This repository contains a compact, end-to-end demo of LaBRADOR — the first practical lattice-based zk-SNARK (CRYPTO 2023) - built on top of ICICLE v4. LaBRADOR produces ~50 kB proofs without a trusted setup and is secure under the Module-SIS assumption, making it resistant to both classical and quantum attacks.
ICICLE ships highly-tuned GPU and CPU kernels for FFT/NTT, polynomial arithmetic and lattice primitives. Thanks to those kernels the prover can run unchanged on a laptop CPU or a CUDA-capable GPU and enjoy order-of-magnitude speed-ups.
For additional background see the original paper and our detailed blog post.
To run the program on CPU use
./run.sh
To run on GPU, run
./run.sh -d CUDA
This script will automatically download the ICICLE CUDA backend (v4.0.0) when running with the CUDA option, build the necessary components, and run the prover.
You can also specify a custom backend installation directory with:
./run.sh -d CUDA -b /path/to/backend
The main program runs a simple benchmarking program for which the parameters can be set here:
std::vector<std::tuple<size_t, size_t>> arr_nr{{1 << 6, 1 << 3}};
std::vector<std::tuple<size_t, size_t>> num_constraint{{10, 10}};
size_t NUM_REP = 1;
bool SKIP_VERIF = false;
benchmark_program(arr_nr, num_constraint, NUM_REP, SKIP_VERIF);You can also run the function prover_verifier_trace in main. This function runs 3 recursion iterations of the Labrador protocol for a random instance.
The following flag in prover.h can be used to control the program output:
// SHOW_STEPS creates a print output listing every step performed by the Prover and the time taken
constexpr bool SHOW_STEPS = true;All functions and objects are documented in code.
