We experience some interesting behaviour on a Intel HD 4600 with the 10.18.10.3383 driver version at a client site.
We have a compute shader with one input and one output. We bind the input texture to slot t0 and the output to slot u0. The output texture has twice the size as the input texture. In the CPU code we first bind the input then the output. When requesting the dimensions of the textures using (RW)Texture2D::GetDimensions(), we get the size of the output texture for both the input and the output. If we bind the output texture to u1, we get the expected results. Is this a known bug? And if so in what driver version was this fixed?
This is running on Windows 7.1 and here is some information about the graphics card: