代码

// fileName: test_O2_Transpose_Unroll_loops_8_type_1.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];

inline void Swap(UI &x, UI &y) { UI t = x; x = y; y = t; }
void Mul() {
	for (int i = 1; i < N; ++i)
		for (int j = 0; j < i; ++j)
			Swap(B[i][j], B[j][i]);
	for (int i = 0; i < N; ++i)
		for (int j = 0; j < N; ++j)
			for (int k = 0; k < N; k += 8) {
				C[i][j] = (C[i][j] + (ULL)A[i][k    ] * B[j][k    ]) % Mod;
				C[i][j] = (C[i][j] + (ULL)A[i][k + 1] * B[j][k + 1]) % Mod;
				C[i][j] = (C[i][j] + (ULL)A[i][k + 2] * B[j][k + 2]) % Mod;
				C[i][j] = (C[i][j] + (ULL)A[i][k + 3] * B[j][k + 3]) % Mod;
				C[i][j] = (C[i][j] + (ULL)A[i][k + 4] * B[j][k + 4]) % Mod;
				C[i][j] = (C[i][j] + (ULL)A[i][k + 5] * B[j][k + 5]) % Mod;
				C[i][j] = (C[i][j] + (ULL)A[i][k + 6] * B[j][k + 6]) % Mod;
				C[i][j] = (C[i][j] + (ULL)A[i][k + 7] * B[j][k + 7]) % 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