__global__ void VecAdd(float* A, float* B, float* C, int N)
{
    int i = blockDim.x * blockIdx.x + threadIdx.x;
    if (i < N)
        C[i] = A[i] + B[i];
 }

int main()
{
    int N = ...;
    size_t size = N * sizeof(float);

    float* h_A = (float*)malloc(size);
    float* h_B = (float*)malloc(size);

    //(wypełnienie wektorów danymi)

    float* d_A; cudaMalloc(&d_A, size);
    float* d_B; cudaMalloc(&d_B, size);
    float* d_C; cudaMalloc(&d_C, size);

    cudaMemcpy(d_A, h_A, size, cudaMemcpyHostToDevice);
    cudaMemcpy(d_B, h_B, size, cudaMemcpyHostToDevice);

    int threadsPerBlock = 256;
    int blocksPerGrid = (N + threadsPerBlock – 1) / threadsPerBlock;

    VecAdd<<<blocksPerGrid, threadsPerBlock>>>(d_A, d_B, d_C, N);

    cudaMemcpy(h_C, d_C, size, cudaMemcpyDeviceToHost);

    cudaFree(d_A); cudaFree(d_B); cudaFree(d_C);
}