Commit | Line | Data |
---|---|---|
14bfc3f5 ILT |
1 | // stringpool.cc -- a string pool for gold |
2 | ||
6cb15b7f ILT |
3 | // Copyright 2006, 2007 Free Software Foundation, Inc. |
4 | // Written by Ian Lance Taylor <iant@google.com>. | |
5 | ||
6 | // This file is part of gold. | |
7 | ||
8 | // This program is free software; you can redistribute it and/or modify | |
9 | // it under the terms of the GNU General Public License as published by | |
10 | // the Free Software Foundation; either version 3 of the License, or | |
11 | // (at your option) any later version. | |
12 | ||
13 | // This program is distributed in the hope that it will be useful, | |
14 | // but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
16 | // GNU General Public License for more details. | |
17 | ||
18 | // You should have received a copy of the GNU General Public License | |
19 | // along with this program; if not, write to the Free Software | |
20 | // Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, | |
21 | // MA 02110-1301, USA. | |
22 | ||
14bfc3f5 ILT |
23 | #include "gold.h" |
24 | ||
14bfc3f5 | 25 | #include <cstring> |
61ba1cf9 ILT |
26 | #include <algorithm> |
27 | #include <vector> | |
14bfc3f5 | 28 | |
61ba1cf9 | 29 | #include "output.h" |
377caf49 | 30 | #include "parameters.h" |
14bfc3f5 ILT |
31 | #include "stringpool.h" |
32 | ||
33 | namespace gold | |
34 | { | |
35 | ||
b8e6aad9 | 36 | template<typename Stringpool_char> |
a8b2552e | 37 | Stringpool_template<Stringpool_char>::Stringpool_template() |
b8e6aad9 | 38 | : string_set_(), strings_(), strtab_size_(0), next_index_(1), |
a8b2552e | 39 | zero_null_(true) |
14bfc3f5 ILT |
40 | { |
41 | } | |
42 | ||
b8e6aad9 ILT |
43 | template<typename Stringpool_char> |
44 | Stringpool_template<Stringpool_char>::~Stringpool_template() | |
14bfc3f5 | 45 | { |
b8e6aad9 | 46 | for (typename std::list<Stringdata*>::iterator p = this->strings_.begin(); |
14bfc3f5 ILT |
47 | p != this->strings_.end(); |
48 | ++p) | |
49 | delete[] reinterpret_cast<char*>(*p); | |
50 | } | |
51 | ||
b8e6aad9 ILT |
52 | // Return the length of a string of arbitrary character type. |
53 | ||
54 | template<typename Stringpool_char> | |
55 | size_t | |
56 | Stringpool_template<Stringpool_char>::string_length(const Stringpool_char* p) | |
57 | { | |
58 | size_t len = 0; | |
59 | for (; *p != 0; ++p) | |
60 | ++len; | |
61 | return len; | |
62 | } | |
63 | ||
64 | // Specialize string_length for char. Maybe we could just use | |
65 | // std::char_traits<>::length? | |
66 | ||
67 | template<> | |
68 | inline size_t | |
69 | Stringpool_template<char>::string_length(const char* p) | |
70 | { | |
71 | return strlen(p); | |
72 | } | |
73 | ||
74 | // Equality comparison function. | |
75 | ||
76 | template<typename Stringpool_char> | |
77 | bool | |
78 | Stringpool_template<Stringpool_char>::Stringpool_eq::operator()( | |
79 | const Stringpool_char* s1, | |
80 | const Stringpool_char* s2) const | |
81 | { | |
82 | while (*s1 != 0) | |
83 | if (*s1++ != *s2++) | |
84 | return false; | |
85 | return *s2 == 0; | |
86 | } | |
87 | ||
88 | // Specialize equality comparison for char. | |
89 | ||
90 | template<> | |
91 | bool | |
92 | Stringpool_template<char>::Stringpool_eq::operator()(const char* s1, | |
93 | const char* s2) const | |
94 | { | |
95 | return strcmp(s1, s2) == 0; | |
