SwiftCU

SwiftCU is a wrapper for CUDA runtime API's (exposed as cxxCU) with extra utilities for device management, memory ops and kernel execution, along with a robust suite of tests. Repo is tested on newest (v12.5) CUDA runtime API on both Linux and Windows.

Operating System	Swift Version	CUDA Version	Supported
Linux	6.0	12.*	✅
Windows 11	6.0	12.*	✅

Installation

To include SwiftCU in your Swift project, add the following line to your Package.swift file:

.package(url: "https://github.com/machineko/SwiftCU", branch: "main")

Documentation

Docc generated for Swift wrapped API SwiftCU

CUDA runtime CUDART

Examples

import SwiftCU

let deviceStatus = CUDADevice(index: 0).setDevice()
if deviceStatus {
    print("Device initialized successfully")
} else {
    print("Failed to initialize device")
}

var pointer: UnsafeMutableRawPointer?
let allocationStatus = pointer.cudaMemoryAllocate(1024)
if allocationStatus.isSuccessful {
    print("Memory allocated successfully")
} else {
    print("Memory allocation failed")
}

let kernel = CUDAKernel(functionPointer: pointerToCUDAKernel)
let launchStatus = kernel.launch(arguments: kernelArgs, blockDim: blockDim, gridDim: gridDim)
if launchStatus.isSuccessful {
    print("Kernel launched successfully")
} else {
    print("Kernel launch failed")
}

For more usage examples check SwiftCUTests

Simple Package overview

import PackageDescription
import Foundation

let packageDir = URL(fileURLWithPath: #file).deletingLastPathComponent().path
#if os(Windows)
    let cuPath = ProcessInfo.processInfo.environment["CUDA_HOME"] ?? "C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.5"
    let cuIncludePath = "-I\(cuPath)\\include"
#elseif os(Linux)
    let cuPath = ProcessInfo.processInfo.environment["CUDA_HOME"] ?? "/usr/local/cuda"
    let cuIncludePath = "-I\(cuPath)/include"
#else
    fatalError("OS not supported \(os)")
#endif

let package = Package(
    name: "SwiftCU-example",
    dependencies: 
    [
        .package(url: "https://github.com/machineko/SwiftCU", branch: "main"),
    ],
    targets: [
        .target(
            name: "cudaKernel",
            publicHeadersPath: "include",
            cxxSettings: [
                .headerSearchPath(cuIncludePath)
            ]
        ),
        .executableTarget(
            name: "SwiftCU-example",
            dependencies: [
                "cudaKernel",
                .product(name: "SwiftCU", package: "SwiftCU"),
            ],
             swiftSettings: [
                .interoperabilityMode(.Cxx),
                .unsafeFlags(
                    ["-Xcc", cuIncludePath]
                )
            ]
        ),
    ]
)

Example build without package/env flags

swift build -Xswiftc -L/usr/local/cuda/lib64 -Xswiftc -I/usr/local/cuda/include

Testing

Current version of SwiftCU is tested with Swift 6.0 development branch using swift-testing package and CUDA v12.5