https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85072
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |compile-time-hog,
| |memory-hog, ra
Target| |x86_64-*-*
Status|UNCONFIRMED |NEW
Last reconfirmed| |2018-03-26
CC| |vmakarov at gcc dot gnu.org
Component|c++ |rtl-optimization
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. It's LRA not finishing lra_assign () but eating memory in this
loop:
#2 0x0000000000a2e7f1 in init_live_reload_and_inheritance_pseudos ()
at /space/rguenther/src/svn/gcc-7-branch/gcc/lra-assigns.c:436
436 bitmap_set_bit (&live_reload_and_inheritance_pseudos[p], i);
(gdb) l
431
&live_reload_and_inheritance_pseudos_bitmap_obstack);
432 for (i = lra_constraint_new_regno_start; i < max_regno; i++)
433 {
434 for (r = lra_reg_info[i].live_ranges; r != NULL; r = r->next)
435 for (p = r->start; p <= r->finish; p++)
436 bitmap_set_bit (&live_reload_and_inheritance_pseudos[p], i);
437 }
with
(gdb) p lra_constraint_new_regno_start
$2 = 505484
(gdb) p max_regno
$3 = 1010691
and just an example for one iteration:
(gdb) p r->start
$4 = 149038
(gdb) p r->finish
$5 = 757257
(gdb) p lra_live_max_point
$7 = 1013502
Looks like representing live_reload_and_inheritance_pseudos explicitely
for a testcase like this is a bad idea.
Vlad? Is there a way to make LRA cheaper at -O1, like turning off
inheritance? Note the basic-blocks in this testcase are huge.
