> Does ghdl make use of more that one processor? > Until the event-driven / delta-cycle simulator is re-tooled to use extra processors explicitly if they are available, probably the only practical way would be to compile it with a compiler that can parallelize things like loops and other common data structures.
The Sun compiler can do this, and I have sped up other software that would normally run on only a single processor to use many more automatically, but then Amdahl's law takes hold and you reach a certain point of diminishing returns. However we are talking about 8 times speedup for code that is well-behaved. However given that GHDL grafts itself onto gcc, which is then compiled with an ada-supported gcc compiler itself, I don't know whether this is even doable using Sun WorkShop/Forte compiler rather than using gcc. The best I can do at the moment is to crank up the optimizations within a single thread, ie. instruction-level parallelism. I just recently managed to get GHDL 0.25 to compile on Solaris 8, and yes I have wondered about how to extract more speed out of it. I think that if the event-driven simulation engine were somehow re-designed to use multiple processors if they are available using standard Unix mechanisms, then a means might exist to bring GHDL up to the performance of commercial simulators. My understanding is that many commercial simulators are indeed faster, though they cannot complete with the price and elegance of GHDL's overall design. Tristan is extremely gifted to have taken GHDL as far as he has, mostly on his own. S. _______________________________________________ Ghdl-discuss mailing list [email protected] https://mail.gna.org/listinfo/ghdl-discuss
