3.7 OpenCL C++ Wapper向量加法

Khronos组织也在后续的OpenCL标准中定义了一套C++ Wapper API。C++ API与C API是对应的(比如，cl::Memory对应cl_mem)，不过C++需要对异常和类进行处理。下面的代码对应的与代码清单3.4中的C代码相对应。

  1. #define __CL_ENABLE_EXCEPTIONS
  3. #include <CL/cl.hpp>
  4. #include <iostream>
  5. #include <fstream>
  6. #include <string>
  7. #include <vector>
  9. int main(){
  10.   const int elements = 2048;
  11.   size_t datasize = sizeof(int) * elements;
  13.   int *A = new int[elements];
  14.   int *B = new int[elements];
  15.   int *C = new int[elements];
  17.   for (int i = 0; i < elements; i++){
  18.     A[i] = i;
  19.     B[i] = i;
  20.   }
  22.   try{
  23.     // Query for platforms
  24.     std::vector<cl::Platform> platforms;
  25.     cl::Platform::get(&platforms);
  27.     // Get a list of devices on this platform
  28.     std::vector<cl::Device> devices;
  30.     platforms[0].getDevices(CL_DEVICE_TYPE_ALL, &devices);
  32.     // Create a context for the devices
  33.     cl::Context context(devices);
  35.     // Create a command-queue for the first device
  36.     cl::CommandQueue queue = cl::CommandQueue(context, devices[0]);
  38.     // Create the memory buffers
  39.     cl::Buffer bufferA = cl::Buffer(context, CL_MEM_READ_ONLY, datasize);
  40.     cl::Buffer bufferB = cl::Buffer(context, CL_MEM_READ_ONLY, datasize);
  41.     cl::Buffer bufferC = cl::Buffer(context, CL_MEM_WRITE_ONLY, datasize);
  43.     // Copy the input data to the input buffers using the
  44.     // command-queue for the first device
  45.     queue.enqueueWriteBuffer(bufferA, CL_TRUE, 0, datasize, A);
  46.     queue.enqueueWriteBuffer(bufferB, CL_TRUE, 0, datasize, B);
  48.     // Read the program source
  49.     std::ifstream sourceFile("vector_add_kernel.cl");
  50.     std::string sourceCode(std::istreambuf_iterator<char>(sourceFile), (std::istreambuf_iterator<char>()));
  51.     cl::Program::Source source(1, std::make_pair(sourceCode.c_str(), sourceCode.length() + 1);
  53.     // Create the program from the source code
  54.     cl::Program program = cl::Program(context, source);
  56.     // Build the program for the devices
  57.     program.build(devices);
  59.     // Create the kernel
  60.     cl::Kernel vecadd_kernel(program, "vecadd");
  62.     // Set the kernel arguments
  63.     vecadd_kernel.setArg(0, bufferA);
  64.     vecadd_kernel.setArg(1, bufferB);
  65.     vecadd_kernel.setArg(2, bufferC);
  67.     // Execute the kernel
  68.     cl::NDRange gloabl(elements);
  69.     cl::NDRange local(256);
  71.     queue.enqueueNDRangeKernel(vecadd_kernel, cl::NullRange, gloabl, local);
  73.     // Copy the output data back to the host
  74.     queue.enqueueReadBuffer(bufferC, CL_TRUE, 0, datasize, C);
  75.   } catch(cl::Error error){
  76.     std::cout << error.what() << "(" << error.err() << ")" << std::endl;
  77.   }
  78. }

代码清单3.5 使用C++ Wapper实现的OpenCL向量相加