Re: [boost] serialization performance

From: Martin Slater (mslater_at_[hidden])
Date: 2005-11-12 18:04:10

  > I believe the intel machine has hardware instructions which implement
> strcmp and that compilers support them.  So even if strcmp is the
> bottleneck, I wouldn't expect it to show up on the profiler unless
> some sort of inlining were turned off.  Or maybe the vtune profiler
> has special provision for these cases somewhere.

VTune doesn't have any special provision for this, if a function is 
inlined it will just show up in the function it inlined it to. Under VC 
by default strcmp is just a regular function call but you can enable it 
as a compiler intrinsic (#pragma intrinsic(strcmp) ) and the compiler 
may well then generate much better code (I know enabling memcpy this way 
can reduce memcpy(&a, &b, sizeof(int)) to a simple register mov in places).

> 
> I did check to verify that the strcmp in the type-id lookup has been
> removed.  Instead we just make sure there is only one instance
> of a particular extended_type_info record so that we can
> just compare the addresses.  There are still some optimizations

This is very good, I was looking at how to do this myself so am very 
happy I now don't have to;)

> to be implemented - but I can't predict how much they will
> speed up anything.

Predication in optimisation I have found to be nigh on impossible, 
without a profiler or at the very least extemely heavy instrumentation 
within you code you will always get a shock as to where the time is 
spent. VC6 was a nightmare in this regard as for example it would not 
inline some trivial functions without being given __forceinline for that 
function casuing some potentially extrememly fast code to run 
pathetically slowly.

If your interested in vtune they do an evaluation verion at 
https://registrationcenter.intel.com/EvalCenter/EvalForm.aspx?ProductID=319
It is simply the best profiling tool I have ever used.

I'd be more than happy to help out with any profiling and optimisation I 
can.

Martin.

-- 
No virus found in this outgoing message.
Checked by AVG Free Edition.
Version: 7.1.362 / Virus Database: 267.13.0/167 - Release Date: 11/11/2005