Texture binding with "unpitched" memory


I would like to allocate an array of, say, 14 x 13 floating point values and bind a texture to it.

If I use cudaMallocPitch to allocate the array, I can bind a texture to the array using cudaBindTexture2D and all operations on the texture work just fine (e.g., interpolation)
If I use cudaMalloc (that is the width of the array is not pitched to 64 bytes), I can still bind a texture using cudaBindTexture2D but most operations on the texture do not provide the correct result.

According to this reference


allocating with cudaMalloc should not be a problem. Any idea?

Thanks for your help,

You are probably using the wrong height/pitch.

try this:
width = 14;
height = 13;
pitch = width*sizeof(float);

Also be sure that your cudaChannelFormatDesc is set up correctly. You should be using:

cudaChannelFormatDesc desc = cudaCreateChannelDesc();

Alternatively you can copy the linear data to a 2D cuArray, and then bind that.

Textures impose certain alignment restrictions on the underlying memory, which frequently also requires padding for each row. Using either cudaArray, or using pitch-linear memory allocated with cudaMallocPicth() guarantees that all requirements are met. You can get things to work with cudaMalloc(), but have to take care of alignment and padding yourself. I am not sure all requirements are publicly documented, so I would not recommend going down that path.

What about copying linear data to a cudaArray or 2d/3d allocated memory, and using that for texture binding?

I’ve been able to copy 1d data to a 3D/2DLayered/2D/1DLayered cudaArray without any issues as far as I can tell. Could this be problematic for certain data types?

Thanks. I have reached the same conclusion too. I will restrict texture binding to pitched memory to make sure that it is properly aligned.