memcpy性能如何? - 瀚海星空

2010-12-01

http://abloz.com 2010.12.1
比较好奇memcpy对性能有多大影响。所以写了个测试程序。源码如下。对1024个字节进行memcpy，拷贝100万次。将b放在全局区是因为如果放栈区会导致栈空间不足。

zhouhh@zhh64:~$ cat sms/test/testmemcpy.c

//////////////////////////////////////////
//author 	zhouhh
//date 		2010.12.1
//notes       http://abloz.com
//history
//copyright( 2010 ) allright reserved!
/////////////////////////////////////////

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#define BUF_LEN 1024
char b[1000000][BUF_LEN]={0};
void main()
{
char a[BUF_LEN]={"1234567890 abcdefg higklmn opq rst uvw xyz"};
a[BUF_LEN]='\0';
int i = 0;
int max = 1000000;
struct timespec ts,te;
printf("a=%sn",a);
clock_gettime (CLOCK_REALTIME, &ts) ;
printf("begin:%ld s,%ld nsn",ts.tv_sec,ts.tv_nsec);
for(i = 0; i< max; i++)
{
memcpy(b[i],a,BUF_LEN);
}

printf("memcpy %d times,buflen=%dn",max,BUF_LEN);
clock_gettime (CLOCK_REALTIME, &te) ;
printf("end:%ld s,%ld nsn",te.tv_sec,te.tv_nsec);
long nsec = te.tv_nsec-ts.tv_nsec;
int sec = te.tv_sec - ts.tv_sec;
if(nsec < 0)
{
nsec += 1000000000;
sec -= 1;
}

printf("spend:%.3f secn",sec+nsec/1000000000.0);
printf("every memcpy spend:%.3f usecn",(sec*1000000000+nsec)/(max*1000.0));

printf("chek b buf:%sn",b[max-1]);
}

zhouhh@zhh64:~/sms/test$ gcc -o tm -lrt testmemcpy.c
zhouhh@zhh64:~/sms/test$ ./tm
a=1234567890 abcdefg higklmn opq rst uvw xyz
begin:1291188463 s,18960481 ns
memcpy 1000000 times,buflen=1024
end:1291188464 s,200238201 ns
spend:1.181 sec
every memcpy spend:1.181 usec
chek b buf:1234567890 abcdefg higklmn opq rst uvw xyz

由此可见，1024个字节，拷贝100万次，花费了1.181秒。测试CPU： AMD Athlon(tm) 64 X2 Dual Core Processor 5200+,1000MHZ

另一个测试服务器： CPU ：Intel(R) Xeon(R) CPU X5680 @ 3.33GHz 24核。

# ./tm
a=1234567890 abcdefg higklmn opq rst uvw xyz
begin:1291188745 s,642300000 ns
memcpy 1000000 times,buflen=1024
end:1291188746 s,904286000 ns
spend:1.262 sec
every memcpy spend:1.262 usec
chek b buf:1234567890 abcdefg higklmn opq rst uvw xyz

总花费1.26秒。与我工作机器用时并无太多差别。平均大概1次memcpy 1024个字节，花费1微秒多点。

Technorati 标签: memcpy

如非注明转载, 均为原创. 本站遵循知识共享CC协议,转载请注明来源

FEATURED TAGS

css vc6 http automake linux make makefile voip 乱码 awk flash vista vi vim javascript pietty putty ssh posix subversion svn windows 删除编译多线程 wxwidgets ie ubuntu 开源 c python bash 备份性能 scp 汉字 log ruby 中文 bug msn nginx php shell wordpress mqueue android eclipse java mac ios html5 js mysql protobuf apache hadoop install iocp twisted centos mapreduce hbase thrift tutorial hive erlang lucene hdfs sqoop utf8 filter 草原 yarn ganglia 恢复 scrapy django fsimage flume tail flume-ng mining scala go kafka gradle cassandra baas spring postgres maven mybatis mongodb https nodejs 镜像心理学机器学习 Keras theano anaconda docker spark akka-http json 群论区块链加密抽象代数离散对数同余欧拉函数扩展欧几里德算法 ES6 node-inspect debug win10 vscode 挖矿

FEATURED TAGS

FRIENDS