1 /* Implement a cached obstack.
2 Written by Fred Fish <fnf@cygnus.com>
3 Rewritten by Jim Blandy <jimb@cygnus.com>
5 Copyright (C) 1999-2013 Free Software Foundation, Inc.
7 This file is part of GDB.
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
11 the Free Software Foundation; either version 3 of the License, or
12 (at your option) any later version.
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.
19 You should have received a copy of the GNU General Public License
20 along with this program. If not, see <http://www.gnu.org/licenses/>. */
23 #include "gdb_obstack.h"
25 #include "gdb_string.h" /* For memcpy declaration */
26 #include "gdb_assert.h"
31 /* The type used to hold a single bcache string. The user data is
32 stored in d.data. Since it can be any type, it needs to have the
33 same alignment as the most strict alignment of any type on the host
34 machine. I don't know of any really correct way to do this in
35 stock ANSI C, so just do it the same way obstack.h does. */
41 /* Assume the data length is no more than 64k. */
42 unsigned short length
;
43 /* The half hash hack. This contains the upper 16 bits of the hash
44 value and is used as a pre-check when comparing two strings and
45 avoids the need to do length or memcmp calls. It proves to be
46 roughly 100% effective. */
47 unsigned short half_hash
;
58 /* The structure for a bcache itself. The bcache is initialized, in
59 bcache_xmalloc(), by filling it with zeros and then setting the
60 corresponding obstack's malloc() and free() methods. */
64 /* All the bstrings are allocated here. */
67 /* How many hash buckets we're using. */
68 unsigned int num_buckets
;
70 /* Hash buckets. This table is allocated using malloc, so when we
71 grow the table we can return the old table to the system. */
72 struct bstring
**bucket
;
75 unsigned long unique_count
; /* number of unique strings */
76 long total_count
; /* total number of strings cached, including dups */
77 long unique_size
; /* size of unique strings, in bytes */
78 long total_size
; /* total number of bytes cached, including dups */
79 long structure_size
; /* total size of bcache, including infrastructure */
80 /* Number of times that the hash table is expanded and hence
81 re-built, and the corresponding number of times that a string is
82 [re]hashed as part of entering it into the expanded table. The
83 total number of hashes can be computed by adding TOTAL_COUNT to
85 unsigned long expand_count
;
86 unsigned long expand_hash_count
;
87 /* Number of times that the half-hash compare hit (compare the upper
88 16 bits of hash values) hit, but the corresponding combined
89 length/data compare missed. */
90 unsigned long half_hash_miss_count
;
92 /* Hash function to be used for this bcache object. */
93 unsigned long (*hash_function
)(const void *addr
, int length
);
95 /* Compare function to be used for this bcache object. */
96 int (*compare_function
)(const void *, const void *, int length
);
99 /* The old hash function was stolen from SDBM. This is what DB 3.0
100 uses now, and is better than the old one. */
103 hash(const void *addr
, int length
)
105 return hash_continue (addr
, length
, 0);
108 /* Continue the calculation of the hash H at the given address. */
111 hash_continue (const void *addr
, int length
, unsigned long h
)
113 const unsigned char *k
, *e
;
115 k
= (const unsigned char *)addr
;
125 /* Growing the bcache's hash table. */
127 /* If the average chain length grows beyond this, then we want to
128 resize our hash table. */
129 #define CHAIN_LENGTH_THRESHOLD (5)
132 expand_hash_table (struct bcache
*bcache
)
134 /* A table of good hash table sizes. Whenever we grow, we pick the
135 next larger size from this table. sizes[i] is close to 1 << (i+10),
136 so we roughly double the table size each time. After we fall off
137 the end of this table, we just double. Don't laugh --- there have
138 been executables sighted with a gigabyte of debug info. */
139 static unsigned long sizes
[] = {
140 1021, 2053, 4099, 8191, 16381, 32771,
141 65537, 131071, 262144, 524287, 1048573, 2097143,
142 4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
143 268435459, 536870923, 1073741827, 2147483659UL
145 unsigned int new_num_buckets
;
146 struct bstring
**new_buckets
;
149 /* Count the stats. Every unique item needs to be re-hashed and
151 bcache
->expand_count
++;
152 bcache
->expand_hash_count
+= bcache
->unique_count
;
