This is a fork of the original rustlantis, modified to work within the no_std environment of a GPU. Instead of comparing different backends / optimization levels, it runs the code on a CPU and on a CUDA-capable GPU.
The size of the generated samples has also been adjusted, to avoid overwhelming GPU drivers(this increases fuzzing speed, at the cost of a lower fuzz quality).
Install the nvptx64-nvidia-cuda target, and a CUDA runtime.
Go into the fuzzer directory, and run:
cargo run --bin fuzzer --release -- --start 0 --count 1000
The starting seed and the fuzz campaign size can be adjusted. The found samples will be dumped at target/fuzz.