opencl linux

#include <CL/cl.h>
#include <stdio.h>
#include <stdlib.h>

#define ARRAY_SIZE 1024

int main() {
    cl_platform_id platform;
    cl_device_id device;
    cl_context context;
    cl_command_queue queue;
    cl_program program;
    cl_kernel kernel;
    cl_mem inputA, inputB, output;
    int inputArrayA[ARRAY_SIZE], inputArrayB[ARRAY_SIZE], outputArray[ARRAY_SIZE];
    size_t globalSize = ARRAY_SIZE;

    // Initialize input arrays
    for (int i = 0; i < ARRAY_SIZE; i++) {
        inputArrayA[i] = i;
        inputArrayB[i] = i * 2;
    }

    // Get platform and device
    clGetPlatformIDs(1, &platform, NULL);
    clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL);

    // Create context and command queue
    context = clCreateContext(NULL, 1, &device, NULL, NULL, NULL);
    queue = clCreateCommandQueue(context, device, 0, NULL);

    // Create buffers
    inputA = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(int) * ARRAY_SIZE, inputArrayA, NULL);
    inputB = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, sizeof(int) * ARRAY_SIZE, inputArrayB, NULL);
    output = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof(int) * ARRAY_SIZE, NULL, NULL);

    // Create and build program
    const char *source = "__kernel void add(__global const int *A, __global const int *B, __global int *C) { \
                          int id = get_global_id(0); \
                          C[id] = A[id] + B[id]; \
                        }";
    program = clCreateProgramWithSource(context, 1, &source, NULL, NULL);
    clBuildProgram(program, 1, &device, NULL, NULL, NULL);

    // Create kernel and set arguments
    kernel = clCreateKernel(program, "add", NULL);
    clSetKernelArg(kernel, 0, sizeof(cl_mem), &inputA);
    clSetKernelArg(kernel, 1, sizeof(cl_mem), &inputB);
    clSetKernelArg(kernel, 2, sizeof(cl_mem), &output);

    // Execute kernel
    clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &globalSize, NULL, 0, NULL, NULL);

    // Read output
    clEnqueueReadBuffer(queue, output, CL_TRUE, 0, sizeof(int) * ARRAY_SIZE, outputArray, 0, NULL, NULL);

    // Print result
    for (int i = 0; i < ARRAY_SIZE; i++) {
        printf("%d + %d = %d
", inputArrayA[i], inputArrayB[i], outputArray[i]);
    }

    // Clean up
    clReleaseMemObject(inputA);
    clReleaseMemObject(inputB);
    clReleaseMemObject(output);
    clReleaseKernel(kernel);
    clReleaseProgram(program);
    clReleaseCommandQueue(queue);
    clReleaseContext(context);

    return 0;
}