代码

// fileName: test_O2_Interchanging_loops_Unroll_loops_14.cpp
#pragma GCC optimize("O2")

#include <cstdio>
#include <random>
#include <chrono>

std::mt19937 rng(std::chrono::steady_clock::now().time_since_epoch().count());

typedef unsigned int UI;
typedef unsigned long long ULL;

const UI Mod = 998244353;
const int N = 1400;

UI A[N][N], B[N][N], C[N][N];

void Mul() {
	for (int i = 0; i < N; ++i)
		for (int k = 0; k < N; ++k)
			for (int j = 0; j < N; j += 14) {
				C[i][j    ] = (C[i][j    ] + (ULL)A[i][k] * B[k][j    ]) % Mod;
				C[i][j + 1] = (C[i][j + 1] + (ULL)A[i][k] * B[k][j + 1]) % Mod;
				C[i][j + 2] = (C[i][j + 2] + (ULL)A[i][k] * B[k][j + 2]) % Mod;
				C[i][j + 3] = (C[i][j + 3] + (ULL)A[i][k] * B[k][j + 3]) % Mod;
				C[i][j + 4] = (C[i][j + 4] + (ULL)A[i][k] * B[k][j + 4]) % Mod;
				C[i][j + 5] = (C[i][j + 5] + (ULL)A[i][k] * B[k][j + 5]) % Mod;
				C[i][j + 6] = (C[i][j + 6] + (ULL)A[i][k] * B[k][j + 6]) % Mod;
				C[i][j + 7] = (C[i][j + 7] + (ULL)A[i][k] * B[k][j + 7]) % Mod;
				C[i][j + 8] = (C[i][j + 8] + (ULL)A[i][k] * B[k][j + 8]) % Mod;
				C[i][j + 9] = (C[i][j + 9] + (ULL)A[i][k] * B[k][j + 9]) % Mod;
				C[i][j + 10] = (C[i][j + 10] + (ULL)A[i][k] * B[k][j + 10]) % Mod;
				C[i][j + 11] = (C[i][j + 11] + (ULL)A[i][k] * B[k][j + 11]) % Mod;
				C[i][j + 12] = (C[i][j + 12] + (ULL)A[i][k] * B[k][j + 12]) % Mod;
				C[i][j + 13] = (C[i][j + 13] + (ULL)A[i][k] * B[k][j + 13]) % Mod;
			}
}

#ifdef PKRBT
int main() {
#else
void matrix_multiply(int, const double *, const double *, double *) {
#endif
	for (int i = 0; i < N; ++i) {
		for (int j = 0; j < N; ++j) {
			A[i][j] = rng() % Mod;
			B[i][j] = rng() % Mod;
			C[i][j] = 0;
		}
	}
	Mul();
	UI Ans = 0;
	for (int i = 0; i < N; ++i)
		for (int j = 0; j < N; ++j)
			Ans ^= C[i][j];
	printf("%u\n", Ans);
#ifdef PKRBT
	return 0;
#endif
}

评测结果