Hi,
Another thing I noticed is that df_note_* code shows quite top in the
profiles. This patch micro optimize two quite obvious things I noticed
while glancing over the code.
One thing I do not understand is all the logic about removing REG notes
and adding them back.
We already manage freelist for notes so just removing them to freelist
and constructing again should be fine and actually faster since the
code is trying to match old ones for no apparent reason?
Bootstrapped/regtested x86_64-linux, OK?
Honza
* df-problems.c (df_create_unused_note, df_note_bb_compute):
micro-optimize the checks when to add new note.