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, 2000, 2002, 2003, 2007, 2008
6 Free Software Foundation, Inc.
8 This file is part of GDB.
10 This program is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 3 of the License, or
13 (at your option) any later version.
15 This program is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
20 You should have received a copy of the GNU General Public License
21 along with this program. If not, see <http://www.gnu.org/licenses/>. */
24 #include "gdb_obstack.h"
26 #include "gdb_string.h" /* For memcpy declaration */
27 #include "gdb_assert.h"
32 /* The type used to hold a single bcache string. The user data is
33 stored in d.data. Since it can be any type, it needs to have the
34 same alignment as the most strict alignment of any type on the host
35 machine. I don't know of any really correct way to do this in
36 stock ANSI C, so just do it the same way obstack.h does. */
42 /* Assume the data length is no more than 64k. */
43 unsigned short length
;
44 /* The half hash hack. This contains the upper 16 bits of the hash
45 value and is used as a pre-check when comparing two strings and
46 avoids the need to do length or memcmp calls. It proves to be
47 roughly 100% effective. */
48 unsigned short half_hash
;
59 /* The structure for a bcache itself. The bcache is initialized, in
60 bcache_xmalloc(), by filling it with zeros and then setting the
61 corresponding obstack's malloc() and free() methods. */
65 /* All the bstrings are allocated here. */
68 /* How many hash buckets we're using. */
69 unsigned int num_buckets
;
71 /* Hash buckets. This table is allocated using malloc, so when we
72 grow the table we can return the old table to the system. */
73 struct bstring
**bucket
;
76 unsigned long unique_count
; /* number of unique strings */
77 long total_count
; /* total number of strings cached, including dups */
78 long unique_size
; /* size of unique strings, in bytes */
79 long total_size
; /* total number of bytes cached, including dups */
80 long structure_size
; /* total size of bcache, including infrastructure */
81 /* Number of times that the hash table is expanded and hence
82 re-built, and the corresponding number of times that a string is
83 [re]hashed as part of entering it into the expanded table. The
84 total number of hashes can be computed by adding TOTAL_COUNT to
86 unsigned long expand_count
;
87 unsigned long expand_hash_count
;
88 /* Number of times that the half-hash compare hit (compare the upper
89 16 bits of hash values) hit, but the corresponding combined
90 length/data compare missed. */
91 unsigned long half_hash_miss_count
;
94 /* The old hash function was stolen from SDBM. This is what DB 3.0 uses now,
95 * and is better than the old one.
99 hash(const void *addr
, int length
)
101 const unsigned char *k
, *e
;
104 k
= (const unsigned char *)addr
;
114 /* Growing the bcache's hash table. */
116 /* If the average chain length grows beyond this, then we want to
117 resize our hash table. */
118 #define CHAIN_LENGTH_THRESHOLD (5)
121 expand_hash_table (struct bcache
*bcache
)
123 /* A table of good hash table sizes. Whenever we grow, we pick the
124 next larger size from this table. sizes[i] is close to 1 << (i+10),
125 so we roughly double the table size each time. After we fall off
126 the end of this table, we just double. Don't laugh --- there have
127 been executables sighted with a gigabyte of debug info. */
128 static unsigned long sizes
[] = {
129 1021, 2053, 4099, 8191, 16381, 32771,
130 65537, 131071, 262144, 524287, 1048573, 2097143,
131 4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
132 268435459, 536870923, 1073741827, 2147483659UL
134 unsigned int new_num_buckets
;
135 struct bstring
**new_buckets
;
138 /* Count the stats. Every unique item needs to be re-hashed and
140 bcache
->expand_count
++;
141 bcache
->expand_hash_count
+= bcache
->unique_count
;
143 /* Find the next size. */
144 new_num_buckets
= bcache
->num_buckets
* 2;
145 for (i
= 0; i
< (sizeof (sizes
) / sizeof (sizes
[0])); i
++)
146 if (sizes
[i
