daily update
[binutils.git] / gold / stringpool.cc
blob434b2d62565d7e178bb559be45609f84c4be988d
1 // stringpool.cc -- a string pool for gold
3 // Copyright 2006, 2007, 2008 Free Software Foundation, Inc.
4 // Written by Ian Lance Taylor <iant@google.com>.
6 // This file is part of gold.
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.
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.
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.
23 #include "gold.h"
25 #include <cstring>
26 #include <algorithm>
27 #include <vector>
29 #include "output.h"
30 #include "parameters.h"
31 #include "stringpool.h"
33 namespace gold
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), optimize_(false), offset_(sizeof(Stringpool_char))
41 if (parameters->options_valid() && parameters->options().optimize() >= 2)
42 this->optimize_ = true;
45 template<typename Stringpool_char>
46 void
47 Stringpool_template<Stringpool_char>::clear()
49 for (typename std::list<Stringdata*>::iterator p = this->strings_.begin();
50 p != this->strings_.end();
51 ++p)
52 delete[] reinterpret_cast<char*>(*p);
53 this->strings_.clear();
54 this->key_to_offset_.clear();
55 this->string_set_.clear();
58 template<typename Stringpool_char>
59 Stringpool_template<Stringpool_char>::~Stringpool_template()
61 this->clear();
64 // Resize the internal hashtable with the expectation we'll get n new
65 // elements. Note that the hashtable constructor takes a "number of
66 // buckets you'd like," rather than "number of elements you'd like,"
67 // but that's the best we can do.
69 template<typename Stringpool_char>
70 void
71 Stringpool_template<Stringpool_char>::reserve(unsigned int n)
73 this->key_to_offset_.reserve(n);
75 #if defined(HAVE_TR1_UNORDERED_MAP)
76 // rehash() implementation is broken in gcc 4.0.3's stl
77 //this->string_set_.rehash(this->string_set_.size() + n);
78 //return;
79 #elif defined(HAVE_EXT_HASH_MAP)
80 this->string_set_.resize(this->string_set_.size() + n);
81 return;
82 #endif
84 // This is the generic "reserve" code, if no #ifdef above triggers.
85 String_set_type new_string_set(this->string_set_.size() + n);
86 new_string_set.insert(this->string_set_.begin(), this->string_set_.end());
87 this->string_set_.swap(new_string_set);
90 // Compare two strings of arbitrary character type for equality.
92 template<typename Stringpool_char>
93 bool
94 Stringpool_template<Stringpool_char>::string_equal(const Stringpool_char* s1,
95 const Stringpool_char* s2)
97 while (*s1 != 0)
98 if (*s1++ != *s2++)
99 return false;
100 return *s2 == 0;
103 // Specialize string_equal for char.
105 template<>
106 inline bool
107 Stringpool_template<char>::string_equal(const char* s1, const char* s2)
109 return strcmp(s1, s2) == 0;
112 // Equality comparison function for the hash table.
114 template<typename Stringpool_char>
115 inline bool
116 Stringpool_template<Stringpool_char>::Stringpool_eq::operator()(
117 const Hashkey& h1,
118 const Hashkey& h2) const
120 return (h1.hash_code == h2.hash_code
121 && h1.length == h2.length
122 && (h1.string == h2.string
123 || memcmp(h1.string, h2.string,
124 h1.length * sizeof(Stringpool_char)) == 0));
127 // Hash function. The length is in characters, not bytes.
129 template<typename Stringpool_char>
130 size_t
131 Stringpool_template<Stringpool_char>::string_hash(const Stringpool_char* s,
132 size_t length)
134 return gold::string_hash<Stringpool_char>(s, length);
137 // Add the string S to the list of canonical strings. Return a
138 // pointer to the canonical string. If PKEY is not NULL, set *PKEY to
139 // the key. LENGTH is the length of S in characters. Note that S may
140 // not be NUL terminated.
