https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110789
--- Comment #6 from Peter Lezoch <pledr at gmx dot net> ---
In the passed days, I have further investigated the issue:
the backtrace leads to: 0xba137d real_from_string(real_value*, char const*)
../../gcc/real.cc:2110 which is the following line
of code:
--> inexact = mpfr_strtofr (m, str, NULL, 10,
MPFR_RNDZ);
--> MPFR -- GNU Multiple Precision Floating-Point
Reliable Library
so I rebuilt mpfr-4.2.0 and found in the configure step a strong connection
to
--> GMP -- GNU Multiple Precision Arithmetic
Library (GMP)
In
https://www.linuxfromscratch.org/lfs/view/stable-systemd/chapter08/gmp.html
I found TWO remarkable NOTEs:
----------------------------------------------------------
Note
The default settings of GMP produce libraries optimized for the host processor.
If libraries suitable for processors less capable than the host's CPU are
desired, generic libraries can be created by running the following:
cp -v configfsf.guess config.guess
cp -v configfsf.sub config.sub
----------------------------------------------------------
and
----------------------------------------------------------
Caution
The code in gmp is highly optimized for the processor where it is built.
Occasionally, the code that detects the processor misidentifies the system
capabilities and there will be errors in the tests or other applications using
the gmp libraries with the message "Illegal instruction". In this case, gmp
should be reconfigured with the option --build=x86_64-pc-linux-gnu and rebuilt.
----------------------------------------------------------
BINGO !!!!
So rebuilding of gmp-6.2.1 with above options and precautions and
rebuilding of mpfr-42.0 and
rebuilding of gcc-12.2 without --disable-avx2
solved all the observed issues !!
So thanks to all for there efforts !
>From my point of view the Bug 110789 is solved and can be closed !!!
