Optimize rendering

renderer.h

@@ -10,6 +10,9 @@
 
   Some comments:
 
+  - The module uses small3dlib, a tiny software rasterization library. This
+    module knows nothing about I/O (windows, canvases, ...), it just say where
+    to draw pixels and what colors they should have.
   - The map 3D model is divided into 4x4x4 chunks, i.e. 64 in total, out of
     which only 8 are loaded at any time, depending on where the camera is and
     where it is looking. This is to save resources, we don't draw the far away
@@ -17,6 +20,10 @@
   - Extremely simple LOD of far away chunks is implemented: we keep an 8x8x8
     bit array of where there is empty space and where there is "something", then
     for far away areas with "something" we just draw some 2D rectangles.
+  - RENDERING IS THE BOTTLENECK OF PERFORMANCE, it takes even much more time
+    than physics simulation, i.e. care should be taken to make code here very
+    optimized, namely the _LCR_pixelFunc3D function, as it is called for every
+    rasterized pixel.
 */
 
 #define S3L_RESOLUTION_X LCR_EFFECTIVE_RESOLUTION_X
@@ -108,12 +115,12 @@ struct
   // pixel function precomputed values:
   uint32_t previousTriID;
   int triUVs[6];
-  int texSubsampleCount;
-  unsigned int flatAndTransparent; /**< If non-zero, transparent (dithered)
-                                        polygons will be drawn without texture,
-                                        with color stored in this variable. */
+  uint_fast16_t texSubsampleCount;
+  uint_fast16_t flatAndTransparent; /**< If non-zero, transparent (dithered)
+                                         polygons will be drawn without texture,
+                                         with color stored in this variable. */
 #if LCR_SETTING_PARTICLES
-  uint16_t particleColor;          /**< 0x0000 means no particles active. */
+  uint_fast16_t particleColor;      /**< 0x0000 means no particles active. */
 #endif
 } LCR_renderer;
 
@@ -338,7 +345,7 @@ void _LCR_pixelFunc3D(S3L_PixelInfo *pixel)
   LCR_gameDrawPixelXYUnsafe(pixel->x,pixel->y,LCR_renderer.flatAndTransparent);
 
 #else // LCR_SETTING_POTATO_GRAPHICS
-  // once we get a new triangle, we precompute things for it:
+  // New triangle? Precompute stuff for it:
   if (pixel->triangleID != LCR_renderer.previousTriID)
   {
     LCR_renderer.previousTriID = pixel->triangleID;
@@ -483,10 +490,14 @@ void _LCR_pixelFunc3D(S3L_PixelInfo *pixel)
     }
   }
 
+  /* Bottleneck: code from here below will be ran for every rasterized pixel,
+     optimizing it may significantly improve rendering performance. */
+
   if (LCR_renderer.flatAndTransparent)
   {
-    if (pixel->x % 2 == pixel->y % 2)
-      LCR_gameDrawPixelXYUnsafe(pixel->x,pixel->y,LCR_renderer.flatAndTransparent);
+    if ((pixel->x % 2) == (pixel->y % 2))
+      LCR_gameDrawPixelXYUnsafe(
+        pixel->x,pixel->y,LCR_renderer.flatAndTransparent);
     else
       S3L_zBufferWrite(pixel->x,pixel->y,S3L_MAX_DEPTH);
       /* ^ Clear z-buffer if we didn't draw the pixel. Without this further
@@ -1034,8 +1045,7 @@ uint8_t _LCR_buildMapModel(void)
                 LCR_IMAGE_WALL_CONCRETE : LCR_IMAGE_WALL_WOOD;
             }
             else
-            { // TODO: tidy this mess?
-
+            {
               if (LCR_mapBlockIsAccelerator(blockType))
                 triData |= LCR_IMAGE_GROUND_ACCEL;
               else if (LCR_mapBlockIsFan(blockType))