#include <arm_neon.h>
#include <stdio.h>
#include <time.h>
#include <stdlib.h>
#define SIZE 1024

void matmul_optimized(float** A, float** B, float** C, int n);
int main() {
	int n = SIZE;

	// 分配内存空间
	float** A = (float**)malloc(sizeof(float*) * n);
	float** B = (float**)malloc(sizeof(float*) * n);
	float** C = (float**)malloc(sizeof(float*) * n);
	int i, j;
	for(i=0; i<n; i++) {
		A[i] = (float*)malloc(sizeof(float) * n);
		B[i] = (float*)malloc(sizeof(float) * n);
		C[i] = (float*)malloc(sizeof(float) * n);
	}
	
	// 随机化
	srand(time(0));
	for(i=0; i<n; i++)
	for(j=0; j<n; j++) {
		A[i][j] = rand() % 100;
		B[i][j] = rand() % 100;
	}

	// 计算并计时
	clock_t start = clock();
	matmul_optimized(A, B, C, SIZE);
	clock_t end = clock();
	printf("使用 NEON 优化稠密矩阵乘法时间：%lf\n", (double)(end-start) / CLOCKS_PER_SEC);

	// 释放内存空间
	for(i=0; i<n; i++) {
		free(A[i]);
		free(B[i]);
		free(C[i]);
	}
	free(A); free(B); free(C);
}

void matmul_optimized(float** A, float** B, float** C, int n) {
	int i, j, k;
	float32x4_t vecA, vecB, vecC;
	for(i=0; i<n; i++)
	for(j=0; j<n; j++) {
		vecC = vdupq_n_f32(0.0);
		for(k=0; k<n; k+=4) {
			vecA = vld1q_f32(&A[i][k]);
			vecB = vld1q_f32(&A[k][j]);
			vecC = vmlaq_f32(vecC, vecA, vecB);
		}
		C[i][j] = vgetq_lane_f32(vecC, 0) + vgetq_lane_f32(vecC, 1) + 
		          vgetq_lane_f32(vecC, 2) + vgetq_lane_f32(vecC, 3);
	}
}