diff --git a/cs677/pa2/threaded.cc b/cs677/pa2/threaded.cc
index edca7ca..0506721 100644
--- a/cs677/pa2/threaded.cc
+++ b/cs677/pa2/threaded.cc
@@ -91,7 +91,7 @@ int main(int argc, char * argv[])
     s = &files[0];
     t = &files[1];
     pthread_t * threads = new pthread_t[num_threads];
-    matrix = new int[files[0].size() * files[1].size()];
+    matrix = new int[(files[0].size() + 1) * (files[1].size() + 1)];
     pthread_barrier_init(&barrier, NULL, num_threads);
 
     for (int i = 0; i < num_threads; i++)
@@ -124,10 +124,12 @@ bool readFile(char * fileName, vector<char> & v)
     ifstream in(fileName);
     if (!in.is_open())
         return false;
-    while (!in.eof())
+    for(;;)
     {
         char chr;
         in >> chr;
+        if (in.eof())
+            break;
         v.push_back(chr);
     }
     return true;
@@ -143,16 +145,16 @@ void * calcSimMatrixThread(void * arg)
     int (*F)[s_size][t_size] = (int (*) [s_size][t_size]) matrix;
     int max_x = 0, max_y = 0, max_val = 0;
     int first_task_id, num_tasks;
-    taskAllocate(t_size, num_threads, id, &first_task_id, &num_tasks);
+    taskAllocate(t_size+1, num_threads, id, &first_task_id, &num_tasks);
     for (int i = 0, idx = first_task_id; i < num_tasks; i++, idx++)
         (*F)[0][idx] = 0;               /* set first row to 0's */
-    taskAllocate(s_size, num_threads, id, &first_task_id, &num_tasks);
+    taskAllocate(s_size+1, num_threads, id, &first_task_id, &num_tasks);
     for (int i = 0, idx = first_task_id; i < num_tasks; i++, idx++)
         (*F)[idx][0] = 0;               /* set first column to 0's */
     pthread_barrier_wait(&barrier);
-    for (int i = 1; i < s_size; i++)
+    for (int i = 1; i <= s_size; i++)
     {
-        for (int j = 1; j < t_size; j++)
+        for (int j = 1; j <= t_size; j++)
         {
             /* Compute the value for the matrix */
             (*F)[i][j] =
@@ -160,7 +162,7 @@ void * calcSimMatrixThread(void * arg)
                     max(
                         (*F)[i][j-1] - 2,
                         (*F)[i-1][j-1] +
-                            (eq(s->at(i), t->at(j)) ? 1 : -1)
+                            (eq(s->at(i-1), t->at(j-1)) ? 1 : -1)
                     ),
                     max(
                         (*F)[i-1][j] - 2,