// stringpool.cc -- a string pool for gold
-// Copyright 2006, 2007 Free Software Foundation, Inc.
+// Copyright (C) 2006-2016 Free Software Foundation, Inc.
// Written by Ian Lance Taylor <iant@google.com>.
// This file is part of gold.
{
template<typename Stringpool_char>
-Stringpool_template<Stringpool_char>::Stringpool_template()
- : string_set_(), strings_(), strtab_size_(0), next_index_(1),
- next_uncopied_key_(-1), zero_null_(true)
+Stringpool_template<Stringpool_char>::Stringpool_template(uint64_t addralign)
+ : string_set_(), key_to_offset_(), strings_(), strtab_size_(0),
+ zero_null_(true), optimize_(false), offset_(sizeof(Stringpool_char)),
+ addralign_(addralign)
{
+ if (parameters->options_valid()
+ && parameters->options().optimize() >= 2
+ && addralign <= sizeof(Stringpool_char))
+ this->optimize_ = true;
}
template<typename Stringpool_char>
++p)
delete[] reinterpret_cast<char*>(*p);
this->strings_.clear();
+ this->key_to_offset_.clear();
this->string_set_.clear();
}
this->clear();
}
-// Return the length of a string of arbitrary character type.
+// Resize the internal hashtable with the expectation we'll get n new
+// elements. Note that the hashtable constructor takes a "number of
+// buckets you'd like," rather than "number of elements you'd like,"
+// but that's the best we can do.
template<typename Stringpool_char>
-size_t
-Stringpool_template<Stringpool_char>::string_length(const Stringpool_char* p)
+void
+Stringpool_template<Stringpool_char>::reserve(unsigned int n)
{
- size_t len = 0;
- for (; *p != 0; ++p)
- ++len;
- return len;
-}
-
-// Specialize string_length for char. Maybe we could just use
-// std::char_traits<>::length?
+ this->key_to_offset_.reserve(n);
+
+#if defined(HAVE_UNORDERED_MAP)
+ this->string_set_.rehash(this->string_set_.size() + n);
+ return;
+#elif defined(HAVE_TR1_UNORDERED_MAP)
+ // rehash() implementation is broken in gcc 4.0.3's stl
+ //this->string_set_.rehash(this->string_set_.size() + n);
+ //return;
+#elif defined(HAVE_EXT_HASH_MAP)
+ this->string_set_.resize(this->string_set_.size() + n);
+ return;
+#endif
-template<>
-inline size_t
-Stringpool_template<char>::string_length(const char* p)
-{
- return strlen(p);
+ // This is the generic "reserve" code, if no #ifdef above triggers.
+ String_set_type new_string_set(this->string_set_.size() + n);
+ new_string_set.insert(this->string_set_.begin(), this->string_set_.end());
+ this->string_set_.swap(new_string_set);
}
-// Equality comparison function.
+// Compare two strings of arbitrary character type for equality.
template<typename Stringpool_char>
bool
-Stringpool_template<Stringpool_char>::Stringpool_eq::operator()(
- const Stringpool_char* s1,
- const Stringpool_char* s2) const
+Stringpool_template<Stringpool_char>::string_equal(const Stringpool_char* s1,
+ const Stringpool_char* s2)
{
while (*s1 != 0)
if (*s1++ != *s2++)
return *s2 == 0;
}
-// Specialize equality comparison for char.
+// Specialize string_equal for char.
template<>
-bool
-Stringpool_template<char>::Stringpool_eq::operator()(const char* s1,
- const char* s2) const
+inline bool
+Stringpool_template<char>::string_equal(const char* s1, const char* s2)
{
return strcmp(s1, s2) == 0;
}
-// Hash function.
+// Equality comparison function for the hash table.
+
+template<typename Stringpool_char>
+inline bool
+Stringpool_template<Stringpool_char>::Stringpool_eq::operator()(
+ const Hashkey& h1,
+ const Hashkey& h2) const
+{
+ return (h1.hash_code == h2.hash_code
+ && h1.length == h2.length
+ && (h1.string == h2.string
+ || memcmp(h1.string, h2.string,
+ h1.length * sizeof(Stringpool_char)) == 0));
+}
+
+// Hash function. The length is in characters, not bytes.
template<typename Stringpool_char>
size_t
-Stringpool_template<Stringpool_char>::Stringpool_hash::operator()(
- const Stringpool_char* s) const
+Stringpool_template<Stringpool_char>::string_hash(const Stringpool_char* s,
+ size_t length)
{
- // Fowler/Noll/Vo (FNV) hash (type FNV-1a).
