| 1 | // stringpool.cc -- a string pool for gold |
| 2 | |
| 3 | // Copyright 2006, 2007, 2008 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 | |
| 23 | #include "gold.h" |
| 24 | |
| 25 | #include <cstring> |
| 26 | #include <algorithm> |
| 27 | #include <vector> |
| 28 | |
| 29 | #include "output.h" |
| 30 | #include "parameters.h" |
| 31 | #include "stringpool.h" |
| 32 | |
| 33 | namespace gold |
| 34 | { |
| 35 | |
| 36 | template<typename Stringpool_char> |
| 37 | Stringpool_template<Stringpool_char>::Stringpool_template() |
| 38 | : string_set_(), key_to_offset_(), strings_(), strtab_size_(0), |
| 39 | zero_null_(true) |
| 40 | { |
| 41 | } |
| 42 | |
| 43 | template<typename Stringpool_char> |
| 44 | void |
| 45 | Stringpool_template<Stringpool_char>::clear() |
| 46 | { |
| 47 | for (typename std::list<Stringdata*>::iterator p = this->strings_.begin(); |
| 48 | p != this->strings_.end(); |
| 49 | ++p) |
| 50 | delete[] reinterpret_cast<char*>(*p); |
| 51 | this->strings_.clear(); |
| 52 | this->key_to_offset_.clear(); |
| 53 | this->string_set_.clear(); |
| 54 | } |
| 55 | |
| 56 | template<typename Stringpool_char> |
| 57 | Stringpool_template<Stringpool_char>::~Stringpool_template() |
| 58 | { |
| 59 | this->clear(); |
| 60 | } |
| 61 | |
| 62 | // Resize the internal hashtable with the expectation we'll get n new |
| 63 | // elements. Note that the hashtable constructor takes a "number of |
| 64 | // buckets you'd like," rather than "number of elements you'd like," |
| 65 | // but that's the best we can do. |
| 66 | |
| 67 | template<typename Stringpool_char> |
| 68 | void |
| 69 | Stringpool_template<Stringpool_char>::reserve(unsigned int n) |
| 70 | { |
| 71 | this->key_to_offset_.reserve(n); |
| 72 | |
| 73 | #if defined(HAVE_TR1_UNORDERED_MAP) |
| 74 | // rehash() implementation is broken in gcc 4.0.3's stl |
| 75 | //this->string_set_.rehash(this->string_set_.size() + n); |
| 76 | //return; |
| 77 | #elif defined(HAVE_EXT_HASH_MAP) |
| 78 | this->string_set_.resize(this->string_set_.size() + n); |
| 79 | return; |
| 80 | #endif |
| 81 | |
| 82 | // This is the generic "reserve" code, if no #ifdef above triggers. |
| 83 | String_set_type new_string_set(this->string_set_.size() + n); |
| 84 | new_string_set.insert(this->string_set_.begin(), this->string_set_.end()); |
| 85 | this->string_set_.swap(new_string_set); |
| 86 | } |
| 87 | |
| 88 | // Return the length of a string of arbitrary character type. |
| 89 | |
| 90 | template<typename Stringpool_char> |
| 91 | size_t |
| 92 | Stringpool_template<Stringpool_char>::string_length(const Stringpool_char* p) |
| 93 | { |
| 94 | size_t len = 0; |
| 95 | for (; *p != 0; ++p) |
| 96 | ++len; |
| 97 | return len; |
| 98 | } |
| 99 | |
| 100 | // Specialize string_length for char. Maybe we could just use |
| 101 | // std::char_traits<>::length? |
| 102 | |
| 103 | template<> |
| 104 | inline size_t |
| 105 | Stringpool_template<char>::string_length(const char* p) |
| 106 | { |
| 107 | return strlen(p); |
| 108 | } |
| 109 | |
| 110 | // Compare two strings of arbitrary character type for equality. |
| 111 | |
| 112 | template<typename Stringpool_char> |
