Re: swap (a, b) - test

[Eric Buddington <eric at sparrow dot vgernet dot net>]

On Sat, Jan 03, 1998 at 08:12:51PM -0800, Richard Henderson wrote:

> If it is faster at all I'll warrent it is only so for depressingly
> register starved machines.

Well, I'm on an Intel 486 - that probably qualifies.

Now, I *did* get great advantage from this before, in more complex
code (quicksort), under gcc-2.7.2.3.

However, a quick test now is now running slower with the XOR swap. I
don't see why three XOR operations on two ints would be slower
than three assignments among three ints...

My test code (included below) ran about 35% faster with the old swap.

Maybe this isn't Nobel-prize material after all :)

-Eric

------------ test code -----------------

#include <stdlib.h>

inline void 
swap1 (int & a, int & b)
{
  int tmp = a; a = b; b = tmp;
}

inline void 
swap2 (int & a, int & b)
{
  if (&a != &b) { a ^= b; b ^= a; a ^= b; }
}

int 
main (int argc, char** argv)
{
  int data [10240];
  for (int i=0; i < 10240; i++) data[i] = rand();
  for (int j=0; j < 10240; j++)
    for (int i=0; i < 10239; i++)
#ifndef FAST_SWAP
      swap1 (data[i], data[i+1]);
#else
      swap2 (data[i], data[i+1]);
#endif
}