96 | } | |
97 | ||
14bfc3f5 ILT |
98 | // Hash function. |
99 | ||
b8e6aad9 | 100 | template<typename Stringpool_char> |
14bfc3f5 | 101 | size_t |
b8e6aad9 ILT |
102 | Stringpool_template<Stringpool_char>::Stringpool_hash::operator()( |
103 | const Stringpool_char* s) const | |
14bfc3f5 ILT |
104 | { |
105 | // Fowler/Noll/Vo (FNV) hash (type FNV-1a). | |
106 | if (sizeof(size_t) == 8) | |
107 | { | |
b3168e9d | 108 | size_t result = static_cast<size_t>(14695981039346656037ULL); |
b8e6aad9 | 109 | while (*s != 0) |
14bfc3f5 | 110 | { |
b8e6aad9 ILT |
111 | const char* p = reinterpret_cast<const char*>(s); |
112 | for (size_t i = 0; i < sizeof(Stringpool_char); ++i) | |
113 | { | |
114 | result &= (size_t) *p++; | |
115 | result *= 1099511628211ULL; | |
116 | } | |
117 | ++s; | |
14bfc3f5 ILT |
118 | } |
119 | return result; | |
120 | } | |
121 | else | |
122 | { | |
123 | size_t result = 2166136261UL; | |
b8e6aad9 | 124 | while (*s != 0) |
14bfc3f5 | 125 | { |
b8e6aad9 ILT |
126 | const char* p = reinterpret_cast<const char*>(s); |
127 | for (size_t i = 0; i < sizeof(Stringpool_char); ++i) | |
128 | { | |
129 | result ^= (size_t) *p++; | |
130 | result *= 16777619UL; | |
131 | } | |
132 | ++s; | |
14bfc3f5 ILT |
133 | } |
134 | return result; | |
135 | } | |
136 | } | |
137 | ||
138 | // Add a string to the list of canonical strings. Return a pointer to | |
f0641a0b | 139 | // the canonical string. If PKEY is not NULL, set *PKEY to the key. |
14bfc3f5 | 140 | |
b8e6aad9 ILT |
141 | template<typename Stringpool_char> |
142 | const Stringpool_char* | |
143 | Stringpool_template<Stringpool_char>::add_string(const Stringpool_char* s, | |
144 | Key* pkey) | |
14bfc3f5 | 145 | { |
a3ad94ed ILT |
146 | // We are in trouble if we've already computed the string offsets. |
147 | gold_assert(this->strtab_size_ == 0); | |
148 | ||
f0641a0b | 149 | // The size we allocate for a new Stringdata. |
14bfc3f5 | 150 | const size_t buffer_size = 1000; |
f0641a0b ILT |
151 | // The amount we multiply the Stringdata index when calculating the |
152 | // key. | |
153 | const size_t key_mult = 1024; | |
a3ad94ed | 154 | gold_assert(key_mult >= buffer_size); |
f0641a0b | 155 | |
b8e6aad9 | 156 | size_t len = (string_length(s) + 1) * sizeof(Stringpool_char); |
14bfc3f5 ILT |
157 | |
158 | size_t alc; | |
159 | bool front = true; | |
b8e6aad9 | 160 | if (len > buffer_size) |
14bfc3f5 | 161 | { |
61ba1cf9 | 162 | alc = sizeof(Stringdata) + len; |
14bfc3f5 ILT |
163 | front = false; |
164 | } | |
165 | else if (this->strings_.empty()) | |
61ba1cf9 | 166 | alc = sizeof(Stringdata) + buffer_size; |
14bfc3f5 ILT |
167 | else |
168 | { | |
61ba1cf9 | 169 | Stringdata *psd = this->strings_.front(); |
b8e6aad9 | 170 | if (len > psd->alc - psd->len) |
61ba1cf9 | 171 | alc = sizeof(Stringdata) + buffer_size; |
14bfc3f5 ILT |
172 | else |
173 | { | |
174 | char* ret = psd->data + psd->len; | |
b8e6aad9 | 175 | memcpy(ret, s, len); |
f0641a0b ILT |
176 | |
177 | if (pkey != NULL) | |
178 | *pkey = psd->index * key_mult + psd->len; | |
179 | ||
b8e6aad9 | 180 | psd->len += len; |
f0641a0b | 181 | |
b8e6aad9 | 182 | return reinterpret_cast<const Stringpool_char*>(ret); |
14bfc3f5 ILT |
183 | } |
184 | } | |
185 | ||
61ba1cf9 ILT |
186 | Stringdata *psd = reinterpret_cast<Stringdata*>(new char[alc]); |
