RunMat
GitHub

RunMat Changelog

What's new across the RunMat runtime, cloud, and sandbox. For technical runtime details and commit diffs, see GitHub Releases (runtime only).

v0.4.0

April 13, 2026

Runtime

Added

New plot types:

  • Add stem with GPU-accelerated rendering
  • Add errorbar visualization
  • Add area filled area plots
  • Add contour / contourf with GPU-resident path rendering
  • Add plot3 3D line plots
  • Add imagesc scaled image display
  • Add pie charts
  • Add histogram plotting
  • Add semilogx, semilogy, loglog log-scale axes
  • Add plot(y) shorthand — single-argument plot

Plotting infrastructure:

  • Add graphics handles and per-axes subplot state
  • Add axes helpers: xlim, ylim, zlim, caxis/clim
  • Add common signature forms: bar(x,y), surf(Z), mesh(Z), surfc(Z), meshc(Z), stairs(y)
  • Add get/set for plot handles with shared property access
  • Add annotation and legend builtins: title, xlabel, ylabel, zlabel, legend
  • Add 3D view controls: view(az, el)
  • Add world-space text annotations

Numerical computing:

  • Implement correct matrix division semantics — \, /, .\, ./ now have distinct, correct semantics through parser, HIR, bytecode, VM, and GPU paths. mldivide and mrdivide support triangular solves, dense square solves, tall least-squares, wide minimum-norm solves, transpose variants, and Cholesky-backed solves with GPU-resident F32 execution
  • Implement full FFT/IFFT family — fft, ifft, fft2, ifft2, fftn, ifftn now fully implemented with underlying RustFFT for CPU paths and high-performance staged GPU shaders supporting power-of-two, radix-3, radix-5, mixed compositions, and Bluestein fallback for non-smooth/prime lengths. Complex array indexing now works throughout (Y = fft(x); Y = Y(1:N/2))
  • Add signal processing helpers — hann, hamming, blackman windowing functions and nextpow2, all with GPU acceleration. Supports symmetric/periodic modes and output type selection

Other:

  • Add clear, clc, close all — session management commands now work in browser and native
  • Add datetime — MATLAB-compatible construction, formatting, string conversion, and subtraction
  • Add duration — display, arithmetic, and datetime interop
  • Add peaks builtin — GPU-accelerated with mixed-residency tensor support and type inference

Changed

  • Expand GPU fusion coverage — sign, fix, hypot, pow2, asinh, acosh, atanh, mod, rem now fuse into single GPU kernels automatically

Fixed

  • Fix struct field indexing — s.arr(k) and s.arr(1:n) now work with MATLAB-compatible semantics
  • Fix implicit struct creation — assigning to r.x = 10 on an uninitialized variable now materializes a struct
  • Fix fprintf compatibility for literal dollar signs and casted numeric inputs
  • Fix atan2 stack underflow that crashed scripts calling atan2 with compound expressions
  • Fix leading-dot floats — .5 parsed as 0.5
  • Fix GPU tensor to f64 conversion for iterative solve workflows
  • Fix range-indexing type errors
  • Fix GPU memory cleanup on QR and mrdivide errors
  • Fix matmul parameter caching
  • Fix histogram and bar charts not displaying
  • Fix right bar chart persisting after generating new charts
  • Fix contourf Z dimension mismatch
  • Fix area X/Y length mismatch
  • Fix image surface data handling
  • Fix stairs 1-argument form
  • Fix Z-axis label styling
  • Fix plot 3-argument overload
  • Fix stem shader offset when baseline is hidden

Sandbox

Added

  • Add notebook editor infrastructure — markdown cells alongside code cells (not yet user-facing)
  • Add RunMat Agent infrastructure — typed protocol, config, ops, and event model (not yet user-facing)

Fixed

  • Fix artifact loading to return newest entry when an ID is reused
  • Fix browser terminal clear/clc handling
  • Fix runtime state leaking between notebook run sessions

Changed

  • Improve path handling with safe segment escaping for dot components
  • Improve auth flow — email verification now redirects directly to the sandbox with an auth modal
  • Improve Auth0 email login flow
  • Improve stdout deduplication and clear event handling