- if (sizeof(size_t) > 4)
- {
- size_t result = static_cast<size_t>(14695981039346656037ULL);
- while (*s != 0)
- {
- const char* p = reinterpret_cast<const char*>(s);
- for (size_t i = 0; i < sizeof(Stringpool_char); ++i)
- {
- result ^= (size_t) *p++;
- result *= 1099511628211ULL;
- }
- ++s;
- }
- return result;
- }
- else
- {
- size_t result = 2166136261UL;
- while (*s != 0)
- {
- const char* p = reinterpret_cast<const char*>(s);
- for (size_t i = 0; i < sizeof(Stringpool_char); ++i)
- {
- result ^= (size_t) *p++;
- result *= 16777619UL;
- }
- ++s;
- }
- return result;
- }
+ return gold::string_hash<Stringpool_char>(s, length);
}
-// Add a string to the list of canonical strings. Return a pointer to
-// the canonical string. If PKEY is not NULL, set *PKEY to the key.
+// Add the string S to the list of canonical strings. Return a
+// pointer to the canonical string. If PKEY is not NULL, set *PKEY to
+// the key. LENGTH is the length of S in characters. Note that S may
+// not be NUL terminated.
template<typename Stringpool_char>
const Stringpool_char*
Stringpool_template<Stringpool_char>::add_string(const Stringpool_char* s,
- Key* pkey)
+ size_t len)
{
// We are in trouble if we've already computed the string offsets.
gold_assert(this->strtab_size_ == 0);
const size_t key_mult = 1024;
gold_assert(key_mult >= buffer_size);
- size_t len = (string_length(s) + 1) * sizeof(Stringpool_char);
+ // Convert len to the number of bytes we need to allocate, including
+ // the null character.
+ len = (len + 1) * sizeof(Stringpool_char);
size_t alc;
bool front = true;
alc = sizeof(Stringdata) + buffer_size;
else
{
- Stringdata *psd = this->strings_.front();
+ Stringdata* psd = this->strings_.front();
if (len > psd->alc - psd->len)
alc = sizeof(Stringdata) + buffer_size;
else
{
char* ret = psd->data + psd->len;
- memcpy(ret, s, len);
-
- if (pkey != NULL)
- *pkey = psd->index * key_mult + psd->len;
+ memcpy(ret, s, len - sizeof(Stringpool_char));
+ memset(ret + len - sizeof(Stringpool_char), 0,
+ sizeof(Stringpool_char));
psd->len += len;
}
}
- Stringdata *psd = reinterpret_cast<Stringdata*>(new char[alc]);
+ Stringdata* psd = reinterpret_cast<Stringdata*>(new char[alc]);
psd->alc = alc - sizeof(Stringdata);
- memcpy(psd->data, s, len);
+ memcpy(psd->data, s, len - sizeof(Stringpool_char));
+ memset(psd->data + len - sizeof(Stringpool_char), 0,
+ sizeof(Stringpool_char));
psd->len = len;
- psd->index = this->next_index_;
- ++this->next_index_;
-
- if (pkey != NULL)
- *pkey = psd->index * key_mult;
if (front)
this->strings_.push_front(psd);
Stringpool_template<Stringpool_char>::add(const Stringpool_char* s, bool copy,
Key* pkey)
{
- // FIXME: This will look up the entry twice in the hash table. The
- // problem is that we can't insert S before we canonicalize it. I
- // don't think there is a way to handle this correctly with
- // unordered_map, so this should be replaced with custom code to do
- // what we need, which is to return the empty slot.
+ return this->add_with_length(s, string_length(s), copy, pkey);
+}
- typename String_set_type::const_iterator p = this->string_set_.find(s);
- if (p != this->string_set_.end())
+// Add a new key offset entry.
+
+template<typename Stringpool_char>
+void
+Stringpool_template<Stringpool_char>::new_key_offset(size_t length)
+{
+ section_offset_type offset;
+ if (this->zero_null_ && length == 0)
+ offset = 0;
+ else
+ {
+ offset = this->offset_;
+ // Align strings.
+ offset = align_address(offset, this->addralign_);
+ this->offset_ = offset + (length + 1) * sizeof(Stringpool_char);
+ }
+ this->key_to_offset_.push_back(offset);
+}
+
+template<typename Stringpool_char>
+const Stringpool_char*
+Stringpool_template<Stringpool_char>::add_with_length(const Stringpool_char* s,
+ size_t length,
+ bool copy,
+ Key* pkey)
+{
+ typedef std::pair<typename String_set_type::iterator, bool> Insert_type;
+
+ // We add 1 so that 0 is always invalid.
+ const Key k = this->key_to_offset_.size() + 1;
+
+ if (!copy)
{
+ // When we don't need to copy the string, we can call insert
+ // directly.