187 | psd->alc = alc - sizeof(Stringdata); | |
b8e6aad9 ILT |
188 | memcpy(psd->data, s, len); |
189 | psd->len = len; | |
f0641a0b ILT |
190 | psd->index = this->next_index_; |
191 | ++this->next_index_; | |
192 | ||
193 | if (pkey != NULL) | |
194 | *pkey = psd->index * key_mult; | |
195 | ||
14bfc3f5 ILT |
196 | if (front) |
197 | this->strings_.push_front(psd); | |
198 | else | |
199 | this->strings_.push_back(psd); | |
f0641a0b | 200 | |
b8e6aad9 | 201 | return reinterpret_cast<const Stringpool_char*>(psd->data); |
14bfc3f5 ILT |
202 | } |
203 | ||
204 | // Add a string to a string pool. | |
205 | ||
b8e6aad9 ILT |
206 | template<typename Stringpool_char> |
207 | const Stringpool_char* | |
208 | Stringpool_template<Stringpool_char>::add(const Stringpool_char* s, Key* pkey) | |
14bfc3f5 ILT |
209 | { |
210 | // FIXME: This will look up the entry twice in the hash table. The | |
211 | // problem is that we can't insert S before we canonicalize it. I | |
a2fb1b05 | 212 | // don't think there is a way to handle this correctly with |
61ba1cf9 | 213 | // unordered_map, so this should be replaced with custom code to do |
14bfc3f5 ILT |
214 | // what we need, which is to return the empty slot. |
215 | ||
b8e6aad9 | 216 | typename String_set_type::const_iterator p = this->string_set_.find(s); |
14bfc3f5 | 217 | if (p != this->string_set_.end()) |
f0641a0b ILT |
218 | { |
219 | if (pkey != NULL) | |
220 | *pkey = p->second.first; | |
221 | return p->first; | |
222 | } | |
14bfc3f5 | 223 | |
f0641a0b | 224 | Key k; |
b8e6aad9 | 225 | const Stringpool_char* ret = this->add_string(s, &k); |
f0641a0b ILT |
226 | |
227 | const off_t ozero = 0; | |
b8e6aad9 ILT |
228 | std::pair<const Stringpool_char*, Val> element(ret, |
229 | std::make_pair(k, ozero)); | |
230 | std::pair<typename String_set_type::iterator, bool> ins = | |
f0641a0b | 231 | this->string_set_.insert(element); |
a3ad94ed | 232 | gold_assert(ins.second); |
f0641a0b ILT |
233 | |
234 | if (pkey != NULL) | |
235 | *pkey = k; | |
236 | ||
14bfc3f5 ILT |
237 | return ret; |
238 | } | |
239 | ||
240 | // Add a prefix of a string to a string pool. | |
241 | ||
b8e6aad9 ILT |
242 | template<typename Stringpool_char> |
243 | const Stringpool_char* | |
244 | Stringpool_template<Stringpool_char>::add(const Stringpool_char* s, size_t len, | |
245 | Key* pkey) | |
14bfc3f5 ILT |
246 | { |
247 | // FIXME: This implementation should be rewritten when we rewrite | |
248 | // the hash table to avoid copying. | |
b8e6aad9 | 249 | std::basic_string<Stringpool_char> st(s, len); |
f0641a0b | 250 | return this->add(st, pkey); |
14bfc3f5 ILT |
251 | } |
252 | ||
b8e6aad9 ILT |
253 | template<typename Stringpool_char> |
254 | const Stringpool_char* | |
255 | Stringpool_template<Stringpool_char>::find(const Stringpool_char* s, | |
256 | Key* pkey) const | |
61ba1cf9 | 257 | { |
b8e6aad9 | 258 | typename String_set_type::const_iterator p = this->string_set_.find(s); |
61ba1cf9 ILT |
259 | if (p == this->string_set_.end()) |
260 | return NULL; | |
f0641a0b ILT |
261 | |
262 | if (pkey != NULL) | |
263 | *pkey = p->second.first; | |
264 | ||
61ba1cf9 ILT |
265 | return p->first; |
266 | } | |
267 | ||
268 | // Comparison routine used when sorting into an ELF strtab. We want | |
269 | // to sort this so that when one string is a suffix of another, we | |
