FACTS instead of FUD about gcc3

johan j.h.veenstra at stud.tue.nl
Wed May 8 09:29:50 PDT 2002


Ian Molton wrote:

> On Wed, 8 May 2002 16:41:59 +0200
> Matthias Benkmann <matthias at winterdrache.de> wrote:
> 
>> After reading a 2nd-hand account about GCC3's 20% performance
>> regression I thought you might be interested in the actual facts.
> 
> Er. theres nothing below I didnt mention.
> 
> And bear in mind that aside from media processing, 90% or more
> operations are integer based.
> 
> Its great that the problem can be worked around, but we shouldnt use
> something until the problems are FIXED, or we should provide patches
> for all the regressions it causes in other packages (unrealistic).

As it turns out, the performance regression had nothing to do with 
integer based operations, it was all about not-inlining memset() from 
libc: 

M. Lespinasse: "OK, so I worked more to find the cause of the 
slowdown, and I figured out its all because of memset(). This 
function seems to be about twice slower than in 2.95"

Without this inlining problem:

M. Lespinasse: "now that I've worked around the two issues I had with 
inlining and with memset, the 3.1 snapshot does provide superior 
performance on my libmpeg2 codebase"

So:

J. Hubicka: "I will send a patch"

All this happened around the 24th of April, so you can expect it to 
be in CVS for some time now.

Johan Veenstra

-- 
Unsubscribe: send email to listar at linuxfromscratch.org
and put 'unsubscribe lfs-dev' in the subject header of the message



More information about the lfs-dev mailing list