optimizations for 3x speedup in ld

[Robert O'Callahan]

I work on Mozilla/Firefox and spend a lot of time linking libgklayout.so
(93MB with debug info). I did some work to speed up ld on this workload
and got a 3x improvement. See here for details:
http://weblogs.mozillazine.org/roc/archives/2005/02/optimizing_gnu.html
I've taken some care to make sure that the changes are localized and
there should be absolutely no impact on output (e.g., sections are
processed in exactly the same order as before). I reran the regression
tests and I pass exactly the same set of tests that CVS HEAD passes
(which actually means 4 unexpected failures ... this is x86-64).

So ... how do I push this upstream? Should I submit the patch as an
attachment? It's 69 lines deleted, 458 lines added in ld.h, ldlang.c,
ldlang.h.

Thanks,
Rob