I want to share my work-in-progress systems language with a v0.1.0 release of Lockstep. It is a data-oriented systems programming language designed for high-throughput, deterministic compute pipelines.

I built Lockstep to bridge the gap between the productivity of C and the execution efficiency of GPU compute shaders. Instead of traditional control flow, Lockstep enforces straight-line SIMD execution. You will not find any if, for, or while statements inside compute kernels; branching is entirely replaced by hardware-native masking and stream-splitting.

Memory is handled via a static arena provided by the Host. There is no malloc, no hidden threads, and no garbage collection, which guarantees predictable performance and eliminates race conditions by construction.

Under the hood, Lockstep targets LLVM IR directly to leverage industrial-grade optimization passes. It also generates a C-compatible header for easy integration with host applications written in C, C++, Rust, or Zig.

v0.1.0 includes a compiler with LLVM IR and C header emission, a CLI simulator for validating pipeline wiring and cardinality on small datasets and an opt-in LSP server for real-time editor diagnostics, hover type info, and autocompletion.

You can check out the repository to see the syntax, and the roadmap outlines where the project is heading next, including parameterized SIMD widths and multi-stage pipeline composition.

I would love to hear feedback on the language semantics, the type system, and the overall architecture!