Topic: -+{ Challenge }+- Serialize Xdata

[Kerry]

Thanks Dan, I'll work through it eventually

The time difference astounds me  !!  .... 8 seconds vs 0.7 seconds.

[pkohut]

Test results of supplied drawing (500,000 points), 'doit' middle 5 test results from  10 samples.

Only changes was to move the switch statement logic in CTypedValue::CTypedValue( const resbuf *buf )
to a class that holds a jump table of functions.  The class is called like

Code: [Select]

   short type = acdbGroupCodeToType(buf->restype);
   blaze1::op(buf, _value, _valueType, vString);
Because the compiler generated indirect calls through the jump table the speed boost I was looking for didn't
really happen.  Would need to restructor more of the code and try to keep parameters in functions to
3 or less, and have a jmp instead of a call.

Code: [Select]

                 Paul            Dan
              0.677111       0.690452
              0.678350       0.691959
              0.679616       0.695678
              0.680110       0.698469
              0.680415       0.699986

Average       0.679120       0.695309

Finally, Dan's runtime he posted of 0.53? for his i7 machine is what I get when I don't archive the data.

[It's Alive!]

I played with  acdbGroupCodeToType too, I created a switch with all the codes including  ones > 5000, but it made little difference.

I'm still amazed at how fast CArchive is compared to serializing with .NET   

[It's Alive!]

I'm still pondering what kind of serialization I want to use, it seems most like some sort of XML format.... 

[pkohut]

Funny, I was just looking at Boost Fusion earlier as a possible XML solution.
Got as far as reading the preface and quick start pages.

http://www.boost.org/doc/libs/1_42_0/libs/fusion/doc/html/fusion/quick_start.html