[Bug rtl-optimization/127889] Excessive compile time at -O3: -fno-inline is 185x faster

[rguenth at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127889

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
And with callgrind update_lives () is shown to cause the majority of
bitmap_clear/set_bit calls.

  for (r = lra_reg_info[regno].live_ranges; r != NULL; r = r->next)
    {
      for (p = r->start; p <= r->finish; p++)
        if (free_p)
          bitmap_clear_bit (&live_hard_reg_pseudos[p], regno);
        else
          {
            bitmap_set_bit (&live_hard_reg_pseudos[p], regno);
            insert_in_live_range_start_chain (regno);
          }

with the main callers being from spill_for and the biggest offender

      /* Restore the live hard reg pseudo info for spilled pseudos.  */
      EXECUTE_IF_SET_IN_BITMAP (&spill_pseudos_bitmap, 0, spill_regno, bi)
        update_lives (spill_regno, false);

I wonder if tracking "spilled pseudos" separately and when querying
live_hard_reg_pseudo[] masking those out would be cheaper than updating
live_hard_reg_pseudo[].


the perf blamed spot in lra_assign is the 2nd hottest part.