] > bcache
->num_buckets
)
148 new_num_buckets
= sizes
[i
];
152 /* Allocate the new table. */
154 size_t new_size
= new_num_buckets
* sizeof (new_buckets
[0]);
155 new_buckets
= (struct bstring
**) xmalloc (new_size
);
156 memset (new_buckets
, 0, new_size
);
158 bcache
->structure_size
-= (bcache
->num_buckets
159 * sizeof (bcache
->bucket
[0]));
160 bcache
->structure_size
+= new_size
;
163 /* Rehash all existing strings. */
164 for (i
= 0; i
< bcache
->num_buckets
; i
++)
166 struct bstring
*s
, *next
;
168 for (s
= bcache
->bucket
[i
]; s
; s
= next
)
170 struct bstring
**new_bucket
;
173 new_bucket
= &new_buckets
[(hash (&s
->d
.data
, s
->length
)
175 s
->next
= *new_bucket
;
180 /* Plug in the new table. */
182 xfree (bcache
->bucket
);
183 bcache
->bucket
= new_buckets
;
184 bcache
->num_buckets
= new_num_buckets
;
188 /* Looking up things in the bcache. */
190 /* The number of bytes needed to allocate a struct bstring whose data
192 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
194 /* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has
195 never seen those bytes before, add a copy of them to BCACHE. In
196 either case, return a pointer to BCACHE's copy of that string. */
198 bcache (const void *addr
, int length
, struct bcache
*bcache
)
200 return bcache_full (addr
, length
, bcache
, NULL
);
203 /* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has
204 never seen those bytes before, add a copy of them to BCACHE. In
205 either case, return a pointer to BCACHE's copy of that string. If
206 optional ADDED is not NULL, return 1 in case of new entry or 0 if
207 returning an old entry. */
210 bcache_full (const void *addr
, int length
, struct bcache
*bcache
, int *added
)
212 unsigned long full_hash
;
213 unsigned short half_hash
;
220 /* Lazily initialize the obstack. This can save quite a bit of
221 memory in some cases. */
222 if (bcache
->total_count
== 0)
224 /* We could use obstack_specify_allocation here instead, but
225 gdb_obstack.h specifies the allocation/deallocation
227 obstack_init (&bcache
->cache
);
230 /* If our average chain length is too high, expand the hash table. */
231 if (bcache
->unique_count
>= bcache
->num_buckets
* CHAIN_LENGTH_THRESHOLD
)
232 expand_hash_table (bcache
);
234 bcache
->total_count
++;
235 bcache
->total_size
+= length
;
237 full_hash
= hash (addr
, length
);
238 half_hash
= (full_hash
>> 16);
239 hash_index
= full_hash
% bcache
->num_buckets
;
241 /* Search the hash bucket for a string identical to the caller's.
242 As a short-circuit first compare the upper part of each hash
244 for (s
= bcache
->bucket
[hash_index
]; s
; s
= s
->next
)
246 if (s
->half_hash
== half_hash
)
248 if (s
->length
== length
249 && ! memcmp (&s
->d
.data
, addr
, length
))
252 bcache
->half_hash_miss_count
++;
256 /* The user's string isn't in the list. Insert it after *ps. */
259 = obstack_alloc (&bcache
->cache
, BSTRING_SIZE (length
));
260 memcpy (&new->d
.data
, addr
, length
);
261 new->length
= length
;
262 new->next
= bcache
->bucket
[hash_index
];
263 new->half_hash
= half_hash
;
264 bcache
->bucket
[hash_index
] = new;
266 bcache
->unique_count
++;
267 bcache
->unique_size
+= length
;
268 bcache
->structure_size
+= BSTRING_SIZE (length
);
277 /* Allocating and freeing bcaches. */
280 bcache_xmalloc (void)
282 /* Allocate the bcache pre-zeroed. */
283 struct bcache
*b
= XCALLOC (1, struct bcache
);
287 /* Free all the storage associated with BCACHE. */
289 bcache_xfree (struct bcache
*bcache
)
293 /* Only free the obstack if we actually initialized it. */
294 if (bcache
->total_count
> 0)
295 obstack_free (&bcache
->cache
, 0);
296 xfree (bcache
->bucket
);
302 /* Printing statistics. */
305 compare_ints (const void *ap
, const void *bp
)
307 /* Because we know we're comparing two ints which are positive,
308 there's no danger of overflow here. */
309 return * (int *) ap
- * (int *) bp
;
314 print_percentage (int portion
, int total
)
317 /* i18n: Like "Percentage of duplicates, by count: (not applicable)" */
318 printf_filtered (_("(not applicable)\n"));
