Some Texture Type Ommited in CUDA2.0beta?

The following texture type was able to be used in CUDA1.1.
However, it is not possible to use it with CUDA2.0beta.

Texture<float4, 2, cudaReadModeNormalizedFloat> tex;

Is this a specification change?
Or, is it a bug?

In the CUDA 1.1 programming guide:

You have Type=float4 which is not a 16 or 8-bit integer type, so I think that CUDA 2.0 is checking against the specification better. What are your expected semantics of reading a float as a “normalized float”?