This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: C Compiler benchmark: gcc 4.6.3 vs. Intel v11 and others
- From: Uros Bizjak <ubizjak at gmail dot com>
- To: GCC Development <gcc at gcc dot gnu dot org>
- Cc: Richard Guenther <rguenther at suse dot de>, "willus.com" <willus dot com at willus dot com>
- Date: Thu, 19 Jan 2012 15:20:39 +0100
- Subject: Re: C Compiler benchmark: gcc 4.6.3 vs. Intel v11 and others
Hello!
> > For the math functions, this is normally more a libc feature, so you might
> > get very different results on different OS. Then again, by using
> > -ffast-math, you allow the math functions to return any random value, so I
> > can think of ways to make it even faster ;-)
>
> Also for math functions you can simply substitute the Intel compilers one
> (GCC uses the Microsoft ones) by linking against libimf. You can also make
> use of their vectorized variants from GCC by specifying -mveclibabi=svml
> and link against libimf (the GCC autovectorizer will then use the routines
> from the Intel compiler math library). That makes a huge difference for
> code using functions from math.h.
However, even with -mveclibabi=svml, we still don't vectorize the call
to sincos ...
>From [1], it looks that the transcendental performance mostly depends
on math library code. I wonder if libimf handles calls with AVX
arguments to vectorized functions, so we can extend libimf support
code in gcc.
[1] http://willus.com/ccomp_benchmark2.shtml?p19
Uros.