Specify alignment for string literal.

Specify alignment for string literal.

[Kaz Kylheku]

Hi all,

Is there any way to specify the alignment of a literal, with 
__attribute__
((align ...))

That is to say:

/* wchar_t is 16 bits and 2 byte aligned on this
   platform; I want 4 byte alignment */

wchar_t *foo = L"...";

The attribute has to go on the literal object, not on foo, obviously.

If there is such a feature, it's not obviously documented in the
manual.

I have other ways of solving the problem I'm working on, but this
would be ideal.

Re: Specify alignment for string literal.

[Ian Lance Taylor]

Kaz Kylheku <kaz@kylheku.com> writes:

> Is there any way to specify the alignment of a literal, with
> __attribute__
> ((align ...))

Not as such, no.



> That is to say:
>
> /* wchar_t is 16 bits and 2 byte aligned on this
>   platform; I want 4 byte alignment */
>
> wchar_t *foo = L"...";
>
> The attribute has to go on the literal object, not on foo, obviously.

Perhaps you could write

static wchar_t foo_str[] __attribute__ ((aligned (4))) = L"...";
wchar_t *foo = foo-str;

Ian

Re: Specify alignment for string literal.

[Kaz Kylheku]

On Sun, 09 Oct 2011 22:11:29 -0700, Ian Lance Taylor <iant@google.com>
wrote:
> Kaz Kylheku <kaz@kylheku.com> writes:
>> /* wchar_t is 16 bits and 2 byte aligned on this
>>   platform; I want 4 byte alignment */
>>
>> wchar_t *foo = L"...";
>>
>> The attribute has to go on the literal object, not on foo, obviously.
> 
> Perhaps you could write
> 
> static wchar_t foo_str[] __attribute__ ((aligned (4))) = L"...";
> wchar_t *foo = foo-str;
> 
> Ian

Hi Ian,

That's not a bad suggestion. In many cases it could be wrapped
with a macro. E.g.

  aligned_literal("abc")

produces a GCC evaluation block:

 ({ static wchar_t f_o_o[] .... = L"abc"; f_o_o; })

The approach I went with is now documented here:

http://git.savannah.gnu.org/cgit/txr.git/tree/HACKING

In section 2.4.2

I'm adding tag bits to string literals so they can look like
objects in a system of dynamic types.

Turns out we can add a two-bit tag to a pointer even if it
is only two byte aligned, if we can ensure that no two such
objects land into the same four-byte word, and if we have a
way to find the real start of the object after ripping off
the tag.

Since the objects are C strings, it's easy to achieve both
goals with null padding.

Cheers ...