Open jleightcap opened 1 year ago
Note: the potential performance benefit mentioned is incorrect. The current singular tiny86
@function
is optimal. CC @lisaoverall, I'm not sure I understood completely why that is the case.
The current singular
tiny86
@function
is optimal.
At this point, this is a conjecture by our collaborators based on (I think) I/O performance in how their proof systems handle @function
.
For targets that require a fully inlined circuit, yosys itself can perform complete inlining. See: https://yosyshq.readthedocs.io/projects/yosys/en/latest/cmd/flatten.html
For targets with function calls (like the primary IR0+), a non-flattened circuit with
@function
s at module boundaries could yield more efficient circuits at verification time.