I have located the vhdl-200x versions of math_real (package and body) at http://www.eda.org/vhdl-200x/vhdl-200x-ft/packages/files.html
The math_real files appear to compile cleanly, but I could not locate the testbenches, so I cannot test exhaustively. However it does at least work in place of the version that currently ships with ghdl. I have not tried math_complex, but this is also available at the vhdl-200x site. Since the vhdl-200x function implementations are native VHDL, they will be slower than the current versions, which are mostly foreign links to the native -lm functions. However, I personally favour correctness over speed by default, so this would not be a deterrent to me. I would suggest the vhdl-200x versions be shipped with ghdl in future, except the copyright message on the header appears to restrict distribution in source form. It looks like the best alternative might be to patch the current versions. I will look at doing that. Best regards, Ivan
