Hello,

I'm trying to read the raw texture values of textures in DirectX 11 and I've run into a bit of a weird problem. My code seems to work fine for 512x512 textures but as soon as I try reading data from a 1024x1024 texture, I get a crash at the memcpy() line.

Here's relevant code:

// set up the loaded textures as subresources
D3D11_SUBRESOURCE_DATA* subresources = new D3D11_SUBRESOURCE_DATA[texCount];
 
// copy the loaded texture data into the subresource structures
for(UINT i = 0; i < texCount; ++i) {
D3D11_MAPPED_SUBRESOURCE mappedSubResource;
 
HRESULT mapResult = p_deviceContext_->Map(textures[i], 0, D3D11_MAP_READ, 0, &mappedSubResource);
 
if(FAILED(mapResult)) {
MessageBox(NULL, L"Failed mapping a subresource.", L"Error", 0);
return;
}
 
// size of the texture's data defined by the length of a single row of data * the total number of rows (texture height)
subresources[i].pSysMem = new BYTE[mappedSubResource.RowPitch * texHeight];
memcpy(const_cast<void*>(subresources[i].pSysMem), mappedSubResource.pData, mappedSubResource.RowPitch * texHeight);
subresources[i].SysMemPitch = mappedSubResource.RowPitch;
subresources[i].SysMemSlicePitch = mappedSubResource.DepthPitch;
 
p_deviceContext_->Unmap(textures[i], 0);
 
}

Maybe an issue with the last row. RowPitch tells you the offset between rows, it doesn't say anywhere in the spec that last row should be RowPitch in size.

I suggest you try to loop on each row, and copy exactly (PixelSizeInBytes*TexWidth), maybe it will help.

BTW, this is a good advice in any case - you then allocate less system memory, and everything in consecutive in memory, resulting in better cache behavior.

Maybe an issue with the last row. RowPitch tells you the offset between rows, it doesn't say anywhere in the spec that last row should be RowPitch in size.

I suggest you try to loop on each row, and copy exactly (PixelSizeInBytes*TexWidth), maybe it will help.

BTW, this is a good advice in any case - you then allocate less system memory, and everything in consecutive in memory, resulting in better cache behavior.

Thanks for your quick reply! I'm not quite sure how to loop row by row, the following code crashes around j = 344:

UINT pixelSize = mappedSubResource.RowPitch / texWidth;
BYTE** pixels = new BYTE*[texHeight];
void* dataPointer = mappedSubResource.pData;
 
for(int j = 0; j < texHeight; ++j) {
pixels[j] = new BYTE[pixelSize * texWidth];
memcpy(pixels[j], dataPointer, pixelSize * texWidth);
dataPointer = static_cast<char*>(dataPointer) + (pixelSize * texWidth);
}

Yes, it's because you're using BC2. The data in a BC2 texture is effectively one byte per pixel. It also rounds the width and height up to a multiple of four.

For the details how how BC2 textures encode their data look at: http://msdn.microsoft.com/en-us/library/windows/desktop/bb694531%28v=vs.85%29.aspx#BC2

When I was loading the textures, I used ZeroMemory() to fill in the D3DX11_IMAGE_LOAD_INFO struct and only set the format, usage and CPU access flags myself. However, this caused the full mipmap chain to be created because

MipLevels
Type: UINT
The maximum number of mipmap levels in the texture. See the remarks in D3D11_TEX1D_SRV. Using 0 or D3DX11_DEFAULT will cause a full mipmap chain to be created.

Once I changed the MipLevels value to 1, my original copying code worked. Oops!

Thanks for your replies though!