154 /* Find the next size. */
155 new_num_buckets
= bcache
->num_buckets
* 2;
156 for (i
= 0; i
< (sizeof (sizes
) / sizeof (sizes
[0])); i
++)
157 if (sizes
[i
] > bcache
->num_buckets
)
159 new_num_buckets
= sizes
[i
];
163 /* Allocate the new table. */
165 size_t new_size
= new_num_buckets
* sizeof (new_buckets
[0]);
167 new_buckets
= (struct bstring
**) xmalloc (new_size
);
168 memset (new_buckets
, 0, new_size
);
170 bcache
->structure_size
-= (bcache
->num_buckets
171 * sizeof (bcache
->bucket
[0]));
172 bcache
->structure_size
+= new_size
;
175 /* Rehash all existing strings. */
176 for (i
= 0; i
< bcache
->num_buckets
; i
++)
178 struct bstring
*s
, *next
;
180 for (s
= bcache
->bucket
[i
]; s
; s
= next
)
182 struct bstring
**new_bucket
;
185 new_bucket
= &new_buckets
[(bcache
->hash_function (&s
->d
.data
,
188 s
->next
= *new_bucket
;
193 /* Plug in the new table. */
195 xfree (bcache
->bucket
);
196 bcache
->bucket
= new_buckets
;
197 bcache
->num_buckets
= new_num_buckets
;
201 /* Looking up things in the bcache. */
203 /* The number of bytes needed to allocate a struct bstring whose data
205 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
207 /* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has
208 never seen those bytes before, add a copy of them to BCACHE. In
209 either case, return a pointer to BCACHE's copy of that string. */
211 bcache (const void *addr
, int length
, struct bcache
*cache
)
213 return bcache_full (addr
, length
, cache
, NULL
);
216 /* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has
217 never seen those bytes before, add a copy of them to BCACHE. In
218 either case, return a pointer to BCACHE's copy of that string. If
219 optional ADDED is not NULL, return 1 in case of new entry or 0 if
220 returning an old entry. */
223 bcache_full (const void *addr
, int length
, struct bcache
*bcache
, int *added
)
225 unsigned long full_hash
;
226 unsigned short half_hash
;
233 /* Lazily initialize the obstack. This can save quite a bit of
234 memory in some cases. */
235 if (bcache
->total_count
== 0)
237 /* We could use obstack_specify_allocation here instead, but
238 gdb_obstack.h specifies the allocation/deallocation
240 obstack_init (&bcache
->cache
);
243 /* If our average chain length is too high, expand the hash table. */
244 if (bcache
->unique_count
>= bcache
->num_buckets
* CHAIN_LENGTH_THRESHOLD
)
245 expand_hash_table (bcache
);
247 bcache
->total_count
++;
248 bcache
->total_size
+= length
;
250 full_hash
= bcache
->hash_function (addr
, length
);
252 half_hash
= (full_hash
>> 16);
253 hash_index
= full_hash
% bcache
->num_buckets
;
255 /* Search the hash bucket for a string identical to the caller's.
256 As a short-circuit first compare the upper part of each hash
258 for (s
= bcache
->bucket
[hash_index
]; s
; s
= s
->next
)
260 if (s
->half_hash
== half_hash
)
262 if (s
->length
== length
263 && bcache
->compare_function (&s
->d
.data
, addr
, length
))
266 bcache
->half_hash_miss_count
++;
270 /* The user's string isn't in the list. Insert it after *ps. */
273 = obstack_alloc (&bcache
->cache
, BSTRING_SIZE (length
));
275 memcpy (&new->d
.data
, addr
, length
);
276 new->length
= length
;
277 new->next
= bcache
->bucket
[hash_index
];
278 new->half_hash
= half_hash
;
279 bcache
->bucket
[hash_index
] = new;
281 bcache
->unique_count
++;
282 bcache
->unique_size
+= length
;
283 bcache
->structure_size
+= BSTRING_SIZE (length
);
293 /* Compare the byte string at ADDR1 of lenght LENGHT to the
294 string at ADDR2. Return 1 if they are equal. */
297 bcache_compare (const void *addr1
, const void *addr2
, int length
)
299 return memcmp (addr1
, addr2
, length
) == 0;
302 /* Allocating and freeing bcaches. */
304 /* Allocated a bcache. HASH_FUNCTION and COMPARE_FUNCTION can be used
305 to pass in custom hash, and compare functions to be used by this
306 bcache. If HASH_FUNCTION is NULL hash() is used and if
307 COMPARE_FUNCTION is NULL memcmp() is used. */
310 bcache_xmalloc (unsigned long (*hash_function
)(const void *, int length
),
311 int (*compare_function
)(const void *,
315 /* Allocate the bcache pre-zeroed. */
316 struct bcache
*b
= XCALLOC (1, struct bcache
);
319 b
->hash_function
= hash_function
;
321 b
->hash_function
= hash
;
323 if (compare_function
)
324 b
->compare_function
= compare_function
;
326 b
->compare_function
= bcache_compare
;
330 /* Free all the storage associated with BCACHE. */
