Closed cynthi8 closed 2 years ago
Thank you Erin for pointing this out. There was a typo in the code. The issue has been fixed. Please let me know if you have other observations. Thanks!
Thanks!
Is it still intended behavior for the XOR gates to be excluded from each other's neighborhood?
With the latest change, I still see XOR = 1 for the XOR gates - but I would expect this to be 3, as XOR B and XOR C are also within two hops of XOR A.
PI | PO | KEY | XOR | XNOR | AND | OR | NAND | NOR | INV | BUF | ADDF | AOI | OAI | MXIT | AO1B | AOI2XB | AO | OA | OAI2XB | in_degree | out_degree | TIELO | TIEHI | RF2R | RF1R | PREICG | POSTICG | M | A | FRICG | MXT | MX | ADDH |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1 | 0 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 2 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
Erin
Hi Erin, Correct. The way we have implemented is as follows: Starting at the target node, we get its output. Then, for the outputs, we also check their outputs (2 hops). Then, we look at the inputs of the target node and their inputs as well (2 hops). An example showing this corner case was not included in the paper. We release the codes so that everyone can access the exact implementation.
Hi Erin, I believe that Fig. 5 in the GNNUnlock+ paper shows a similar example. https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9529342
Incorrect Neighborhood Count
When the circuit from Fig. 3b is parsed using
SFLL_Verilog_to_graph.pl
, the feature vector does not match the one shown in DATE 2021.Neighborhood Count Dependent on Key/Primary Inputs
Neighborhood vs Input/Output Cone of Dependency
Although no neighborhood examples were provided in the paper, the feature vector is described as
This feature vector for an XOR node also shows PO = 1 which is incorrect.
Reproduction Steps: