SHA3 Library – High-Performance Parallel SHA3-256

This library provides a parallel, SIMD-accelerated implementation of SHA3-256 optimized for fixed-size messages.

Build

mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release \
      -DSHA3_BUILD_EXAMPLES=ON \
      -DCMAKE_C_FLAGS="-O3 -march=native -funroll-loops" ..
make -j$(nproc)

API

Include the public header:

#include "sha3.h"

Hash N messages of constant length msg_len (≤ block size):

int rc = sha3_hash_parallel_len(
    SHA3_256,   // hash type
    data,       // input buffer: N * msg_len bytes
    msg_len,    // constant message length
    out,        // output buffer: N * digest_size bytes
    N           // number of messages
);

For exactly 64-byte messages:

sha3_hash_parallel(SHA3_256, data, out, N);

Benchmark

Measure pad-and-hash throughput for 1 000 000 messages of 64 bytes:

build/bin/sha3_parallel_benchmark 1000000 64

Example output on AVX2/AVX-512 hardware:

sha3_hash_parallel_len: 1000000 messages of 64 bytes in 0.009 s
Throughput: 113e6 msgs/s (6.9 GiB/s) on 16 threads

All-in-One Build & Benchmark Script

For an end-to-end build, self-test, and performance run (parallel hashing and Merkle-tree build), use the included bench.sh script:

chmod +x bench.sh
./bench.sh [PAR_BENCH_COUNT] [MERKLE_BENCH_LEAVES]

• PAR_BENCH_COUNT (default: 10000000) – number of 64-byte messages to hash in parallel
• MERKLE_BENCH_LEAVES (default: 10000000) – number of 32-byte leaves in the 4-ary Merkle tree

Example run on a 16-core AVX-512 machine:

=== SHA3 Parallel Benchmark ===
sha3_hash_parallel_len: 10000000 messages of 64 bytes in 0.126 s
Throughput: 79e6 msgs/s (4.8 GiB/s) on 16 threads

=== Merkle Tree Benchmark ===
Building 4-ary Merkle tree with 10000000 leaves...
Built in 0.194 s -> 52 M hashes/s
All done.

License

Apache-2.0