320 printf_filtered ("%3d%%\n", (int) (portion
* 100.0 / total
));
324 /* Print statistics on BCACHE's memory usage and efficacity at
325 eliminating duplication. NAME should describe the kind of data
326 BCACHE holds. Statistics are printed using `printf_filtered' and
329 print_bcache_statistics (struct bcache
*c
, char *type
)
331 int occupied_buckets
;
332 int max_chain_length
;
333 int median_chain_length
;
335 int median_entry_size
;
337 /* Count the number of occupied buckets, tally the various string
338 lengths, and measure chain lengths. */
341 int *chain_length
= XCALLOC (c
->num_buckets
+ 1, int);
342 int *entry_size
= XCALLOC (c
->unique_count
+ 1, int);
345 occupied_buckets
= 0;
347 for (b
= 0; b
< c
->num_buckets
; b
++)
349 struct bstring
*s
= c
->bucket
[b
];
359 gdb_assert (b
< c
->num_buckets
);
361 gdb_assert (stringi
< c
->unique_count
);
362 entry_size
[stringi
++] = s
->length
;
368 /* To compute the median, we need the set of chain lengths sorted. */
369 qsort (chain_length
, c
->num_buckets
, sizeof (chain_length
[0]),
371 qsort (entry_size
, c
->unique_count
, sizeof (entry_size
[0]),
374 if (c
->num_buckets
> 0)
376 max_chain_length
= chain_length
[c
->num_buckets
- 1];
377 median_chain_length
= chain_length
[c
->num_buckets
/ 2];
381 max_chain_length
= 0;
382 median_chain_length
= 0;
384 if (c
->unique_count
> 0)
386 max_entry_size
= entry_size
[c
->unique_count
- 1];
387 median_entry_size
= entry_size
[c
->unique_count
/ 2];
392 median_entry_size
= 0;
395 xfree (chain_length
);
399 printf_filtered (_(" Cached '%s' statistics:\n"), type
);
400 printf_filtered (_(" Total object count: %ld\n"), c
->total_count
);
401 printf_filtered (_(" Unique object count: %lu\n"), c
->unique_count
);
402 printf_filtered (_(" Percentage of duplicates, by count: "));
403 print_percentage (c
->total_count
- c
->unique_count
, c
->total_count
);
404 printf_filtered ("\n");
406 printf_filtered (_(" Total object size: %ld\n"), c
->total_size
);
407 printf_filtered (_(" Unique object size: %ld\n"), c
->unique_size
);
408 printf_filtered (_(" Percentage of duplicates, by size: "));
409 print_percentage (c
->total_size
- c
->unique_size
, c
->total_size
);
410 printf_filtered ("\n");
412 printf_filtered (_(" Max entry size: %d\n"), max_entry_size
);
413 printf_filtered (_(" Average entry size: "));
414 if (c
->unique_count
> 0)
415 printf_filtered ("%ld\n", c
->unique_size
/ c
->unique_count
);
417 /* i18n: "Average entry size: (not applicable)" */
418 printf_filtered (_("(not applicable)\n"));
419 printf_filtered (_(" Median entry size: %d\n"), median_entry_size
);
420 printf_filtered ("\n");
422 printf_filtered (_(" Total memory used by bcache, including overhead: %ld\n"),
424 printf_filtered (_(" Percentage memory overhead: "));
425 print_percentage (c
->structure_size
- c
->unique_size
, c
->unique_size
);
426 printf_filtered (_(" Net memory savings: "));
427 print_percentage (c
->total_size
- c
->structure_size
, c
->total_size
);
428 printf_filtered ("\n");
430 printf_filtered (_(" Hash table size: %3d\n"), c
->num_buckets
);
431 printf_filtered (_(" Hash table expands: %lu\n"),
433 printf_filtered (_(" Hash table hashes: %lu\n"),
434 c
->total_count
+ c
->expand_hash_count
);
435 printf_filtered (_(" Half hash misses: %lu\n"),
436 c
->half_hash_miss_count
);
437 printf_filtered (_(" Hash table population: "));
438 print_percentage (occupied_buckets
, c
->num_buckets
);
439 printf_filtered (_(" Median hash chain length: %3d\n"),
440 median_chain_length
);
441 printf_filtered (_(" Average hash chain length: "));
442 if (c
->num_buckets
> 0)
443 printf_filtered ("%3lu\n", c
->unique_count
/ c
->num_buckets
);
445 /* i18n: "Average hash chain length: (not applicable)" */
446 printf_filtered (_("(not applicable)\n"));
447 printf_filtered (_(" Maximum hash chain length: %3d\n"), max_chain_length
);
448 printf_filtered ("\n");
452 bcache_memory_used (struct bcache
*bcache
)
454 if (bcache
->total_count
== 0)
456 return obstack_memory_used (&bcache
->cache
);
This page took 0.037883 seconds and 4 git commands to generate.