nudt-compiler-cpp/test/test_case/testdata2026/performance/conv2d-3.sy

const int N = 2048;
const int KSIZE = 5;

int state;
int repeat_factor;
int N_eff;

int get_random() {
    int n = state % 2048;
    int i = 0;
    while (i < n) {
        state = state + 128;
        state = state % 65535;
        i = i + 1;
    }
    state = state % 65535;
    return state;
}

int idx(int r, int c, int n) {
    return r * n + c;
}

void init_matrix(int A[]) {
    int r = 0;
    while (r < N_eff) {
        int c = 0;
        if (r < N_eff / 2) {
            while (c < N_eff) {
                A[idx(r, c, N_eff)] = get_random() % 65535;
                c = c + 1;
            }
        } else {
            while (c < N_eff) {
                A[idx(r, c, N_eff)] = -1;
                c = c + 1;
            }
        }
        r = r + 1;
    }
}

void init_kernel(int K[]) {
    int i = 0;
    while (i < KSIZE * KSIZE) {
        K[i] = (i % 3) - 1;
        i = i + 1;
    }
}

void conv2d(int In[], int Out[], int K[]) {
    int pad = KSIZE / 2;
    int repeat = 0;
    while (repeat < repeat_factor) {
        int r = 0;
        while (r < N_eff) {
            int c = 0;
            while (c < N_eff) {
                int sum = 0;
                int kr = 0;
                while (kr < KSIZE) {
                    int kc = 0;
                    int rr = r + kr - pad;
                    while (kc < KSIZE) {
                        int cc = c + kc - pad;
                        if (rr >= 0 && rr < N_eff && cc >= 0 && cc < N_eff) {
                            sum = sum + In[idx(rr, cc, N_eff)] * K[idx(kr, kc, KSIZE)];
                        }
                        kc = kc + 1;
                    }
                    kr = kr + 1;
                }
                Out[idx(r, c, N_eff)] = sum;
                c = c + 1;
            }
            r = r + 1;
        }
        repeat = repeat + 1;
    }
}

void nonlinear(int A[]) {
    int i = 0;
    while (i < N_eff * N_eff) {
        int v = A[i];
        int t1 = v * v;
        int t2 = t1 + (3 * v);
        int t3 = t2 - 7;
        int t4 = t3 % 97;
        A[i] = t4;
        i = i + 1;
    }
}

void row_reduce(int A[]) {
    int r = 0;
    while (r < N_eff) {
        int sum = 0;
        int c = 0;
        while (c < N_eff) {
            sum = sum + A[idx(r, c, N_eff)];
            c = c + 1;
        }
        c = 0;
        while (c < N_eff) {
            A[idx(r, c, N_eff)] = A[idx(r, c, N_eff)] - sum;
            c = c + 1;
        }
        r = r + 1;
    }
}

int checksum(int A[]) {
    int s = 0;
    int i = 0;
    while (i < N_eff * N_eff) {
        s = s + A[i];
        i = i + 1;
    }
    return s;
}

int In[2048 * 2048];  // 只操作前N_eff部分
int Out[2048 * 2048];
int K[25];

int main() {
    state = getint();
    repeat_factor = getint();

    // 控制 N_eff 使得矩阵规模可调，范围64~576
    N_eff = (state % 513) + 64;

    starttime();

    init_matrix(In);
    init_kernel(K);

    conv2d(In, Out, K);
    nonlinear(Out);
    row_reduce(Out);

    int sum = checksum(Out);

    stoptime();

    putint(sum);
    putch(10);
    return 0;
}