Fast

.gitignore

@@ -0,0 +1 @@
+bin

Makefile

@@ -1,9 +1,8 @@
 bilbowa: src/bilbowa.v0.1.c 
 	mkdir -p bin
-	gcc src/bilbowa.v0.1.c -g -o bin/bilbowa -lm -pthread -march=native -funroll-loops -w 
+	gcc -std=c99 -Wall -O2 -march=native -funroll-loops src/bilbowa.v0.1.c -o bin/bilbowa -lm -pthread
 bidist: src/bidist.c
-	gcc src/bidist.c -g -o bin/bidist -lm -march=native -funroll-loops -w
+	gcc -std=c99 -Wall -O2 -march=native -funroll-loops src/bidist.c -o bin/bidist -lm
 all: bilbowa bidist
 clean:
-	rm bin/bilbowa bin/bidist
-
+	rm -fv bin/*

src/bidist.c

@@ -54,7 +54,7 @@ int main(int argc, char **argv) {
     for (a = 0; a < size_lang1; a++) fread(&M_lang1[a + b * size_lang1], sizeof(float), 1, f);
     len = 0;
     for (a = 0; a < size_lang1; a++) len += M_lang1[a + b * size_lang1] * M_lang1[a + b * size_lang1];
-    len = sqrt(len);
+    len = sqrtf(len);
     for (a = 0; a < size_lang1; a++) M_lang1[a + b * size_lang1] /= len;
   }
   fclose(f);
@@ -79,7 +79,7 @@ int main(int argc, char **argv) {
     for (a = 0; a < size_lang2; a++) fread(&M_lang2[a + b * size_lang2], sizeof(float), 1, f);
     len = 0;
     for (a = 0; a < size_lang2; a++) len += M_lang2[a + b * size_lang2] * M_lang2[a + b * size_lang2];
-    len = sqrt(len);
+    len = sqrtf(len);
     for (a = 0; a < size_lang2; a++) M_lang2[a + b * size_lang2] /= len;
   }
   fclose(f);
@@ -134,7 +134,7 @@ int main(int argc, char **argv) {
     }
     len = 0;
     for (a = 0; a < size_lang2; a++) len += vec[a] * vec[a];
-    len = sqrt(len);
+    len = sqrtf(len);
     for (a = 0; a < size_lang2; a++) vec[a] /= len;
     for (a = 0; a < N; a++) bestd[a] = 0;
     for (a = 0; a < N; a++) bestw[a][0] = 0;

src/bilbowa.v0.1.c

@@ -12,12 +12,15 @@
 //  See the License for the specific language governing permissions and
 //  limitations under the License.
 
+#define _POSIX_C_SOURCE 200809L // for posix_memalign()
+
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <math.h>
 #include <pthread.h>
 #include <unistd.h>
+#include <stdint.h>
 
 #define MAX_STRING 100
 #define EXP_TABLE_SIZE 1000
@@ -110,7 +113,8 @@ void ReadWord(char *word, FILE *fin) {
 /* Returns hash value of a word */
 int GetWordHash(char *word) {
   unsigned long long a, hash = 0;
-  for (a = 0; a < strlen(word); a++) hash = hash * 257 + word[a];
+  unsigned int len = strlen(word);
+  for (a = 0; a < len; a++) hash = hash * 257 + word[a];
   hash = hash % vocab_hash_size;
   return hash;
 }
@@ -352,7 +356,7 @@ void InitNet(int lang_id) {
 
 char SubSample(int lang_id, long long word_id) {
   long long count = vocabs[lang_id][word_id].cn;
-  real thresh = (sqrt(count / (sample * train_words[lang_id])) + 1) * 
+  real thresh = (sqrtf(count / (sample * train_words[lang_id])) + 1) *
     (sample * train_words[lang_id]) / count;
   next_random = next_random * (unsigned long long)25214903917 + 11;
   if ((next_random & 0xFFFF) / (real)65536 > thresh) return 1;
@@ -391,7 +395,7 @@ void UpdateEmbeddings(real *embeddings, real *grads, int offset,
     if (adagrad) {
       // Use Adagrad for automatic learning rate selection
       grads[offset + a] += (deltas[a] * deltas[a]);
-      step = (alpha / fmax(epsilon, sqrt(grads[offset + a]))) * deltas[a];
+      step = (alpha / fmaxf(epsilon, sqrtf(grads[offset + a]))) * deltas[a];
     } else {
       // Regular SGD
       step = alpha * deltas[a];
@@ -453,7 +457,7 @@ void BuildCDF(real *cdf, long long *sen, int lang_id, int len) {
     }
     else if (PAR_SAMPLE && sample > 0) {  // subsample
       count = vocabs[lang_id][word].cn;
-      threshold = (sqrt(count / (sample * train_words[lang_id])) + 1) * (sample *
+      threshold = (sqrtf(count / (sample * train_words[lang_id])) + 1) * (sample *
         train_words[lang_id]) / count; 
       if (threshold < 0) threshold = 0.0;
     } 
@@ -502,7 +506,7 @@ void *BilbowaThread(void *id) {
   // TODO: Change this for more than two languages
   int lang_id1 = 0, lang_id2 = 1; 
   // Each thread will be responsible for reading a portion of both lang_id1 and lang_id2 files. portion size is: file_size/num_threads
-  int thread_id = (int)id % num_threads; // total_sampled;
+  int thread_id = ((int)(uintptr_t) id) % num_threads; // total_sampled;
   long long par_sen1[MAX_SEN_LEN], par_sen2[MAX_SEN_LEN],
     //        sampled_sen1[10], sampled_sen2[10],
             updates_l1 = 1, updates_l2 = 1,
@@ -603,7 +607,8 @@ void *MonoModelThread(void *id) {
   long long word_count = 0, last_word_count = 0, all_train_words = 0;
   long long mono_sen[MAX_SEN_LEN + 1];
   long long l1, l2, c, target, label;
-  int lang_id = (int)id / num_threads, thread_id = (int)id % num_threads;
+  int id_int = (int)(uintptr_t) id;
+  int lang_id = id_int / num_threads, thread_id = id_int % num_threads;
   char *train_file = mono_train_files[lang_id];
   long long vocab_size = vocab_sizes[lang_id];
   real f, g;
@@ -674,9 +679,9 @@ void *MonoModelThread(void *id) {
     }
     if (EARLY_STOP) {
       if (word_count_actual > EARLY_STOP) {
-        fprintf(stderr, "EARLY STOP point reached (thread %d)\n", (int)id);
+        fprintf(stderr, "EARLY STOP point reached (thread %d)\n", (int)(uintptr_t) id);
         break;
-    }
+      }
     }
     word = mono_sen[sentence_position];
     if (word == -1) continue;
@@ -800,7 +805,7 @@ void TrainModel() {
   expTable = malloc((EXP_TABLE_SIZE + 1) * sizeof(real));
   for (i = 0; i < EXP_TABLE_SIZE; i++) {
     // Precompute the exp() table
-    expTable[i] = exp((i / (real)EXP_TABLE_SIZE * 2 - 1) * MAX_EXP);
+    expTable[i] = expf((i / (real)EXP_TABLE_SIZE * 2 - 1) * MAX_EXP);
     // Precompute sigmoid f(x) = x / (x + 1)
     expTable[i] = expTable[i] / (expTable[i] + 1);
   }