fixup_var_refs trivia

["Zack Weinberg" <zackw at stanford dot edu>]

Some people have been wondering what's become of the fixup_var_refs
performance improvements I promised about a month ago.

I have a printout of the routines that still need to be rewritten on
my floor right now.  In 7 point Courier.

purge_addressof_1 and optimize_bit_field, together, are four and a
half feet long.

fixup_var_refs_insn, find_fixup_replacement, fixup_memory_subreg, and
walk_fixup_memory subreg are a total of three feet long.

fixup_var_refs_1 is just shy of six feet, all by itself.

Incidentally, purge_addressof_1 is presently structured like this:

  if (A)
    four lines
  else if (B && C)
    twenty lines
  else if (D)
    ~200 lines
  else if (B)
    two lines
  else if (A)
    the same four lines that were at the top

zw