[Bug tree-optimization/82518] [8 regression] gfortran.fortran-torture/execute/in-pack.f90 fails on armeb since r252917

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82518

Aldy Hernandez <aldyh at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2018-02-09
     Ever confirmed|0                           |1

--- Comment #40 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
*sigh*  Confirmed with:

$ sudo yum install qemu
$ git clone https://git.linaro.org/toolchain/abe.git
$ (cd abe; git checkout stable)
$ mkdir build-cortex-a9
$ cd build-cortex-a9
$ ../abe/configure
$../abe/abe.sh --target armeb-linux-gnueabihf --set languages=fortran --build
gcc,glibc,binutils gcc=gcc.git~master --set cpu=cortex-a9

I avoided --build all because gdb fails to build with --set cpu=cortex-a9.

Then I ran tests like this:

../abe/abe.sh --target armeb-linux-gnueabihf --set
runtestflags=execute.exp=in-pack.f90 --build gcc --check gcc 
gcc=gcc.git~master --disable update

which actually passed, because there doesn't seem to be a way to pass to pass
-mfpu= to either the abe.sh configury, or the "runtestflags=" line.

So... then you do:

$ find . -name gfortran.log
./builds/x86_64-unknown-linux-gnu/armeb-linux-gnueabihf/gcc.git~master-stage2/gcc/testsuite/gfortran/gfortran.log

and fish out the gcc command line to run manually on the command line with
-mcpu=cortex-a9 -mfpu=neon-fp16 -O3 -o i-wanna-kill-myself.exe

And finally:

$ qemu-armeb -cpu any -R 0 -L
/home/cygnus/aldyh/bld/arm-hell/build-cortex-a9/sysroots/armeb-linux-gnueabihf
i-wanna-kill-myself.exe
Program aborted. Backtrace:
qemu: uncaught target signal 6 (Aborted) - core dumped

I've verified with -O2 or without -mfpu=neon-fp16 succeeds, so the test isn't
just dying for all flags :).

Sigh (have I said that before?).  I will now see if I can reduce the test and
see where we are dying.

p.s. gcc.git~master directory seems to have a pretty up to date upstream GCC
(as of today). I'll double check dropping in an actual FSF gcc doesn't change
the results.