Solved by pull request #1213
Benchmarks used: GitHub - adriannic/openfhe-bench: Personal benchmark suite for openfhe · GitHub
Benchmark results: OpenFHE Benchmarks - Google Sheets
While running benchmarks to make a performance comparison against OpenFHE and OpenFHE-HEXL I noticed a performance degradation especially in less computationally intensive operations like addition or multiplication CKKS primitives, reaching slowdowns of up to 4-5x compared to v1.4.0, due to a high page fault/kernel overhead as mentioned in pull request #1213.
Context:
OpenFHE v1.5.1 9e3a29724e48314ec21e52995c12870cea0fa728
OpenFHE v1.4.0.0 99d9762b9cb61f4e645a267f8eedfd95c7e471d7
(both from openfhe-configurator)
Os: Arch Linux
Kernel: 7.0.14
Compiler: AMD clang version 22.0.0git (/srcdest/rocm-llvm f58b06dce1f9c15707c5f808fd002e18c2accf7e) (from package rocm-llvm 2:7.2.4-1)
CMake configuration: in the benchmark repo (CMakeLists.txt) referenced in the spreadsheet, CMake v4.3.4
To run the benchmarks point the OPENFHE makefile variable to a folder containing openfhe and openfhe-hexl installed in separate folders. Then, run either “make openfhe” or “make openfhe-hexl” followed by “make bench-12” or any other “bench” configuration. The “bench-all” target outputs the results to three json files inside the “build” directory, the rest just output to stdout.