RE: Re: Fusion-ifying proto parse trees

Joel de Guzman wrote: 


> > Regarding your question about taking the minimum result, I 
> lifted this 
> > code from John Maddock's regex performance shoot-out. This 
> is how he 
> > collects his timing. Yes, this is a black art -- I don't really 
> > understand it.
> 
> I recall Dave A mention some benchmarking strategies that 
> take off the unwanted effects. I'm still looking for the 
> links. You can see in the VC8.0 tests that fusion list is 
> winning. My guess is that this is a problem with VC7.1's 
> optimizer that gives fusion lists a great disadvantage. My 
> guess is that if that effect is known and isolated, fusion 
> lists can be optimized for VC7.1 to mitigate this effect. I 
> believe that what we are seeing is an effect; not really due 
> to the raw-algorithmic advantage of segmentation. IOTW, I 
> think you were just lucky to have found a good formulation 
> for 7.1 and for 7.1 alone.

I'm actually planning to have a look at the generated code for different
compilers, but will have time for that this weekend only. I hope to get some
insides from this. The question is, what really get's optimized and what
not. Additionally I want to run the tests through the Intel compiler which
usually gets the small scale pipelining optimization better than vc -
probably an advantage in our scenarious.

Regards Hartmut




-------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642