Re: Forced alignment of arrays

["Richard Guenther" <richard dot guenther at gmail dot com>]

On 5/5/07, Mark Shinwell <shinwell@codesourcery.com> wrote:
> I'd like to test the water on the patch below, which was developed
> internally against a 4.2-based branch.  It introduces a new flag that
> forces the minimum alignment of array variables to be the largest power
> of two less than or equal to their total storage size, or the biggest
> alignment used on the machine, whichever is smaller.
>
> This patch was developed in response to a request from a customer who
> has a large body of code using type punning in non-strictly-conforming
> (in the sense of the C standard) ways such as this:
>
>         char foo[4];
>         *(int *) foo = 42;
>
> Use of the new -falign-arrays option clearly does not guarantee that
> all code using dubious type punning will work; it just provides
> alignment guarantees about the start of arrays.
>
> Note that this patch requires that callers of DATA_ALIGNMENT and
> LOCAL_ALIGNMENT now go via calculate_data_alignment and
> calculate_local_alignment respectively rather than using the macros
> directly.
>
> Is the consensus that this patch, suitably tested on mainline, is
> acceptable -- or not?

I don't think we want such a thing.  People will use it for things it was
not designed for (like forcing alignment for vectorizing, trying to force
alignment on stack locals, etc.).

Richard.