Re: [xml] libxml2 performance

Hi Daniel, All,

  thanks, applied, but it seems to not change much:

Yeah, that was my first suspicion. In such a little routine
like isolat1ToUTF8, a competent optimizer may produce 
roughly the same code for any not completely broken source.

But it seems not hurt and may give benefits for
some (hey Igor, what's the MSVC numbers here?).

When we have optimized everything else to take only
a quarter of the current time, it may be worth to return
to isolat1ToUTF8.

 I'm surprized by the overall changes, well maybe I can't trust gprof much

Another possible cause is code alignment. If your compiler doesn't start
the most heavily used routines on your processor's prefererred 
boundaries, every change of the code may introduce unforeseeable 
changes in timing.

I've seen the same, but worst for Intel P4. So I normally don't benchmark 
on a P4 and hope that in the long run the intentional changes in timing 
outweight the unintentional ones. And if it that doesn't help, the 
production code may have be to compiled with the Intel compiler.

Peter Jacobi 

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]