https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69609
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |stevenb.gcc at gmail dot com
--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
Top offender is now
df live&initialized regs: 12.06 (21%) usr 0.00 ( 0%) sys 12.10 (21%) wall
0 kB ( 0%) ggc
PRE : 15.53 (27%) usr 0.29 (26%) sys 15.83 (27%) wall
793 kB ( 0%) ggc
with perf reporting
11.95% cc1 cc1 [.] df_live_bb_local_compute(unsigned int)
6.69% cc1 cc1 [.] compute_transp(rtx_def const*, int,
simple_
4.44% cc1 cc1 [.] df_worklist_dataflow(dataflow*,
bitmap_head
3.10% cc1 cc1 [.]
df_simulate_one_insn_forwards(basic_block_d
2.99% cc1 cc1 [.] bitmap_set_bit(bitmap_head*, int)
I suspect the DF part is PRE as well. PRE already has some limiting on
CFG complexity that appearantly doesn't trigger here. See
gcse_or_cprop_is_too_expensive. OTOH we still call df_analyze before
checking that (oops?!).
