What is the format of DXGI_FORMAT_D24_UNORM_S8_UINT?

Question

I'm trying to read the values in a depth texture of type DXGI_FORMAT_D24_UNORM_S8_UINT.

I know this means "24 bits for depth, 8 bits for stencil" "A 32-bit z-buffer format that supports 24 bits for depth and 8 bits for stencil.", but how do you interpret those 24 bits?

It's clearly not going to be a 32-bit int, and it's not going to be a 32-bit float. If it is an integer value, how "far away" is a value of "1" in the depth texture?

Answer 1

It's an Unsigned NORMalized float, which is a floating-point value between 0.0 and 1.0.

EDIT: To clarify, the Depth component is accessed as a float, and the Stencil component is accessed as an unsigned integer.

Answer 2

Here is some demo code which shows how it works:

// create a temporary texture 
D3D11_TEXTURE2D_DESC desc; 
depthStencilTexture->GetDesc(&desc); 
desc.Usage = D3D11_USAGE_STAGING; 
desc.BindFlags = 0; 
desc.CPUAccessFlags = D3D11_CPU_ACCESS_READ; 
desc.MiscFlags = 0; 

ID3D11Texture2D* tmpTexture = nullptr; 
HRESULT hr = getDevice()->CreateTexture2D(&desc, nullptr, &tmpTexture); 

// Copy depth buffer
getImmediateContextD3D()->CopyResource(tmpTexture, depthStencilTexture); 

// access depth buffer
D3D11_MAPPED_SUBRESOURCE mappedRes; 
getImmediateContextD3D()->Map(tmpTexture, 0, D3D11_MAP_READ, 0, &mappedRes); 

unsigned int* color = (unsigned int*)mappedRes.pData;

// Extract 24 depth bits
float depth = static_cast<float>(color & 0x00FFFFFF); 
depth /= 16777216f; // divide bei 2^24

// compute a grayscale value [0;255]
unsigned char colorValue = static_cast<unsigned char>(depth * 255.0f);

Answer 3

Generally for your texture (or rather the SRV you'll create) you use a format in the same "type" family, in this case DXGI_FORMAT_R24G8_TYPELESS - then when you sample it in your shader you just use a .r swizzle to get the depth value out (in 0..1 range). You should not, of course, be reading it back to the CPU - how to interpret the data is going to be the least of your problems if you do.

Answer 4

I solved this problem by using DXGI_FORMAT_D32_FLOAT in my depth texture instead. It turned out that the z-buffer is all floats between 0.0f and 1.0f.