memkit-gpu

Backend-agnostic GPU memory management for the memkit ecosystem.

Version: 0.1.0-alpha.1

Overview

memkit-gpu provides a unified GPU memory management API that works across different graphics backends. Write your code once, run it on Vulkan, Metal, or DirectX 12.

Features

• Backend-Agnostic — Same API works across all GPU backends
• Staging Buffers — Efficient CPU → GPU data transfers
• Device Buffers — Fast device-local memory
• Batch Transfers — Coalesce multiple transfers for efficiency
• Memory Pools — Reuse GPU buffers to reduce allocation overhead

Quick Start

use memkit_gpu::{MkGpu, MkBufferUsage, DummyBackend};

// Create GPU allocator (using dummy backend for testing)
let gpu = MkGpu::new(DummyBackend::new());

// Create staging buffer and upload data
let vertices: &[f32] = &[0.0, 1.0, 0.0, 1.0, 0.0, 0.0];
let staging = gpu.staging_buffer_with_data(vertices).unwrap();

// Create device-local buffer
let device = gpu.create_device_buffer(
    std::mem::size_of_val(vertices),
    MkBufferUsage::VERTEX | MkBufferUsage::TRANSFER_DST,
).unwrap();

// Transfer data
gpu.transfer(&staging, &device).unwrap();

Backend-Agnostic Code

use memkit_gpu::{MkGpu, MkGpuBackend, MkBufferUsage};

// Generic function works with ANY backend
fn upload_mesh<B: MkGpuBackend>(
    gpu: &MkGpu<B>,
    vertices: &[Vertex],
) -> Result<MkDeviceBuffer<B>, B::Error> {
    let staging = gpu.staging_buffer_with_data(vertices)?;
    let device = gpu.create_device_buffer(
        std::mem::size_of_val(vertices),
        MkBufferUsage::VERTEX | MkBufferUsage::TRANSFER_DST,
    )?;
    gpu.transfer(&staging, &device)?;
    Ok(device)
}

Backends

Dummy Backend

The dummy backend simulates GPU operations in CPU memory. Useful for:

• Unit testing without a GPU
• Running on systems without GPU support
• Prototyping before implementing a real backend

use memkit_gpu::{DummyBackend, DummyBackendConfig};

// Default config
let backend = DummyBackend::new();

// Custom config
let config = DummyBackendConfig {
    max_buffer_size: 64 * 1024 * 1024, // 64 MB
    simulate_device_local: true,       // Device-local buffers not mappable
    transfer_delay_us: 0,              // Simulate transfer latency
};
let backend = DummyBackend::with_config(config);

Vulkan Backend

Requires the vulkan feature flag. Uses ash and gpu-allocator.

[dependencies]
memkit-gpu = { version = "0.1.0-alpha.1", features = ["vulkan"] }

use memkit_gpu::{MkGpu, VulkanBackend, VulkanConfig};

let config = VulkanConfig {
    app_name: "My App".to_string(),
    validation: cfg!(debug_assertions),
    ..Default::default()
};

let backend = VulkanBackend::new(config)?;
let gpu = MkGpu::new(backend);

Types

License

This project is licensed under the Mozilla Public License 2.0 - see the LICENSE.md file for details.