RE: optimization/8474: -mcpu=i686 produces much faster code than -mcpu=pentium4 on a P4 machine

The following reply was made to PR optimization/8474; it has been noted by GNATS.

From: Dara Hazeghi <[email protected]>
To: Dave <[email protected]>
Cc: [email protected]
Subject: RE: optimization/8474: -mcpu=i686 produces much faster code than -mcpu=pentium4 on a P4 machine
Date: Mon, 19 May 2003 17:38:50 -0700 (PDT)

 --- Dave <[email protected]> wrote:
 > New results are attached - sorry for the delay (been
 > busy lately).
 
 No problem. It doesn't look like the problem is
 solved, so I guess we'll leave the PR open, and hope
 somebody notices it...
 
 Thanks for your feedback,
 
 Dara
 
 > 
 > These were run on Red Hat 9 (I installed that rather
 > than Mandrake v9.1)
 > rather than the original RH 8 system.
 > 
 > Doesn't seem to be any improvement for P4 or with
 > gcc 3.4.
 > 
 > - Dave
 > 
 > On Wed, 2003-05-14 at 19:43, Dara Hazeghi wrote:
 > > Doh, I forgot to statically link the 3.4 ones (new
 > > version of libstdc++). Just a sec...
 > > 
 > > Here, I statically linked them (the 3.4 ones) this
 > > time.
 > > 
 > > Dara
 > > 
 > whetstone took: 1.05 secs for 952 MFLOPS (w/  math
 > lib)
 > whetstone took: 0.39 secs for 2564 MFLOPS (w/o math
 > lib)
 > 
 > whet33p4
 > whetstone took: 1.04 secs for 962 MFLOPS (w/  math
 > lib)
 > whetstone took: 0.40 secs for 2500 MFLOPS (w/o math
 > lib)
 > 
 > whet34p3
 > whetstone took: 1.05 secs for 952 MFLOPS (w/  math
 > lib)
 > whetstone took: 0.40 secs for 2500 MFLOPS (w/o math
 > lib)
 > 
 > whet34p4
 > whetstone took: 1.06 secs for 943 MFLOPS (w/  math
 > lib)
 > whetstone took: 0.39 secs for 2564 MFLOPS (w/o math
 > lib)
 > 
 > 
 
 __________________________________
 Do you Yahoo!?
 The New Yahoo! Search - Faster. Easier. Bingo.
 http://search.yahoo.com