target/6456: builtin memset broken on x86 with -minline-all-stringops

[roger@eyesopen.com]

>Number:         6456
>Category:       target
>Synopsis:       builtin memset broken on x86 with -minline-all-stringops
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Thu Apr 25 12:16:01 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     Roger Sayle <roger@eyesopen.com>
>Release:        mainline CVS 20020425 (probably also 3.1 prerelease)
>Organization:
>Environment:
i686-pc-linux-gnu (RedHat 7.2)
>Description:
The x86 backend is currently generates incorrect code for
builtin memset, when using "-minline-all-stringops".  This
is specific to the ia-32 backend.  Interestingly, GCC uses
different code sequences for "-march=athlon" which doesn't
exhibit the problem.
>How-To-Repeat:
The problem can be reproduced by compiling and running the
memset correctness test, gcc.c-torture/execute/memset-1.c,
with command line options "-O2 -minline-all-stringops".

Once I get a PR number, I'll submit a new gcc.dg testcase
to gcc-patches to check for this problem on i?86-*-*.
>Fix:
For the time being, don't specify "-minline-all-stringops".
>Release-Note:
>Audit-Trail:
>Unformatted: