Foundation.OpenCL 1.0.8

Foundation.OpenCL ⚡

OpenCL binding for DotNet, idiomatic C#, zero cost abstraction.

Performance Density Matters - 100 kernel chain executes in 11ms

Foundation.OpenCL is a high-performance, modern OpenCL binding for .NET that combines the full power of OpenCL 3.0 with idiomatic C# design patterns. Built from the ground up for performance and reliability.

Foundation.OpenCL.Extensions.Intel

🚀 Why Foundation.OpenCL?

Zero-Cost Abstractions

Get type safety and modern C# features without performance penalties:

// Traditional bindings: clSetKernelArg(kernel, 0, sizeof(cl_mem), &buffer)
// Foundation: Type-safe, zero-overhead
kernel.SetArgBuffer(0, buffer);  // Compiles to optimal native calls

Idiomatic C#

Work with spans, generics, and events instead of raw pointers:

// Modern memory patterns
Span<float> data = stackalloc float[1024];
queue.EnqueueWriteBuffer(buffer, 0, data);

Proven Performance

// 100-kernel chain executes in 18ms
var events = new Event[100];
for (int i = 0; i < 100; i++)
{
    events[i] = queue.EnqueueNdRangeKernel(kernel, globalSize, localSize, 
                 i > 0 ? events[i-1] : default);
}

💡 Quick Start

Installation

dotnet add package Foundation.OpenCL

Your First OpenCL Program in 5 Minutes

using Foundation.OpenCL;

// 1. Discover platforms and devices
var platform = Platform.GetPlatforms()[0];
var device = platform.GetDevices(DeviceType.Gpu)[0];

// 2. Create context and command queue
using var context = platform.CreateContext(new[] { device });
using var queue = context.CreateCommandQueue(device);

// 3. Build kernel from source
string kernelSource = """
__kernel void vector_add(__global const float* a, 
                        __global const float* b, 
                        __global float* c) {
    int i = get_global_id(0);
    c[i] = a[i] + b[i];
}
""";

using var program = context.CreateWithSource(kernelSource);
program.Build(new[] { device });

// 4. Create and execute kernel
using var kernel = program.CreateKernel("vector_add");

// 5. Manage memory with type safety
using var aBuffer = context.CreateBuffer(MemFlags.ReadOnly, 1024 * sizeof(float));
using var bBuffer = context.CreateBuffer(MemFlags.ReadOnly, 1024 * sizeof(float)); 
using var cBuffer = context.CreateBuffer(MemFlags.WriteOnly, 1024 * sizeof(float));

// 6. Execute with proper event synchronization
var kernelEvent = queue.EnqueueNdRangeKernel(kernel, 
    globalSize: new[] { (nuint)1024 },
    localSize: new[] { (nuint)64 });

kernelEvent.Wait(); // Blocks until completion

🎯 Advanced Features

Intel USM Extensions

using Foundation.OpenCL.Extensions.Intel;

// Unified Shared Memory allocation
void* deviceMemory = context.AllocateDeviceMemory(device, 1024 * sizeof(float));

// Set kernel arguments directly from USM pointers
kernel.SetArgMemPointer(0, deviceMemory);

Multi-GPU Ready

var devices = Platform.GetPlatforms().SelectMany(p => p.GetDevices()).ToArray();
var contexts = devices.Select(d => Context.CreateContext(new[] { d })).ToArray();

// Distribute work across all available GPUs
Parallel.For(0, devices.Length, i =>
{
    var queue = contexts[i].CreateCommandQueue(devices[i]);
    queue.EnqueueNdRangeKernel(kernel, globalSize, localSize);
});

Sophisticated Event System

// Multicast event callbacks
var completionEvent = queue.EnqueueNdRangeKernel(kernel, globalSize, localSize);

completionEvent.OnComplete += () => Console.WriteLine("Kernel completed!");
completionEvent.OnRunning += () => Console.WriteLine("Kernel started execution!");

// Complex dependency chains
var eventA = queue.EnqueueKernel(kernelA, size);
var eventB = queue.EnqueueKernel(kernelB, size, waitFor: eventA); 
var eventC = queue.EnqueueKernel(kernelC, size, waitFor: new[] { eventA, eventB });

🏗️ Architecture Benefits

Type-Safe Handles

// Compile-time safety: no mixing of handle types
Handle<Context> contextHandle = context.Handle;
Handle<Device> deviceHandle = device.Handle;

// This won't compile - caught at build time!
// contextHandle = deviceHandle; // Error!

Resource Management

// Automatic reference counting and disposal
using var context = Context.CreateContext(devices);
using var queue = context.CreateCommandQueue(device);
using var buffer = context.CreateBuffer(MemFlags.ReadWrite, size);

// Proper cleanup even with complex object graphs

Performance Optimizations

• Stack allocation for temporary buffers
• Aggressive inlining of hot paths
• Span-based APIs for zero-copy operations
• Generic specialization for common types

📊 Performance Showcase

Kernel Chaining Benchmark

100 sequential 32x32 gemm kernels with event dependencies Typical result: 18ms on Intel Arc Pro B60

🛠 Installation

Package Manager

Install-Package Foundation.OpenCL
Install-Package Foundation.OpenCL.Extensions.Intel

.NET CLI

dotnet add package Foundation.OpenCL
dotnet add package Foundation.OpenCL.Extensions.Intel

Project Reference

<PackageReference Include="Foundation.OpenCL" Version="1.0.2" />
<PackageReference Include="Foundation.OpenCL.Extensions.Intel" Version="1.0.2" />

🔧 Requirements

• .NET 9.0 or later
• OpenCL 1.2+ compatible drivers, full support for OpenCL 3.0
• Windows/Linux/macOS with GPU support

🤝 Collaboration

This project was developed through human-AI collaboration:

• Project Lead & Architect: Samuel Alexandre Vidal
• AI Collaborators:
  • DeepSeek (implementation, code review, technical expertise)
  • Qwen3-Next-80B-A3B (help with native API mapping)
  • Gemini 2.5 (initial concept collaboration)

The entire development process featured continuous collaboration between human expertise and multiple AI systems, each contributing unique strengths to create a production-quality library.

📄 License

MIT License - see LICENSE.txt for details.

🚀 Getting Involved

• Found a bug? Open an issue
• Have a feature request? Start a discussion
• Want to contribute? Check our contribution guidelines

Ready to accelerate your GPU computing using idiomatic C#? Get started with Foundation.OpenCL today! ⚡