__global__ void addKernel(float *POSX, float *POSY, float *POSR_X, float *POSR_Y, float *POSR)
{
    int first = threadIdx.x;
    float  den;
    POSR_X[first] = 0; POSR_Y[first] = 0;
    for (int next = 0; next < N; next++) {
        if ((first != next) && ((POSX[next] != POSX[first]) || (POSY[next] != POSY[first]))) {
            den = 100000 * sqrt((POSX[next] - POSX[first])*(POSX[next] - POSX[first]) + (POSY[next] - POSY[first])*(POSY[next] - POSY[first]));
            POSR_X[first] += (POSX[next] - POSX[first]) / den;
            POSR_Y[first] += (POSY[next] - POSY[first]) / den;
        }
    }
    POSR[first] += POSR_X[first];
    POSR[first + N] += POSR_Y[first];
}

void GPU_Math()
{
    float *CU_POSR, *CU_POSX, *CU_POSY, *CU_POSR_X, *CU_POSR_Y;

    cudaMalloc((void**)&CU_POSX, N * sizeof(int));
    cudaMalloc((void**)&CU_POSY, N * sizeof(int));
    cudaMalloc((void**)&CU_POSR_X, N * sizeof(int));
    cudaMalloc((void**)&CU_POSR_Y, N * sizeof(int));
    cudaMalloc((void**)&CU_POSR, 2 * N * sizeof(int));

    cudaMemcpy(CU_POSX, POSX, N * sizeof(int), cudaMemcpyKind::cudaMemcpyHostToDevice);
    cudaMemcpy(CU_POSY, POSY, N * sizeof(int), cudaMemcpyKind::cudaMemcpyHostToDevice);
    cudaMemcpy(CU_POSR_X, AX_M, N * sizeof(int), cudaMemcpyKind::cudaMemcpyHostToDevice);
    cudaMemcpy(CU_POSR_Y, AY_M, N * sizeof(int), cudaMemcpyKind::cudaMemcpyHostToDevice);
    cudaMemcpy(CU_POSR, POSR, 2 * N * sizeof(int), cudaMemcpyKind::cudaMemcpyHostToDevice);

    addKernel << <1, N >> >(CU_POSX, CU_POSY, CU_POSR_X, CU_POSR_Y, CU_POSR);

    cudaMemcpy(AX_M, CU_POSR_X, N * sizeof(int), cudaMemcpyKind::cudaMemcpyDeviceToHost);
    cudaMemcpy(AY_M, CU_POSR_Y, N * sizeof(int), cudaMemcpyKind::cudaMemcpyDeviceToHost);
    cudaMemcpy(POSX, CU_POSX, N * sizeof(int), cudaMemcpyKind::cudaMemcpyDeviceToHost);
    cudaMemcpy(POSY, CU_POSY, N * sizeof(int), cudaMemcpyKind::cudaMemcpyDeviceToHost);
    cudaMemcpy(POSR, CU_POSR, 2 * N * sizeof(int), cudaMemcpyKind::cudaMemcpyDeviceToHost);

    cudaFree(CU_POSX);
    cudaFree(CU_POSY);
    cudaFree(CU_POSR_X);
    cudaFree(CU_POSR_Y);
}