Versions:
MLRift, published by Pantelis Christou and currently available at version v1.1.0 (the single release to date), is a self-hosted, zero-dependency systems language and compiler designed for machine-learning workloads. Forked from KernRift, the package installs two portable command-line tools: mlrc, the compiler itself, and mlr, a fat-binary runner. Positioned within the compiler and developer-tools category, MLRift targets users who need native code generation for ML applications without reliance on external toolchains or heavyweight runtime dependencies. The compiler emits native x86_64 and ARM64 binaries across Linux, Windows, macOS, and Android, making it suitable for cross-platform deployment of performance-sensitive machine-learning software on desktops, servers, and mobile devices alike. On the language side, MLRift extends the intermediate-representation backend with ML-specific primitives, including tensors, event streams, and sparse CSR operations, enabling the direct expression of common machine-learning data structures and computational patterns at the systems level. A distinguishing feature is its native AMDGCN GPU backend, which communicates with the /dev/kfd kernel interface directly, eliminating any dependency on the ROCm stack for GPU acceleration on AMD hardware. Because the project is self-hosted, the MLRift compiler is written in MLRift itself, and its zero-dependency design simplifies installation and portability across supported operating systems. Typical use cases include compiling tensor-based workloads to native executables, building ML applications that exploit sparse matrix operations, running compiled fat binaries through the mlr runner, and targeting AMD GPUs for compute tasks without installing vendor runtime frameworks. As a version 1.x release in the programming-language and compiler tooling category, MLRift provides a compact, self-contained option for developers seeking a unified systems language with built-in machine-learning primitives and direct hardware access across CPU and GPU targets.
Tags: