Uses Python to generate Solana-compatible LLVM IR.

Generates LLVM IR using Python (llvmlite) → compiles to Solana sBPF → deploys to Solana.

The transpiler is hand-coded (not parsing arbitrary Python yet...), but it proves the concept works.

LLVM IR (Intermediate Representation) is a low-level, strongly typed, and platform-independent assembly-like language used by the LLVM compiler framework to represent programs during compilation. It acts as a common representation layer between high-level source languages and machine code, enabling advanced optimization and code generation for multiple architectures.

# Setup
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

# Build and deploy
chmod +x build.sh
./build.sh

You'll get a Program ID. Test it:

npm install # just for the first time
node invoke.js <YOUR_PROGRAM_ID>

You should see: Program log: Hello World from Python!

transpiler.py (llvmlite)
    ↓ generates LLVM IR
program.ll
    ↓ llvm-as
program.bc (LLVM bitcode)
    ↓ sbpf-linker --cpu v3
hello_world.so (Solana program)
    ↓ solana deploy
Runs on-chain ✓

# LLVM tools (macOS)
brew install llvm

# sbpf-linker (converts LLVM bitcode to Solana sBPF)
cargo install sbpf-linker

Solana syscalls are function pointers at specific addresses. For example, sol_log_ is at 0x207559bd.

We use LLVM's inttoptr to cast that address to a function pointer:

syscall_hash = ir.Constant(i64, 544561597)  # 0x207559bd
sol_log_ptr = builder.inttoptr(syscall_hash, func_ptr_type_ptr)
builder.call(sol_log_ptr, [string_ptr, message_len])

In the end we're just generating LLVM IR that Solana can understand, and then compiling it to sBPF with sbpf-linker.

• LLVM IR
• sbpf-linker
• llvmlite