minigpu
1.9.1, released 5 weeks ago by practicalxr.com
Multiplatform GPU compute for Dart and Flutter using WebGPU and Dawn.
WASM-ready
What it does
Brings cross-platform GPU compute shaders to Dart and Flutter applications. It integrates WGSL and WebGPU directly utilizing a Dawn backend across mobile, desktop, and web.
AI-written summary of version 1.9.1, based on the package's README and metadata. Report an inaccuracy
Use it when
- You need to run high-performance compute shaders or AI-ML math models on the GPU across multiple platforms.
- You want to process video frames or perform direct GPU buffer copies without CPU round-trips.
Skip it when
- You are building a standard UI app that doesn't require low-level GPU acceleration.
- You cannot handle complex native build setups involving CMake and Dawn.
Strengths
- Cross-platform support including Windows, Linux, macOS, Web, Android, and iOS.
- Built-in persistent shader caching to mitigate compilation overhead.
- Direct video frame import and interop for zero-copy GPU processing.
Weaknesses
- Dawn backend can take a long time to build and requires a recent version of CMake and Emscripten for web.
- Experimental library with potential build friction on native platforms.
Setup
- Kind
- Flutter widgets
- Effort
- High
- Code generation
- No
- Direct dependencies
- 3
Requires CMake 3.22+ and Emscripten for web. Dawn is cloned and compiled automatically on first build, which may take time. Android requires a manually supplied shader cache directory.
final gpu = Minigpu();
await gpu.init();
final shader = gpu.createComputeShader();
shader.loadKernelString('''
@group(0) @binding(0) var<storage, read_write> inp: array<f32>;
@compute @workgroup_size(256)
fn main(@builtin(global_invocation_id) id: vec3<u32>) {}
''');Alternatives
Packages that do the same job and could replace minigpu.
- flutter_edge_ai_litertlm
LiteRT-LM on-device engine for flutter_edge_ai supporting .litertlm models and embeddings.
87Compare