https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80960
--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> ---
With --param max-combine-insns=2 compile-time and memory use at -O1 shrinks
considerably for me (still too high IMHO).
-O1 --param max-combine-insns=2: 70.48user 0.63system 1:11.12elapsed 99%CPU
(0avgtext+0avgdata 2385992maxresident)k
-O1: 167.06user 0.96system 2:48.04elapsed 99%CPU (0avgtext+0avgdata
3401780maxresident)k
RTL expansion shows the biggest BB having ~70000 instructions.
with limiting to 2 insn combinations -ftime-report shows
alias stmt walking : 10.89 (16%) usr 0.10 (12%) sys 10.97 (15%) wall
3 kB ( 0%) ggc
dead store elim2 : 9.69 (14%) usr 0.10 (12%) sys 9.77 (14%) wall
1828350 kB (57%) ggc
combiner : 14.93 (21%) usr 0.37 (43%) sys 15.29 (22%) wall
1155538 kB (36%) ggc
integrated RA : 8.84 (13%) usr 0.03 ( 3%) sys 8.87 (12%) wall
34845 kB ( 1%) ggc
LRA hard reg assignment : 14.80 (21%) usr 0.07 ( 8%) sys 14.91 (21%) wall
0 kB ( 0%) ggc
TOTAL : 70.16 0.86 71.04
3181288 kB
so it's still doing sth (as expected).