Implement iteration optimization

Instead of checking every element of a
heightmap for eligibility to process,
only the elements neighbouring just
processed elements are checked.
This cuts the time to roughly a third.

diff --git a/povodi.py b/povodi.py
index 86477ef2685fa302d227dd1ca0afd62c506eef4f..08cb2afac0153f91aae08eac9b2cc6a235db57af 100644 (file)
--- a/povodi.py
+++ b/povodi.py
@@ -5,8 +5,7 @@ def     is_noncalced_max(height_map, calced, multi_ind):
        high_x = min(height_map.shape[0] - 1, multi_ind[0] + 1)
        low_y = max(0, multi_ind[1] - 1)
        high_y = min(height_map.shape[1] - 1, multi_ind[1] + 1)
-       masked_subarr = np.ma.masked_array(height_map[low_x:high_x + 1, low_y:high_y + 1],
-                       mask=calced[low_x:high_x + 1, low_y:high_y + 1])
+       masked_subarr = np.ma.masked_array(height_map[low_x:high_x + 1, low_y:high_y + 1], mask=calced[low_x:high_x + 1, low_y:high_y + 1])
        return (masked_subarr.max() == height_map[multi_ind])
 
 def    get_recipient(height_map, multi_ind):
@@ -26,19 +25,38 @@ def get_recipient(height_map, multi_ind):
        min_ind_small = np.unravel_index(sub_arr.argmin(), sub_arr.shape)
        return ((min_ind_small[0] + multi_ind[0] - sub_x, min_ind_small[1] + multi_ind[1] - sub_y))
 
+def    append_eligible_neighbours(to_calc, height_map, calced, multi_ind):
+       high_x = min(height_map.shape[0] - 1, multi_ind[0] + 1)
+       low_x = max(0, multi_ind[0] - 1)
+       high_y = min(height_map.shape[1] - 1, multi_ind[1] + 1)
+       low_y = max(0, multi_ind[1] - 1)
+       for x in range(low_x, high_x + 1):
+               for y in range(low_y, high_y + 1):
+                       if (is_noncalced_max(height_map, calced, (x, y))):
+                               to_calc.append((x, y))
+       return
+
 def povodi(height_map):
        height_map = np.array(height_map)
        calced = height_map.astype('bool')
        calced.fill(0)
        result = height_map.astype("uint32")
        result.fill(1)
-       while (not(calced.all())):
-               print(calced.sum() / calced.size)
-               it = np.nditer(height_map, flags=['multi_index'])
-               for el in it:
-                       if (not(calced[it.multi_index]) and is_noncalced_max(height_map, calced, it.multi_index)):
-                               calced[it.multi_index] = 1
-                               recipient_ind = get_recipient(height_map, it.multi_index)
-                               if (recipient_ind != it.multi_index):
-                                       result[recipient_ind] += result[it.multi_index]
+       to_calc = []
+       it = np.nditer(height_map, flags=['multi_index'])
+       for el in it:
+               if (is_noncalced_max(height_map, calced, it.multi_index)):
+                       to_calc.append(it.multi_index)
+       i = 0
+       while (to_calc != []):
+               multi_ind = to_calc.pop()
+               if (calced[multi_ind]):
+                       continue
+               print(i / height_map.size)
+               calced[multi_ind] = 1
+               recipient_ind = get_recipient(height_map, multi_ind)
+               if (recipient_ind != multi_ind):
+                       result[recipient_ind] += result[multi_ind]
+               append_eligible_neighbours(to_calc, height_map, calced, multi_ind)
+               i += 1
        return (result)