| 113 | bool |
| 114 | Stringpool_template<Stringpool_char>::string_equal(const Stringpool_char* s1, |
| 115 | const Stringpool_char* s2) |
| 116 | { |
| 117 | while (*s1 != 0) |
| 118 | if (*s1++ != *s2++) |
| 119 | return false; |
| 120 | return *s2 == 0; |
| 121 | } |
| 122 | |
| 123 | // Specialize string_equal for char. |
| 124 | |
| 125 | template<> |
| 126 | inline bool |
| 127 | Stringpool_template<char>::string_equal(const char* s1, const char* s2) |
| 128 | { |
| 129 | return strcmp(s1, s2) == 0; |
| 130 | } |
| 131 | |
| 132 | // Equality comparison function for the hash table. |
| 133 | |
| 134 | template<typename Stringpool_char> |
| 135 | inline bool |
| 136 | Stringpool_template<Stringpool_char>::Stringpool_eq::operator()( |
| 137 | const Hashkey& h1, |
| 138 | const Hashkey& h2) const |
| 139 | { |
| 140 | return (h1.hash_code == h2.hash_code |
| 141 | && h1.length == h2.length |
| 142 | && (h1.string == h2.string |
| 143 | || memcmp(h1.string, h2.string, |
| 144 | h1.length * sizeof(Stringpool_char)) == 0)); |
| 145 | } |
| 146 | |
| 147 | // Hash function. The length is in characters, not bytes. |
| 148 | |
| 149 | template<typename Stringpool_char> |
| 150 | size_t |
| 151 | Stringpool_template<Stringpool_char>::string_hash(const Stringpool_char* s, |
| 152 | size_t length) |
| 153 | { |
| 154 | // This is the hash function used by the dynamic linker for |
| 155 | // DT_GNU_HASH entries. I compared this to a Fowler/Noll/Vo hash |
| 156 | // for a C++ program with 385,775 global symbols. This hash |
| 157 | // function was very slightly worse. However, it is much faster to |
| 158 | // compute. Overall wall clock time was a win. |
| 159 | const unsigned char* p = reinterpret_cast<const unsigned char*>(s); |
| 160 | size_t h = 5381; |
| 161 | for (size_t i = 0; i < length * sizeof(Stringpool_char); ++i) |
| 162 | h = h * 33 + *p++; |
| 163 | return h; |
| 164 | } |
| 165 | |
| 166 | // Add the string S to the list of canonical strings. Return a |
| 167 | // pointer to the canonical string. If PKEY is not NULL, set *PKEY to |
| 168 | // the key. LENGTH is the length of S in characters. Note that S may |
| 169 | // not be NUL terminated. |
| 170 | |
| 171 | template<typename Stringpool_char> |
| 172 | const Stringpool_char* |
| 173 | Stringpool_template<Stringpool_char>::add_string(const Stringpool_char* s, |
| 174 | size_t len) |
| 175 | { |
| 176 | // We are in trouble if we've already computed the string offsets. |
| 177 | gold_assert(this->strtab_size_ == 0); |
| 178 | |
| 179 | // The size we allocate for a new Stringdata. |
| 180 | const size_t buffer_size = 1000; |
| 181 | // The amount we multiply the Stringdata index when calculating the |
| 182 | // key. |
| 183 | const size_t key_mult = 1024; |
| 184 | gold_assert(key_mult >= buffer_size); |
| 185 | |
| 186 | // Convert len to the number of bytes we need to allocate, including |
| 187 | // the null character. |
| 188 | len = (len + 1) * sizeof(Stringpool_char); |
| 189 | |
| 190 | size_t alc; |
| 191 | bool front = true; |
| 192 | if (len > buffer_size) |
| 193 | { |
| 194 | alc = sizeof(Stringdata) + len; |
