One eval() method per clock domain ?
in multiple clock domains designs, I am wondering if it would be more efficient to have one eval() method per clock (or even two : rising and falling edges ?).
To generate clocks with arbitrary frequency, I have this special C++ object : http://github.com/fredrequin/fpga_1943/tree/master/verilator/clock_gen .
If I connect the eval() methods to the clock generator through call-backs, I should get some nice runtime improvement.
Any thoughts ?
RE: One eval() method per clock domain ? - Added by Wilson Snyder over 1 year ago
Multiple evals would save some branch instructions. There's two problems though, first is any logic downstream which feeds from multiple edges would be in multiple evals. Second multiple clocks may arrive at the exact same time (e.g. gated clocks) and for correctness (to avoid races) must be part of the same eval.