Re: PR 6212

[kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)]

    I think the following will do the job.  It adds the alignment
    based on the type of the dereference, not based on the form of
    the index.

That's somewhat of a change in strategy from the way it's been done,
but I agree it's best we can do now.  For 3.2, I'd like to do all the
computations based on expressions, not types, but this seems like a
good compromise for now.

    Bar one, the other uses of highest_pow2_factor are simple field
    offsets, and so do not need adjustment.  

You have the issue of nestings of field and array references and the
tracking of the alignments through all that.  I can't prove it, but I
think what you are doing will catch most of the common case.

    but it's for sure better to guess low than high when talking about
    alignment.

Agreed.