https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83726
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |vmakarov at gcc dot gnu.org
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Wonder if this isn't a LRA bug though, the insn in question is:
(define_insn "*movti_aarch64"
[(set (match_operand:TI 0
"nonimmediate_operand" "=r, w,r,w,r,m,m,w,m")
(match_operand:TI 1
"aarch64_movti_operand" " rn,r,w,w,m,r,Z,m,w"))]
"(register_operand (operands[0], TImode)
|| aarch64_reg_or_zero (operands[1], TImode))"
and aarch64_movti_operand doesn't allow CONST_WIDE_INT:
(define_predicate "aarch64_movti_operand"
(and (match_code "reg,subreg,mem,const_int")
(ior (match_operand 0 "register_operand")
(ior (match_operand 0 "memory_operand")
(match_operand 0 "const_int_operand")))))
and the legitimate_constant_p target hook doesn't allow that either, yet
simplify_operand_subreg will do
else if (CONSTANT_P (reg))
{
/* Try to simplify subreg of constant. It is usually result of
equivalence substitution. */
if (innermode == VOIDmode
&& (innermode = original_subreg_reg_mode[nop]) == VOIDmode)
innermode = curr_static_id->operand[nop].mode;
if ((new_reg = simplify_subreg (mode, reg, innermode,
SUBREG_BYTE (operand))) != NULL_RTX)
{
*curr_id->operand_loc[nop] = new_reg;
return true;
}
}
and nothing calls the predicate nor the legitimate_constant_p target hook for
it.
Is that because LRA operates only on constraints, not predicates, and seeing
"n" defined as:
(define_constraint "n"
"Matches a non-symbolic integer constant."
(and (match_test "CONST_SCALAR_INT_P (op)")
(match_test "!flag_pic || LEGITIMATE_PIC_OPERAND_P (op)")))
doesn't indeed require anything?
If so, shouldn't *movti_aarch64 use a different constraint, that either
requires arbitrary "const_int", so just
(define_constraint "whatever"
"..."
(match_code "const_int"))
or requires a scalar int that satisfies target_legitimate_constant_p?