142 template<typename Stringpool_char>
143 const Stringpool_char*
144 Stringpool_template<Stringpool_char>::add_string(const Stringpool_char* s,
145 size_t len)
147 // We are in trouble if we've already computed the string offsets.
148 gold_assert(this->strtab_size_ == 0);
150 // The size we allocate for a new Stringdata.
151 const size_t buffer_size = 1000;
152 // The amount we multiply the Stringdata index when calculating the
153 // key.
154 const size_t key_mult = 1024;
155 gold_assert(key_mult >= buffer_size);
157 // Convert len to the number of bytes we need to allocate, including
158 // the null character.
159 len = (len + 1) * sizeof(Stringpool_char);
161 size_t alc;
162 bool front = true;
163 if (len > buffer_size)
165 alc = sizeof(Stringdata) + len;
166 front = false;
168 else if (this->strings_.empty())
169 alc = sizeof(Stringdata) + buffer_size;
170 else
172 Stringdata* psd = this->strings_.front();
173 if (len > psd->alc - psd->len)
174 alc = sizeof(Stringdata) + buffer_size;
175 else
177 char* ret = psd->data + psd->len;
178 memcpy(ret, s, len - sizeof(Stringpool_char));
179 memset(ret + len - sizeof(Stringpool_char), 0,
180 sizeof(Stringpool_char));
182 psd->len += len;
184 return reinterpret_cast<const Stringpool_char*>(ret);
188 Stringdata* psd = reinterpret_cast<Stringdata*>(new char[alc]);
189 psd->alc = alc - sizeof(Stringdata);
190 memcpy(psd->data, s, len - sizeof(Stringpool_char));
191 memset(psd->data + len - sizeof(Stringpool_char), 0,
192 sizeof(Stringpool_char));
193 psd->len = len;
195 if (front)
196 this->strings_.push_front(psd);
197 else
198 this->strings_.push_back(psd);
200 return reinterpret_cast<const Stringpool_char*>(psd->data);
203 // Add a string to a string pool.
205 template<typename Stringpool_char>
206 const Stringpool_char*
207 Stringpool_template<Stringpool_char>::add(const Stringpool_char* s, bool copy,
208 Key* pkey)
210 return this->add_with_length(s, string_length(s), copy, pkey);
213 // Add a new key offset entry.
215 template<typename Stringpool_char>
216 void
217 Stringpool_template<Stringpool_char>::new_key_offset(size_t length)
219 section_offset_type offset;
220 if (this->zero_null_ && length == 0)
221 offset = 0;
222 else
224 offset = this->offset_;
225 this->offset_ += (length + 1) * sizeof(Stringpool_char);
227 this->key_to_offset_.push_back(offset);
230 template<typename Stringpool_char>
231 const Stringpool_char*
232 Stringpool_template<Stringpool_char>::add_with_length(const Stringpool_char* s,
233 size_t length,
234 bool copy,
235 Key* pkey)
237 typedef std::pair<typename String_set_type::iterator, bool> Insert_type;
239 // We add 1 so that 0 is always invalid.
240 const Key k = this->key_to_offset_.size() + 1;
242 if (!copy)
244 // When we don't need to copy the string, we can call insert
245 // directly.
247 std::pair<Hashkey, Hashval> element(Hashkey(s, length), k);
249 Insert_type ins = this->string_set_.insert(element);
251 typename String_set_type::const_iterator p = ins.first;
253 if (ins.second)
255 // We just added the string. The key value has now been
256 // used.
257 this->new_key_offset(length);
259 else
261 gold_assert(k != p->second);
264 if (pkey != NULL)
265 *pkey = p->second;
266 return p->first.string;
269 // When we have to copy the string, we look it up twice in the hash
270 // table. The problem is that we can't insert S before we
271 // canonicalize it by copying it into the canonical list. The hash
272 // code will only be computed once.
274 Hashkey hk(s, length);
275 typename String_set_type::const_iterator p = this->string_set_.find(hk);
276 if (p != this->string_set_.end())
278 if (pkey != NULL)
279 *pkey = p->second;
280 return p->first.string;
283 this->new_key_offset(length);
285 hk.string = this->add_string(s, length);
286 // The contents of the string stay the same, so we don't need to
287 // adjust hk.hash_code or hk.length.
