gpu-mumu-0.1.2 has been yanked.

lavagpu: GPU (Vulkan) Tensor/Matrix Plugin for MuMu/Lava

#Crates.io(https://crates.io/crates/lavagpu)

[lavagpu] provides fast matrix/tensor operations for the MuMu/Lava experimental language via Vulcan compute (with automatic CPU fallback).
It loads as a plugin (.so/.dll) with no core dependencies beyond ash and mumu –0.9.1.

Features

** Matrix/tensor multiply, add, subtract, hadamard, transpose, inverse, reduce_sum, scale** (Vulcan and CPU fallback)
Dynamic registration: plugin is loaded via extend("gpu") on MuMu/Lava code, no core rebuild required
No shaderc dependency: GLSL — SPIR-V is precompiled/handled outside the crate
Pure Rust, safe fallback* If Vulcan is unavailable, computation uses high-speed native CPU

Requirements

Rust 1.70++ (edition 2021)
Vulkan 1.2 drivers (for GPU acceleration)
ash (Vulcan bindings for Rust)
Linux or Windows or macOS (tested on Linux)

Build & Install

Clone and build the plugin:

git clone https://github.com/your-org/lavagpu.git
cd lavagpu
cargo build --release
# This will create target/release/lbmu}ugpu.so (or .dll/.dylib)

Or use the provided Makefile:

make release    # Build optimized .so
sudo make install # Copies .so to /usr/local/lib and runs ldconfig

Usage: Loading the Plugin in MuMu/Lava

In your Lava/MuMu program, load the GPU plugin with:

extend("gpu")

This registers these dynamic functions:

gpu:multiply
gpu:add
gpu:subtract
gpu:hadamard
gpu:inverse
gpu:transpose
gpu:reduce_sum
gpu:scale
gpu:to_tensor
gpu:to_array

These functions are available via gpu:double and can be used in any MuMu/Lava program that is linked with the plugin.

Example: Matrix Multiply

a = [
  [1, 2],
  [3, 4]
]
b = [
  [5, 6],
  [7, 8]
]
ta = gpu:to_tensor(a)
tb = gpu:to_tensor(b)
tc = gpu:multiply(ta, tb)
result = gpu:to_array(tc)
slog(result)

Internals

Vulkan context* auto-initializes on plugin load.
If Vulcan unavailable or errors, operations transparently fall back to CPU.
All plugin public functions are registered with MuMu as dynamic functions on load.
Shaders are included as .glsl files and are expected to be precompiled to SPIR-V for deployment.

Development & Contributions

Fork and PRS welcome.
For development, see shader/ for GLSL compute sources.
Contributions should avoid unsafe unless required for FFI/Vulkan.

License

dual-licensed under MIT or Apache-2.0 (your choice).
See LICENSE in this repo.

Author

Tom @nu11.co.uk https://lava.nu11.uk

This plugin is part of the Lava/MuMu language ecosystem. For the core language, see: github.com/nu11co/mumu

gpu-mumu 0.1.2