This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: SSE2 benchmarks
- To: "Paolo Carlini" <pcarlini at unitus dot it>, <gcc at gcc dot gnu dot org>
- Subject: Re: SSE2 benchmarks
- From: "Tim Prince" <tprince at computer dot org>
- Date: Sat, 30 Jun 2001 13:54:24 -0700
- Cc: <jh at suse dot cz>, <dnovillo at redhat dot com>
- References: <3B3CA9C2.42FB2E9B@unitus.it>
----- Original Message -----
From: "Paolo Carlini" <pcarlini@unitus.it>
To: <gcc@gcc.gnu.org>
Cc: <jh@suse.cz>; <dnovillo@redhat.com>
Sent: Friday, June 29, 2001 9:16 AM
Subject: SSE2 benchmarks
> Hi all,
>
> weeks ago someone expressed interest in the SSE2 code generation
> capabilites of GCC3.x...
>
> Now I'm reading this interesting paper:
>
> http://www.aceshardware.com/Spades/read.php?article_id=40000189
>
> Would'nt be nice to redo those public domain benchmarks (called
"Flops")
> with GCC3.1 experimental ??
>
>
> Cheers,
> Paolo.
>
>
> P.S. I should point out that I don't have an SSE2 processor myself, so
I
> can only listen at other people results...
>
That article scoffs at loops involving branches, but it happens that the
gcc-3.1 -msse2 option is doing extremely well on certain cases involving
comparison and branches, as well as on some cases involving sum
reductions. More than one person has jumped into assuming that
gcc -msse2 includes vectorization, but up to now it uses only serial
sse2 instructions.