Crate borsalino 

Borsalino — Thin GPU Compute Abstraction

One trait, two backends, zero ceremony.

Borsalino provides a minimal synchronous GPU compute interface for the Industrial Algebra ecosystem. Write WGSL compute kernels, dispatch them on Metal or Vulkan hardware, read the results back — no bind groups, no pipeline layouts, no descriptor sets.

Design

• Synchronous: dispatch() blocks until the GPU finishes. No async runtime, no callback hell, no completion handlers.
• WGSL-first: Kernels are authored in WGSL (WebGPU Shading Language). The Metal backend translates to MSL via naga; the Vulkan backend translates to SPIR-V via naga.
• Minimal surface area: Four operations — create buffers, compile shaders, dispatch, read results. That’s it.
• Zero-cost abstraction: No allocation, no validation, no safety checks beyond what the GPU driver provides.

Quick Start

use borsalino::GpuBackend;

// WGSL compute kernel
let wgsl = r#"
    @group(0) @binding(0) var<storage, read> input: array<f32>;
    @group(0) @binding(1) var<storage, read_write> output: array<f32>;

    @compute @workgroup_size(256)
    fn add_one(@builtin(global_invocation_id) gid: vec3<u32>) {
        let i = gid.x;
        output[i] = input[i] + 1.0;
    }
"#;

let mut gpu = borsalino::init()?;
let pipeline = gpu.compile("add_one", wgsl)?;
let input = gpu.create_buffer(&[1.0f32, 2.0, 3.0, 4.0])?;
let output = gpu.create_buffer_uninit::<f32>(4)?;
gpu.dispatch(&pipeline, &[&input, &output], (1, 1, 1))?;
let result = gpu.read_buffer(&output)?;
assert_eq!(result, vec![2.0, 3.0, 4.0, 5.0]);

Backends

Feature	Platform	Status
metal	macOS	✅ Active
vulkan	Linux, Windows	✅ Active

The Metal backend requires no external dependencies beyond naga — it calls the Metal framework directly via objc_msgSend. The Vulkan backend uses ash for raw Vulkan FFI.

Safety

The GPU driver layer uses unsafe for FFI. The public API is safe — buffer bounds are checked, shader compilation errors are surfaced as GpuError, and dispatch parameters are validated before hitting the driver.

Structs

ComputePipeline

Handle to a compiled compute pipeline.

DispatchSpec

Spec for a single dispatch within GpuBackend::dispatch_many.

GpuBuffer

Handle to a GPU buffer.

NoBackendStub

Stub backend — no GPU backend compiled for this target.

Pulse

Handle to an in-flight asynchronous dispatch.

Enums

GpuError

Errors that can occur in GPU operations.

MemoryStrategy

Memory allocation strategy for GPU buffers.

Traits

GpuBackend

Backend-agnostic GPU compute interface.

Functions

init

Stub backend — returned by init() when no GPU backend is available. Exists solely to give Result<impl GpuBackend> a concrete type in the fallback compilation path.

Type Aliases

Result

Result type alias for Borsalino operations.