Skip to content

Dictionary: Improve Dictionary Search Efficiency #491

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
dhdaines merged 2 commits into from
Aug 26, 2026
Merged

Dictionary: Improve Dictionary Search Efficiency #491

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
36098a5
Perf: Replace linear root channel search with lookup table
nithin-aashik-mcw Aug 6, 2026
3166290
Fix: root lookup table memory leak
nithin-aashik-mcw Aug 17, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions src/ngram_search.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -237,6 +237,7 @@ struct ngram_search_s {
int32 n_nonroot_chan; /**< Number of valid non-root channels */
int32 max_nonroot_chan; /**< Maximum possible number of non-root channels */
root_chan_t *rhmm_1ph; /**< Root HMMs for single-phone words */
root_chan_t ***root_lookup; /**< Root channel lookup table */

/**
* Channels associated with a given word (only used for right
Expand Down
26 changes: 17 additions & 9 deletions src/ngram_search_fwdtree.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,6 +79,10 @@ init_search_tree(ngram_search_t *ngs)
ndiph = 0;
ngs->n_1ph_words = 0;
n_ci = bin_mdef_n_ciphone(ps_search_acmod(ngs)->mdef);
/* Allocate and zero-initialize the root channel lookup table indexed by
* (ciphone, ci2phone) for constant-time root channel lookup.
*/
ngs->root_lookup = (root_chan_t ***)ckd_calloc_2d(n_ci, n_ci, sizeof(root_chan_t *));
/* Allocate a bitvector with flags for each possible diphone. */
dimap = bitvec_alloc(n_ci * n_ci);
for (w = 0; w < n_words; w++) {
Expand Down Expand Up @@ -175,7 +179,7 @@ create_search_channels(ngram_search_t *ngs)
{
chan_t *hmm;
root_chan_t *rhmm;
int32 w, i, j, p, ph, tmatid;
int32 w, j, p, ph, tmatid;
int32 n_words;
dict_t *dict = ps_search_dict(ngs);
dict2pid_t *d2p = ps_search_dict2pid(ngs);
Expand Down Expand Up @@ -210,23 +214,25 @@ create_search_channels(ngram_search_t *ngs)
* allocate one if not found. */
ciphone = dict_first_phone(dict, w);
ci2phone = dict_second_phone(dict, w);
for (i = 0; i < ngs->n_root_chan; ++i) {
if (ngs->root_chan[i].ciphone == ciphone
&& ngs->root_chan[i].ci2phone == ci2phone)
break;
}
if (i == ngs->n_root_chan) {

/* Lookup an existing root channel for the first two CI phones.
* This replaces the previous O(n_root_chan) linear search with
* an O(1) table lookup.
*/
rhmm = ngs->root_lookup[ciphone][ci2phone];

if (rhmm == NULL) {
rhmm = &(ngs->root_chan[ngs->n_root_chan]);
rhmm->hmm.tmatid = bin_mdef_pid2tmatid(ps_search_acmod(ngs)->mdef, ciphone);
/* Begin with CI phone? Not sure this makes a difference... */
hmm_mpx_ssid(&rhmm->hmm, 0) =
bin_mdef_pid2ssid(ps_search_acmod(ngs)->mdef, ciphone);
rhmm->ciphone = ciphone;
rhmm->ci2phone = ci2phone;
/* Cache the newly created root channel for constant-time lookups. */
ngs->root_lookup[ciphone][ci2phone] = rhmm;
ngs->n_root_chan++;
}
else
rhmm = &(ngs->root_chan[i]);

E_DEBUG("word %s rhmm %d\n", dict_wordstr(dict, w), rhmm - ngs->root_chan);
/* Now, rhmm = root channel for w. Go on to remaining phones */
Expand Down Expand Up @@ -415,6 +421,8 @@ deinit_search_tree(ngram_search_t *ngs)
ngs->single_phone_wid = NULL;
ckd_free(ngs->homophone_set);
ngs->homophone_set = NULL;
ckd_free_2d(ngs->root_lookup);
ngs->root_lookup = NULL;
}

void
Expand Down