tig-binary

A Rust crate that wraps an algorithm from tig-algorithm for compilation into a shared object.

TIG uses a forked version of LLVM compiler for injecting runtime signature and fuel tracking into the shared object.

For CUDA, TIG uses nvcc to generate ptx using target version sm_70/compute_70, before injecting runtime signature and fuel tracking into the ptx file.

Getting Started

Run the appropiate development docker image. Available flavours are:

amd64 (x86_64 compatible)
aarch64
amd64-cuda12.6.3 (x86_64 compatible)
aarch64-cuda12.6.3

# example
docker run -it -v $(pwd):/app ghcr.io/tig-foundation/tig-monorepo/dev:0.0.1-aarch64
# scripts build_so and build_ptx are on PATH

Build shared object using build_so script:

Expects tig_algorithm::<CHALLENGE>::<ALGORITHM>::solve_challenge to be importable
Outputs to tig-algorithms/lib/<CHALLENGE>/<ARCH>/<ALGORITHM>.so, where ARCH is aarch64 or amd64

# add '--cuda' flag if building cuda algorithm
build_so $CHALLENGE $ALGORITHM

(Optional) Build ptx using build_ptx script:

Expects tig_algorithm/src/<CHALLENGE>/<ALGORITHM>.cu or tig_algorithm/src/<CHALLENGE>/<ALGORITHM>/benchmarker_outbound.cu file to exist
Outputs to tig-algorithms/lib/<CHALLENGE>/ptx/<ALGORITHM>.ptx

build_ptx $CHALLENGE $ALGORITHM

License

End User License Agreement

1.6 KiB Raw Permalink Blame History

tig-binary

Getting Started

License

1.6 KiB

Raw Permalink Blame History