What to do with new-ra for GCC 4.0

[Richard Kenner]

    To avoid spilling, it must try to satisfy constraints before
    allocating.  I am fairly sure this strategy would also help the
    existing allocator.

But of course, the existing register allocator *is* trying to satisfy
constraints: that's what the whole register class preferencing logic is
all about.

    > and its core algorithm is comprehensible enough 

    I disagree ;-)

The *core* algorithm is indeed quite comprehensible.  It's the add-ons that
basically try to implement a register allocator that are very hard to
comprehend (I can't anymore either).