代码

#define n 1024
void matrix_multiply(int _, const double* A, const double* B, double* C) {
	double a = 0, b = 0, c = 0, d = 0;
	for (int i = 0; i < n; i++) {
		for (int j = 0; j < n; j++) {
			a = b = c = d = 0;
			for (int k = 0; k < n; k += 4) {
				a += A[(i << 10) | k] * B[(k << 10) | j];
				b += A[(i << 10) | k | 1] * B[((k | 1) << 10) | j];
				c += A[(i << 10) | k | 2] * B[((k | 2) << 10) | j];
				d += A[(i << 10) | k | 3] * B[((k | 3) << 10) | j];
			}
			C[(i << 10) | j] = a + b + c + d;
		}
	}
}

评测结果