#include <stdio.h>
#include <ctime>
#include <stdlib.h>
#include <arm_neon.h> //启用 NEON 指令

#define SIZE 102400000

// 基础向量加法函数
void vector_add(float* A, float* B, float* C, int size) 
{
    for (int i = 0; i < size; ++i) 
	{
        C[i] = A[i] + B[i];
    }
}
//优化的向量加法函数
void vector_add_optimized(float* A, float* B, float* C, int size)
{
	for(int i=0;i<SIZE;i+=4)
	{
		//加载A和B向量的4个浮点数到NEO寄存器 
		float32x4_t vecA=vld1q_f32(&A[i]);
		float32x4_t vecB=vld1q_f32(&B[i]);
        //执行向量加法 
        float32x4_t vecC=vaddq_f32(vecA,vecB);
        //将结果存储到C向量 
        vst1q_f32(&C[i],vecC);
	}
}

//基础的矩阵乘法
#define M  2000
#define N  2000
#define Q  2000
//A(M*N),B(N*Q),C(M*Q)对于矩阵维度的说明
void matmul(float** A, float** B, float** C)
{
	for (int i = 0; i < M; ++i) 
	{	
		for (int j = 0; j < Q; ++j) 
		{
			C[i][j] = 0;
			for (int k = 0; k < N; ++k) 
			{	
				C[i][j] += A[i][k] * B[k][j];
			}
		}
	}
}

//转置函数
void transposeMatrix(float** matrix, float** transposed) {
    for (int i = 0; i < N; i++) {
        for (int j = 0; j < Q; j++) {
            transposed[j][i] = matrix[i][j];
        }
    }
}
//优化的乘法函数
void matmul_optimized(float** A, float** B, float** C) {
    for (int i = 0; i < M; i++) {
        for (int j = 0; j < Q; j++) {
            float sum = 0.0f;
            for (int k = 0; k < N; k += 4) 
            {
                float32x4_t vecA, vecB, vecC;
                if (k + 4 <= N) 
                {
                    // 加载A和B的4个元素，进行向量化计算
                    vecA = vld1q_f32(&A[i][k]);
                    vecB = vld1q_f32(&B[j][k]);
                    // 向量化乘法并累加结果
                    vecC = vmulq_f32(vecA, vecB);
                    sum += vgetq_lane_f32(vecC, 0) + vgetq_lane_f32(vecC, 1) +
                           vgetq_lane_f32(vecC, 2) + vgetq_lane_f32(vecC, 3);
                } 
                else 
                {
                    // 处理剩余的元素
                    for (int m = k; m < N; m++) 
                    {
                        sum += A[i][m] * B[j][m];
                    }
                }
            }
            C[i][j] = sum;
        }
    }
}
int main() 
{
    srand(time(NULL));

    // 动态分配内存，并检查是否分配成功
    float *A = (float *)malloc(SIZE * sizeof(float));
    float *B = (float *)malloc(SIZE * sizeof(float));
    float *C = (float *)malloc(SIZE * sizeof(float));
    // 随机初始化，生成0到99之间的随机浮点数
    for (int i = 0; i < SIZE; i++) 
	{
        A[i] = (float)(rand() % 100) / 100.0f;
        B[i] = (float)(rand() % 100) / 100.0f;
    }

    printf("当SIZE取%d时\n", SIZE);
    clock_t start1 = clock();
    vector_add(A, B, C, SIZE);
    clock_t end1 = clock();
    // 计算并输出向量加法的时间
    double basical_time_spent = double(end1 - start1) / CLOCKS_PER_SEC;
    printf("使用基础的向量加法用时%lf秒\n", basical_time_spent);

    clock_t start2 = clock();
    vector_add_optimized(A, B, C, SIZE);
    clock_t end2 = clock();
    // 计算并输出向量加法的时间
    double time_spent = double(end2 - start2) / CLOCKS_PER_SEC;
    printf("使用NEON 优化向量加法用时%lf秒\n", time_spent);
    // 释放动态分配的内存
    free(A);free(B);free(C);

    // 分配矩阵内存
    float** A2 = (float**)malloc(M * sizeof(float*));
    float** B2 = (float**)malloc(N * sizeof(float*));
    float** C2 = (float**)malloc(M * sizeof(float*));
    for (int i = 0; i < M; ++i) 
    {
        A2[i] = (float*)malloc(N * sizeof(float));
        C2[i] = (float*)malloc(Q * sizeof(float));
    }
    for (int i = 0; i < N; ++i) 
    {
        B2[i] = (float*)malloc(Q * sizeof(float));
    }
    // 初始化矩阵数据
    for (int i = 0; i < M; i++) {
        for (int j = 0; j < N; j++) {
            A2[i][j] = (float)(rand() % 100) / 100.0f;
        }
    }
    for (int i = 0; i < N; i++) {
        for (int j = 0; j < Q; j++) {
            B2[i][j] = (float)(rand() % 100) / 100.0f;
        }
    }

    printf("\n");
    printf("当矩阵A2的维度为%d*%d,矩阵B2的维度为%d*%d时\n",M,N,N,Q);
    //计算基础用时
    clock_t start3 = clock();
    matmul(A2, B2, C2);
    clock_t end3 = clock();
    double multiply_time_spent = double(end3 - start3) / CLOCKS_PER_SEC;
    printf("使用基础的矩阵乘法用时：%lf秒\n",multiply_time_spent);
    //计算 优化用时
    float** transposed = (float**)malloc(Q * sizeof(float*));
    for (int i = 0; i < Q; ++i) 
    {
        transposed[i] = (float*)malloc(N * sizeof(float));
    }
    transposeMatrix(B2, transposed);
    clock_t start4 = clock();
    matmul_optimized(A2, transposed, C2);
    clock_t end4 = clock();
    double optimized_multiply_time_spent = double(end4 - start4) / CLOCKS_PER_SEC;
    printf("使用优化的矩阵乘法用时：%lf秒\n", optimized_multiply_time_spent);

    // 释放动态分配的内存
    for (int i = 0; i < M; ++i) {
        free(A2[i]);free(C2[i]);
    }
    for (int i = 0; i < N ;++i) {
        free(B2[i]);
    }
    for (int i = 0; i < Q ;++i) {
        free(transposed[i]);
    }
    free(A2); free(B2); free(C2);free(transposed);
}