332 bcache_xfree (struct bcache
*bcache
)
336 /* Only free the obstack if we actually initialized it. */
337 if (bcache
->total_count
> 0)
338 obstack_free (&bcache
->cache
, 0);
339 xfree (bcache
->bucket
);
345 /* Printing statistics. */
348 print_percentage (int portion
, int total
)
351 /* i18n: Like "Percentage of duplicates, by count: (not applicable)". */
352 printf_filtered (_("(not applicable)\n"));
354 printf_filtered ("%3d%%\n", (int) (portion
* 100.0 / total
));
358 /* Print statistics on BCACHE's memory usage and efficacity at
359 eliminating duplication. NAME should describe the kind of data
360 BCACHE holds. Statistics are printed using `printf_filtered' and
363 print_bcache_statistics (struct bcache
*c
, char *type
)
365 int occupied_buckets
;
366 int max_chain_length
;
367 int median_chain_length
;
369 int median_entry_size
;
371 /* Count the number of occupied buckets, tally the various string
372 lengths, and measure chain lengths. */
375 int *chain_length
= XCALLOC (c
->num_buckets
+ 1, int);
376 int *entry_size
= XCALLOC (c
->unique_count
+ 1, int);
379 occupied_buckets
= 0;
381 for (b
= 0; b
< c
->num_buckets
; b
++)
383 struct bstring
*s
= c
->bucket
[b
];
393 gdb_assert (b
< c
->num_buckets
);
395 gdb_assert (stringi
< c
->unique_count
);
396 entry_size
[stringi
++] = s
->length
;
402 /* To compute the median, we need the set of chain lengths
404 qsort (chain_length
, c
->num_buckets
, sizeof (chain_length
[0]),
405 compare_positive_ints
);
406 qsort (entry_size
, c
->unique_count
, sizeof (entry_size
[0]),
407 compare_positive_ints
);
409 if (c
->num_buckets
> 0)
411 max_chain_length
= chain_length
[c
->num_buckets
- 1];
412 median_chain_length
= chain_length
[c
->num_buckets
/ 2];
416 max_chain_length
= 0;
417 median_chain_length
= 0;
419 if (c
->unique_count
> 0)
421 max_entry_size
= entry_size
[c
->unique_count
- 1];
422 median_entry_size
= entry_size
[c
->unique_count
/ 2];
427 median_entry_size
= 0;
430 xfree (chain_length
);
434 printf_filtered (_(" Cached '%s' statistics:\n"), type
);
435 printf_filtered (_(" Total object count: %ld\n"), c
->total_count
);
436 printf_filtered (_(" Unique object count: %lu\n"), c
->unique_count
);
437 printf_filtered (_(" Percentage of duplicates, by count: "));
438 print_percentage (c
->total_count
- c
->unique_count
, c
->total_count
);
439 printf_filtered ("\n");
441 printf_filtered (_(" Total object size: %ld\n"), c
->total_size
);
442 printf_filtered (_(" Unique object size: %ld\n"), c
->unique_size
);
443 printf_filtered (_(" Percentage of duplicates, by size: "));
444 print_percentage (c
->total_size
- c
->unique_size
, c
->total_size
);
445 printf_filtered ("\n");
447 printf_filtered (_(" Max entry size: %d\n"), max_entry_size
);
448 printf_filtered (_(" Average entry size: "));
449 if (c
->unique_count
> 0)
450 printf_filtered ("%ld\n", c
->unique_size
/ c
->unique_count
);
452 /* i18n: "Average entry size: (not applicable)". */
453 printf_filtered (_("(not applicable)\n"));
454 printf_filtered (_(" Median entry size: %d\n"), median_entry_size
);
455 printf_filtered ("\n");
457 printf_filtered (_(" \
458 Total memory used by bcache, including overhead: %ld\n"),
460 printf_filtered (_(" Percentage memory overhead: "));
461 print_percentage (c
->structure_size
- c
->unique_size
, c
->unique_size
);
462 printf_filtered (_(" Net memory savings: "));
463 print_percentage (c
->total_size
- c
->structure_size
, c
->total_size
);
464 printf_filtered ("\n");
466 printf_filtered (_(" Hash table size: %3d\n"),
468 printf_filtered (_(" Hash table expands: %lu\n"),
470 printf_filtered (_(" Hash table hashes: %lu\n"),
471 c
->total_count
+ c
->expand_hash_count
);
472 printf_filtered (_(" Half hash misses: %lu\n"),
473 c
->half_hash_miss_count
);
474 printf_filtered (_(" Hash table population: "));
475 print_percentage (occupied_buckets
, c
->num_buckets
);
476 printf_filtered (_(" Median hash chain length: %3d\n"),
477 median_chain_length
);
478 printf_filtered (_(" Average hash chain length: "));
479 if (c
->num_buckets
> 0)
480 printf_filtered ("%3lu\n", c
->unique_count
/ c
->num_buckets
);
482 /* i18n: "Average hash chain length: (not applicable)". */
483 printf_filtered (_("(not applicable)\n"));
484 printf_filtered (_(" Maximum hash chain length: %3d\n"),
486 printf_filtered ("\n");
490 bcache_memory_used (struct bcache
*bcache
)
492 if (bcache
->total_count
== 0)
494 return obstack_memory_used (&bcache
->cache
);