Xref: news2.mv.net comp.os.msdos.djgpp:5225
From: richard DOT young AT crc DOT doc DOT ca (Richard Young)
Newsgroups: comp.os.msdos.djgpp
Subject: Re: Speed optimization: memcpy() or for loop ??  ** Some actual test results **
Date: 19 Jun 1996 19:25:25 GMT
Organization: Communications Research Centre
Lines: 72
Message-ID: <4q9k75$bgg@crc-news.doc.ca>
References: <4pmlrp$p7u AT crc-news DOT doc DOT ca> <31C2577D DOT 22FD99BD AT alcyone DOT com> <4q679q$4op AT snlsu1>
NNTP-Posting-Host: yaker.vpcs.doc.ca
Mime-Version: 1.0
Content-Type: Text/Plain; charset=US-ASCII
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp

> Richard Young wrote:
> > A question for the optimization experts:
> > For moving data, is it faster to use
> > 
> > a) memcpy(x,y,n*sizeof(x[0]))
> > or
> > b) for (i = 0; i < n; i++) x[i] = y[i];
> > 
> > or are they basically the same speed.

Well, I wrote a test program to test several data move options:

1) memcpy() 
2) an index based for loop
3) a ptr based for loop

The results are:  NO DIFFERENCE.

On my 120 MHz DEC Pentium executing a data move of 100000 floats 2000 times 
took roughly 45 seconds for all options.

I'm a little surprised that memcpy() wasn't faster.  Am I getting an inlined 
"C" version instead of an assembly coded library function.

My test program below was compiled with 

gcc -O3 -m486 -otestmemc testmemc.cpp

//--------------------------------------------------------
// testmemc.cpp

#include <time.h>
#include <string.h>
#include <stdio.h>

#define I 100000
#define N 2000

float x[I];
float y[I];
   
void main()
{
   time_t t1 = time(NULL);
   for (int n = 0; n < N; n++)
      memcpy(&y[0],&x[0],I*sizeof(float));
   time_t t2 = time(NULL);
   printf("memcpy time     = %f\n",difftime(t2,t1));
   
   time_t t3 = time(NULL);
   for (int n = 0; n < N; n++)
      for (int i = 0; i < I; i++)
         y[i] = x[i];
   time_t t4 = time(NULL);
   printf("Index loop time = %f\n",difftime(t4,t3));

   time_t t5 = time(NULL);
   for (int n = 0; n < N; n++)
   {
      float* xp = &x[0];
      float* yp = &y[0];
      for (;xp <= &x[I];)
         *yp++ = *xp++;
   }
   time_t t6 = time(NULL);
   printf("Ptr loop time   = %f\n",difftime(t6,t5));
}

--------------------------------------------------------
Richard Young
CRC, Ottawa, Canada