Linear filtering for integer textures

Accelerated Computing CUDA CUDA Programming and Performance
1

Are there any plans to support linear filtering for integer textures?

2

It depends what you mean by integer textures - CUDA supports filtering of 8-bit normalized textures, just like the graphics APIs, but the hardware doesn’t support filtering of unnormalized integer textures:

[url=“http://developer.download.nvidia.com/opengl/specs/GL_EXT_texture_integer.txt”]http://developer.download.nvidia.com/openg...ure_integer.txt[/url]

I don’t think we have any plans to add hardware support for filtering integer formats since these were mainly designed for storing data, not images.

If you want a higher precision texture format that is filterable you can use floating point.

3

Medical imaging usually stores the images in integer (2 bytes) format. How can I use texturing without explicitely converting to floating point format?

4

Ah, you should be able to use filtering on 16-bit formats (just not 32-bit) - you just need to set the read mode to normalized float, i.e.

texture<ushort4, Dim, cudaReadModeNormalizedFloat> texRef;