openCL程序分为两个部份，一部份是内核代码，负责具体算法。另一部份是主程序负责初始化OpenCL和准备数据。主程序加载内核代码，并按照既定方法进行运算。
下面通过一个向量相加的程序来了解OpenCL . 有A，B两个四维向量，相加后值存在C向量里，openCL根据用户提供的维数，将向量分解成多个任务分发给DSP进行并行计算
openCL程序分为两个部份，一部份是内核代码，负责具体算法。另一部份是主程序负责初始化OpenCL和准备数据。主程序加载内核代码，并按照既定方法进行运算
kernel代码如下
__kernel voidvectoradd(__global int *a, __global int *b, __global int *c)
{
int id = get_global_id(0);
c[id] = a[id] + b[id];
}
__kernel 指明这是一个openCL内核，__global 说明指针指向的是全局的设备内存空间
HOST端代码如下
int get_ocl_string(const char *file_name, char *ocl_string)
{
FILE *fp;
int file_length;
int status = 0;
fp = fopen(file_name, "r");
if (fp == NULL)
return -1;
fseek(fp, 0, SEEK_END);
file_length = ftell(fp);
fseek(fp, 0, SEEK_SET);
status = fread(ocl_string, 1, file_length,fp);
if (status == -1)
return -1;
return file_length;
}
int main(void)
{
int array_a[10] = {0, 1, 2, 3, 4, 5, 6, 7,8, 9};
int array_b[10] = {9, 8, 7, 6, 5, 4, 3, 2,1, 0};
int array_c[10] = {0, 0, 0, 0, 0, 0, 0, 0,0, 0};
size_t datasize = 10 * sizeof(int);
size_t ocl_string_size;
char *ocl_string;
ocl_string = (char *)malloc(1024*1024);
cl_platform_id platform_id;
cl_device_id device_id;
cl_context context;
cl_command_queue command_queue;
cl_mem buffer_a, buffer_b, buffer_c;
cl_program program;
cl_kernel kernel;
clGetPlatformIDs(1, &platform_id,NULL);
clGetDeviceIDs(platform_id,CL_DEVICE_TYPE_ACCELERATOR, 1, &device_id, NULL);
//创建上下文
context = clCreateContext(NULL, 1,&device_id, NULL, NULL, NULL);
command_queue = clCreateCommandQueue(context,device_id, 0, NULL);
//分配内存
buffer_a = clCreateBuffer(context,CL_MEM_READ_ONLY, datasize, NULL, NULL);
buffer_b = clCreateBuffer(context,CL_MEM_READ_ONLY, datasize, NULL, NULL);
buffer_c = clCreateBuffer(context,CL_MEM_READ_ONLY, datasize, NULL, NULL);
//读取核函数，并且上传到DSP端
ocl_string_size =get_ocl_string("vectoradd.cl", ocl_string);
clEnqueueWriteBuffer(command_queue,buffer_a, CL_FALSE, 0, datasize, array_a, 0, NULL, NULL);
clEnqueueWriteBuffer(command_queue,buffer_b, CL_FALSE, 0, datasize, array_b, 0, NULL, NULL);
program =clCreateProgramWithSource(context, 1, (const char **)&ocl_string,&ocl_string_size, NULL);
clBuildProgram(program, 1, &device_id,NULL, NULL, NULL);
kernel = clCreateKernel(program,"vectoradd", NULL);
//传递参数
clSetKernelArg(kernel, 0,sizeof(cl_mem),&buffer_a);
clSetKernelArg(kernel, 1,sizeof(cl_mem),&buffer_b);
clSetKernelArg(kernel, 2,sizeof(cl_mem),&buffer_c);
size_t global_work_size[1] = {10};
//执行核函数
clEnqueueNDRangeKernel(command_queue,kernel, 1, NULL, global_work_size, NULL, 0, NULL, NULL);
//从核函数取回计算结果
clEnqueueReadBuffer(command_queue,buffer_c, CL_TRUE, 0, datasize, array_c, 0, NULL, NULL);
for (int i = 0 ; i < 10; i ++) {
printf("%d ", array_a);
}
printf("n");
for (int i = 0 ; i < 10; i ++) {
printf("%d ", array_b);
}
printf("n");
for (int i = 0 ; i < 10; i ++) {
printf("%d ", array_c);
}
printf("n");
clReleaseKernel(kernel);
clReleaseProgram(program);
clReleaseCommandQueue(command_queue);
clReleaseMemObject(buffer_a);
clReleaseMemObject(buffer_b);
clReleaseMemObject(buffer_c);
clReleaseContext(context);
return 0;
}
将上述程序编译，结果如下
0 1 2 3 4 5 6 7 89
9 8 7 6 5 4 3 2 10
9 9 9 9 9 9 9 9 99