1 /* Implement a cached obstack.
2 Written by Fred Fish <fnf@cygnus.com>
3 Rewritten by Jim Blandy <jimb@cygnus.com>
5 Copyright (C) 1999-2014 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 <string.h> /* For memcpy declaration */
26 #include "gdb_assert.h"
30 /* The type used to hold a single bcache string. The user data is
31 stored in d.data. Since it can be any type, it needs to have the
32 same alignment as the most strict alignment of any type on the host
33 machine. I don't know of any really correct way to do this in
34 stock ANSI C, so just do it the same way obstack.h does. */
40 /* Assume the data length is no more than 64k. */
41 unsigned short length
;
42 /* The half hash hack. This contains the upper 16 bits of the hash
43 value and is used as a pre-check when comparing two strings and
44 avoids the need to do length or memcmp calls. It proves to be
45 roughly 100% effective. */
46 unsigned short half_hash
;
57 /* The structure for a bcache itself. The bcache is initialized, in
58 bcache_xmalloc(), by filling it with zeros and then setting the
59 corresponding obstack's malloc() and free() methods. */
63 /* All the bstrings are allocated here. */
66 /* How many hash buckets we're using. */
67 unsigned int num_buckets
;
69 /* Hash buckets. This table is allocated using malloc, so when we
70 grow the table we can return the old table to the system. */
71 struct bstring
**bucket
;
74 unsigned long unique_count
; /* number of unique strings */
75 long total_count
; /* total number of strings cached, including dups */
76 long unique_size
; /* size of unique strings, in bytes */
77 long total_size
; /* total number of bytes cached, including dups */
78 long structure_size
; /* total size of bcache, including infrastructure */
79 /* Number of times that the hash table is expanded and hence
80 re-built, and the corresponding number of times that a string is
81 [re]hashed as part of entering it into the expanded table. The
82 total number of hashes can be computed by adding TOTAL_COUNT to
84 unsigned long expand_count
;
85 unsigned long expand_hash_count
;
86 /* Number of times that the half-hash compare hit (compare the upper
87 16 bits of hash values) hit, but the corresponding combined
88 length/data compare missed. */
89 unsigned long half_hash_miss_count
;
91 /* Hash function to be used for this bcache object. */
92 unsigned long (*hash_function
)(const void *addr
, int length
);
94 /* Compare function to be used for this bcache object. */
95 int (*compare_function
)(const void *, const void *, int length
);
98 /* The old hash function was stolen from SDBM. This is what DB 3.0
99 uses now, and is better than the old one. */
102 hash(const void *addr
, int length
)
104 return hash_continue (addr
, length
, 0);
107 /* Continue the calculation of the hash H at the given address. */
110 hash_continue (const void *addr
, int length
, unsigned long h
)
112 const unsigned char *k
, *e
;
114 k
= (const unsigned char *)addr
;
124 /* Growing the bcache's hash table. */
126 /* If the average chain length grows beyond this, then we want to
127 resize our hash table. */
128 #define CHAIN_LENGTH_THRESHOLD (5)
131 expand_hash_table (struct bcache
*bcache
)
133 /* A table of good hash table sizes. Whenever we grow, we pick the
134 next larger size from this table. sizes[i] is close to 1 << (i+10),
135 so we roughly double the table size each time. After we fall off
136 the end of this table, we just double. Don't laugh --- there have
137 been executables sighted with a gigabyte of debug info. */
138 static unsigned long sizes
[] = {
139 1021, 2053, 4099, 8191, 16381, 32771,
140 65537, 131071, 262144, 524287, 1048573, 2097143,
141 4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
142 268435459, 536870923, 1073741827, 2147483659UL
144 unsigned int new_num_buckets
;
145 struct bstring
**new_buckets
;
148 /* Count the stats. Every unique item needs to be re-hashed and
150 bcache
->expand_count
++;
151 bcache
->expand_hash_count
+= bcache
->unique_count
;
153 /* Find the next size. */
154 new_num_buckets
= bcache
->num_buckets
* 2;
155 for (i
= 0; i
< (sizeof (sizes
) / sizeof (sizes
[0])); i
++)
156 if (sizes
[i
] > bcache
->num_buckets
)
158 new_num_buckets
= sizes
[i
];
162 /* Allocate the new table. */
164 size_t new_size
