fortran/9974: [3.4 regression] Test g77.f-torture/noncompile/check0.f fails

[Roger Sayle]

> This regression was introduced between 21:00 and 22:30 UTC on
> 25 Jan 2003 by patch
>
> 2003-01-25  Roger Sayle  <roger@eyesopen.com>
>
>         * gcse.c (bypass_last_basic_block): New global variable.
>         (bypass_block):  Use redirect_edge_and_branch_force to redirect
>         fall-through edges.  Use bypass_last_basic_block to determine
>         which blocks have valid PRE information.
>         (bypass_conditional_jumps): Initialize bypass_last_basic_block.

This GNATS PR may take a while to track down.  The problem is a
miscompilation of f771 itself on mips.  Hence the failure doesn't
show up in either a cross-compiler to mips-sgi-irix6.5 compiler,
or in the native f771 built with the MIPSPro compilers or a
non-mainline GCC.

Worse is that there were no new C regressions on mips-sgi-irix6.5
when the above patch was introduced, (according to the test results
mailing list) so there isn't a simple test failure that can be
analysed.

Worse still is that the gdb and dbx debuggers on IRIX leave something
to be desired, so I'm currently left trying to debug f771 using no
more than printf, bootstrapping each iteration, on a painfully slow
SGI machine.

The errant code could be anywhere, and its just the reorganization
of the control flow graph by my "aggressive jump bypassing" patch
that has exposed it within the fortran front-end.


I'm still working on it, but I doubt I'll be able to provide my
usual 24 hour turn-around on this one.  Hope to have better news
soon.

Roger
--
Roger Sayle,                         E-mail: roger@eyesopen.com
OpenEye Scientific Software,         WWW: http://www.eyesopen.com/
Suite 1107, 3600 Cerrillos Road,     Tel: (+1) 505-473-7385
Santa Fe, New Mexico, 87507.         Fax: (+1) 505-473-0833