better performance on 64-bit mingw

[Timothy C Prince]

-----Original Message-----
From: "Timothy C Prince" <tprince@myrealbox.com>
To: fortran@gcc.gnu.org
Date: Tue, 04 Dec 2007 18:16:38 +0000
Subject: better performance on 64-bit mingw

I noticed that eoshift, pack, and unpack run several times as fast on mingw 64-bit, compared with cygwin 32-bit or linux 32 and 64-bit (on Core 2 Duo)..  Profiling on linux 64-bit, I see the time is spent in non-optimized memcpy().  Is this a recognized finding?
Maybe it won't be so bad on other CPUs.  I will try to see if it is possible to build libgfortran with another memcpy().
________________________-
The performance is gained simply by making a local copy of memcpy() built by gcc -O3 -funroll-loops. Any vectorizing compiler will do.
Tim Prince