[C++-sig] [ANN] arbitrary arity support / new callback interface
Toon Knapen
toon.knapen at si-lab.org
Fri Apr 12 15:46:38 CEST 2002
> Note that some of us (numeric crunchers) are stuck to 2.95.x because, for
> example, gcc-3.x.x built ATLAS library is known to be slower to when using
> 2.95.x and therefore 2.95.x compiler will not be obsolete for some time.
I did not know it had such a big effect on ATLAS (thanks for the info).
gcc-3.0.3 should be overall faster (e.g. due to type-based alias analysis),
but still is less performant on loop-unrolling etc. In my last simple test,
gcc-3.0.4 almost obtained similar performance as 2.95.3 on numerical codes
to. So I guess 3.1 should definitly (hopefully) make it obsolete.
BTW, since Intel is stopping the KAI Compiler, Los Alomos apparantly is
heavily interested again in gcc which can only boost gcc's performance (heard
it on the POOMA ml)
> I am far from asking full support for gcc-2.95.x of BPL as on can always
> keep two compilers around for different software but it is a bit
> inconvinient...
I also have 2.95.3 still around for numerical performance reasons. But for
this area, exception handling is'nt super important anyway.
t
More information about the Cplusplus-sig
mailing list