> >
> > Unless the package has special assembler instructions like Mesa you
> > shouldn't. In a K6/2 (notice this is not a plain K6) code compiled
> > for K6 is slower that code you let alone ie compiled for 386.
> >
>
>
> Jean
> Are you sure about this? I'm using gcc 2.95 which has special options for the
> K6 family.
>
In fact what I told was valid for gcc 2.96. For gcc 2.95 at O2
optimization level and using the Byte benchmarks the code generated
with arch=k6 (optimize for k6, compiler is free to pick k6 specific
instructions) is: 1% slower for memory-intensive tests, 1.5% faster
for integer tests, 2% slower at floating point tests respective to
code targetted at 386. Tests were run on a K6/2 450. In othzer words
either gcc 2.95 does not amke a good job at optimizing specifically
for K6 or the K6/2 is internally different enough from the K6 to make
K6 optimizations ineffective.
--
Jean Francois Martinez
Project Independence: Linux for the Masses
http://www.independence.seul.org
_______________________________________________
Redhat-devel-list mailing list
[EMAIL PROTECTED]
https://listman.redhat.com/mailman/listinfo/redhat-devel-list
