About the service
I will develop high-performance systems on Terra and Lua, as well as code generators for your task. Terra is a statically typed low-level companion to Lua: Lua is used for metaprogramming, and Terra generates native machine code via LLVM. As a result, you get C-level speed and the flexibility of a scripting language. This combination is invaluable when you need to generate specialized code at runtime, fuse hot loops, or embed a domain language without interpreter overhead.
What I do:
— JIT kernels for numerical computations: matrix operations, signal processing, simulation loops, specialized for data shape, type, and target processor.
— Code generators: Lua metaprograms that produce optimized Terra functions from your schemas, templates, and configurations.
— Embedded DSLs: clean high-level syntax compiles directly to native Terra with staging and partial evaluation.
— Interop and bindings: connecting Terra to C/C++ libraries, access via Lua or pure C ABI, integration with Node or Socket IO service.
— Optimization: SIMD, loop unrolling, memory layout and cache work, plus benchmarks confirming speedup.
You get clean, documented, and tested code, reproducible build scripts, benchmark reports, and a brief overview so your team can develop the generators themselves. Correctness first, then speed — each kernel comes with tests. Describe your task, target hardware, and constraints, and I will propose a fast and maintainable solution.
What I do:
— JIT kernels for numerical computations: matrix operations, signal processing, simulation loops, specialized for data shape, type, and target processor.
— Code generators: Lua metaprograms that produce optimized Terra functions from your schemas, templates, and configurations.
— Embedded DSLs: clean high-level syntax compiles directly to native Terra with staging and partial evaluation.
— Interop and bindings: connecting Terra to C/C++ libraries, access via Lua or pure C ABI, integration with Node or Socket IO service.
— Optimization: SIMD, loop unrolling, memory layout and cache work, plus benchmarks confirming speedup.
You get clean, documented, and tested code, reproducible build scripts, benchmark reports, and a brief overview so your team can develop the generators themselves. Correctness first, then speed — each kernel comes with tests. Describe your task, target hardware, and constraints, and I will propose a fast and maintainable solution.
