https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70689
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Most likely that just uncovered a latent RA or machine description RA related
bug.
The
(define_insn "*float<SWI48x:mode><MODEF:mode>2_i387"
[(set (match_operand:MODEF 0 "register_operand" "=f")
(float:MODEF (match_operand:SWI48x 1 "nonimmediate_operand" "m")))]
insn allows nonimmediate_operand - thus both MEM and REG before reload, and has
just "m" constraint, so we should force that into memory.
In *.ira that is using REG, which has REG_EQUIV of (plus (frame) (const_int
-4)).
Vlad, could you please have a look at this?