Cloud

Added

  • Add durable persistence and replay for run artifacts

v0.3.2

March 24, 2026

Deployment and infrastructure improvements for macOS, Windows, and Linux — no runtime or engine changes. Covers v0.3.1 through v0.3.2.

v0.3.0

March 24, 2026

Runtime

Added

  • WebAssembly (WASM) compile target — RunMat now runs entirely in the browser with WebGPU acceleration, published as the runmat npm package
  • Add fused GPU rendering pipeline for 2D and 3D plots with zero-copy surface data path
  • Add 3D depth camera with reversed-Z and dynamic clip planes
  • Add type inference — context-aware shape resolvers track tensor shapes through the compiler
  • Add builtins: int32, uint16, isgpuarray, magic, empty, frewind, rand(m,n,p), uint8 elementwise ops, isequal, logical, cellfun handle support, fullfile, erase, atanh, tempname
  • Add full call stack in error diagnostics with stack depth limit and source location tracking
  • Add MATLAB compatibility mode

Changed

  • Migrate the entire VM, all builtins, and the GPU provider to async
  • Change default error namespace to RunMat (from MATLAB)

Fixed

  • Fix slice indexing shape preservation, conv2 kernel flip, chol multi-output control flow, load auto-assignment, dlmread/csvread header parsing, cummax/cummin alignment, fgets character limit, native auto-init race condition, and GPU compute dispatch stack layout

Sandbox

Added

Fixed

  • Fix save file behavior
  • Fix figure state reset on re-run
  • Fix variable inspector pagination — large vectors now paginate correctly
  • Fix variable inspector caching — stale values no longer shown after re-runs
  • Fix data file viewer error handling

Changed

  • Show struct values before tensor values in the output panel
  • Improve variable inspector layout

Cloud

Added

  • Add organizations, projects, and team memberships

v0.2.8

December 22, 2025

Added

  • Add complex number support — complex arithmetic works throughout the runtime
  • Add non-conjugate transpose (.') — apostrophe handling now distinguishes from conjugate transpose
  • Add Homebrew installation — brew install runmat (see CLI docs)

Changed

  • Replace manual REPL input with rustyline for line editing, command history, and formatting
  • Persist ans variable across expressions
  • Display struct content in the REPL
  • Improve multi-line formatting for ls and matrices
  • Suppress output for assignments ending in ;

Fixed

  • Fix save() stack overflow bug

v0.2.7

December 2, 2025

Added

  • Add anonymous telemetry to help improve RunMat. No code is ever captured — only internal error codes. See Telemetry for details

Fixed

  • Fix CLI version display

v0.2.6

November 21 – 24, 2025

Covers v0.2.0 through v0.2.6.

Added

Accelerate (GPU backend):

  • Add wgpu-based GPU backend — Metal (macOS), DirectX 12 (Windows), Vulkan (Linux)
  • Add cost model — runtime profiling routes work to CPU or GPU based on data size and transfer cost
  • Add f32/f64 compute shaders for a broad set of builtins — ones, zeros, rand, reductions, and many more now dispatch to the GPU automatically

Fusion engine:

  • Add computation graph pattern scanning — the runtime analyzes your code's computation graph, matches sequences against a library of fusible patterns, and replaces them with optimized GPU kernels automatically
  • Add 5–6 initial fusion operations including elementwise math chains, where multiple operations collapse into a single GPU kernel eliminating intermediate memory traffic

v0.0.3

August 25, 2025

Added

  • Add Jupyter kernel support — ZMQ transport so RunMat can serve as a Jupyter kernel
  • Add PowerShell installer for Windows

v0.0.2

August 19, 2025

Added

  • Implement interpretation for the core MATLAB language grammar and semantic surface, across the parser, HIR, and VM. See Language Coverage for the full matrix of implemented coverage

v0.0.1

August 10, 2025

Added

  • Initial release — lexer, architecture plan, project scaffolding
  • Add release workflow with macOS signing, cross-compilation, crates.io publishing