| 195 | front = false; |
| 196 | } |
| 197 | else if (this->strings_.empty()) |
| 198 | alc = sizeof(Stringdata) + buffer_size; |
| 199 | else |
| 200 | { |
| 201 | Stringdata *psd = this->strings_.front(); |
| 202 | if (len > psd->alc - psd->len) |
| 203 | alc = sizeof(Stringdata) + buffer_size; |
| 204 | else |
| 205 | { |
| 206 | char* ret = psd->data + psd->len; |
| 207 | memcpy(ret, s, len - sizeof(Stringpool_char)); |
| 208 | memset(ret + len - sizeof(Stringpool_char), 0, |
| 209 | sizeof(Stringpool_char)); |
| 210 | |
| 211 | psd->len += len; |
| 212 | |
| 213 | return reinterpret_cast<const Stringpool_char*>(ret); |
| 214 | } |
| 215 | } |
| 216 | |
| 217 | Stringdata *psd = reinterpret_cast<Stringdata*>(new char[alc]); |
| 218 | psd->alc = alc - sizeof(Stringdata); |
| 219 | memcpy(psd->data, s, len - sizeof(Stringpool_char)); |
| 220 | memset(psd->data + len - sizeof(Stringpool_char), 0, |
| 221 | sizeof(Stringpool_char)); |
| 222 | psd->len = len; |
| 223 | |
| 224 | if (front) |
| 225 | this->strings_.push_front(psd); |
| 226 | else |
| 227 | this->strings_.push_back(psd); |
| 228 | |
| 229 | return reinterpret_cast<const Stringpool_char*>(psd->data); |
| 230 | } |
| 231 | |
| 232 | // Add a string to a string pool. |
| 233 | |
| 234 | template<typename Stringpool_char> |
| 235 | const Stringpool_char* |
| 236 | Stringpool_template<Stringpool_char>::add(const Stringpool_char* s, bool copy, |
| 237 | Key* pkey) |
| 238 | { |
| 239 | return this->add_with_length(s, string_length(s), copy, pkey); |
| 240 | } |
| 241 | |
| 242 | template<typename Stringpool_char> |
| 243 | const Stringpool_char* |
| 244 | Stringpool_template<Stringpool_char>::add_with_length(const Stringpool_char* s, |
| 245 | size_t length, |
| 246 | bool copy, |
| 247 | Key* pkey) |
| 248 | { |
| 249 | typedef std::pair<typename String_set_type::iterator, bool> Insert_type; |
| 250 | |
| 251 | const Key k = this->key_to_offset_.size(); |
| 252 | |
| 253 | if (!copy) |
| 254 | { |
| 255 | // When we don't need to copy the string, we can call insert |
| 256 | // directly. |
| 257 | |
| 258 | std::pair<Hashkey, Hashval> element(Hashkey(s, length), k); |
| 259 | |
| 260 | Insert_type ins = this->string_set_.insert(element); |
| 261 | |
| 262 | typename String_set_type::const_iterator p = ins.first; |
| 263 | |
| 264 | if (ins.second) |
| 265 | { |
| 266 | // We just added the string. The key value has now been |
| 267 | // used. |
| 268 | this->key_to_offset_.push_back(0); |
| 269 | } |
| 270 | else |
| 271 | { |
| 272 | gold_assert(k != p->second); |
| 273 | } |
| 274 | |
| 275 | if (pkey != NULL) |
| 276 | *pkey = p->second; |
| 277 | return p->first.string; |
| 278 | } |
| 279 | |
| 280 | // When we have to copy the string, we look it up twice in the hash |
| 281 | // table. The problem is that we can't insert S before we |
| 282 | // canonicalize it by copying it into the canonical list. The hash |
| 283 | // code will only be computed once. |
| 284 | |
| 285 | Hashkey hk(s, length); |
| 286 | typename String_set_type::const_iterator p = this->string_set_.find(hk); |
| 287 | if (p != this->string_set_.end()) |
| 288 | { |
| 289 | if (pkey != NULL) |
