[patch] ld speedup 1/3 (suffix merge)

[Bryce McKinlay]

Hi Guys,

I can report that this patch reduces link time for libgcj.so from over 
1:30 to 9 seconds on an AMD XP 1800! This was actually something I was 
planning on looking into after noticing that functions in merge.c 
accounted for >98% of the linker's runtime, but you guys have beaten me 
to it. Thanks!

I can confirm there is a problem with hash collisions however - libgcj 
has over 1M .debug_str strings. Fitting these all in a hashtable with 
~4000 entries is sure to be inefficient. Sure enough, changing 
sec_merge_init to use bfd_hash_table_init_n with a size of ~50000 
almost halved the link time again to ~4.5 seconds.

Is there a way to work out (or estimate) the number of entries in the 
strings section, and set the size of the hash table appropriately?

Regards

Bryce.