diff --git a/rotation.cpp b/rotation.cpp
index 305409d..a6f69b4 100644
--- a/rotation.cpp
+++ b/rotation.cpp
@@ -250,6 +250,59 @@ struct TiledImage : public Image {
       }
   }
 
+  uint8_t*
+  access_pixel(unsigned int x, unsigned int y)
+  {
+    if (x >= width || y >= height)
+      return nullptr;
+
+    unsigned int const tile_width = tile_w * 3;
+
+    unsigned int const tile_index = (y / tile_h) * nb_col_tile + (x / tile_w);
+    uint8_t* tile = tiles[tile_index];
+    unsigned int const tile_j = y % tile_h;
+    unsigned int const tile_i = x % tile_w;
+    return tile + tile_j * tile_width + (tile_i * 3);
+  }
+
+  uint8_t const*
+  access_pixel(unsigned int x, unsigned int y) const
+  {
+    if (x >= width || y >= height)
+      return nullptr;
+
+    unsigned int const tile_width = tile_w * 3;
+
+    unsigned int const tile_index = (y / tile_h) * nb_col_tile + (x / tile_w);
+    //cout << "tile index: " << tile_index << endl;
+    uint8_t* tile = tiles[tile_index];
+    unsigned int const tile_j = y % tile_h;
+    unsigned int const tile_i = x % tile_w;
+    return tile + tile_j * tile_width + (tile_i * 3);
+  }
+
+  void
+  print_tile(unsigned int index) const
+  {
+    cout << "Tile[" << index << "]" << endl;
+    uint8_t const* tile = tiles[index];
+    unsigned int const tile_width = tile_w * 3;
+    for (unsigned int j = 0; j < tile_h; ++j)
+    {
+      for (unsigned int i = 0; i < tile_w; ++i)
+      {
+        if (i != 0)
+          cout << ", ";
+        uint8_t const* p = tile + j * tile_width + i * 3;
+        cout << (int) *p << " " << (int) *(p + 1) << " " << (int) *(p + 2);
+
+      }
+      cout << endl;
+    }
+    cout << endl;
+  }
+
+
   protected:
     void allocate_memory(unsigned int w, unsigned int h)
     {
@@ -277,19 +330,11 @@ struct TiledImage : public Image {
     {
       this->allocate_memory(width, height);
 
-      unsigned int const tile_width = tile_w * 3;
-
       // Pixel loading
       for (unsigned int j = 0; j < height; ++j)
         for (unsigned int i = 0; i < width; ++i)
         {
-          unsigned int const tile_index = (j / tile_h) * nb_col_tile + (i / tile_w);
-          uint8_t* tile = tiles[tile_index];
-          unsigned int const tile_j = j % tile_h;
-          unsigned int const tile_i = i % tile_w;
-          // if (i % 3 == 0)
-          //   cout << "pixel (" << i / 3 << ", " << j << ") goes in tile (" << i / (tile_w * 3) << ", " << j / tile_h << ")" << endl;
-          tile +=  tile_j * tile_width + (tile_i * 3);
+          uint8_t* tile = this->access_pixel(i, j);
           *(tile++) = istr.get();
           *(tile++) = istr.get();
           *(tile++) = istr.get();
@@ -300,16 +345,10 @@ struct TiledImage : public Image {
 
     virtual bool write_body(std::ofstream& ostr) const override
     {
-      unsigned int const tile_width = tile_w * 3;
-
       for (unsigned int j = 0; j < height; ++j)
         for (unsigned int i = 0; i < width; ++i)
         {
-          unsigned int const tile_index = (j / tile_h) * nb_col_tile + (i / tile_w);
-          uint8_t* tile = tiles[tile_index];
-          unsigned int const tile_j = j % tile_h;
-          unsigned int const tile_i = i % tile_w;
-          tile +=  tile_j * tile_width + (tile_i * 3);
+          uint8_t const* tile = this->access_pixel(i, j);
           ostr << (char) *(tile++);
           ostr << (char) *(tile++);
           ostr << (char) *(tile++);
@@ -532,6 +571,9 @@ void round_if_very_small(double& d)
 {
   if (abs(d) < 1.0e-10)
     d = 0.0;
+
+  if (abs(d - 1) < 1.0e-10)
+    d = 1.0;
 }
 
 
@@ -576,6 +618,13 @@ void rotate_pixel(Image const& src, Image& rotated,
   double y_delta = src_rotated_point.y - floor(src_rotated_point.y);
   round_if_very_small(y_delta);
 
+  // special case if we can directly map the src to the dest
+  if (x_delta == 0 && y_delta == 0)
+  {
+    memcpy(&rotated.buffer[rot_index], &src.buffer[src_index], 3 * sizeof (uint8_t));
+    return;
+  }
+
   // SIMD
   __m128 const x_d = _mm_set_ps1(x_delta);
   __m128 const inv_x_d = _mm_set_ps1(1 - x_delta);
@@ -692,6 +741,136 @@ Image rotate(Image const& src, double angle)
   return rotated;
 }
 
