+ handle boundary cases in 3x3 convolution

include/mapnik/image_filter.hpp

@@ -196,7 +196,11 @@ void process_channel (Src const& src, Dst & dst, mapnik::filter::sobel)
 template <typename Src, typename Dst, typename FilterTag>
 void apply_convolution_3x3(Src const& src_view, Dst & dst_view, FilterTag filter_tag)
 {
-    typename Src::xy_locator src_loc = src_view.xy_at(1,1);
+    // p0 p1 p2
+    // p3 p4 p5
+    // p6 p7 p8
+
+    typename Src::xy_locator src_loc = src_view.xy_at(0,0);
     typename Src::xy_locator::cached_location_t loc00 = src_loc.cache_location(-1,-1);
     typename Src::xy_locator::cached_location_t loc10 = src_loc.cache_location( 0,-1);
     typename Src::xy_locator::cached_location_t loc20 = src_loc.cache_location( 1,-1);
@@ -207,34 +211,147 @@ void apply_convolution_3x3(Src const& src_view, Dst & dst_view, FilterTag filter
     typename Src::xy_locator::cached_location_t loc12 = src_loc.cache_location( 0, 1);
     typename Src::xy_locator::cached_location_t loc22 = src_loc.cache_location( 1, 1);
 
+    typename Src::x_iterator dst_it = dst_view.row_begin(0);
+
+    // top row
+    for (int x = 0 ; x < src_view.width(); ++x)
+    {
+        *dst_it = src_loc[loc11];
+        for (int i = 0; i < 3; ++i)
+        {
+            bits32f p[9];
+
+            p[4] = src_loc[loc11][i];
+            p[7] = src_loc[loc12][i];
+
+            if (x == 0)
+            {
+                p[3] = p[4];
+                p[6] = p[7];
+            }
+            else
+            {
+                p[3] = src_loc[loc01][i];
+                p[6] = src_loc[loc02][i];
+            }
+
+            if ( x == src_view.width()-1)
+            {
+                p[5] = p[4];
+                p[8] = p[7];
+            }
+            else
+            {
+                p[5] = src_loc[loc21][i];
+                p[8] = src_loc[loc22][i];
+            }
+
+
+            p[0] = p[6];
+            p[1] = p[7];
+            p[2] = p[8];
+
+            process_channel(p, (*dst_it)[i], filter_tag);
+        }
+        ++src_loc.x();
+        ++dst_it;
+    }
+    // carrige-return
+    src_loc += point2<std::ptrdiff_t>(-src_view.width(),1);
+
+    // 1... height-1 rows
     for (int y = 1; y<src_view.height()-1; ++y)
     {
-        typename Src::x_iterator dst_it = dst_view.row_begin(y);
+        for (int x = 0; x < src_view.width(); ++x)
-        for (int x = 0; x<src_view.width(); ++x) 
         {
             *dst_it = src_loc[loc11];
             for (int i = 0; i < 3; ++i)
             {
                 bits32f p[9];
-                p[0] = src_loc[loc00][i];
+
                 p[1] = src_loc[loc10][i];
-                p[2] = src_loc[loc20][i];
-                p[3] = src_loc[loc01][i];            
                 p[4] = src_loc[loc11][i];
-                p[5] = src_loc[loc21][i];
-                p[6] = src_loc[loc02][i];            
                 p[7] = src_loc[loc12][i];
-                p[8] = src_loc[loc22][i];                
+
+                if (x == 0)
+                {
+                    p[0] = p[1];
+                    p[3] = p[4];
+                    p[6] = p[7];
+                }
+                else
+                {
+                    p[0] = src_loc[loc00][i];
+                    p[3] = src_loc[loc01][i];
+                    p[6] = src_loc[loc02][i];
+                }
+
+                if ( x == src_view.width() - 1)
+                {
+                    p[2] = p[1];
+                    p[5] = p[4];
+                    p[8] = p[7];
+                }
+                else
+                {
+                    p[2] = src_loc[loc20][i];
+                    p[5] = src_loc[loc21][i];
+                    p[8] = src_loc[loc22][i];
+                }
                 process_channel(p, (*dst_it)[i], filter_tag);
             }
-            
             ++dst_it;
             ++src_loc.x();
         }
+        // carrige-return
         src_loc += point2<std::ptrdiff_t>(-src_view.width(),1);
     }
-}
 
+    // bottom row
+    //src_loc = src_view.xy_at(0,src_view.height()-1);
+    for (int x = 0 ; x < src_view.width(); ++x)
+    {
+        *dst_it = src_loc[loc11];
+        for (int i = 0; i < 3; ++i)
+        {
+            bits32f p[9];
+
+            p[1] = src_loc[loc10][i];
+            p[4] = src_loc[loc11][i];
+
+            if (x == 0)
+            {
+                p[0] = p[1];
+                p[3] = p[4];
+            }
+            else
+            {
+                p[0] = src_loc[loc00][i];
+                p[3] = src_loc[loc01][i];
+            }
+
+            if ( x == src_view.width()-1)
+            {
+                p[2] = p[1];
+                p[5] = p[4];
+
+            }
+            else
+            {
+                p[2] = src_loc[loc20][i];
+                p[5] = src_loc[loc21][i];
+            }
+
+            p[6] = p[0];
+            p[7] = p[1];
+            p[8] = p[2];
+
+            process_channel(p, (*dst_it)[i], filter_tag);
+        }
+        ++src_loc.x();
+        ++dst_it;
+    }
+}
 
 template <typename Src, typename Dst,typename FilterTag>
 void apply_filter(Src const& src, Dst & dst, FilterTag filter_tag)
@@ -368,7 +485,7 @@ template <typename Src>
 struct filter_visitor : boost::static_visitor<void>
 {
     filter_visitor(Src & src)
-        : src_(src) {} 
+    : src_(src) {}
 
     template <typename T>
     void operator () (T filter_tag)