Client Performance degradation on large maps about endlessclient

[ethanmoffat]

Used the Visual Studio profiler, and it found most of the effort in running the game was in EOMapRenderer::_doMapDrawing and EOMapRenderer::_doMapRenderTargetDrawing (as expected). The plan to update this is as follows:

1. Render fill tile/ground layer to one target, this will be cached for performance gains
2. Render other map gfx, players, NPCs on separate render target (referred to as target2 here)
   1. Unfortunately, cannot do players/map objects separately/using blending because they would be clipped incorrectly in certain places
      • so will have to redraw each time an update is required for a player, npc, or door due to some kind of action.
   2. Additional performance gains will have to come from only drawing within certain range of main character
3. Render main player to target2 with following blending rules:
   1. If target alpha > 0 (ie, visible pixels), player alpha = 50% (or whatever looks good)
   2. If player is hidden, don't bother with blending
4. Render ground layer to back buffer
5. Render map items to back buffer
6. Render target2 to back buffer
7. profit

The order for drawing any object (players, npcs, and map objects included) will start at 0, 0 and go through each layer row by row, with increasing columns drawn later, so the objects appear to be rendered in the proper order (this is currently being done).

[ethanmoffat]

New performance measure for large maps:

Pretty much idles at 20% on the largest maps. Note that CPU usage still scales with map size. This is to be expected. Overall, drastically improved from the original case. Note increased memory usage as well - this is a byproduct of switching between more maps as well as improving caching of GFX.