https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97144
rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2020-12-30
Ever confirmed|0 |1
--- Comment #1 from rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org>
---
Confirmed. Seems to be a bug in constrain_operands
(which was a surprise, given how much of a workhorse it is).
Alternative 5 of adddi3_poly_1 allows (and requires)
a matching operand but alternative 6 has an earlyclobber.
When enforcing the earlyclobber, constrain_operands tries
to check whether a match is explicitly allowed for the pair
of operands that it's considering, but it wrongly picks up
the matching constraint for alternative 5.
As a result, postreload makes the invalid transformation from:
(set (reg/f:DI 0 x0 [orig:96 _15 ] [96])
(plus:DI (reg/f:DI 2 x2 [110])
(const_poly_int:DI [32, 16])))
to:
(set (reg/f:DI 0 x0 [orig:96 _15 ] [96])
(plus:DI (reg/f:DI 0 x0 [139])
(const_poly_int:DI [48, 16])))