On 32-bit HWI host calling
simplify_binary_operation (MINUS, DImode, (reg/v:DI 60 [ x ]), (const_int
-2147483648 [0x80000000]))
returns
(plus:DI (reg/v:DI 60 [ x ])
(const_int -2147483648 [0x80000000]))
which is wrong. I guess we need to avoid the INTVAL wrap in that case.
2041 /* Don't let a relocatable value get a negative coeff. */
2042 if (CONST_INT_P (op1) && GET_MODE (op0) != VOIDmode)
2043 return simplify_gen_binary (PLUS, mode,
2044 op0,
2045 neg_const_int (mode, op1));
and:
66/* Negate a CONST_INT rtx, truncating (because a conversion from a
67 maximally negative number can overflow). */
68static rtx
69neg_const_int (enum machine_mode mode, const_rtx i)
70{
71 return gen_int_mode (- INTVAL (i), mode);
72}
if (GET_MODE_BITSIZE (mode) > HOST_BITS_PER_WIDE_INT), I guess we want to avoid
this "optimization" if op1 != const0_rtx && (unsigned HOST_WIDE_INT) INTVAL
(op1) == -(unsigned HOST_WIDE_INT) INTVAL (op1). Also, I wonder if
neg_const_int we don't need to use - (unsigned HOST_WIDE_INT) INTVAL (i), as
otherwise we e.g. for -ftrapv built gcc we could trap.
--
Summary: simplify_binary_operation_1 MINUS x const_int issue
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: rtl-optimization
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: jakub at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43226
