.DDS File Format – Pixel format

DirectDraw Surface

The DDS_PIXELFORMAT structure sits inside the header, at offset 76 from the start of the file. It either names a format with a four-character code or describes uncompressed pixels with bit masks.

Fields

File offsetSizeNameDescription
764dwSizeStructure size: 32.
804dwFlagsWhat kind of data the surface holds; see below.
844dwFourCCA four-character code for compressed or custom formats. DX10 means a DX10 header follows.
884dwRGBBitCountBits per pixel, valid with DDPF_RGB, DDPF_LUMINANCE or DDPF_YUV.
924dwRBitMaskMask for red, or luminance, or Y.
964dwGBitMaskMask for green, or U.
1004dwBBitMaskMask for blue, or V.
1044dwABitMaskMask for alpha.

dwFlags

FlagValueMeaning
DDPF_ALPHAPIXELS0x1The texture has alpha data, and dwABitMask is valid.
DDPF_ALPHA0x2Older files: alpha-only uncompressed data.
DDPF_FOURCC0x4dwFourCC is valid.
DDPF_RGB0x40Uncompressed RGB data; the bit count and masks are valid.
DDPF_YUV0x200Older files: uncompressed YUV data, with Y, U and V in the three masks.
DDPF_LUMINANCE0x20000Older files: single-channel uncompressed data. Can be combined with DDPF_ALPHAPIXELS for two channels.

Four-character codes

Microsoft’s table gives the following. Codes marked “legacy” are ones Microsoft says a robust reader must be able to handle, while writers should prefer the DX10 header.

dwFourCCFormatNote
DXT1DXGI_FORMAT_BC1_UNORM, D3DFMT_DXT1
DXT2D3DFMT_DXT2Premultiplied alpha
DXT3DXGI_FORMAT_BC2_UNORM, D3DFMT_DXT3
DXT4D3DFMT_DXT4Premultiplied alpha
DXT5DXGI_FORMAT_BC3_UNORM, D3DFMT_DXT5
BC4UDXGI_FORMAT_BC4_UNORMLegacy
BC4SDXGI_FORMAT_BC4_SNORMLegacy
ATI2DXGI_FORMAT_BC5_UNORMLegacy
BC5SDXGI_FORMAT_BC5_SNORMLegacy
RGBGDXGI_FORMAT_R8G8_B8G8_UNORM, D3DFMT_R8G8_B8G8
GRGBDXGI_FORMAT_G8R8_G8B8_UNORM, D3DFMT_G8R8_G8B8
UYVYD3DFMT_UYVY
YUY2D3DFMT_YUY2
DX10Any DXGI formatThe real format is in the DX10 header

Numeric codes

A common variant sets dwFourCC directly to a D3DFORMAT or DXGI_FORMAT enumeration value. There is no way to tell which enumeration a number belongs to, so this is not recommended, but these values are widely written and Microsoft says readers must handle them:

ValueFormat
36DXGI_FORMAT_R16G16B16A16_UNORM, D3DFMT_A16B16G16R16
110DXGI_FORMAT_R16G16B16A16_SNORM, D3DFMT_Q16W16V16U16
111DXGI_FORMAT_R16_FLOAT, D3DFMT_R16F
112DXGI_FORMAT_R16G16_FLOAT, D3DFMT_G16R16F
113DXGI_FORMAT_R16G16B16A16_FLOAT, D3DFMT_A16B16G16R16F
114DXGI_FORMAT_R32_FLOAT, D3DFMT_R32F
115DXGI_FORMAT_R32G32_FLOAT, D3DFMT_G32R32F
116DXGI_FORMAT_R32G32B32A32_FLOAT, D3DFMT_A32B32G32R32F
117D3DFMT_CxV8U8

Uncompressed formats by mask

In these rows, dwFlags combines the flags above: RGB is DDPF_RGB, RGBA adds DDPF_ALPHAPIXELS, ALPHA is DDPF_ALPHA and LUMINANCE is DDPF_LUMINANCE.

FormatdwFlagsBitsRed maskGreen maskBlue maskAlpha mask
DXGI_FORMAT_R8G8B8A8_UNORM, D3DFMT_A8B8G8R8RGBA320xFF0xFF000xFF00000xFF000000
DXGI_FORMAT_R16G16_UNORM, D3DFMT_G16R16RGBA, RGB320xFFFF0xFFFF0000
DXGI_FORMAT_R10G10B10A2_UNORM, D3DFMT_A2B10G10R10RGBA320x3FF0xFFC000x3FF00000
DXGI_FORMAT_B5G5R5A1_UNORM, D3DFMT_A1R5G5B5RGBA160x7C000x3E00x1F0x8000
DXGI_FORMAT_B5G6R5_UNORM, D3DFMT_R5G6B5RGB160xF8000x7E00x1F
DXGI_FORMAT_A8_UNORM, D3DFMT_A8ALPHA80xFF
D3DFMT_A8R8G8B8RGBA320xFF00000xFF000xFF0xFF000000
D3DFMT_X8R8G8B8RGB320xFF00000xFF000xFF
D3DFMT_X8B8G8R8RGB320xFF0xFF000xFF0000
D3DFMT_A2R10G10B10RGBA320x3FF000000xFFC000x3FF0xC0000000
D3DFMT_R8G8B8RGB240xFF00000xFF000xFF
D3DFMT_X1R5G5B5RGB160x7C000x3E00x1F
D3DFMT_A4R4G4B4RGBA160xF000xF00xF0xF000
D3DFMT_X4R4G4B4RGB160xF000xF00xF
D3DFMT_A8R3G3B2RGBA160xE00x1C0x30xFF00
D3DFMT_A8L8LUMINANCE160xFF0xFF00
D3DFMT_L16LUMINANCE160xFFFF
D3DFMT_L8LUMINANCE80xFF
D3DFMT_A4L4LUMINANCE80xF0xF0

For 10:10:10:2 data, Microsoft says the most robust choice is the DX10 header with format 24, because of long-standing inconsistencies in how other software reads the masks. D3DFMT_A2R10G10B10 data should be converted first.