[Bug fortran/65450] New: [5.0 Regression]: Unaligned access with -O3 -mtune=k8

[ubizjak at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65450

            Bug ID: 65450
           Summary: [5.0 Regression]: Unaligned access with -O3 -mtune=k8
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ubizjak at gmail dot com

The compiler generates unaligned access for Polyhedron channel.f90 test when
compiled with -O2 -mtune=k8:

/home/uros/gcc-build/gcc/gfortran -B/home/uros/gcc-build/gcc
-B/home/uros/gcc-build/x86_64-unknown-linux-gnu/libgfortran/ 
-B/home/uros/gcc-build/x86_64-unknown-linux-gnu/libquadmath/.libs
-L/home/uros/gcc-build/x86_64-unknown-linux-gnu/libgfortran/.libs
-L-L/home/uros/gcc-build/x86_64-unknown-linux-gnu/libquadmath/.libs -O2
-ffast-math -mtune=k8 channel.f90

LD_LIBRARY_PATH=/home/uros/gcc-build/gcc:/home/uros/gcc-build/x86_64-unknown-linux-gnu/libgfortran/.libs:/home/uros/gcc-build/x86_64-unknown-linux-gnu/libquadmath/.libs
./a.out

...
Program received signal SIGSEGV: Segmentation fault - invalid memory reference.

Backtrace for this error:
#0  0x2AC9058B3C27
#1  0x2AC9058B2E20
#2  0x31CDC3002F
#3  0x402AFB in ddx at channel.f90:247
Segmentation fault

(gdb) r
...
Program received signal SIGSEGV, Segmentation fault.
0x0000000000402afb in ddx (array=..., __result=...) at channel.f90:247
247         ddx(2:I-1,1:J) = array(3:I,1:J)-array(1:I-2,1:J)    ! interior
points

(gdb) bt
#0  0x0000000000402afb in ddx (array=..., __result=...) at channel.f90:247
#1  sw () at channel.f90:148
#2  0x0000000000404cfd in main (argc=<optimized out>, argv=<optimized out>) at
channel.f90:234
#3  0x00000031cdc1d9f4 in __libc_start_main () from /lib64/libc.so.6
#4  0x0000000000400909 in _start ()

(gdb) disass
   ...
   0x0000000000402acc <+7644>:  movaps %xmm4,-0x20(%rdx)
   0x0000000000402ad0 <+7648>:  movlpd -0x10(%rax),%xmm0
   0x0000000000402ad5 <+7653>:  movhpd -0x8(%rax),%xmm0
   0x0000000000402ada <+7658>:  movapd %xmm0,%xmm4
   0x0000000000402ade <+7662>:  subpd  %xmm3,%xmm4
   0x0000000000402ae2 <+7666>:  movaps %xmm4,-0x10(%rdx)
   0x0000000000402ae6 <+7670>:  cmp    $0xf4,%rdi
   0x0000000000402aed <+7677>:  jne    0x402a71 <sw+7553>
   0x0000000000402aef <+7679>:  lea    0x20(%rcx),%rdi
   0x0000000000402af3 <+7683>:  add    $0x20,%rax
   0x0000000000402af7 <+7687>:  add    $0x20,%rdx
=> 0x0000000000402afb <+7691>:  movapd -0x20(%rax),%xmm3
   0x0000000000402b00 <+7696>:  mov    $0xf6,%r11d
   0x0000000000402b06 <+7702>:  xor    %ecx,%ecx
   0x0000000000402b08 <+7704>:  movapd %xmm3,%xmm4
   0x0000000000402b0c <+7708>:  subpd  %xmm0,%xmm4
   ...

(gdb) p/x $rax
$3 = 0x8c5a38

The failure can also be seen on SUSE gcc tester [1].

[1] http://gcc.opensuse.org/c++bench/polyhedron/polyhedron-performance-latest