[Developers] ifndef in DXYDG_guts.h

[Erik Schnetter]

On Mar 14, 2007, at 11:10:17, John Shalf wrote:

> It is certainly interesting to perform the experiment (perhaps even
> using standalone_BSSN) to ensure that our assumptions about the
> quality of branch-prediction are indeed true.  But the intuition
> about modern branch predictors is probably true (they are pretty
> damned good these days).

Let me clarify: We did in fact perform this experiment last December,  
and the result was that the overall performance increase was lost in  
the noise (somewhere around 1% or 2%, I cannot remember the exact  
number).  We did not have a good intuition about modern branch  
predictors: we actually assumed that we could speed up the code a  
lot, but the experiment proved us wrong.  There is nothing like a  
dose of reality.

This was the work of Jennifer Seiler, a physics student at the AEI.   
She specialised parts of BSSN_MoL, removing many if statements, not  
just those for finite differencing operators.  She also removed if  
statements choosing gauge conditions and selecting formulations.  We  
decided in the end that the performance gain (if any) was not worth  
the added complication.

-erik

-- 
Erik Schnetter <schnetter at cct.lsu.edu>

My email is as private as my paper mail.  I therefore support encrypting
and signing email messages.  Get my PGP key from www.keyserver.net.



-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 186 bytes
Desc: This is a digitally signed message part
Url : http://www.cactuscode.org/pipermail/developers/attachments/20070314/4a442022/attachment.bin