https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88723
--- Comment #9 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
> That said, I think this exact UNSPEC should be delegitimized by the target
> hook,
> (define_insn "movsi_pic_gotdata_op"
> [(set (match_operand:SI 0 "register_operand" "=r")
> (unspec:SI [(match_operand:SI 1 "register_operand" "r")
> (match_operand:SI 2 "register_operand" "r")
> (match_operand 3 "symbolic_operand" "")]
> UNSPEC_MOVE_GOTDATA))]
> "flag_pic && check_pic (1)"
> {
> #ifdef HAVE_AS_SPARC_GOTDATA_OP
> return "ld\t[%1 + %2], %0, %%gdop(%a3)";
> #else
> return "ld\t[%1 + %2], %0";
> #endif
> }
> suggests that (unspec:MODE [(reg:MODE) (reg:MODE) (symbol_ref)]) is either
> (mem:MODE (plus:MODE (reg:MODE) (reg:MODE))) or perhaps the third operand
> itself?
Yes, operand #3 is the {LABEL,SYMBOL}_REF that we want to return in this case.
