Quoting Paul Welton <[email protected]>: Hi,
> I have been experimenting with using GHDL to simulate a processor > design. The good news is that GHDL did compile and simulate the design > so that a comprehensive test ran to completion and produced the same > result. Small changes to the VHDL were required, but these appeared to > be moving closer to the VHDL standard. These were: > > > * Some declarations of std_logic_vector constants were found with > implicit dimensions. These needed to be given dimensions. > > < constant VALID0 : std_logic_vector := "10"; > > --- > > > constant VALID0 : std_logic_vector (1 downto 0) := "10"; This is supposed to work (and a few bugs have been fixed for that). What was the issue or the error message ? > * Some uses of the mod operator were ambiguous as to the > interpretation of the type of the operands, which required a cast. > > < case value'length mod 4 is > > --- > > > case integer'(value'length) mod 4 is Interesting. I think I could reproduce the issue. > * One of the text files which is read the simulation required > hexadecimal values to be padded with zeros to a multiple of four digits > to be read successfully by the procedure "hread". > > > > Some statistics: > > Active lines of code (excluding lines with only comments or > whitespace) 186210 > > Final executable size > 34 Mbytes > > Memory requirement when running > 484 Mbytes Woow, pretty large design! > The bad news is that the simulation took two hours, whereas on a > proprietary simulator it completes in around three minutes. My question > is, would it be possible to speed it up, assuming that significant > resource was available? Would there be interest in doing this? If it was > less than five times slower than the proprietary simulator, it would be > cost effective to buy five times the computers in order to use GHDL and > save on simulator licenses. Yes, of course it would be interesting to speed it up. But this is not an easy task. First we need to find why GHDL is so slow. The best method is to isolate tiny design that are very slow compared to proprietary simulator. You can help here. Then we need to understand why and fix GHDL. Thomas has recently spotted out slowness in non-sensitized processes and I am working currently on fixing this. So, in one word we need your input! Tristan. _______________________________________________ Ghdl-discuss mailing list [email protected] https://mail.gna.org/listinfo/ghdl-discuss
