[Bug driver/114046] New: Aarch64: Os,O1,02,O3 optimisations generates unaligned access in short members of structure

[Bug driver/114046] New: Aarch64: Os,O1,02,O3 optimisations generates unaligned access in short members of structure

[stephane at syena dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114046

            Bug ID: 114046
           Summary: Aarch64: Os,O1,02,O3 optimisations generates unaligned
                    access in short members of structure
           Product: gcc
           Version: 13.2.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: driver
          Assignee: unassigned at gcc dot gnu.org
          Reporter: stephane at syena dot net
  Target Milestone: ---

The issuer is reproduced with the following (simplified) code:
struct test {
        short a;
        short b;
        short c;
};

void function( struct test *p, int n ) {
        p->a = n;
        p->b = 0;
        p->c = 0;
}

# gcc -Os -S unaligned_store.c && cat unaligned_store.s
        .arch armv8-a
        .file   "unaligned_store.c"
        .text
        .align  2
        .global function
        .type   function, %function
function:
        strh    w1, [x0]
        str     wzr, [x0, 2]   <<<<<<<<< Alignment fault
        ret
        .size   function, .-function
        .ident  "GCC: (Arm GNU Toolchain 13.2.rel1 (Build arm-13.7)) 13.2.1
20231009"


The second instruction generates an unaligned 32-bits store at a 16-bits
aligned address.
I suspect that the optimizer detects that consecutive short members b and c are
both set to zero and group them to generate a single long access but somehow
fails to detect that the resulting access is not aligned.

[Bug driver/114046] Aarch64: Os,O1,02,O3 optimisations generates unaligned access in short members of structure

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114046

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |INVALID

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
You need to use -mstrict-align here if you are writing (early) firmware where
the page table is not setup and marked as unaligned is ok for normal memory.

If you writing to non-normal memory, you need to use the volatile keyword so
GCC does not produce unaligned memory access in general.