An illegal memory access bug that leads to Wrong PSI Results or Segmentation fault

[Ni-kaFruit]

Hi, I am trying to run your code with different parameters presented in Table 1 of your paper.
However, there is an illegal memory access bug that leads to Wrong PSI Results or Segmentation fault.

The parameters I used are as follows:
senderSize = 1 << 16; receiverSize = 1 << 10; width = 632; logHeight = 11; height = 1.2 * receiverSize; hashLengthInBytes = 9;
The results are as follows:

I have fixed this bug and opened a pull request.

Thx :)

[peihanmiao]

Hello,

Thank you for pointing this out.

Our code doesn't support the scenario of "height = 1.2 * receiverSize" (height is always set as 1 << logHeight and we always set logHeight = rs). This is because in all the experiments in our paper, we set m = n (where m is the height and n is the receiverSize) as it achieves nearly optimal communication among all choices of m and allows for optimal computation. The code is also designed and optimized for the setting where the height is a power of 2 (1 << logHeight).

I you would like to try other parameters, I just wanted to mention that other changes may need to be made (e.g. /PSI_test/main.cpp) and you may need to be careful.

Thanks,
Peihan

[Ni-kaFruit]

Hi

Indeed, I'm now trying to run your code with arbitrary parameters that it wasn't originally designed to support (senderSize, height, and receiverSize are not always set as a power of 2 and the code for kkrt16 supports this scenario). Our tests show that your code is able to handle this situation with the revision of #7. I believe that supporting this arbitrary parameter scenario can bolster the impact of your open-source code.

I am grateful for your advice. I have implemented the calculation mentioned in "Section 3.3 Parameter Analysis" for a suitable width. If you wish, I could submit another pull request to add support for this part.

thx, :)