[PING] Optimize manual byte swap implementations using the GCC bswap builtin

[Richard Guenther]

On Fri, Jan 9, 2009 at 9:01 AM, Andreas Krebbel
<krebbel@linux.vnet.ibm.com> wrote:
> Hello,
>
> any opinions about this one:
> http://gcc.gnu.org/ml/gcc-patches/2008-11/msg01439.html

How about building (accumulated) bit-transfer matrices starting off
interesting statements by simulating them and then
base the recognition on them?

That may even work to optimize things the SSE pack/unpack/shuffle
sequences in general (and thus as a framework be useful for the
vectorizer?)

Richard.