- The DSimpleCanvas constructor now fills MemBuffer with zeros.

- Fixed: If the FBTexture wasn't exactly the same size as the screen,
  D3DFB::PaintToWindow() would still lock it with D3DLOCK_DISCARD. Alas,
  I saw no speedup for using a dirty region. (Side note: The Radeons are
  apparently slower compared to DirectDraw because they must do
  power-of-2 textures. If they ever add non-power-of-2 support like nvidia,
  I assume they will also see a speed gain.)
- Changed fb_d3d9.cpp so that instead of trying to compensate for Geforce
  off-by-one errors in the pixel shader, it automatically detects where
  the error occurs and modifies the way the palette is uploaded to
  compensate. Palette color 255 is then represented using the texture
  border color instead of actually being part of the palette. This should
  work correctly with all cards, since I had a report of an FX where the
  off-by-one occurred in a different spot from the place where I observed
  it on a 6 and 7 series cards. Since the shader now has one fewer
  instruction, I notice a very marginal speedup. (Interestingly, removing
  the flash blending from the shader had no perceivable performance gain.)


SVN r399 (trunk)

src/v_video.cpp

@@ -595,6 +595,7 @@ DSimpleCanvas::DSimpleCanvas (int width, int height)
 		}
 	}
 	MemBuffer = new BYTE[Pitch * height];
+	memset (MemBuffer, 0, Pitch * height);
 }
 
 DSimpleCanvas::~DSimpleCanvas ()