270 | // always see the shorter string immediately after the longer string. | |
271 | // For example, we want to see these strings in this order: | |
272 | // abcd | |
273 | // cd | |
274 | // d | |
275 | // When strings are not suffixes, we don't care what order they are | |
276 | // in, but we need to ensure that suffixes wind up next to each other. | |
277 | // So we do a reversed lexicographic sort on the reversed string. | |
278 | ||
b8e6aad9 | 279 | template<typename Stringpool_char> |
61ba1cf9 | 280 | bool |
b8e6aad9 | 281 | Stringpool_template<Stringpool_char>::Stringpool_sort_comparison::operator()( |
614f30a2 ILT |
282 | const Stringpool_sort_info& sort_info1, |
283 | const Stringpool_sort_info& sort_info2) const | |
61ba1cf9 | 284 | { |
614f30a2 ILT |
285 | const Stringpool_char* s1 = sort_info1.it->first; |
286 | const Stringpool_char* s2 = sort_info2.it->first; | |
287 | const size_t len1 = sort_info1.string_length; | |
288 | const size_t len2 = sort_info2.string_length; | |
289 | const size_t minlen = len1 < len2 ? len1 : len2; | |
b8e6aad9 ILT |
290 | const Stringpool_char* p1 = s1 + len1 - 1; |
291 | const Stringpool_char* p2 = s2 + len2 - 1; | |
292 | for (size_t i = minlen; i > 0; --i, --p1, --p2) | |
61ba1cf9 ILT |
293 | { |
294 | if (*p1 != *p2) | |
295 | return *p1 > *p2; | |
296 | } | |
297 | return len1 > len2; | |
298 | } | |
299 | ||
300 | // Return whether s1 is a suffix of s2. | |
301 | ||
b8e6aad9 | 302 | template<typename Stringpool_char> |
61ba1cf9 | 303 | bool |
b8e6aad9 | 304 | Stringpool_template<Stringpool_char>::is_suffix(const Stringpool_char* s1, |
614f30a2 ILT |
305 | size_t len1, |
306 | const Stringpool_char* s2, | |
307 | size_t len2) | |
61ba1cf9 | 308 | { |
61ba1cf9 ILT |
309 | if (len1 > len2) |
310 | return false; | |
b8e6aad9 | 311 | return memcmp(s1, s2 + len2 - len1, len1 * sizeof(Stringpool_char)) == 0; |
61ba1cf9 ILT |
312 | } |
313 | ||
314 | // Turn the stringpool into an ELF strtab: determine the offsets of | |
315 | // each string in the table. | |
316 | ||
b8e6aad9 | 317 | template<typename Stringpool_char> |
61ba1cf9 | 318 | void |
b8e6aad9 | 319 | Stringpool_template<Stringpool_char>::set_string_offsets() |
61ba1cf9 | 320 | { |
a3ad94ed ILT |
321 | if (this->strtab_size_ != 0) |
322 | { | |
323 | // We've already computed the offsets. | |
324 | return; | |
325 | } | |
326 | ||
b8e6aad9 ILT |
327 | const size_t charsize = sizeof(Stringpool_char); |
328 | ||
329 | // Offset 0 may be reserved for the empty string. | |
330 | off_t offset = this->zero_null_ ? charsize : 0; | |
614f30a2 | 331 | |
377caf49 ILT |
332 | // Sorting to find suffixes can take over 25% of the total CPU time |
333 | // used by the linker. Since it's merely an optimization to reduce | |
334 | // the strtab size, and gives a relatively small benefit (it's | |
335 | // typically rare for a symbol to be a suffix of another), we only | |
336 | // take the time to sort when the user asks for heavy optimization. | |
337 | if (parameters->optimization_level() < 2) | |
61ba1cf9 | 338 | { |
377caf49 ILT |
339 | for (typename String_set_type::iterator curr = this->string_set_.begin(); |
340 | curr != this->string_set_.end(); | |
341 | curr++) | |
342 | { | |
343 | if (this->zero_null_ && curr->first[0] == 0) | |
344 | curr->second.second = 0; | |
345 | else | |
346 | { | |
