16 bit bitmaps

You don''t have to match color depths, but there are some things to be aware of:

In 32 bit modes, it can be faster to use 16, and/or 8 bit textures. This is because there is less data to fetch. For example, greyscale images can be loaded as L8 (luminance, ie: r=g=b=l, a=1), P8 (paletted), or X8R8G8B8 depending on your hardware. Each will produce a perfect 256 level grey image. Whether that''s an 8 bit image, or 32 bit, is partly your problem as a developer, and partly the end user''s problem based on what card they have. Similarly, 256 level alpha masks could be loaded as A8, or P8, or A8L8, or A8R8G8B8 depending on what your hardware supports.

Even for color images, if you don''t require subtle gradients in the source texture, use a 16 bit (A1R5G5B5, X1R5G5B5, R5G6B5) texture. You save GPU memory, and precious clock cycles. Most of your subtle gradients are going to come from lighting most likely anyway, and the source bitmap''s depth doesn''t matter.

In 16 bit modes you may want to use 32 bit images. Lets say you have a fullcolor image with alpha. You want this to be a single texture. You want to mix this image with another image. Your options for the first image are A4R4G4B4 (which most card''s can''t do), A1R5G5B5 (only 1 bit of alpha, and limited color), or A8R8G8B8, which gives plenty of alpha and color information to blend with.

Then again, most times you don''t want to blend the texture with anything, in which case anything over 16 bit would be wasteful.

Look into using DDS textures with the DXTn formats. They are compressed, and most modern hardware knows how to deal with them. They save both RAM and bandwidth. If your hardware doesn''t know how to deal with DXT, D3DXCreateTextureFromFile[InMemory][Ex] decompresses them for you. Some images may look bad as DXT (it IS lossy), so keep those images 16, or 32 bit.

Anyway, there are no short answers. Every project is different. But you don''t have to worry about matching formats.


About the tiles: If you create an Ortho projection matrix, no depth scaling will occur. Another option, if you''ve already got your DDraw engine going, is to disable the Z buffer (or not make one in the first place), and draw all the objects back to front just like you used to.

The advantage of the depth buffer and orthogonal projection is that you can batch together several sets of tiles at several depths and draw them at once, then switch textures, and drawing all tiles that need the new texture in the next draw call. Don''t worry about batching yet, just be aware that there is something to learn once you''re past the basics, which can provide a nice speed boost.

Any tiles that have semi-transparent parts must be draw back to front for alpha blending to work correctly. This is something the depth buffer can''t help you out with. (Blended with alpha 0, or alpha 255 is fine, but blending with alpha of 10, 50, 127, etc. are potential problems)

Wow... that''s a load of helpful information. I think that pretty much covers what I needed to know. so color loss won''t be a problem with mixing formats then? Well, thank you for that response! I''m going to have to print it out.