289 std::pair<Hashkey, Hashval> element(hk, k);
291 Insert_type ins = this->string_set_.insert(element);
292 gold_assert(ins.second);
294 if (pkey != NULL)
295 *pkey = k;
296 return hk.string;
299 template<typename Stringpool_char>
300 const Stringpool_char*
301 Stringpool_template<Stringpool_char>::find(const Stringpool_char* s,
302 Key* pkey) const
304 Hashkey hk(s);
305 typename String_set_type::const_iterator p = this->string_set_.find(hk);
306 if (p == this->string_set_.end())
307 return NULL;
309 if (pkey != NULL)
310 *pkey = p->second;
312 return p->first.string;
315 // Comparison routine used when sorting into an ELF strtab. We want
316 // to sort this so that when one string is a suffix of another, we
317 // always see the shorter string immediately after the longer string.
318 // For example, we want to see these strings in this order:
319 // abcd
320 // cd
321 // d
322 // When strings are not suffixes, we don't care what order they are
323 // in, but we need to ensure that suffixes wind up next to each other.
324 // So we do a reversed lexicographic sort on the reversed string.
326 template<typename Stringpool_char>
327 bool
328 Stringpool_template<Stringpool_char>::Stringpool_sort_comparison::operator()(
329 const Stringpool_sort_info& sort_info1,
330 const Stringpool_sort_info& sort_info2) const
332 const Hashkey& h1(sort_info1->first);
333 const Hashkey& h2(sort_info2->first);
334 const Stringpool_char* s1 = h1.string;
335 const Stringpool_char* s2 = h2.string;
336 const size_t len1 = h1.length;
337 const size_t len2 = h2.length;
338 const size_t minlen = len1 < len2 ? len1 : len2;
339 const Stringpool_char* p1 = s1 + len1 - 1;
340 const Stringpool_char* p2 = s2 + len2 - 1;
341 for (size_t i = minlen; i > 0; --i, --p1, --p2)
343 if (*p1 != *p2)
344 return *p1 > *p2;
346 return len1 > len2;
349 // Return whether s1 is a suffix of s2.
351 template<typename Stringpool_char>
352 bool
353 Stringpool_template<Stringpool_char>::is_suffix(const Stringpool_char* s1,
354 size_t len1,
355 const Stringpool_char* s2,
356 size_t len2)
358 if (len1 > len2)
359 return false;
360 return memcmp(s1, s2 + len2 - len1, len1 * sizeof(Stringpool_char)) == 0;
363 // Turn the stringpool into an ELF strtab: determine the offsets of
364 // each string in the table.
366 template<typename Stringpool_char>
367 void
368 Stringpool_template<Stringpool_char>::set_string_offsets()
370 if (this->strtab_size_ != 0)
372 // We've already computed the offsets.
373 return;
376 const size_t charsize = sizeof(Stringpool_char);
378 // Offset 0 may be reserved for the empty string.
379 section_offset_type offset = this->zero_null_ ? charsize : 0;
381 // Sorting to find suffixes can take over 25% of the total CPU time
382 // used by the linker. Since it's merely an optimization to reduce
383 // the strtab size, and gives a relatively small benefit (it's
384 // typically rare for a symbol to be a suffix of another), we only
385 // take the time to sort when the user asks for heavy optimization.
386 if (!this->optimize_)
388 // If we are not optimizing, the offsets are already assigned.
