Improved Swaping Method In sort.c

Improved Swaping Method In sort.c

[Soumyadip Das Mahapatra]

Thanks Jan,
Actually it should be
static void u32_swap(void *a, void *b)
{
    *(u32 *)b ^= *(u32 *)a;
    *(u32 *)a ^= *(u32 *)b;
    *(u32 *)b ^= *(u32 *)a;
}
But this code saves memory equaling two integers. What about that ?




      Meet people who discuss and share your passions. Go to http://in.promos.yahoo.com/groups/bestofyahoo/

Re: Improved Swaping Method In sort.c

[Andi Kleen]

Soumyadip Das Mahapatra <dip_kernel@yahoo.co.in> writes:

> But this code saves memory equaling two integers. What about that ?

It won't because the compiler uses registers not memory for the
temporary of course. In fact your version will cause more memory
accesses because the compiler doesn't know that a and b do not alias
each other, so it cannot cache the immediate value and has to always
go through memory. The difference will not be large on x86 because L1
cache accesses are reasonably fast, but they're still much slower than
a register access.

-Andi

Re: Improved Swaping Method In sort.c

[Ray Lee]

On Wed, Apr 30, 2008 at 11:09 AM, Soumyadip Das Mahapatra
<dip_kernel@yahoo.co.in> wrote:
> Thanks Jan,
>  Actually it should be
>  static void u32_swap(void *a, void *b)
>  {
>     *(u32 *)b ^= *(u32 *)a;
>     *(u32 *)a ^= *(u32 *)b;
>     *(u32 *)b ^= *(u32 *)a;
>  }
>  But this code saves memory equaling two integers. What about that ?

The above prevents gcc from storing the intermediary in a register. So
you're trading fast register access for slow memory access.

Really, you shouldn't be prosing optimizations unless you do two
things: check the output of gcc (on at least one platform, but really
should be several), and include timings of before/after.