https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85560
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |missed-optimization
Status|UNCONFIRMED |NEW
Last reconfirmed| |2018-04-30
Component|rtl-optimization |middle-end
Summary|Missed optimization in |Missed optimization in
|bit-by-bit variable zeroing |niter analysis for
| |bit-by-bit variable zeroing
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. GCC fails to prove the loop is finite. This is a niter / SCEV
analysis issue. It's a bit convoluted case so I'm not sure how we can handle
this easily. GIMPLE:
<bb 3> [local count: 955630224]:
# a_10 = PHI <a_5(2), a_6(3)>
# bits_11 = PHI <1(2), bits_7(3)>
_1 = ~bits_11;
a_6 = _1 & a_10;
bits_7 = bits_11 << 1;
if (a_6 == 0)
goto <bb 4>; [11.00%]
else
goto <bb 3>; [89.00%]
we do have the force-simulation code which in theory could also be used
to compute an upper bound rather than an exact niter count. But then for
this case it needs to maintain fuzzy lattices (for bits).