+
+ std::pair<Hashkey, Hashval> element(Hashkey(s, length), k);
+
+ Insert_type ins = this->string_set_.insert(element);
+
+ typename String_set_type::const_iterator p = ins.first;
+
+ if (ins.second)
+ {
+ // We just added the string. The key value has now been
+ // used.
+ this->new_key_offset(length);
+ }
+ else
+ {
+ gold_assert(k != p->second);
+ }
+
if (pkey != NULL)
- *pkey = p->second.first;
- return p->first;
+ *pkey = p->second;
+ return p->first.string;
}
- Key k;
- const Stringpool_char* ret;
- if (copy)
- ret = this->add_string(s, &k);
- else
+ // When we have to copy the string, we look it up twice in the hash
+ // table. The problem is that we can't insert S before we
+ // canonicalize it by copying it into the canonical list. The hash
+ // code will only be computed once.
+
+ Hashkey hk(s, length);
+ typename String_set_type::const_iterator p = this->string_set_.find(hk);
+ if (p != this->string_set_.end())
{
- ret = s;
- k = this->next_uncopied_key_;
- --this->next_uncopied_key_;
+ if (pkey != NULL)
+ *pkey = p->second;
+ return p->first.string;
}
- const off_t ozero = 0;
- std::pair<const Stringpool_char*, Val> element(ret,
- std::make_pair(k, ozero));
- std::pair<typename String_set_type::iterator, bool> ins =
- this->string_set_.insert(element);
- gold_assert(ins.second);
+ this->new_key_offset(length);
- if (pkey != NULL)
- *pkey = k;
+ hk.string = this->add_string(s, length);
+ // The contents of the string stay the same, so we don't need to
+ // adjust hk.hash_code or hk.length.
- return ret;
-}
+ std::pair<Hashkey, Hashval> element(hk, k);
-// Add a prefix of a string to a string pool.
+ Insert_type ins = this->string_set_.insert(element);
+ gold_assert(ins.second);
-template<typename Stringpool_char>
-const Stringpool_char*
-Stringpool_template<Stringpool_char>::add_prefix(const Stringpool_char* s,
- size_t len,
- Key* pkey)
-{
- // FIXME: This implementation should be rewritten when we rewrite
- // the hash table to avoid copying.
- std::basic_string<Stringpool_char> st(s, len);
- return this->add(st.c_str(), true, pkey);
+ if (pkey != NULL)
+ *pkey = k;
+ return hk.string;
}
template<typename Stringpool_char>
Stringpool_template<Stringpool_char>::find(const Stringpool_char* s,
Key* pkey) const
{
- typename String_set_type::const_iterator p = this->string_set_.find(s);
+ Hashkey hk(s);
+ typename String_set_type::const_iterator p = this->string_set_.find(hk);
if (p == this->string_set_.end())
return NULL;
if (pkey != NULL)
- *pkey = p->second.first;
+ *pkey = p->second;
- return p->first;
+ return p->first.string;
}
// Comparison routine used when sorting into an ELF strtab. We want
const Stringpool_sort_info& sort_info1,
const Stringpool_sort_info& sort_info2) const
{
- const Stringpool_char* s1 = sort_info1.it->first;
- const Stringpool_char* s2 = sort_info2.it->first;
- const size_t len1 = sort_info1.string_length;
- const size_t len2 = sort_info2.string_length;
+ const Hashkey& h1(sort_info1->first);
+ const Hashkey& h2(sort_info2->first);
+ const Stringpool_char* s1 = h1.string;
+ const Stringpool_char* s2 = h2.string;
+ const size_t len1 = h1.length;
+ const size_t len2 = h2.length;
const size_t minlen = len1 < len2 ? len1 : len2;
const Stringpool_char* p1 = s1 + len1 - 1;
const Stringpool_char* p2 = s2 + len2 - 1;
const size_t charsize = sizeof(Stringpool_char);
// Offset 0 may be reserved for the empty string.
- off_t offset = this->zero_null_ ? charsize : 0;
+ section_offset_type offset = this->zero_null_ ? charsize : 0;
// Sorting to find suffixes can take over 25% of the total CPU time
// used by the linker. Since it's merely an optimization to reduce
// the strtab size, and gives a relatively small benefit (it's
// typically rare for a symbol to be a suffix of another), we only
// take the time to sort when the user asks for heavy optimization.
