The new probe system works nicely ! Except when it doesn't... This happens when any sign of backwards connection appears, even when a backwards gate is inserted !
This is caused by the limitations in the way one can drive an entity's output : it has to be triggered by at least a change of one input. The new probe algorithm takes a "safety shortcut" by changing the output only when all the inputs have a valid value. This effectively prevents the gate's output from propagating the "toggle signal" when one input is tied to a backwards gate, which has not yet received the toggle signal...
It's a chicken&egg problem that would not exist if a gate's output was directly controllable by an external code. But the other cause is the AND condition for the new algorithm. It is required to ensure the speed in most circumstances and it performs great in the pure-boolean INC8 and ALU8 units. The "corner cases" test miserably fails though. Worse : all the logic cone after the first gate that receives a backwards signal is also "shadowed" because the trigger signal can't propagate.
And because of the "shadow", it's not possible to know if a "disconnected" gate is really disconnected, receives a backwards signal OR is in the shadow of a backwards-fed gate.
The answer to this question can come from a "preliminary pass" that scans/propagates through all the circuit but instead of sending a chunk of the driving gate's number, it sends the type of the driving gate: input, boolean, backwards... This first pass will already solve the question of which inputs are not connected, so these gates can still propagate other trigger signals during the main probing algorithm.
The new pre-pass requires storage of (temporary ?) extra data for each input (gate & output port), which can be :
- not connected - 'U'
- input port
- gate
- backwards
- (eventually a sequential gate ?)
This can be a single std_logic variable, but it must be allocated somewhere. It can be temporary because unconnected inputs cause the rest of the program to fail, and once the netlist is built, checking the type of the driver is easy :
- input ports are non-positive driver numbers,
- gates, including backwards, have a positive driver number,
- gates have a enum_gate_kind that can be looked up.
(to be continued)
Discussions
Become a Hackaday.io Member
Create an account to leave a comment. Already have an account? Log In.