MLRift Language Support

Editor support for the MLRift systems language — syntax highlighting plus a language server that drives live diagnostics and IntelliSense from the real compiler.

Features

• Syntax highlighting for .mlr files, with a file icon in the explorer and tabs.
• Live diagnostics — errors and warnings from mlrc check, mapped to the exact line and column as you type.
• IntelliSense — completions, hover docs, and go-to-definition for keywords, built-ins, types, functions, structs, and enums.

Highlighting covers:

• Integer types — short aliases u8/u16/u32/u64, i8/i16/i32/i64 and long forms uint8..int64.
• Floating point — f16, f32, f64.
• Built-in functions — pointer ops (load8..64, store8..64), volatile ops (vload/vstore8..64), atomics (atomic_load/store/cas/add/...), bitfield ops (bit_get/set/clear/range/insert), signed compares (signed_lt/gt/le/ge), string output (print_str, println_str), and platform hooks (get_target_os, get_arch_id, syscall_raw, exec_process).
• Keywords including match, loop, defer, for, while, and the declaration set (fn, struct, enum, static, const, device, ...).
• device blocks for MMIO: device NAME at ADDR { FIELD at OFF : TYPE rw }.
• Static and struct arrays (static u8[N] name, Point[10] pts), slice parameters (fn foo([u8] data) with data.len), and method syntax (fn Point.sum(Point self) -> u64).
• Annotations (@export, @noreturn, @naked, @packed, @section("name")) and the #lang directive (#lang stable / #lang experimental).
• String/char literals with escapes, line (//) and block (/* */) comments, auto-closing brackets, indentation, and folding.

Requirements

Syntax highlighting works on its own. Diagnostics and hover docs need the mlrc compiler installed — the language server shells out to mlrc check. If mlrc is not found, highlighting and completions still work, but no errors are reported.

Build the compiler from the repository, and make sure mlrc is on your PATH or set mlrift.compilerPath below.

Extension Settings

About MLRift

MLRift is a self-hosted systems language for machine-learning and GPU workloads. Forked from KernRift, it compiles itself ahead-of-time to native machine code — no VM, no interpreter, no libc — and extends the base language with ML-specific primitives (tensors, event streams, sparse ops) and a native backend targeting AMD GPUs (HIP/AMDGPU) alongside embedded MCUs (ESP32/Xtensa, RISC-V). One .mlrbo fat binary bundles multiple platform slices; a small runner extracts and executes the matching slice at startup. Self-host bootstrap fixed point is verified by CI on every push.

The compiler is written entirely in MLRift and includes an SSA IR backend, graph-coloring register allocator, constant folding / DCE / CSE, and per-target native code emitters.

• GitHub