http://www.ibrtses.com/embedded/logarithms.html

The first part is aligning the number such that the leading bit is one. 
This done by a multiple shift. The second is basically a table. Or a 
table with interpolations.

For small bit sizes, you can implement totally it with a table.

I can provide you with a code and a license to use it in professional 
applications. The code is scalable to 64 Bit wide input vectors and 
works for base "e". In conjunction with a pre calculated value of 
1/ln(2) or 1/ln(10) you can directly multiply it with the result and 
obtain all bases you want to have.

I would charge you with €199,- + VAT, payment by bank wire transfer.

Please drop me a note.

I am interested in this core. Can you sent a kind of demo in shape of a 
ngc for xilinx for instance?