OpenMP && Neon

1.OpenMP

#include <stdio.h>
#include <omp.h>
#include <stdlib.h>
#include<iostream>
#include <chrono>
#include <chrono>
#include <ctime>    
//g++ -fopenmp for_test.cpp -o  f
int main(void) {
	
    // Some computation here

	long  sum = 0;
	double time_cost = 0;
	// 1.define data 
	int array_len=100000;
	int array_a[array_len];
	int array_b[array_len];
	int array_c[array_len];
	//2.ini data 
	for(int i=0;i<array_len;++i)
	{
		array_a[i]=i;
		array_b[i]=10;
		array_c[i]=0;
	}
	auto start = std::chrono::system_clock::now();
	//#pragma omp parallel for-----------------------------------------use or not use openmp
	//for(int i=0;i<10000;++i)//1G
	{
		//#pragma omp parallel for reduction(+:sum)
		for (int i=1; i<=array_len; i++) {
			array_c[i]=array_a[i]*array_b[i];
		}
	}
    auto end = std::chrono::system_clock::now();
    std::chrono::duration<double> elapsed_seconds = end-start;
    std::time_t end_time = std::chrono::system_clock::to_time_t(end);
    std::cout << "finished computation at " << std::ctime(&end_time)
              << "elapsed time: " << elapsed_seconds.count() << "s"
              << std::endl;
	// for(int i=0;i<array_len;++i)
	// printf("Sum:%d\n", array_c[i]);
	return 0;
}

without openmp time cost

without openmp cpu load

with openmp time cost

with openmp cpu load

2 Neon

#include <iostream>
#include <arm_neon.h> //需包含的头文件
#include <stdio.h>
#include <stdlib.h>
#include <omp.h>
#include <stdlib.h>
#include <chrono>
#include <chrono>
#include <ctime>   
#include <cstring>

using namespace std;
#include<time.h>

void add_int_c(int* dst, int* src1, int* src2, int count)
{
  int i;
  for (i = 0; i < count; i++)
    dst[i] = src1[i] + src2[i];
}

void add_int_neon(int* dst, int* src1, int* src2, int count)
{
    int i;
    for (i = 0; i < count; i += 4)
    {
        int32x4_t in1, in2, out;
        in1 = vld1q_s32(src1);
        src1 += 4;
        in2 = vld1q_s32(src2);
        src2 += 4;
        out = vaddq_s32(in1, in2);
        vst1q_s32(dst, out);
        dst += 4;
    }
}

int main()
{
    /* 数据内存分配，初始化 */
    int size =100000;//
    int *dst = (int*)malloc(size * sizeof(int));
    int *src1 = (int*)malloc(size * sizeof(int));
    int *src2 = (int*)malloc(size * sizeof(int));

    // int dst[size];
    // int src1[size];
    // int src2[size];
    for(int i = 0; i < size; i++)
    {
        src1[i] = i;
        src2[i] = i;
    }

    /* 时间初始化 */
    struct timespec time1_img = {0, 0};
    struct timespec time2_img = {0, 0};
    clock_gettime(CLOCK_REALTIME, &time1_img);

   // for(int i=0;i<40;++i)
    {
        for(int i = 0; i < size; i++)
        {
            add_int_c(dst, src1, src2, size);
        }
    }
    clock_gettime(CLOCK_REALTIME, &time2_img);
    printf("C time:%d ms\n", (time2_img.tv_sec - time1_img.tv_sec)*1000 + (time2_img.tv_nsec - time1_img.tv_nsec)/1000000);

    printf("dst[0]:%d\n", dst[0]);
    memset(dst, 0, size);

    clock_gettime(CLOCK_REALTIME, &time1_img);
    //for(int i=0;i<40;++i)
    {
        for(int i = 0; i < size; i++)
        {
            add_int_neon(dst, src1, src2, size);
        }
    }
  
    clock_gettime(CLOCK_REALTIME, &time2_img);
    printf("Neon time:%d ms\n", (time2_img.tv_sec - time1_img.tv_sec)*1000 + (time2_img.tv_nsec - time1_img.tv_nsec)/1000000);
    printf("dst[0]:%d\n", dst[0]);

    free(dst);
    free(src1);
    free(src2);

    return 0;
}