| 290 | *pkey = p->second; |
| 291 | return p->first.string; |
| 292 | } |
| 293 | |
| 294 | this->key_to_offset_.push_back(0); |
| 295 | |
| 296 | hk.string = this->add_string(s, length); |
| 297 | // The contents of the string stay the same, so we don't need to |
| 298 | // adjust hk.hash_code or hk.length. |
| 299 | |
| 300 | std::pair<Hashkey, Hashval> element(hk, k); |
| 301 | |
| 302 | Insert_type ins = this->string_set_.insert(element); |
| 303 | gold_assert(ins.second); |
| 304 | |
| 305 | if (pkey != NULL) |
| 306 | *pkey = k; |
| 307 | return hk.string; |
| 308 | } |
| 309 | |
| 310 | template<typename Stringpool_char> |
| 311 | const Stringpool_char* |
| 312 | Stringpool_template<Stringpool_char>::find(const Stringpool_char* s, |
| 313 | Key* pkey) const |
| 314 | { |
| 315 | Hashkey hk(s); |
| 316 | typename String_set_type::const_iterator p = this->string_set_.find(hk); |
| 317 | if (p == this->string_set_.end()) |
| 318 | return NULL; |
| 319 | |
| 320 | if (pkey != NULL) |
| 321 | *pkey = p->second; |
| 322 | |
| 323 | return p->first.string; |
| 324 | } |
| 325 | |
| 326 | // Comparison routine used when sorting into an ELF strtab. We want |
| 327 | // to sort this so that when one string is a suffix of another, we |
| 328 | // always see the shorter string immediately after the longer string. |
| 329 | // For example, we want to see these strings in this order: |
| 330 | // abcd |
| 331 | // cd |
| 332 | // d |
| 333 | // When strings are not suffixes, we don't care what order they are |
| 334 | // in, but we need to ensure that suffixes wind up next to each other. |
| 335 | // So we do a reversed lexicographic sort on the reversed string. |
| 336 | |
| 337 | template<typename Stringpool_char> |
| 338 | bool |
| 339 | Stringpool_template<Stringpool_char>::Stringpool_sort_comparison::operator()( |
| 340 | const Stringpool_sort_info& sort_info1, |
| 341 | const Stringpool_sort_info& sort_info2) const |
| 342 | { |
| 343 | const Hashkey& h1(sort_info1->first); |
| 344 | const Hashkey& h2(sort_info2->first); |
| 345 | const Stringpool_char* s1 = h1.string; |
| 346 | const Stringpool_char* s2 = h2.string; |
| 347 | const size_t len1 = h1.length; |
| 348 | const size_t len2 = h2.length; |
| 349 | const size_t minlen = len1 < len2 ? len1 : len2; |
| 350 | const Stringpool_char* p1 = s1 + len1 - 1; |
| 351 | const Stringpool_char* p2 = s2 + len2 - 1; |
| 352 | for (size_t i = minlen; i > 0; --i, --p1, --p2) |
| 353 | { |
| 354 | if (*p1 != *p2) |
| 355 | return *p1 > *p2; |
| 356 | } |
| 357 | return len1 > len2; |
| 358 | } |
| 359 | |
| 360 | // Return whether s1 is a suffix of s2. |
| 361 | |
| 362 | template<typename Stringpool_char> |
| 363 | bool |
| 364 | Stringpool_template<Stringpool_char>::is_suffix(const Stringpool_char* s1, |
| 365 | size_t len1, |
| 366 | const Stringpool_char* s2, |
| 367 | size_t len2) |
| 368 | { |
| 369 | if (len1 > len2) |
| 370 | return false; |
| 371 | return memcmp(s1, s2 + len2 - len1, len1 * sizeof(Stringpool_char)) == 0; |
| 372 | } |
| 373 | |
| 374 | // Turn the stringpool into an ELF strtab: determine the offsets of |
| 375 | // each string in the table. |
| 376 | |
| 377 | template<typename Stringpool_char> |
| 378 | void |
| 379 | Stringpool_template<Stringpool_char>::set_string_offsets() |
| 380 | { |
| 381 | if (this->strtab_size_ != 0) |
| 382 | { |
| 383 | // We've already computed the offsets. |
| 384 | return; |
| 385 | } |
| 386 | |
| 387 | const size_t charsize = sizeof(Stringpool_char); |
| 388 | |
| 389 | // Offset 0 may be reserved for the empty string. |
| 390 | section_offset_type offset = this->zero_null_ ? charsize : 0; |
| 391 | |
| 392 | // Sorting to find suffixes can take over 25% of the total CPU time |
| 393 | // used by the linker. Since it's merely an optimization to reduce |
| 394 | // the strtab size, and gives a relatively small benefit (it's |
| 395 | // typically rare for a symbol to be a suffix of another), we only |
| 396 | // take the time to sort when the user asks for heavy optimization. |
| 397 | if (parameters->options().optimize() < 2) |
| 398 | { |
| 399 | for (typename String_set_type::iterator curr = this->string_set_.begin(); |
| 400 | curr != this->string_set_.end(); |
| 401 | curr++) |
| 402 | { |
| 403 | section_offset_type* poff = &this->key_to_offset_[curr->second]; |
| 404 | if (this->zero_null_ && curr->first.string[0] == 0) |
| 405 | *poff = 0; |
| 406 | else |
| 407 | { |
| 408 | *poff = offset; |
| 409 | offset += (curr->first.length + 1) * charsize; |
| 410 | } |
| 411 | } |
| 412 | } |
| 413 | else |
| 414 | { |
| 415 | size_t count = this->string_set_.size(); |
| 416 | |
| 417 | std::vector<Stringpool_sort_info> v; |
| 418 | v.reserve(count); |
| 419 | |
| 420 | for (typename String_set_type::iterator p = this->string_set_.begin(); |
| 421 | p != this->string_set_.end(); |
| 422 | ++p) |
| 423 | v.push_back(Stringpool_sort_info(p)); |
| 424 | |
| 425 | std::sort(v.begin(), v.end(), Stringpool_sort_comparison()); |
| 426 | |
| 427 | section_offset_type last_offset = -1; |
| 428 | for (typename std::vector<Stringpool_sort_info>::iterator last = v.end(), |
| 429 | curr = v.begin(); |
| 430 | curr != v.end(); |
| 431 | last = curr++) |
| 432 | { |
| 433 | section_offset_type this_offset; |
| 434 | if (this->zero_null_ && (*curr)->first.string[0] == 0) |
| 435 | this_offset = 0; |
| 436 | else if (last != v.end() |
| 437 | && is_suffix((*curr)->first.string, |
| 438 | (*curr)->first.length, |
| 439 | (*last)->first.string, |
| 440 | (*last)->first.length)) |
| 441 | this_offset = (last_offset |
| 442 | + (((*last)->first.length - (*curr)->first.length) |
| 443 | * charsize)); |
| 444 | else |
| 445 | { |
| 446 | this_offset = offset; |
| 447 | offset += ((*curr)->first.length + 1) * charsize; |
| 448 | } |
| 449 | this->key_to_offset_[(*curr)->second] = this_offset; |
| 450 | last_offset = this_offset; |
| 451 | } |
| 452 | } |
| 453 | |
| 454 | this->strtab_size_ = offset; |
| 455 | } |
| 456 | |
| 457 | // Get the offset of a string in the ELF strtab. The string must |
| 458 | // exist. |
| 459 | |
| 460 | template<typename Stringpool_char> |
| 461 | section_offset_type |
| 462 | Stringpool_template<Stringpool_char>::get_offset(const Stringpool_char* s) |
| 463 | const |
| 464 | { |
| 465 | return this->get_offset_with_length(s, string_length(s)); |
| 466 | } |
| 467 | |
| 468 | template<typename Stringpool_char> |
| 469 | section_offset_type |
| 470 | Stringpool_template<Stringpool_char>::get_offset_with_length( |
| 471 | const Stringpool_char* s, |
| 472 | size_t length) const |
| 473 | { |
| 474 | gold_assert(this->strtab_size_ != 0); |
| 475 | Hashkey hk(s, length); |
| 476 | typename String_set_type::const_iterator p = this->string_set_.find(hk); |
| 477 | if (p != this->string_set_.end()) |
| 478 | return this->key_to_offset_[p->second]; |
| 479 | gold_unreachable(); |
| 480 | } |
| 481 | |
| 482 | // Write the ELF strtab into the buffer. |
| 483 | |
| 484 | template<typename Stringpool_char> |
| 485 | void |
| 486 | Stringpool_template<Stringpool_char>::write_to_buffer( |
| 487 | unsigned char* buffer, |
| 488 | section_size_type bufsize) |
| 489 | { |
| 490 | gold_assert(this->strtab_size_ != 0); |
| 491 | gold_assert(bufsize >= this->strtab_size_); |
| 492 | if (this->zero_null_) |
| 493 | buffer[0] = '\0'; |
| 494 | for (typename String_set_type::const_iterator p = this->string_set_.begin(); |
| 495 | p != this->string_set_.end(); |
| 496 | ++p) |
| 497 | { |
| 498 | const int len = (p->first.length + 1) * sizeof(Stringpool_char); |
| 499 | const section_offset_type offset = this->key_to_offset_[p->second]; |
| 500 | gold_assert(static_cast<section_size_type>(offset) + len |
| 501 | <= this->strtab_size_); |
| 502 | memcpy(buffer + offset, p->first.string, len); |
| 503 | } |
| 504 | } |
| 505 | |
| 506 | // Write the ELF strtab into the output file at the specified offset. |
| 507 | |
| 508 | template<typename Stringpool_char> |
| 509 | void |
| 510 | Stringpool_template<Stringpool_char>::write(Output_file* of, off_t offset) |
| 511 | { |
| 512 | gold_assert(this->strtab_size_ != 0); |
| 513 | unsigned char* view = of->get_output_view(offset, this->strtab_size_); |
| 514 | this->write_to_buffer(view, this->strtab_size_); |
| 515 | of->write_output_view(offset, this->strtab_size_, view); |
| 516 | } |
| 517 | |
| 518 | // Print statistical information to stderr. This is used for --stats. |
| 519 | |
| 520 | template<typename Stringpool_char> |
| 521 | void |
| 522 | Stringpool_template<Stringpool_char>::print_stats(const char* name) const |
| 523 | { |
| 524 | #if defined(HAVE_TR1_UNORDERED_MAP) || defined(HAVE_EXT_HASH_MAP) |
| 525 | fprintf(stderr, _("%s: %s entries: %zu; buckets: %zu\n"), |
| 526 | program_name, name, this->string_set_.size(), |
| 527 | this->string_set_.bucket_count()); |
| 528 | #else |
| 529 | fprintf(stderr, _("%s: %s entries: %zu\n"), |
| 530 | program_name, name, this->table_.size()); |
| 531 | #endif |
| 532 | fprintf(stderr, _("%s: %s Stringdata structures: %zu\n"), |
| 533 | program_name, name, this->strings_.size()); |
| 534 | } |
| 535 | |
| 536 | // Instantiate the templates we need. |
| 537 | |
| 538 | template |
| 539 | class Stringpool_template<char>; |
| 540 | |
| 541 | template |
| 542 | class Stringpool_template<uint16_t>; |
| 543 | |
| 544 | template |
| 545 | class Stringpool_template<uint32_t>; |
| 546 | |
| 547 | } // End namespace gold. |