https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87543
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2018-10-08
CC| |rguenth at gcc dot gnu.org
Version|unknown |8.2.1
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue at -O2 is that we only unroll loops if we estimate that to not grow
code-size. For f2 the estimate "works" while for f1 it doesn't because of
the awkward IV structure PRE present us with:
Estimating sizes for loop 1
BB: 3, after_exit: 0
size: 1 __init_12 = prephitmp_8 + __init_19;
size: 1 __first_13 = __first_18 + 4;
size: 1 ivtmp_10 = ivtmp_11 - 1;
Induction variable computation will be folded away.
size: 2 if (ivtmp_10 == 0)
Exit condition will be eliminated in peeled copies.
Exit condition will be eliminated in last copy.
Constant conditional.
BB: 4, after_exit: 1
size: 1 pretmp_17 = MEM[(int *)__first_18 + 4B];
size: 6-3, last_iteration: 5-3
Loop size: 6
Estimated size after unrolling: 9
Not unrolling loop 1: size would grow.
<bb 3> [local count: 955630224]:
# __first_18 = PHI <&arr(2), __first_13(4)>
# __init_19 = PHI <0(2), __init_12(4)>
# prephitmp_8 = PHI <1(2), pretmp_17(4)>
# ivtmp_11 = PHI <5(2), ivtmp_10(4)>
__init_12 = prephitmp_8 + __init_19;
__first_13 = __first_18 + 4;
ivtmp_10 = ivtmp_11 - 1;
if (ivtmp_10 == 0)
goto <bb 5>; [12.36%]
else
goto <bb 4>; [87.64%]
<bb 4> [local count: 837518623]:
pretmp_17 = MEM[(int *)__first_18 + 4B];
goto <bb 3>; [100.00%]
<bb 5> [local count: 118111601]:
