Getting started with opencl

Download opencl eBook

Remarks

This section provides an overview of what opencl is, and why a developer might want to use it.

It should also mention any large subjects within opencl, and link out to the related topics. Since the Documentation for opencl is new, you may need to create initial versions of those related topics.


Opencl is an api that puts gpus,cpus and some other accelerators(like a pcie-fpga) into good use of C99-like computations but with a very wide concurrency advantage. Once installation and basic implementation is done, only simple changes in a kernel string(or its file) applies an algorithm to N hardware threads automagically.

A developer might want to use it because it will be much easier to optimize for memory space or speed than doing same thing on opengl or direct-x. Also it is royalty-free. Concurrency within a device is implicit so no need for explicit multi-threading for each device. But for multi-device configurations, a cpu-multi-threading is still needed. For example, when a 1000 threaded job is sent to a cpu, thread synchronization is handled by driver. You just tell it how big a workgroup should be(such as 256 each connected with virtual local memory) and where synchronization points are(only when needed).

Using gpu for general purpose operations is nearly always faster than cpu. You can sort things quicker, multiply matrices 10x faster and left join in-memory sql tables in "no" time. Any 200$ desktop-grade gpu will finish quicker in a physics(finite-element-method fluid) workload than any 200$ cpu. Opencl makes it easier and portable. When you're done working in C#, you can easily move to java-opencl implementation using same kernels and C++ project(ofcourse using JNI with extra C++ compiling).

For the graphics part, you are not always have to send buffers between cpu and gpu. You can work purely on gpu using "interop" option in context creation part. With interop, you can prepare geometries at the limit performance of gpu. No pci-e required for any vertex data. Just a "command" is sent through, and work is done only inside of graphics card. This means no cpu-overhead for data. Opencl prepares geometry data, opengl renders it. CPU becomes released. For example, if a single thread of cpu can build a 32x32 verticed sphere in 10000 cycles, then a gpu with opencl can build 20 spheres in 1000 cycles.


C# implementation of OpenCL 1.2: number of platforms for an AMD system in 64-bit windows

OpenCL is low level api so it must be implemented in "C space" first. For that, one needs to download header files from Khronos' site. My hardware is AMD and capable of version 1.2, downloading

opencl.h 
cl_platform.h 
cl.h 
cl_ext.h 
cl_egl.h 
cl_dx9_media_sharing.h 
cl_d3d10.h 
cl_d3d11.h 
cl_gl.h 
cl_gl_ext.h 
cl.hpp
 

from this page

should be enough for C++ bindings so after adding these files to your project and setting proper binary(and library) file locations(

$(AMDAPPSDKROOT)\lib\x86_64 for 64-bit amd library (amd app sdk's libraries are preferred)

,

C:\Windows\SysWOW64 for 64-bit opencl.dll (.so file if ICD is of a Linux system)

for example but different for Intel-Nvidia), you can start querying a list of platforms(amd,intel,xilinx,nvidia) after installing proper drivers(such as crimson for amd). Drivers are for running opencl application(using ICD), libraries and header files are for development to be in short.

To query platforms:

#define __CL_ENABLE_EXCEPTIONS
#include "stdafx.h"
#include <vector>
#include <CL/cl.hpp>

extern "C"
    {
       // when this class is created, it contains a list of platforms in "platforms" field.
       class OpenClPlatformList
       {
           public:
               std::vector<cl::Platform> platforms;
               int platformNum;
               OpenClPlatformList()
               {
                   platforms= std::vector< cl::Platform>();
                   cl::Platform::get(&platforms);
                   platformNum= platforms.size();
               }
        };


        // this is seen from C# when imported. Creates an object in memory.
        __declspec(dllexport)
            OpenClPlatformList * createPlatformList()
        {
            return new OpenClPlatformList();
        }

        __declspec(dllexport)
            int platformNumber(OpenClPlatformList * hList)
        {
            return hList->platformNum;
        }


        __declspec(dllexport)
            void deletePlatformList(OpenClPlatformList * p)
        {
            if (p != NULL)
                delete p;
            p = NULL;
        }


    }
 

could be built into a dll(such as OCLImplementation.dll)

and to use it from C# side,

using System;
using System.Collections.Generic;
using System.Runtime.InteropServices;


namespace WrapperCSharp
{
    public class WrapperCSharp
    {
        [DllImport("OCLImplementation", CallingConvention = CallingConvention.Cdecl)]
        private static extern IntPtr createPlatformList();

        [DllImport("OCLImplementation", CallingConvention = CallingConvention.Cdecl)]
        private static extern int platformNumber(IntPtr hList);

        [DllImport("OCLImplementation", CallingConvention = CallingConvention.Cdecl)]
        private static extern void deletePlatformList(IntPtr hList);
    }
}
 

ofcourse the dll must be seen by the C# project, simply putting it near executable of project solves it.

Now, if sample computer has at least one opencl-capable platform,

IntPtr platformList = createPlatformList(); // just an address in C-space
int totalPlatforms = platformNumber(platformList); // AMD+NVIDIA systems should have "2"
deletePlatformList(platformList); //
 

totalPlatforms variable must have at least "1" value. Then you can use platforms variable in C-space using additional functions to iterate through all platforms to query all devices such as CPU,GPU and special purpose accelerators such as phi or some fpga.

One does not simply write all these C++ to C# wrappers for time-critical projects. There are many wrappers written for C#, Java and other languages. For java, there is "Aparapi" that is the "java bytecode to opencl-c" converter api that takes what you write purely in java to a gpu-parallel version on the fly so it is somewhat portable.

OpenCL and C#

For C# there exist many wrappers that offer an interface to communicate with OpenCL.

  • OpenCL.NET: This is one of the most low level wrappers out there. It offers a complete implementation of the OpenCL API for C# without adding any abstraction at all. So C\C++ examples are easily ported for this library. The only project page is currently on codeplex, which shuts down on 15.12.2017 but the package is available on NuGet

https://openclnet.codeplex.com/

  • NOpenCL: This library offers an abstract interface between C# and OpenCL.

The short-term goal is providing an easy-to-use abstract layer which provides access to the full capability of OpenCL without sacrificing performance.

https://github.com/tunnelvisionlabs/NOpenCL

  • Cloo:

Cloo is an open source, easy to use, managed library which enables .NET/Mono applications to take full advantage of the OpenCL framework.

https://sourceforge.net/projects/cloo/

Prerequisites

If you have a modern CPU or graphics card (GPU) inside your machine, chances are you have everything ready for first steps in OpenCL. Finding out if your processor is OpenCL capable can be usually done via the manufacturer's homepage, a good first start is the official documentation at

https://www.khronos.org/conformance/adopters/conformant-products#opencl

What is OpenCL?

Open Computing Language (OpenCL) is a framework for writing programs that execute on CPUs, GPUs, and other parallel processors and accelerators.

OpenCL specifies a programming language (based on C) that provides access to named on-chip memory, a model for executing tasks in parallel, and the ability to synchronize those tasks.

What is OpenCL?

OpenCL is short for Open Computing Language. OpenCL is a Framework for parallel programming across heterogeneous platforms, called compute devices, ranging from CPUs over GPUs to more special platforms like FPGAs. OpenCL provides a standard interface for parallel computing on these compute devices but also inter-device parallelism. It specifies a programming language, based on C99, and minimum requirements of basic functions implemented on OpenCL capable devices. OpenCL furthermore describes an abstract computing and memory model, being as general as possible to make the reuse of code between different platforms straightforward.

Stats

117 Contributors: 7
Thursday, June 8, 2017
Licensed under: CC-BY-SA

Not affiliated with Stack Overflow
Rip Tutorial: info@zzzprojects.com

Download eBook