Sampler does not work with cubemaps in 16-bit floating-point pixel format
Submitted by Vic Lee
Assigned to Default DRI bug account
If a cubemap is created using 16-bit component pixel format like PIPE_FORMAT_R16G16B16A16_UNORM or PIPE_FORMAT_R16G16B16A16_FLOAT, it will not be sampled correctly in fragment shader and generate garbage output. Those 16-bit formats works just fine when using in 2D texture, and the cubemap using 8-bit format also works. It's just the combination cubemap+16bit that does not work.