
1175 | hash_table<D> *map = static_cast<hash_table<D> *> (h); |
1176 | gcc_checking_assert (map->m_entries == obj); |
1177 | for (size_t i = 0; i < map->m_size; i++) |
1178 | { |
1179 | typedef hash_table<D> table; |
1180 | if (table::is_empty (map->m_entries[i]) |
1181 | || table::is_deleted (map->m_entries[i])) |
1182 | continue; |
1183 | |
1184 | D::pch_nx (map->m_entries[i], op, cookie); |
1185 | } |
1186 | } |
1187 | |
1188 | template<typename D> |
1189 | static void |
1190 | gt_pch_nx (hash_table<D> *h) |
1191 | { |
1192 | bool success |
1193 | = gt_pch_note_object (h->m_entries, h, hashtab_entry_note_pointers<D>); |
1194 | gcc_checking_assert (success); |
1195 | for (size_t i = 0; i < h->m_size; i++) |
1196 | { |
1197 | if (hash_table<D>::is_empty (h->m_entries[i]) |
1198 | || hash_table<D>::is_deleted (h->m_entries[i])) |
1199 | continue; |
1200 | |
1201 | D::pch_nx (h->m_entries[i]); |
1202 | } |
1203 | } |
1204 | |
1205 | template<typename D> |
1206 | static inline void |
1207 | gt_pch_nx (hash_table<D> *h, gt_pointer_operator op, void *cookie) |
1208 | { |
1209 | op (&h->m_entries, NULL, cookie); |
1210 | } |
1211 | |
1212 | template<typename H> |
1213 | inline void |
1214 | gt_cleare_cache (hash_table<H> *h) |
1215 | { |
1216 | typedef hash_table<H> table; |
1217 | if (!h) |
1218 | return; |
1219 | |
1220 | for (typename table::iterator iter = h->begin (); iter != h->end (); ++iter) |
1221 | if (!table::is_empty (*iter) && !table::is_deleted (*iter)) |
1222 | { |
1223 | int res = H::keep_cache_entry (*iter); |
1224 | if (res == 0) |
1225 | h->clear_slot (&*iter); |
1226 | else if (res != -1) |
1227 | H::ggc_mx (*iter); |
1228 | } |
1229 | } |
1230 | |
1231 | #endif /* TYPED_HASHTAB_H */ |
1232 | |