+//
+//
+// Tile rotation
+//
+
+template<unsigned int W, unsigned int H>
+void rotate_pixel(TiledImage<W, H> const& src, TiledImage<W, H>& rotated,
+                  DPoint const& src_rotated_point,
+                  unsigned int rot_tile_index, unsigned int rot_index)
+{
+  uint8_t const* src_index_1 = src.access_pixel((int) src_rotated_point.x, (int) src_rotated_point.y);
+
+  double x_delta = src_rotated_point.x - (int) src_rotated_point.x;
+  round_if_very_small(x_delta);
+  double y_delta = src_rotated_point.y - (int) src_rotated_point.y;
+  round_if_very_small(y_delta);
+
+  // special case if we can directly map the src to the dest
+  if (x_delta == 0 && y_delta == 0)
+  {
+//    cout << "we can directly map, w00t" << endl;
+    uint8_t* rot_tile = rotated.tiles[rot_tile_index];
+    memcpy(&rot_tile[rot_index], src_index_1, 3 * sizeof (uint8_t));
+    return;
+  }
+//  cout << "src rotated point: " << src_rotated_point << endl;
+//  cout << "src rotated point y: " << src_rotated_point.y << endl;
+//  cout << "src rotated point y int: " << (int) src_rotated_point.y << endl;
+//  cout << "x delta = " << x_delta << endl;
+//  cout << "y delta = " << y_delta << endl;
+
+  uint8_t const* src_index_2 = src.access_pixel((int) src_rotated_point.x + 1, (int) src_rotated_point.y);
+  uint8_t const* src_index_3 = src.access_pixel((int) src_rotated_point.x, (int) src_rotated_point.y + 1);
+  uint8_t const* src_index_4 = src.access_pixel((int) src_rotated_point.x + 1, (int) src_rotated_point.y + 1);
+
+  // FIXME: deal with image border
+  if (!src_index_1 || !src_index_2 || !src_index_3 || !src_index_4)
+    return;
+
+  // SIMD
+  __m128 const x_d = _mm_set_ps1(x_delta);
+  __m128 const inv_x_d = _mm_set_ps1(1 - x_delta);
+  __m128 top_left = _mm_set_ps(*src_index_1, *(src_index_1 + 1), *(src_index_1 + 2), 0.0);
+  __m128 top_right = _mm_set_ps(*src_index_2, *(src_index_2 + 1), *(src_index_2 + 2), 0.0);
+  top_left = _mm_mul_ps(top_left, inv_x_d);
+  top_right = _mm_mul_ps(top_right, x_d);
+  top_left = _mm_add_ps(top_left, top_right);
+
+  __m128 bottom_left = _mm_set_ps(*src_index_3, *(src_index_3 + 1), *(src_index_3 + 2), 0.0);
+  __m128 bottom_right = _mm_set_ps(*src_index_4, *(src_index_4 + 1), *(src_index_4 + 2), 0.0);
+  bottom_left = _mm_mul_ps(bottom_left, inv_x_d);
+  bottom_right = _mm_mul_ps(bottom_right, x_d);
+  bottom_left = _mm_add_ps(bottom_left, bottom_right);
+
+  __m128 const y_d = _mm_set_ps1(y_delta);
+  __m128 const inv_y_d = _mm_set_ps1(1 - y_delta);
+  top_left = _mm_mul_ps(top_left, inv_y_d);
+  bottom_left = _mm_mul_ps(bottom_left, y_d);
+  top_left = _mm_add_ps(top_left, bottom_left);
+
+  // convert float values to uint8_t
+  uint8_t* rot_tile = rotated.tiles[rot_tile_index];
+  rot_tile[rot_index] = top_left[3];
+  rot_tile[rot_index + 1] = top_left[2];
+  rot_tile[rot_index + 2] = top_left[1];
+}
+
+template<unsigned int W, unsigned int H>
+TiledImage<W, H> rotate(TiledImage<W, H> const& src, double angle)
+{
+  double const rotation = (angle / 180.0f) * M_PI;
+  unsigned int w = 0;
+  unsigned int h = 0;
+  compute_output_size(src, rotation, w, h);
+  TiledImage<W, H> rotated(w, h);
+
+  DPoint src_origin = get_mapped_point(rotated, Point(0, 0), -rotation);
+  DPoint src_delta_x = get_mapped_point(rotated, Point(1, 0), -rotation);
+  DPoint src_delta_y = get_mapped_point(rotated, Point(0, 1), -rotation);
+
+  src_delta_x.x = src_delta_x.x - src_origin.x;
+  src_delta_x.y = src_delta_x.y - src_origin.y;
+  round_if_very_small(src_delta_x.x);
+  round_if_very_small(src_delta_x.y);
+  src_delta_y.x = src_delta_y.x - src_origin.x;
+  src_delta_y.y = src_delta_y.y - src_origin.y;
+  round_if_very_small(src_delta_y.x);
+  round_if_very_small(src_delta_y.y);
+
+  DPoint const rot_origin_in_src_grid = get_mapped_point(rotated, Point(0, 0), -rotation);
+  DPoint const rot_origin_in_src = convert_img_coord_precision(src, rot_origin_in_src_grid);
+
+  for (unsigned int y = 0; y < rotated.nb_row_tile; ++y)
+  {
+    for (unsigned int x = 0; x < rotated.nb_col_tile; ++x)
+    {
+      unsigned int const rot_tile_index = y * rotated.nb_col_tile + x;
+
+      for (unsigned int j = 0; j < H; ++j)
+      {
+        int const y_index = y * H + j;
+
+        for (unsigned int i = 0; i < W; ++i)
+        {
+//          cout << "rotated: tile[" << x << ", " << y << "] point(" << i << ", " << j << ")" << endl;
+          unsigned int const rot_index = (j * W + i) * 3;
+          int const x_index = x * W + i;
+          Point const rot_point(x_index, y_index);
+          DPoint src_rotated_point(rot_origin_in_src.x + x_index * src_delta_x.x + y_index * src_delta_y.x,
+                                   rot_origin_in_src.y + x_index * src_delta_x.y + y_index * src_delta_y.y);
+
+//          cout << "rotated tile index: " << rot_tile_index << endl;
+//          cout << "src point: " << src_rotated_point << endl;
+
+          if (src_rotated_point.x < 0 || src_rotated_point.x > src.width
+            || src_rotated_point.y < 0 || src_rotated_point.y > src.height)
+            continue;
+
+          rotate_pixel(src, rotated,
+                       src_rotated_point,
+                       rot_tile_index, rot_index);
+
+        }
+      }
+    }
+  }
+
+  return rotated;
+}
+
 
