summaryrefslogtreecommitdiff
path: root/klm/lm/search_hashed.cc
diff options
context:
space:
mode:
authorKenneth Heafield <github@kheafield.com>2014-01-27 17:42:19 -0800
committerKenneth Heafield <github@kheafield.com>2014-01-27 17:42:19 -0800
commit783c57b2d3312738ddcf992ac55ff750afe7cb47 (patch)
treec4811dab0d916836b8631f3c7df94f284a490b9b /klm/lm/search_hashed.cc
parentf7e051a05d65ef25c2ada0b84cd82bfb375ef265 (diff)
KenLM 5cc905bc2d214efa7de2db56a9a672b749a95591
Diffstat (limited to 'klm/lm/search_hashed.cc')
-rw-r--r--klm/lm/search_hashed.cc33
1 files changed, 19 insertions, 14 deletions
diff --git a/klm/lm/search_hashed.cc b/klm/lm/search_hashed.cc
index 62275d27..354a56b4 100644
--- a/klm/lm/search_hashed.cc
+++ b/klm/lm/search_hashed.cc
@@ -204,9 +204,10 @@ template <class Build, class Activate, class Store> void ReadNGrams(
namespace detail {
template <class Value> uint8_t *HashedSearch<Value>::SetupMemory(uint8_t *start, const std::vector<uint64_t> &counts, const Config &config) {
- std::size_t allocated = Unigram::Size(counts[0]);
- unigram_ = Unigram(start, counts[0], allocated);
- start += allocated;
+ unigram_ = Unigram(start, counts[0]);
+ start += Unigram::Size(counts[0]);
+ std::size_t allocated;
+ middle_.clear();
for (unsigned int n = 2; n < counts.size(); ++n) {
allocated = Middle::Size(counts[n - 1], config.probing_multiplier);
middle_.push_back(Middle(start, allocated));
@@ -218,9 +219,21 @@ template <class Value> uint8_t *HashedSearch<Value>::SetupMemory(uint8_t *start,
return start;
}
-template <class Value> void HashedSearch<Value>::InitializeFromARPA(const char * /*file*/, util::FilePiece &f, const std::vector<uint64_t> &counts, const Config &config, ProbingVocabulary &vocab, Backing &backing) {
- // TODO: fix sorted.
- SetupMemory(GrowForSearch(config, vocab.UnkCountChangePadding(), Size(counts, config), backing), counts, config);
+/*template <class Value> void HashedSearch<Value>::Relocate(uint8_t *start, const std::vector<uint64_t> &counts, const Config &config) {
+ unigram_ = Unigram(start, counts[0]);
+ start += Unigram::Size(counts[0]);
+ for (unsigned int n = 2; n < counts.size(); ++n) {
+ middle[n-2].Relocate(start);
+ start += Middle::Size(counts[n - 1], config.probing_multiplier)
+ }
+ longest_.Relocate(start);
+}*/
+
+template <class Value> void HashedSearch<Value>::InitializeFromARPA(const char * /*file*/, util::FilePiece &f, const std::vector<uint64_t> &counts, const Config &config, ProbingVocabulary &vocab, BinaryFormat &backing) {
+ void *vocab_rebase;
+ void *search_base = backing.GrowForSearch(Size(counts, config), vocab.UnkCountChangePadding(), vocab_rebase);
+ vocab.Relocate(vocab_rebase);
+ SetupMemory(reinterpret_cast<uint8_t*>(search_base), counts, config);
PositiveProbWarn warn(config.positive_log_probability);
Read1Grams(f, counts[0], vocab, unigram_.Raw(), warn);
@@ -277,14 +290,6 @@ template <class Value> template <class Build> void HashedSearch<Value>::ApplyBui
ReadEnd(f);
}
-template <class Value> void HashedSearch<Value>::LoadedBinary() {
- unigram_.LoadedBinary();
- for (typename std::vector<Middle>::iterator i = middle_.begin(); i != middle_.end(); ++i) {
- i->LoadedBinary();
- }
- longest_.LoadedBinary();
-}
-
template class HashedSearch<BackoffValue>;
template class HashedSearch<RestValue>;