[RC5] Linux OGR code not optimized?

Peter Cordes peter at llama.nslug.ns.ca
Mon Sep 25 03:35:59 EDT 2000


On Sat, Sep 23, 2000 at 12:14:06PM -0700, Chase Covello wrote:
> I compiled the test client and tested it.
> My OGR rate is:
> [Sep 23 19:10:01 UTC] Benchmark for OGR core #0 (GARSP 5.13)
>                       0.00:00:16.09 [4,905,396.20 nodes/sec]
> 
> [This is version 2.8008-459-CTR-00050300 for Linux (Linux 2.4.0-test8)]
> 
> The official client says:
> [Sep 23 18:53:59 UTC] Benchmark for OGR core #0 (GARSP 5.13)
>                       0.00:00:16.07 [4,322,121.09 nodes/sec]
> 
> [Version 2.8010-463-CTR-00071321 for Linux (Linux 2.4.0-test8)]
> 
> This tells me either:
> 1. The old code is faster, or

 Notice the nodes/sec numbers.  Your newly compiled version is reporting
4.9Gnodes/sec, but the standard client reports 4.3Gnodes/sec.  The benchmark
dynamically adjusts its workload to run for 16 seconds.

> 2. I'm using a different gcc (mine is 2.95.3)
>
> (BTW, I experimented with different levels of optimization, and they are slower
> than -O2)

 This sometimes happens, because some of gcc's optimizations turned on by higher
-Ox levels don't do so well on the register-starved IA32 architecture.  Of
course, it is always best to try different flags and keep the best results.

 I would like to see a closed-source client that could fetch and send
blocks, and handles all the other non-crunching duties, while an open source
program loads and crunches blocks.  That way, people could choose the best
compile flags for their particular machine, and take advantage of new
compiler releases.  I think this makes it easy to do the kind of messing
around that having closed source clients is intended to minimize, though.
Damn.

-- 
#define X(x,y) x##y
Peter Cordes ;  e-mail: X(peter at llama.nslug. , ns.ca)

"The gods confound the man who first found out how to distinguish the hours!
 Confound him, too, who in this place set up a sundial, to cut and hack
 my day so wretchedly into small pieces!" -- Plautus, 200 BCE

--
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest



More information about the rc5 mailing list