wrong TXL spec in GL_NV_fragment_program2?

According to the the spec, the LOD for the TXL op is computed as follows:

The level of detail is computed as specified in section 3.8.8, except that log_2(rho(x,y)) is given by 2^w, where w is the w component of the vector operand.

However, my experiments show that log_2(rho(x,y)) is given by w, or in other words rho(x,y) is given by 2^w.

Is this a typ-o in the spec, or is something wrong with my implementation.

PS. apparently, although magnification filters don’t have the mipmap option, if you want to use the TXL op on a magnified texture, the minification filter needs to be set to mipmap-filtering.