#include <stdio.h>
#include <time.h>
#include <stdlib.h>
#include <iostream>
#include <ostream>
#include <arm_neon.h>
void sparse_matmul_coo(float **A_denseMatrix,float **B_denseMatrix,float **C_denseMatrix,int MAX){
	for(int i=0;i<MAX;++i){
		for(int j=0;j<MAX;++j){
			float32x4_t vecC_denseMatrix=vdupq_n_f32(0.0);
	        for(int k=0;k<MAX;k+=4){
		        float32x4_t vecA_denseMatrix=vld1q_f32(&A_denseMatrix[i][k]);
		        float32x4_t vecB_denseMatrix=vld1q_f32(&B_denseMatrix[i][k]);
		        vecC_denseMatrix=vmlaq_f32(vecC_denseMatrix,vecA_denseMatrix,vecB_denseMatrix);
		        C_denseMatrix[i][j]=vgetq_lane_f32(vecC_denseMatrix,0)+vgetq_lane_f32(vecC_denseMatrix,1)
				+vgetq_lane_f32(vecC_denseMatrix,2)+vgetq_lane_f32(vecC_denseMatrix,3);
	        }
		}
	}
}
int main(){
	float A_values[]={1,2,3,4,5};
	int A_rowIndex[]={0,0,1,2,2};
	int A_colIndex[]={0,2,1,0,2};
	int A_nonZeroCount=5;
	
	float B_values[]={6,8,7,9};
	int B_rowIndex[]={0,2,1,2};
	int B_colIndex[]={0,0,1,2};
	int B_nonZeroCount=4;
	
	int MAX;
	if(A_nonZeroCount>B_nonZeroCount)MAX=A_nonZeroCount;
	else MAX=B_nonZeroCount;
	
	float C_values[MAX];
	int C_rowIndex[MAX];
	int C_colIndex[MAX];
	int C_nonZeroCount=0;
	
	int ROWS=MAX,COLS=MAX;
	float **A_denseMatrix=new float*[MAX];
	float **B_denseMatrix=new float*[MAX];
	float **C_denseMatrix=new float*[MAX];
	for(int i=0;i<MAX;++i){
		A_denseMatrix[i]=new float[MAX];
		B_denseMatrix[i]=new float[MAX];
		C_denseMatrix[i]=new float[MAX];
	}
	for(int i=0;i<A_nonZeroCount;i++){
		int row=A_rowIndex[i];
		int col=A_colIndex[i];
		A_denseMatrix[row][col]=A_values[i];
	}
	for(int i=0;i<B_nonZeroCount;i++){
		int row=B_rowIndex[i];
		int col=B_colIndex[i];
		B_denseMatrix[row][col]=B_values[i];
	}
	
	
	
	clock_t start=clock();
	sparse_matmul_coo(A_denseMatrix,B_denseMatrix,C_denseMatrix,MAX);
					   
	clock_t end=clock();
	std::cout<<"优化的稀疏矩阵乘法时间："<< double(end-start)/CLOCKS_PER_SEC<<"秒"<<std::endl;
}