= new_num_buckets
* sizeof (new_buckets
[0]);
166 new_buckets
= (struct bstring
**) xmalloc (new_size
);
167 memset (new_buckets
, 0, new_size
);
169 bcache
->structure_size
-= (bcache
->num_buckets
170 * sizeof (bcache
->bucket
[0]));
171 bcache
->structure_size
+= new_size
;
174 /* Rehash all existing strings. */
175 for (i
= 0; i
< bcache
->num_buckets
; i
++)
177 struct bstring
*s
, *next
;
179 for (s
= bcache
->bucket
[i
]; s
; s
= next
)
181 struct bstring
**new_bucket
;
184 new_bucket
= &new_buckets
[(bcache
->hash_function (&s
->d
.data
,
187 s
->next
= *new_bucket
;
192 /* Plug in the new table. */
194 xfree (bcache
->bucket
);
195 bcache
->bucket
= new_buckets
;
196 bcache
->num_buckets
= new_num_buckets
;
200 /* Looking up things in the bcache. */
202 /* The number of bytes needed to allocate a struct bstring whose data
204 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
206 /* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has
207 never seen those bytes before, add a copy of them to BCACHE. In
208 either case, return a pointer to BCACHE's copy of that string. */
210 bcache (const void *addr
, int length
, struct bcache
*cache
)
212 return bcache_full (addr
, length
, cache
, NULL
);
215 /* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has
216 never seen those bytes before, add a copy of them to BCACHE. In
217 either case, return a pointer to BCACHE's copy of that string. If
218 optional ADDED is not NULL, return 1 in case of new entry or 0 if
219 returning an old entry. */
222 bcache_full (const void *addr
, int length
, struct bcache
*bcache
, int *added
)
224 unsigned long full_hash
;
225 unsigned short half_hash
;
232 /* Lazily initialize the obstack. This can save quite a bit of
233 memory in some cases. */
234 if (bcache
->total_count
== 0)
236 /* We could use obstack_specify_allocation here instead, but
237 gdb_obstack.h specifies the allocation/deallocation
239 obstack_init (&bcache
->cache
);
242 /* If our average chain length is too high, expand the hash table. */
243 if (bcache
->unique_count
>= bcache
->num_buckets
* CHAIN_LENGTH_THRESHOLD
)
244 expand_hash_table (bcache
);
246 bcache
->total_count
++;
247 bcache
->total_size
+= length
;
249 full_hash
= bcache
->hash_function (addr
, length
);
251 half_hash
= (full_hash
>> 16);
252 hash_index
= full_hash
% bcache
->num_buckets
;
254 /* Search the hash bucket for a string identical to the caller's.
255 As a short-circuit first compare the upper part of each hash
257 for (s
= bcache
->bucket
[hash_index
]; s
; s
= s
->next
)
259 if (s
->half_hash
== half_hash
)
261 if (s
->length
== length
262 && bcache
->compare_function (&s
->d
.data
, addr
, length
))
265 bcache
->half_hash_miss_count
++;
269 /* The user's string isn't in the list. Insert it after *ps. */
272 = obstack_alloc (&bcache
->cache
, BSTRING_SIZE (length
));
274 memcpy (&new->d
.data
, addr
, length
);
275 new->length
= length
;
276 new->next
= bcache
->bucket
[hash_index
];
277 new->half_hash
= half_hash
;
278 bcache
->bucket
[hash_index
] = new;
280 bcache
->unique_count
++;
281 bcache
->unique_size
+= length
;
282 bcache
->structure_size
+= BSTRING_SIZE (length
);
292 /* Compare the byte string at ADDR1 of lenght LENGHT to the
293 string at ADDR2. Return 1 if they are equal. */
296 bcache_compare (const void *addr1
, const void *addr2
, int length
)
298 return memcmp (addr1
, addr2
, length
) == 0;
301 /* Allocating and freeing bcaches. */
303 /* Allocated a bcache. HASH_FUNCTION and COMPARE_FUNCTION can be used
304 to pass in custom hash, and compare functions to be used by this
305 bcache. If HASH_FUNCTION is NULL hash() is used and if
306 COMPARE_FUNCTION is NULL memcmp() is used. */
309 bcache_xmalloc (unsigned long (*hash_function
)(const void *, int length
),
310 int (*compare_function
)(const void *,
314 /* Allocate the bcache pre-zeroed. */
315 struct bcache
*b
= XCNEW (struct bcache
);
318 b
->hash_function
= hash_function
;
320 b
->hash_function
= hash
;
322 if (compare_function
)
323 b
->compare_function
= compare_function
;
325 b
->compare_function
= bcache_compare
;
329 /* Free all the storage associated with BCACHE. */
331 bcache_xfree (struct bcache
*bcache
)
335 /* Only free the obstack if we actually initialized it. */
336 if (bcache
->total_count
> 0)
337 obstack_free (&bcache
