DDS File Layout for Textures

For simple uncompressed textures, a .dds file should contain DDSD_PITCH and a DDPF_RGB pixel format surface. The pitch value should be DWORD-aligned. In this case, the main image should be (dwPitchOrLinearSize-by-dwHeight) bytes total.

If the file also contains mipmaps, the DDSD_MIPMAPCOUNT, DDSCAPS_MIPMAP, and DDSCAPS_COMPLEX flags should be set, and the dwMipMapCount field should contain the total number of levels defined in the file (for a 256-by-256 image, this field would be 9). The pitch for each mipmap-level image is computed from dwWidth and the pixel format, which is then DWORD-aligned.

A 256-by-256 main image with a pixel format of R8G8B8 (three bytes per pixel) and all mipmap levels would contain the following layers:

For simple compressed textures, the file should contain DDSD_LINEARSIZE and a DDPF_FOURCC pixel format surface. In this case, the image should be dwPitchOrLinearSize bytes total. If the file contains mipmaps, the DDSD_MIPMAPCOUNT, DDSCAPS_COMPLEX, DDSCAPS_MIPMAP, and dwMipMapCount fields should be set.

For DXTn compressed formats, the size of each mipmap level image is one-fourth the size of the previous, with a minimum of 8 (DXT1) or 16 (DXT2-5) bytes (this is true only for square textures). For example, a 256-by-256 main image, with a pixel format of DXT1 and all mipmap levels, would contain the following:

If mipmaps are generated, all levels down to 1-by-1 are usually written. Some tools may not support partial mipmap chains.

When computing DXTn compressed sizes for non-square textures, the following formula should be used at each mipmap level:

max(1, width ÷ 4) x max(1, height ÷ 4) x 8(DXT1) or 16(DXT2-5)

A 256-by-64 main image, with a pixel format of DXT1 and all mipmap levels, would contain the following: