catch22_c

Canonical catch22 C kernel benchmark

Times the upstream catch22 C kernel on the same Gaussian inputs and n schedule as examples/bench_catch22.rs. Pair the two outputs for an apples-to-apples kernel-vs-kernel comparison (no Python wrapper in either column).

Build & run

# 1. Clone the canonical C tree (anywhere outside this repo's src/).
git clone --depth 1 https://github.com/DynamicsAndNeuralSystems/catch22 \
    .reference/catch22-c

# 2. Build the C library.
cd .reference/catch22-c/C
gcc -O3 -c $(ls *.c | grep -v '^main.c$')
ar rcs libcatch22.a *.o

# 3. Build and run the bench harness against it.
gcc -O3 -Wno-format \
    -o bench_c \
    ../../../benchmarks/catch22_c/bench_c.c libcatch22.a -lm
./bench_c

The harness uses the same xorshift64 + Box–Muller RNG as the Rust bench, so the inputs are statistically identical. It runs the full 22-feature panel (z-score + every kernel main.c invokes) per timed iteration and reports the median wall-clock per call.

.reference/ is gitignored — keep the upstream clone there to avoid mixing it into the rust-stats source tree.