347 | curr->second.second = offset; | |
348 | offset += (string_length(curr->first) + 1) * charsize; | |
349 | } | |
350 | } | |
351 | } | |
352 | else | |
353 | { | |
354 | size_t count = this->string_set_.size(); | |
355 | ||
356 | std::vector<Stringpool_sort_info> v; | |
357 | v.reserve(count); | |
358 | ||
359 | for (typename String_set_type::iterator p = this->string_set_.begin(); | |
360 | p != this->string_set_.end(); | |
361 | ++p) | |
362 | v.push_back(Stringpool_sort_info(p, string_length(p->first))); | |
363 | ||
364 | std::sort(v.begin(), v.end(), Stringpool_sort_comparison()); | |
365 | ||
366 | for (typename std::vector<Stringpool_sort_info>::iterator last = v.end(), | |
367 | curr = v.begin(); | |
368 | curr != v.end(); | |
369 | last = curr++) | |
370 | { | |
371 | if (this->zero_null_ && curr->it->first[0] == 0) | |
372 | curr->it->second.second = 0; | |
373 | else if (last != v.end() | |
374 | && is_suffix(curr->it->first, curr->string_length, | |
375 | last->it->first, last->string_length)) | |
376 | curr->it->second.second = (last->it->second.second | |
377 | + ((last->string_length | |
378 | - curr->string_length) | |
379 | * charsize)); | |
380 | else | |
381 | { | |
382 | curr->it->second.second = offset; | |
383 | offset += (curr->string_length + 1) * charsize; | |
384 | } | |
385 | } | |
61ba1cf9 ILT |
386 | } |
387 | ||
388 | this->strtab_size_ = offset; | |
389 | } | |
390 | ||
391 | // Get the offset of a string in the ELF strtab. The string must | |
392 | // exist. | |
393 | ||
b8e6aad9 | 394 | template<typename Stringpool_char> |
61ba1cf9 | 395 | off_t |
b8e6aad9 ILT |
396 | Stringpool_template<Stringpool_char>::get_offset(const Stringpool_char* s) |
397 | const | |
61ba1cf9 | 398 | { |
a3ad94ed | 399 | gold_assert(this->strtab_size_ != 0); |
b8e6aad9 | 400 | typename String_set_type::const_iterator p = this->string_set_.find(s); |
61ba1cf9 | 401 | if (p != this->string_set_.end()) |
f0641a0b | 402 | return p->second.second; |
a3ad94ed | 403 | gold_unreachable(); |
61ba1cf9 ILT |
404 | } |
405 | ||
406 | // Write the ELF strtab into the output file at the specified offset. | |
407 | ||
b8e6aad9 | 408 | template<typename Stringpool_char> |
61ba1cf9 | 409 | void |
b8e6aad9 | 410 | Stringpool_template<Stringpool_char>::write(Output_file* of, off_t offset) |
61ba1cf9 | 411 | { |
a3ad94ed | 412 | gold_assert(this->strtab_size_ != 0); |
61ba1cf9 ILT |
413 | unsigned char* viewu = of->get_output_view(offset, this->strtab_size_); |
414 | char* view = reinterpret_cast<char*>(viewu); | |
b8e6aad9 ILT |
415 | if (this->zero_null_) |
416 | view[0] = '\0'; | |
417 | for (typename String_set_type::const_iterator p = this->string_set_.begin(); | |
61ba1cf9 ILT |
418 | p != this->string_set_.end(); |
419 | ++p) | |
b8e6aad9 ILT |
420 | memcpy(view + p->second.second, p->first, |
421 | (string_length(p->first) + 1) * sizeof(Stringpool_char)); | |
61ba1cf9 ILT |
422 | of->write_output_view(offset, this->strtab_size_, viewu); |
423 | } | |
424 | ||
b8e6aad9 ILT |
425 | // Instantiate the templates we need. |
426 | ||
427 | template | |
428 | class Stringpool_template<char>; | |
429 | ||
430 | template | |
431 | class Stringpool_template<uint16_t>; | |
432 | ||
433 | template | |
434 | class Stringpool_template<uint32_t>; | |
435 | ||
14bfc3f5 | 436 | } // End namespace gold. |