Build from Source

Anvil256 — Build from Source

Prerequisites

Component	Tool	Min version
Contracts	Foundry (forge + cast)	0.2.x
CLI	Rust toolchain	1.78
Kernel (NVIDIA)	CUDA toolkit (nvcc)	12.3
Kernel (AMD/Intel)	OpenCL SDK (opencl-headers)	3.0
Website	Node.js + pnpm	20.x

---

Reproducible Build (Recommended)

docker run --rm -v $PWD:/src ghcr.io/anvil256/builder:latest
ls dist/       # binaries + sha256sums.txt

The Docker image pins exact compiler versions. The output sha256sums.txt must match the official release. Verify:

sha256sum -c dist/sha256sums.txt

---

Native Build

Contracts

cd contracts/
forge install          # OpenZeppelin, forge-std
forge build            # compiles Anvil256.sol + all sub-contracts
forge test -vvv        # full test suite + fuzz runs

Key contracts and their role:

Contract	Role
Anvil256.sol	Core: mine(), ERC-20, halving, fee, epoch counter
CascadePoW.sol	τ + inner + outer Cascade verification
PIController.sol	PI + NCT combined difficulty update
MinerWindow.sol	256-slot circular buffer + uniqueCount + windowFreq
FixedPointMath.sol	Q60.18, 4th-order Maclaurin exp, O(1) log₂
FeeOracle.sol	Chainlink ETH/USD → fee_wei

CLI

cd cli/
cargo build --release    # produces target/release/anvil256-cli
cargo test --release

The CLI is responsible for:

• Reading γ(miner) = minerEpochCount[address] from the contract
• Reading epochEntropy from the contract
• Computing τ = H(miner ∥ γ ∥ genesis) and inner = H(τ ∥ epoch)
• Dispatching the kernel job with the derived inner value (not the raw address)
• Watching for epoch transitions and re-dispatching immediately

Kernel — CUDA (NVIDIA)

cd kernel/
make cuda                    # produces ./miner
make cuda ARCHS="75 80 86 89 90"  # fat binary for multiple SM architectures

Kernel — OpenCL (AMD / Intel / fallback)

cd kernel/
make opencl              # produces ./miner-opencl

Website

cd website/
pnpm install
pnpm dev       # local preview at http://localhost:4321
pnpm build     # produces dist/

---

Cross-Compilation

Target	Method
linux-x86_64-gnu	native or cross Docker
linux-aarch64-gnu	cross build --target aarch64-unknown-linux-gnu
darwin-arm64	on macOS: cargo build --target aarch64-apple-darwin
windows-x64-msvc	cargo build --target x86_64-pc-windows-msvc

CUDA kernel cross-compilation is not supported. Build natively on the target platform.

---

Test Coverage Targets

Suite	Target	Critical property verified
CascadePoW.t.sol	100% branch	Two-pass correctness; γ-binding rejection
PIController.t.sol	100% branch	Steady-state convergence; NCT; anti-windup
MinerWindow.t.sol	100% branch	Circular buffer invariants; uniqueCount accuracy
FixedPointMath.t.sol	100% branch	exp4 error bound; log2Floor exactness
Fuzz	≥ 10,000 runs	Supply cap; epoch monotonicity; fee accounting
Integration (CLI)	End-to-end	γ fetch → kernel → submit → receipt → γ increment