- if (parameters->optimization_level() < 2)
+ if (!this->optimize_)
{
- for (typename String_set_type::iterator curr = this->string_set_.begin();
- curr != this->string_set_.end();
- curr++)
- {
- if (this->zero_null_ && curr->first[0] == 0)
- curr->second.second = 0;
- else
- {
- curr->second.second = offset;
- offset += (string_length(curr->first) + 1) * charsize;
- }
- }
+ // If we are not optimizing, the offsets are already assigned.
+ offset = this->offset_;
}
else
{
for (typename String_set_type::iterator p = this->string_set_.begin();
p != this->string_set_.end();
++p)
- v.push_back(Stringpool_sort_info(p, string_length(p->first)));
+ v.push_back(Stringpool_sort_info(p));
std::sort(v.begin(), v.end(), Stringpool_sort_comparison());
+ section_offset_type last_offset = -1;
for (typename std::vector<Stringpool_sort_info>::iterator last = v.end(),
curr = v.begin();
curr != v.end();
last = curr++)
{
- if (this->zero_null_ && curr->it->first[0] == 0)
- curr->it->second.second = 0;
+ section_offset_type this_offset;
+ if (this->zero_null_ && (*curr)->first.string[0] == 0)
+ this_offset = 0;
else if (last != v.end()
- && is_suffix(curr->it->first, curr->string_length,
- last->it->first, last->string_length))
- curr->it->second.second = (last->it->second.second
- + ((last->string_length
- - curr->string_length)
- * charsize));
+ && ((((*curr)->first.length - (*last)->first.length)
+ % this->addralign_) == 0)
+ && is_suffix((*curr)->first.string,
+ (*curr)->first.length,
+ (*last)->first.string,
+ (*last)->first.length))
+ this_offset = (last_offset
+ + (((*last)->first.length - (*curr)->first.length)
+ * charsize));
else
{
- curr->it->second.second = offset;
- offset += (curr->string_length + 1) * charsize;
+ this_offset = align_address(offset, this->addralign_);
+ offset = this_offset + ((*curr)->first.length + 1) * charsize;
}
+ this->key_to_offset_[(*curr)->second - 1] = this_offset;
+ last_offset = this_offset;
}
}
// exist.
template<typename Stringpool_char>
-off_t
+section_offset_type
Stringpool_template<Stringpool_char>::get_offset(const Stringpool_char* s)
const
+{
+ return this->get_offset_with_length(s, string_length(s));
+}
+
+template<typename Stringpool_char>
+section_offset_type
+Stringpool_template<Stringpool_char>::get_offset_with_length(
+ const Stringpool_char* s,
+ size_t length) const
{
gold_assert(this->strtab_size_ != 0);
- typename String_set_type::const_iterator p = this->string_set_.find(s);
+ Hashkey hk(s, length);
+ typename String_set_type::const_iterator p = this->string_set_.find(hk);
if (p != this->string_set_.end())
- return p->second.second;
+ return this->key_to_offset_[p->second - 1];
gold_unreachable();
}
template<typename Stringpool_char>
void
-Stringpool_template<Stringpool_char>::write_to_buffer(unsigned char* buffer,
- size_t bufsize)
+Stringpool_template<Stringpool_char>::write_to_buffer(
+ unsigned char* buffer,
+ section_size_type bufsize)
{
gold_assert(this->strtab_size_ != 0);
- // Quiet the compiler in opt mode.
- if (bufsize < static_cast<size_t>(this->strtab_size_))
- gold_assert(bufsize >= static_cast<size_t>(this->strtab_size_));
+ gold_assert(bufsize >= this->strtab_size_);
if (this->zero_null_)
buffer[0] = '\0';
for (typename String_set_type::const_iterator p = this->string_set_.begin();
p != this->string_set_.end();
++p)
{
- const int len = (string_length(p->first) + 1) * sizeof(Stringpool_char);
- gold_assert(p->second.second + len <= this->strtab_size_);
- memcpy(buffer + p->second.second, p->first, len);
+ const int len = (p->first.length + 1) * sizeof(Stringpool_char);
+ const section_offset_type offset = this->key_to_offset_[p->second - 1];
+ gold_assert(static_cast<section_size_type>(offset) + len
+ <= this->strtab_size_);
+ memcpy(buffer + offset, p->first.string, len);
}
}
void
Stringpool_template<Stringpool_char>::print_stats(const char* name) const
{
-#if defined(HAVE_TR1_UNORDERED_MAP) || defined(HAVE_EXT_HASH_MAP)
+#if defined(HAVE_UNORDERED_MAP) || defined(HAVE_TR1_UNORDERED_MAP) || defined(HAVE_EXT_HASH_MAP)
fprintf(stderr, _("%s: %s entries: %zu; buckets: %zu\n"),
program_name, name, this->string_set_.size(),
this->string_set_.bucket_count());