->cache
, 0);
338 xfree (bcache
->bucket
);
344 /* Printing statistics. */
347 print_percentage (int portion
, int total
)
350 /* i18n: Like "Percentage of duplicates, by count: (not applicable)". */
351 printf_filtered (_("(not applicable)\n"));
353 printf_filtered ("%3d%%\n", (int) (portion
* 100.0 / total
));
357 /* Print statistics on BCACHE's memory usage and efficacity at
358 eliminating duplication. NAME should describe the kind of data
359 BCACHE holds. Statistics are printed using `printf_filtered' and
362 print_bcache_statistics (struct bcache
*c
, char *type
)
364 int occupied_buckets
;
365 int max_chain_length
;
366 int median_chain_length
;
368 int median_entry_size
;
370 /* Count the number of occupied buckets, tally the various string
371 lengths, and measure chain lengths. */
374 int *chain_length
= XCNEWVEC (int, c
->num_buckets
+ 1);
375 int *entry_size
= XCNEWVEC (int, c
->unique_count
+ 1);
378 occupied_buckets
= 0;
380 for (b
= 0; b
< c
->num_buckets
; b
++)
382 struct bstring
*s
= c
->bucket
[b
];
392 gdb_assert (b
< c
->num_buckets
);
394 gdb_assert (stringi
< c
->unique_count
);
395 entry_size
[stringi
++] = s
->length
;
401 /* To compute the median, we need the set of chain lengths
403 qsort (chain_length
, c
->num_buckets
, sizeof (chain_length
[0]),
404 compare_positive_ints
);
405 qsort (entry_size
, c
->unique_count
, sizeof (entry_size
[0]),
406 compare_positive_ints
);
408 if (c
->num_buckets
> 0)
410 max_chain_length
= chain_length
[c
->num_buckets
- 1];
411 median_chain_length
= chain_length
[c
->num_buckets
/ 2];
415 max_chain_length
= 0;
416 median_chain_length
= 0;
418 if (c
->unique_count
> 0)
420 max_entry_size
= entry_size
[c
->unique_count
- 1];
421 median_entry_size
= entry_size
[c
->unique_count
/ 2];
426 median_entry_size
= 0;
429 xfree (chain_length
);
433 printf_filtered (_(" Cached '%s' statistics:\n"), type
);
434 printf_filtered (_(" Total object count: %ld\n"), c
->total_count
);
435 printf_filtered (_(" Unique object count: %lu\n"), c
->unique_count
);
436 printf_filtered (_(" Percentage of duplicates, by count: "));
437 print_percentage (c
->total_count
- c
->unique_count
, c
->total_count
);
438 printf_filtered ("\n");
440 printf_filtered (_(" Total object size: %ld\n"), c
->total_size
);
441 printf_filtered (_(" Unique object size: %ld\n"), c
->unique_size
);
442 printf_filtered (_(" Percentage of duplicates, by size: "));
443 print_percentage (c
->total_size
- c
->unique_size
, c
->total_size
);
444 printf_filtered ("\n");
446 printf_filtered (_(" Max entry size: %d\n"), max_entry_size
);
447 printf_filtered (_(" Average entry size: "));
448 if (c
->unique_count
> 0)
449 printf_filtered ("%ld\n", c
->unique_size
/ c
->unique_count
);
451 /* i18n: "Average entry size: (not applicable)". */
452 printf_filtered (_("(not applicable)\n"));
453 printf_filtered (_(" Median entry size: %d\n"), median_entry_size
);
454 printf_filtered ("\n");
456 printf_filtered (_(" \
457 Total memory used by bcache, including overhead: %ld\n"),
459 printf_filtered (_(" Percentage memory overhead: "));
460 print_percentage (c
->structure_size
- c
->unique_size
, c
->unique_size
);
461 printf_filtered (_(" Net memory savings: "));
462 print_percentage (c
->total_size
- c
->structure_size
, c
->total_size
);
463 printf_filtered ("\n");
465 printf_filtered (_(" Hash table size: %3d\n"),
467 printf_filtered (_(" Hash table expands: %lu\n"),
469 printf_filtered (_(" Hash table hashes: %lu\n"),
470 c
->total_count
+ c
->expand_hash_count
);
471 printf_filtered (_(" Half hash misses: %lu\n"),
472 c
->half_hash_miss_count
);
473 printf_filtered (_(" Hash table population: "));
474 print_percentage (occupied_buckets
, c
->num_buckets
);
475 printf_filtered (_(" Median hash chain length: %3d\n"),
476 median_chain_length
);
477 printf_filtered (_(" Average hash chain length: "));
478 if (c
->num_buckets
> 0)
479 printf_filtered ("%3lu\n", c
->unique_count
/ c
->num_buckets
);
481 /* i18n: "Average hash chain length: (not applicable)". */
482 printf_filtered (_("(not applicable)\n"));
483 printf_filtered (_(" Maximum hash chain length: %3d\n"),
485 printf_filtered ("\n");
489 bcache_memory_used (struct bcache
*bcache
)
491 if (bcache
->total_count
== 0)
493 return obstack_memory_used (&bcache
->cache
);