Implementation:Alibaba MNN OpenCL CL HPP Header

Knowledge Sources	Alibaba_MNN Khronos OpenCL
Domains	GPU_Computing, OpenCL
Last Updated	2026-02-10 12:00 GMT

Overview

Vendored Khronos OpenCL C++ bindings header (deprecated original cl.hpp version) providing RAII wrapper classes around OpenCL C objects for MNN's GPU backend.

Description

cl.hpp is the original OpenCL C++ bindings header that wraps the low-level C API declared in cl.h with C++ classes offering automatic reference counting, exception handling, and object-oriented interfaces. It defines classes such as cl::Context, cl::CommandQueue, cl::Buffer, cl::Kernel, cl::Program, cl::Device, cl::Platform, cl::Image2D, and cl::Event. MNN's OpenCL backend uses these C++ wrappers to manage GPU resources with deterministic cleanup, simplifying context lifecycle management and reducing boilerplate around OpenCL API calls. This header is the deprecated predecessor to cl2.hpp but remains vendored for backward compatibility.

Usage

This header is included internally by MNN's OpenCL backend. It is not directly imported by end users.

Code Reference

Source Location

• Repository: Alibaba_MNN
• File: 3rd_party/OpenCLHeaders/CL/cl.hpp
• Lines: 1-12952

Signature

namespace cl {

class Device : public detail::Wrapper<cl_device_id> {
public:
    Device() {}
    Device(cl_device_id device) { ... }
    cl_int getInfo(cl_device_info name, ...) const;
};

class Context : public detail::Wrapper<cl_context> {
public:
    Context(const vector<Device>& devices, ...);
    Context(cl_device_type type, ...);
};

class CommandQueue : public detail::Wrapper<cl_command_queue> {
public:
    CommandQueue(const Context& context, const Device& device, ...);
    cl_int enqueueNDRangeKernel(const Kernel& kernel, ...);
};

class Buffer : public Memory {
public:
    Buffer(const Context& context, cl_mem_flags flags, size_t size, ...);
};

class Kernel : public detail::Wrapper<cl_kernel> {
public:
    Kernel(const Program& program, const char* name, cl_int* err = NULL);
    template<typename T> cl_int setArg(cl_uint index, T value);
};

} // namespace cl

Import

#include "3rd_party/OpenCLHeaders/CL/cl.hpp"

I/O Contract

Inputs

Name	Type	Required	Description
N/A	N/A	N/A	Header-only; no runtime inputs

Outputs

Name	Type	Description
API declarations	C++ classes	RAII wrapper classes for OpenCL objects (Context, CommandQueue, Buffer, Kernel, Program, Device, Platform, Event, Image, Sampler)

Usage Examples

// MNN internal usage (not user-facing)
#include "CL/cl.hpp"

// Get platforms and devices
std::vector<cl::Platform> platforms;
cl::Platform::get(&platforms);

std::vector<cl::Device> devices;
platforms[0].getDevices(CL_DEVICE_TYPE_GPU, &devices);

// Create context and queue with RAII
cl::Context context(devices);
cl::CommandQueue queue(context, devices[0]);

// Create buffer and kernel
cl::Buffer buffer(context, CL_MEM_READ_WRITE, 1024);
cl::Program program(context, sourceCode);
program.build(devices);
cl::Kernel kernel(program, "my_kernel");
kernel.setArg(0, buffer);

// Execute
queue.enqueueNDRangeKernel(kernel, cl::NullRange, cl::NDRange(256), cl::NDRange(64));
queue.finish();

Related Pages

• Environment:Alibaba_MNN_GPU_OpenCL_Environment
• Alibaba_MNN_OpenCL_CL_Header - Underlying C API that this header wraps
• Alibaba_MNN_OpenCL_CL2_HPP_Header - Modern replacement C++ bindings (cl2.hpp)
• Alibaba_MNN_OpenCL_CL_Platform_Header - Platform type definitions