__global__ void addKernel(float* cuPOSX, float* cuPOSY, float* cuPOSR, float *cuAX_M, float *cuAY_M)
{
    int first = threadIdx.x;
    for (int next = 0; next < N; next++) {
        if (first != next) {
            cuAX_M[first] = 0.1 * (cuPOSX[next] - cuPOSX[first]) / sqrt((cuPOSX[next] - cuPOSX[first])*(cuPOSX[next] - cuPOSX[first]) + (cuPOSY[next] - cuPOSY[first])*(cuPOSY[next] - cuPOSY[first]));
            cuAY_M[first] = 0.1 * (cuPOSY[next] - cuPOSY[first]) / sqrt((cuPOSX[next] - cuPOSX[first])*(cuPOSX[next] - cuPOSX[first]) + (cuPOSY[next] - cuPOSY[first])*(cuPOSY[next] - cuPOSY[first]));
        }
    }
    cuPOSR[first] += cuAX_M[first];
    cuPOSR[first + N] += cuAY_M[first];
}

void GPU_Math()
{
    float *CU_POSX, *CU_POSY, *CU_POSR, *CU_AX_M, *CU_AY_M;
    int size = N;

    cudaMalloc((void**)&CU_POSX, size);
    cudaMalloc((void**)&CU_POSY, size);
    cudaMalloc((void**)&CU_POSR, *2);
    cudaMalloc((void**)&CU_AX_M, size);
    cudaMalloc((void**)&CU_AY_M, size);

    cudaMemcpy(CU_POSX, POSX, size, cudaMemcpyKind::cudaMemcpyHostToDevice);
    cudaMemcpy(CU_POSY, POSY, size, cudaMemcpyKind::cudaMemcpyHostToDevice);
    cudaMemcpy(CU_POSR, POSR, size*2, cudaMemcpyKind::cudaMemcpyHostToDevice);
    cudaMemcpy(CU_AX_M, AX_M, size, cudaMemcpyKind::cudaMemcpyHostToDevice);
    cudaMemcpy(CU_AY_M, AY_M, size, cudaMemcpyKind::cudaMemcpyHostToDevice);

    addKernel<<<1, size>>>(CU_POSX, CU_POSY, CU_POSR, CU_AX_M, CU_AY_M);

    cudaMemcpy(POSR, CU_POSR, size + size, cudaMemcpyKind::cudaMemcpyDeviceToHost);
    cudaMemcpy(AX_M, CU_AX_M, size, cudaMemcpyKind::cudaMemcpyDeviceToHost);
    cudaMemcpy(AY_M, CU_AY_M, size, cudaMemcpyKind::cudaMemcpyDeviceToHost);

    cudaFree(CU_POSX);
    cudaFree(CU_POSY);
    cudaFree(CU_POSR);
    cudaFree(CU_AX_M);
    cudaFree(CU_AY_M);
}