Re: [PATCH] Optimize sprintf(buffer,"foo")

[Andrew Pinski <pinskia at physics dot uc dot edu>]

On Tuesday, Jun 24, 2003, at 09:19 US/Eastern, Roger Sayle wrote:

> On Mon, 23 Jun 2003, Andrew Pinski wrote:
> > Here is the result on my laptop (a PowerBook 800MHz G4 running Mac OS 
> > X
> > 10.2.6, Darwin 6.6)  I used the PPC TimeBase register as the timer, 
> > also
> > I made the buffer size 102400 plus one for the zero:
> >
> > sprintf 5618
> > strcpy 21422
> > strcpy+strlen(buf2) 25291
> > strlen(strcpy(buf1, buf2)) 25456
> >
> > As you can see that the sprintf wins over just strcpy (without a even
> > strlen), the strlen adds about the same time as how long sprintf 
> > takes.
>
> Can someone else also confirm these figures?  Is the call to strcpy
> getting inlined badly?  What are the times for an equivalent memcpy?
>
> I'm just very confused by these numbers.  Has this "bug" been filed 
> with
> the Apple folks, or should I consider an optimization that converts 
> calls
> to strcpy into calls to sprintf on some platforms?  Hopefully, its just
> a case where we inline a strcpy where we should be calling the larger 
> but
> more efficient system library routine.

Maybe a transformation of strcpy to memcpy with a strlen
which seems to have about the same speed as the sprintf.

Thanks,
Andrew Pinski


> Thanks to Andrew for doing the timings.
>
> Roger
> --