The following plugs possible overflow issues in vn_reference_compute_hash
and possibly in vn_reference_eq. The inchash "integer" adds are a bit
of a mess, but I know overloads with different integer types can get
messy, so not this time. For hashing simply truncate to 64bits.
PR tree-optimization/90242
* tree-ssa-sccvn.cc (vn_reference_compute_hash): Use
poly_offset_int for offset accumulation. For hashing
truncate to 64 bits and also hash 64 bits.
(vn_reference_eq): Likewise.
hashval_t result;
int i;
vn_reference_op_t vro;
- poly_int64 off = -1;
+ poly_offset_int off = -1;
bool deref = false;
FOR_EACH_VEC_ELT (vr1->operands, i, vro)
{
if (maybe_ne (off, -1)
&& maybe_ne (off, 0))
- hstate.add_poly_int (off);
+ hstate.add_poly_hwi (off.force_shwi ());
off = -1;
if (deref
&& vro->opcode == ADDR_EXPR)
j = 0;
do
{
- poly_int64 off1 = 0, off2 = 0;
+ poly_offset_int off1 = 0, off2 = 0;
vn_reference_op_t vro1, vro2;
vn_reference_op_s tem1, tem2;
bool deref1 = false, deref2 = false;