Infrastructure for MEM tracking

[Richard Kenner]

    Nice I was quite impatiently waiting for this patch as I would like to
    do some working that depdends on this bits.  Thanks!

Note that I'll be rapidly adding new functionality to this in the next
several days.

    But I do have different question.  How much work would be to make your
    approach to work for hard registers too?  I would like to track some
    information after reload (currently we do track the original pseudo,
    but I would like to track original pseudo and the offset when it is
    split to multiple registers, so we can update the debug information
    after register renaming or live range splitting is done).

We could do something similar, and replace the original_regno field with
a datastructure that contains that and the offset, I suppose.