 
 //
@@ -851,38 +1030,58 @@ int main(int argc, char* argv[])
 
     if (!check_90(argv[1]))
     {
-      cerr << __LINE__ << " | 90 degrees check failed" << endl;
+      cerr << __LINE__ << " | 90 degrees check failed" << endl << endl;
 //      return 1;
     }
   }
 
-  TiledImage<8, 8> tiled_img(argv[1]);
-  tiled_img.save("tiled.ppm");
-  return 0;
-
   Image img(argv[1]);
+  TiledImage<16, 16> tiled_img(argv[1]);
 
-  for (double rotation = 0; rotation < 360; rotation += 5)
+  for (double rotation = 0; rotation < 360; rotation += 450)
   {
-  //  double rotation = 45;
+    // No tile
     auto const before = chrono::high_resolution_clock::now();
-
-    Image rotated = rotate(img, rotation);
-
+    Image const rotated = rotate(img, rotation);
     auto const after = chrono::high_resolution_clock::now();
     auto const duration_ms = std::chrono::duration_cast<std::chrono::milliseconds>(after - before);
+
+    // Tile
+    auto const before_tiled = chrono::high_resolution_clock::now();
+    auto const rotated_tiled = rotate(tiled_img, rotation);
+    auto const after_tiled = chrono::high_resolution_clock::now();
+    auto const duration_ms_tiled = std::chrono::duration_cast<std::chrono::milliseconds>(after_tiled - before_tiled);
+
     cout << "rotate(" << rotation << "): " << duration_ms.count() << " ms" << endl;
-    // TODO: compare implementation
+    cout << "tiled: " << duration_ms_tiled.count() << " ms" << endl;
+    cout << "speedup: " << (int) (((float) duration_ms.count() / duration_ms_tiled.count() - 1) * 100) << "%" << endl << endl;
+
+    // tile check
+    //tiled_img.print_tile(0);
+    //rotated_tiled.print_tile(0);
 
     stringstream filename;
-    // filename << "/tmp/";
+//    filename << "/tmp/";
     filename << "rotated_";
+
+    stringstream filename_tiled;
+//    filename_tiled << "/tmp/";
+    filename_tiled << "tiled_rotated_";
+
     if (rotation < 100)
+    {
       filename << "0";
+      filename_tiled << "0";
+    }
     if (rotation < 10)
+    {
       filename << "0";
+      filename_tiled << "0";
+    }
     filename << rotation << ".ppm";
+    filename_tiled << rotation << ".ppm";
     rotated.save(filename.str());
+    rotated_tiled.save(filename_tiled.str());
   }
 
   return 0;