Implement correlation on GPU

[J08nY]

Two things are meant under this, (Pearson's) correlation coefficient computation for CPA and cross-correlation computation for alignment and other uses. These are two quite different algorithms, grouped here because both are interesting and use the word "correlation".

TODO

• Pearson's correlation coefficient
• Cross-correlation

Pearson's correlation coefficient

For CPA: https://www.iacr.org/archive/ches2004/31560016/31560016.pdf
where it is used to compute (samplewise) correlation coefficient between hypothesized leakage values and power leakage.

Implemented on GPU here: https://arxiv.org/abs/1412.7682
also here: https://link.springer.com/chapter/10.1007/978-3-642-27257-8_16
also here: https://dl.acm.org/doi/10.1145/2611765.2611775
also here: https://www.semanticscholar.org/paper/Improving-DPA-analysis-with-distributed-computing-Amaral-Inesc-Id/43c150d1dd42a2327e302202784d2b685347e7a4
CPU work here: https://crypto.fit.cvut.cz/sites/default/files/publications/fulltexts/pearson.pdf
Some interesting work: https://informatik.rub.de/wp-content/uploads/2021/11/MA_Klostermann.pdf

More resources:

• http://www.cs.tau.ac.il/~tromer/istvr1516-files/lecture3-power-analysis.pdf
• https://wiki.newae.com/Correlation_Power_Analysis

Cross-correlation

For alignment you take a reference trace and cross-correlate every trace from the trace set with it, then find the peak
value of the result and use that as a shift to align the trace.