Hello,

I’m currently working with PQC, and I noticed that some PQM4 benchmarks show similar performance for ML-KEM decapsulation and encapsulation. In some cases, decapsulation is even faster than encapsulation.

But how is that possible? Since, one step of the decapsulation is performing the encapsulation process.

I tried to check if in the decapsulation the matrix A is cached from the encapsulation, but based on what I’ve understood from the source code, it seems to be generated on the fly.

Best Regards,
João Pedro