389 offset = this->offset_;
391 else
393 size_t count = this->string_set_.size();
395 std::vector<Stringpool_sort_info> v;
396 v.reserve(count);
398 for (typename String_set_type::iterator p = this->string_set_.begin();
399 p != this->string_set_.end();
400 ++p)
401 v.push_back(Stringpool_sort_info(p));
403 std::sort(v.begin(), v.end(), Stringpool_sort_comparison());
405 section_offset_type last_offset = -1;
406 for (typename std::vector<Stringpool_sort_info>::iterator last = v.end(),
407 curr = v.begin();
408 curr != v.end();
409 last = curr++)
411 section_offset_type this_offset;
412 if (this->zero_null_ && (*curr)->first.string[0] == 0)
413 this_offset = 0;
414 else if (last != v.end()
415 && is_suffix((*curr)->first.string,
416 (*curr)->first.length,
417 (*last)->first.string,
418 (*last)->first.length))
419 this_offset = (last_offset
420 + (((*last)->first.length - (*curr)->first.length)
421 * charsize));
422 else
424 this_offset = offset;
425 offset += ((*curr)->first.length + 1) * charsize;
427 this->key_to_offset_[(*curr)->second - 1] = this_offset;
428 last_offset = this_offset;
432 this->strtab_size_ = offset;
435 // Get the offset of a string in the ELF strtab. The string must
436 // exist.
438 template<typename Stringpool_char>
439 section_offset_type
440 Stringpool_template<Stringpool_char>::get_offset(const Stringpool_char* s)
441 const
443 return this->get_offset_with_length(s, string_length(s));
446 template<typename Stringpool_char>
447 section_offset_type
448 Stringpool_template<Stringpool_char>::get_offset_with_length(
449 const Stringpool_char* s,
450 size_t length) const
452 gold_assert(this->strtab_size_ != 0);
453 Hashkey hk(s, length);
454 typename String_set_type::const_iterator p = this->string_set_.find(hk);
455 if (p != this->string_set_.end())
456 return this->key_to_offset_[p->second - 1];
457 gold_unreachable();
460 // Write the ELF strtab into the buffer.
462 template<typename Stringpool_char>
463 void
464 Stringpool_template<Stringpool_char>::write_to_buffer(
465 unsigned char* buffer,
466 section_size_type bufsize)
468 gold_assert(this->strtab_size_ != 0);
469 gold_assert(bufsize >= this->strtab_size_);
470 if (this->zero_null_)
471 buffer[0] = '\0';
472 for (typename String_set_type::const_iterator p = this->string_set_.begin();
473 p != this->string_set_.end();
474 ++p)
476 const int len = (p->first.length + 1) * sizeof(Stringpool_char);
477 const section_offset_type offset = this->key_to_offset_[p->second - 1];
478 gold_assert(static_cast<section_size_type>(offset) + len
479 <= this->strtab_size_);
480 memcpy(buffer + offset, p->first.string, len);
484 // Write the ELF strtab into the output file at the specified offset.
486 template<typename Stringpool_char>
487 void
488 Stringpool_template<Stringpool_char>::write(Output_file* of, off_t offset)
490 gold_assert(this->strtab_size_ != 0);
491 unsigned char* view = of->get_output_view(offset, this->strtab_size_);
492 this->write_to_buffer(view, this->strtab_size_);
493 of->write_output_view(offset, this->strtab_size_, view);
496 // Print statistical information to stderr. This is used for --stats.
498 template<typename Stringpool_char>
499 void
500 Stringpool_template<Stringpool_char>::print_stats(const char* name) const
502 #if defined(HAVE_TR1_UNORDERED_MAP) || defined(HAVE_EXT_HASH_MAP)
503 fprintf(stderr, _("%s: %s entries: %zu; buckets: %zu\n"),
504 program_name, name, this->string_set_.size(),
505 this->string_set_.bucket_count());
506 #else
507 fprintf(stderr, _("%s: %s entries: %zu\n"),
508 program_name, name, this->table_.size());
509 #endif
510 fprintf(stderr, _("%s: %s Stringdata structures: %zu\n"),
511 program_name, name, this->strings_.size());
514 // Instantiate the templates we need.
516 template
517 class Stringpool_template<char>;
519 template
520 class Stringpool_template<uint16_t>;
522 template
523 class Stringpool_template<uint32_t>;
525 } // End namespace gold.