[Campaign-news] New k-centers

[Marc Sosnick]

Kai, Bill:

> You are right, we should probably agree on some default directory  
> for the includes (I think neither mine nor Marc's is) and provide  
> the Makefile correspondingly.

Actually, the way Bill is doing it seems to me quite acceptable for a  
prerelease test version.  There needs to be a list of specific "do  
this to get the software running," instructions that users can follow  
step-by-step.  Among these rules would be "make sure your xxxx  
director[y|ies] [is|are] correct."

To really give it the professional polish, we should think about using  
autoconf.  This would result in an automatic ./configure file to  be  
used before "make install" (i.e. the standard unix installation  
method).  I've already been looking into this, and we can talk about  
it tomorrow.

> The file should of course not be hardcoded in the final release, in  
> fact I don't think that the main() should be provided.   
> kcenterGPU.cu can be included like a header file in users' projects,  
> which of course makes sense for a library.  I think we should later  
> move the main into a separate file as one example on how to use the  
> library.  If you prefer providing a file name on the command line,  
> please go ahead and make any changes you like.

Yes.  However, I do like the way that NVIDIA does it, and may I  
suggest that we follow their lead.  We can explicitly compile out the  
object file (e.g. kcenterGPU.o above) as well as have an executable  
demo file for each clustering algorithm.   For your example, the  
kcenterGPU.cu file should not be included, but rather linked in during  
the link phase, and instead a kcenterGPU.h file included in the users'  
projects.

See everyone tomorrow...

Marc