RISP neuron model fire comparison is incorrectly evaluated

[BGull00]

In risp_neuron.sv, the following line determines whether or not a neuron fires: fire = sum >= (THRESHOLD + !THRESHOLD_INCLUSIVE);

In the case of THRESHOLD=1, THRESHOLD_INCLUSIVE=1, and sum=-1, this comparison incorrectly evaluates to fire=1. This comparison is evaluated correctly upon changing !THRESHOLD_INCLUSIVE to a literal 0. This comparison is also evaluated correctly upon changing !THRESHOLD_INCLUSIVE to ~THRESHOLD_INCLUSIVE.

[keegandent]

Weird, I'm going to see if I can find other places where there is arithmetic with !var expressions and check them too.

[keegandent]

So this wasn't actually an issue with !var math but instead a side effect of #14. In my PR, I did add support for the threshold_inclusive JSON parameter that wasn't there, as well as the non_negative_charge parameter. I will at some point implement fire_like_ravens.

[BGull00]

I'm still having this same issue with my test XOR network upon further evaluation. Here is a screenshot demonstrating the issue. Notice that when the highlighted sum signal for neuron 2 = -1, fire is high despite the threshold being 1 (-1 < 1 so this should not be the case).

[keegandent]

That's so bizarre; Can you tell me what number it says for POTENTIAL_ABS_MAX? If that logic is working I haven't the faintest idea what could be screwing with the right side of that sum comparison.

After that, could you please cherry-pick this commit and see if that fixes it? What is the sim value for THRESHOLD_ACTUAL?

[keegandent]

Something is really funky here because it won't fire sum = 0 but will when sum = -1

[keegandent]

It is occurs to me that asking the synthesis tool to compare two values is equivalent to asking it to subtract the right-hand side and compare to 0. So maybe I should make that explicit and adjust SUM_WIDTH, perhaps renaming sum to something like fire_integral? Just thinking out loud