https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103699

--- Comment #8 from Petr <kobalicek.petr at gmail dot com> ---
My only problem is that A returns a different value compared to B, C, and D:

uint32_t test_u32_a() {
    char array[16] {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
    writeU64be(array + 6, 0xAABBCCDDEEFF1213);
    return readU32be(array + 7);
}

uint32_t test_u32_b() {
    static char array[16] {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
    writeU64be(array + 6, 0xAABBCCDDEEFF1213);
    return readU32be(array + 7);
}

uint32_t test_u32_c() {
    thread_local char array[16] {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
    writeU64be(array + 6, 0xAABBCCDDEEFF1213);
    return readU32be(array + 7);
}

uint32_t test_u32_d(char array[16]) {
    writeU64be(array + 6, 0xAABBCCDDEEFF1213);
    return readU32be(array + 7);
}

And when you compile this, you would actually see that ALL functions evaluate
to a constant (because it's known what the output will be), but only in A case
the constant is different (of course because B, C, D have side effects):

test_u32_a():
        mov     eax, 117967114
        ret
test_u32_b():
        movabs  rax, 1374442237689904042
        mov     QWORD PTR test_u32_b()::array[rip+6], rax
        mov     eax, -1144201746
        ret
test_u32_c():
        movabs  rax, 1374442237689904042
        mov     QWORD PTR fs:test_u32_c()::array@tpoff+6, rax
        mov     eax, -1144201746
        ret
test_u32_d(char*):
        movabs  rax, 1374442237689904042
        mov     QWORD PTR [rdi+6], rax
        mov     eax, -1144201746
        ret

So yeah, we can talk about breaking strict aliasing here, but it's just
inconsistent. I would just expect all functions return the same value.

Reply via email to