Re: Fastcode SpreadsheetCreator 0.8.2

Hi Dennis,


Should blended results calculation have some correction factors so
that weight of each platform is equal or near equal? For example in

Perhaps.

Because of this problem it is important to use the same target PC's in all
challenges and in all releases.


That's difficult.


One could claim that it is proper to give the slowest CPU in a given
challenge a higher weigth.

CompareMem results are like this:
Presler 646
Northwood 1396
Yonah 727
Dothan 668
AMD64 887
AMDX2 581
Northwood has there much bigger weight to total results than other
faster platforms.

I think it is difficult to design a proper system. Perhaps somebody can come
up with some ideas?

In another challenge Northwood can end up being faster than the others. This
rules out using fixed weigths to targets.

Is it ok to calculate target weigths simply by scaling all benchmark results
on a given target such that the fastest function will get a benchmark of say
1000? (only for calculating Blended winners)


In my opinion scaling blended winners in the way you described
sounds good.

Regards,
Jouni

--
The Fastcode Project: http://www.fastcodeproject.org/
.

Quantcast