A relatively longstanding issue is that some of the math_real functions provided by ghdl use a nonstandard function name. I got annoyed by having to maintain two different versions of my testbenches for ghdl and another simulator, so I wrote a two patches to make math_real more standards compliant. ieee-mathreal-nocompat.patch just renames the functions. It will break code that relies on the ghdl function names. ieee-mathreal.patch also contains backward compatible functions that warn once that they will disappear in the future. However, to implement the warn once feature, these functions had to be marked impure, which breaks existing code that relies on these functions being pure. Comments? I'd really like to see ghdl moving towards standards compliant function names... And now for something different: $ ghdl -a t.vhd t.vhd:13:14:warning: universal integer bound must be numeric literal or attribute This warning seems totally bogus, the bounds _are_ numeric literals. What's wrong here? Also, to compile ghdl on ppc64, I had to use the hack in ghdl-ppc64abort.patch. Is there a better solution? Tom
on 2009-04-02 15:38
on 2009-04-03 18:04
Quoting Thomas Sailer <t.sailer@alumni.ethz.ch>: > A relatively longstanding issue is that some of the math_real functions > provided by ghdl use a nonstandard function name. I got annoyed by > having to maintain two different versions of my testbenches for ghdl and > another simulator, so I wrote a two patches to make math_real more > standards compliant. Thank you for working on this issue. > I'd really like to see ghdl moving towards standards compliant function > names... I will have a look on these patches. The current situation is: * official math vhdl sources cannot be distributed freely (but you can get it from the web) * an early draft is available (the one provided with ghdl) but the final draft is somewhat different. > And now for something different: > > $ ghdl -a t.vhd > t.vhd:13:14:warning: universal integer bound must be numeric literal or > attribute > > This warning seems totally bogus, the bounds _are_ numeric literals. > What's wrong here? Well -10 is not a numeric literal. The warning is correct but the vhdl standard is crazy on this point! > Also, to compile ghdl on ppc64, I had to use the hack in > ghdl-ppc64abort.patch. Is there a better solution? Looks not bad. Tristan.
Enable email notification
Enable multi-page view