OpenCl入门getting-started-with-opencl-and-gpu-computing
原文来自于：getting-started-with-opencl-and-gpu-computing/
对整个程序的注释：http://www.kimicat.com/opencl-1/opencl-jiao-xue-yi
但是对CUDA比较熟悉的用户来说，应该不需要看注释就能理解全部的程序
main.cpp
1 #include <stdio.h> 2 #include <stdlib.h> 3 #include <CL/cl.h> 4 #define MAX_SOURCE_SIZE (0x100000) 5 int main(void) 6 { 7 // Create the two input vectors 8 int i; 9 const int LIST_SIZE = 1000; 10 int *A = (int*) malloc(sizeof(int) * LIST_SIZE); 11 int *B = (int*) malloc(sizeof(int) * LIST_SIZE); 12 for (i = 0; i < LIST_SIZE; i++) 13 { 14 A[i] = i; 15 B[i] = LIST_SIZE - i; 16 } 17 18 // Load the kernel source code into the array source_str 19 FILE *fp; 20 char *source_str; 21 size_t source_size; 22 fp = fopen("vector_add_kernel.cl", "r"); 23 24 if (!fp) 25 { 26 fprintf(stderr, "Failed to load kernel.\n"); 27 exit(1); 28 } 29 source_str = (char*) malloc(MAX_SOURCE_SIZE); 30 source_size = fread(source_str, 1, MAX_SOURCE_SIZE, fp); 31 fclose(fp); 32 33 // Get platform and device information 34 cl_platform_id platform_id = NULL; 35 cl_device_id device_id = NULL; 36 cl_uint ret_num_devices; 37 cl_uint ret_num_platforms; 38 cl_int ret = clGetPlatformIDs(1, &platform_id, &ret_num_platforms); 39 ret = clGetDeviceIDs(platform_id, CL_DEVICE_TYPE_DEFAULT, 1, &device_id,&ret_num_devices); 40 41 // Create an OpenCL context 42 cl_context context = clCreateContext(NULL, 1, &device_id, NULL, NULL, &ret); 43 // Create a command queue 44 cl_command_queue command_queue = clCreateCommandQueue(context, device_id, 0, &ret); 45 46 // Create memory buffers on the device for each vector 47 cl_mem a_mem_obj = clCreateBuffer(context, CL_MEM_READ_ONLY,LIST_SIZE * sizeof(int), NULL, &ret); 48 cl_mem b_mem_obj = clCreateBuffer(context, CL_MEM_READ_ONLY,LIST_SIZE * sizeof(int), NULL, &ret); 49 cl_mem c_mem_obj = clCreateBuffer(context, CL_MEM_WRITE_ONLY,LIST_SIZE * sizeof(int), NULL, &ret); 50 51 // Copy the lists A and B to their respective memory buffers 52 ret = clEnqueueWriteBuffer(command_queue, a_mem_obj, CL_TRUE, 0,LIST_SIZE * sizeof(int), A, 0, NULL, NULL); 53 ret = clEnqueueWriteBuffer(command_queue, b_mem_obj, CL_TRUE, 0,LIST_SIZE * sizeof(int), B, 0, NULL, NULL); 54 55 // Create a program from the kernel source 56 cl_program program = clCreateProgramWithSource(context, 1,(const char **) &source_str, (const size_t *) &source_size, &ret); 57 58 // Build the program 59 ret = clBuildProgram(program, 1, &device_id, NULL, NULL, NULL); 60 // Create the OpenCL kernel 61 cl_kernel kernel = clCreateKernel(program, "vector_add", &ret); 62 // Set the arguments of the kernel 63 ret = clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &a_mem_obj); 64 ret = clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &b_mem_obj); 65 ret = clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &c_mem_obj); 66 67 // Execute the OpenCL kernel on the list 68 size_t global_item_size = LIST_SIZE; // Process the entire lists 69 size_t local_item_size = 1; // Process one item at a time 70 ret = clEnqueueNDRangeKernel(command_queue, kernel, 1, NULL,&global_item_size, &local_item_size, 0, NULL, NULL); 71 72 // Read the memory buffer C on the device to the local variable C 73 int *C = (int*) malloc(sizeof(int) * LIST_SIZE); 74 ret = clEnqueueReadBuffer(command_queue, c_mem_obj, CL_TRUE, 0,LIST_SIZE * sizeof(int), C, 0, NULL, NULL); 75 76 // Display the result to the screen 77 for (i = 0; i < LIST_SIZE; i++) 78 printf("%d + %d = %d\n", A[i], B[i], C[i]); 79 80 // Clean up 81 ret = clFlush(command_queue); 82 ret = clFinish(command_queue); 83 ret = clReleaseKernel(kernel); 84 ret = clReleaseProgram(program); 85 ret = clReleaseMemObject(a_mem_obj); 86 ret = clReleaseMemObject(b_mem_obj); 87 ret = clReleaseMemObject(c_mem_obj); 88 ret = clReleaseCommandQueue(command_queue); 89 ret = clReleaseContext(context); 90 91 free(A); 92 free(B); 93 free(C); 94 95 return 0; 96 97 }
vector_add_kernel.cl
__kernel void vector_add(__global const int *A, __global const int *B, __global int *C) { // Get the index of the current element to be processed int i = get_global_id(0); // Do the operation C[i] = A[i] + B[i]; }
之前已经安装好了CUDA的运行环境，这里作者说使用g++ -I/usr/local/cuda/include -L/usr/local/cuda/lib64 -lOpenCL main.cpp -o openclApp命令来执行，结果提示
'clGetPlatformIDs' undefined reference，但是我的include和lib都是正常的，因此，调整编译命令为：
g++ main.cpp -o openclApp -I/usr/local/cuda/include -L/usr/local/cuda/lib64 -lOpenCL
编译通过并运行通过，因此gcc编译选项的顺序也对程序有一定影响（理论上不应该有这个问题）。但是，这个问题使用clang编译就没有任何影响。
