> povray is compiled with 
> march=i586. This is why the official binary is so slow : it doesn't take 
> advantage of SSE/SSE2/3DNow2 instructions which are really faster than 
> standards ones

	To take advantage of such specialized instructions, you have to
make your program _use_ them, which is not the case in povray because its
code is mostly doing (double precision) floating-point arithmetics.

> If you have an athlon thunderbird and compile povray with -march=athlon 
> instead of -march=i586, you'll get a povray running about 30% faster 
> than the original one.

	Please report results of running benchmark.ini with a version
compiled using -march=i586 and another with -march=athlon, using the
same compiler of course.

 > [snip]
> the CXXFLAGS line should be replaced by CXXFLAGS="${CFLAGS}"
> That's all !

	"Thanks for teaching" (private joke)  :-)

	- NC

Post a reply to this message