[lug] Various Arch/Compiler Binaries living together

[Ed Hill]

On Sun, 2002-06-23 at 00:49, Timothy C. Klein wrote:
> 
> So, if I were to compile KDE with 2.95, arch=i686, and everything else with
> 3.0, arch=athlon, what do you think would happen?  My gut tells me
> this would be a bad thing, and would probably end up causing KDE to
> break, or worse, crash all the time.  But I don't really know.  
> 
> Any thoughts?


Does anyone remember the Stampede Linux distro?  One of their:

  http://www.stampede.org/

main goals was to provide a distro that was compiled with all
optimizations (-O3 + architecture-specific stuff) turned on.  They
targeted the i686 and Athlon platforms rather than providing more
general i[3-5]86 binaries.  I met a few of the main developers at a
conference in NYC 2-3 years ago and, at the time, I thought it was a
fairly neat idea.

Since then, I've become convinced that fussing with platform-specific
compilation and optimization tweaks is almost always a waste of time.  I
don't have any URLs at the moment to back this up, but here it is:  If
you aren't doing CPU-intensive work (eg. number crunching, audio/video
en-/de-coding, etc.) then the optimization tweaks are a *total waste*. 
Depending upon the tweaks used, they will bloat the code while giving
you <10% improvement (any typically less than 1-3%) on the vast majority
of open source projects.  From an investment point of view, you won't
even get your re-compilation time and effort back in amortized run-time
savings!

If you really want to speed up some code, buy a faster machine or pick a
project that you think is "slow" and work with the developers to
implement better algorithms and/or data structures.  Yes, thats either
expensive or a lot of work.  But both of those approaches will actually
accomplish something whereas re-compiling stuff with optimization tweaks
is (again, for most general codes) an exercise in futility...

Ed

ps - I spent a fair chunk of my research work writing codes to do 
     matrix-vector calculations for simulation and optimization 
     codes.  In my experience, the choice and quality-of-
     implementation of the data structures and algorithms are 
     orders-of-magnitude more important than any optimization 
     tweaks used with the gcc compiler.


-- 
Edward H. Hill III, PhD 
Post-Doctoral Researcher   |  Emails:      ed at eh3.com, ehill at mines.edu
Division of ESE            |  URL:         http://www.eh3.com
Colorado School of Mines   |  Phone:       303-273-3483
Golden, CO  80401          |  Fax:         303-273-3311
Key fingerprint = 5BDE 4DA1 66BE 4F7B BC17  3A0C 932B 7266 1E76 F123