Re: AMD64 64bit vs 32bit speed increase
On Saturday 14 February 2004 12:10 am, C. Scott Ananian wrote:
> But 50% performance improvement on oggenc is really quite impressive.
> Media processing is not exactly the first thing that pops into mind as an
> AMD64 application.
AMD64 gcc generates (scalar) sse for FP code by default whereas ix86 generates
traditional i387 code by default. This could explain the large speedup as the
sse instructions are generaly much more efficient end easier to optimize. It
may be possible to get a similar speedup in the 32-bit code by using gcc
-msse2 -mfpmath=sse2. It's possible that oggenc also has custom sse based
routines which are not enabled when compiled for i386 (the default x86 debian
arch).
Paul
Reply to: