This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

gcc-ss-3.5-20040327 / gcc 3.3.3 20040217 / icc on bench++


On Saturday 27 March 2004 13:23, Gerald Pfeifer wrote:
> Yup, there's no typo in the subject.
>
> This morning I created the first GCC 3.5 snapshot from CVS mainline. As
> an added bonus, to make transition for those following GCC 3.4 easier,
> I added diff sets for the latest nine (sic!) GCC 3.4 snapshots.
>

Cool :-))

I just got the snapshot, built it and ran it against bench++-1.1_linux4. 
Additionally I compared it with my systems default compiler gcc 3.3.3 
20040217 (gentoo) and the icc 8.0.55.

For both gcc I built the benchmark using -O3 -march=pentium4 -funroll-loops 
-fprefetch-loop-arrays -ftracer -ffast-math; for icc I used -O3 
-march=pentium4 -mcpu=pentium4 -Ob2 -axWN -ipo

The results ARE interesting! Taking runtime from gcc 3.5 as 100%, on average 
it beats gcc 3.3.3 with 118.40% (min 22.72%, max 633.58%, ) and even icc with 
108.50% (min 0.28%, max 530.05%). Congrats, gcc-team!!!

Only if I take the sum of absolute time (which in my opinion is a bit of 
nonsense) the both gcc are about equal and icc is ahead about 40% - so 
everything where icc is better then gcc, it is REALLY better :-)

Of course, the values differ for every compiler and every test largely (and 
let me dream of a compiler which wins every test +g+).

If anyone is interested in the detailed benchmark logs please send me a mail.

Best regards,
Thomas


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]