[Bug middle-end/70871] questionable optimisation in fold-const.c

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70871

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2016-04-29
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Assembler w/o asan:

somefunc:
.LFB0:
        .cfi_startproc
        subq    $24, %rsp
        .cfi_def_cfa_offset 32
        movq    $0, 8(%rsp)
        leaq    8(%rsp), %rdi
        call    call
        movabsq $280379743272960, %rdx
        movq    8(%rsp), %rax
        testq   %rdx, 40(%rax)
        setne   %al
        movzbl  %al, %eax
        addq    $24, %rsp
        .cfi_def_cfa_offset 8
        ret

if my math is correct this loads 8 bytes from &px->i and accesses two stray
bytes after the object.  That's actually fine as alignment guarantees of
the type ensure this won't cross a page boundary.  If we misalign it by
making it packed the folding doesn't apply.

So while the folding is strange it isn't incorrect but it messes with ASAN.