https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122502
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |amacleod at redhat dot com,
| |rguenth at gcc dot gnu.org
--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The problem is this loop:
```
FOR_EACH_IMM_USE_STMT (use_stmt, imm_iter, rslt)
{
gimple_stmt_iterator gsi = gsi_for_stmt (use_stmt);
if (!stmt_can_throw_internal (cfun, use_stmt)
&& fold_stmt (&gsi, follow_all_ssa_edges))
update_stmt (gsi_stmt (gsi));
}
```
vs this loop (with the same SSA_NAME, s4_31):
```
// Loop over each immediate use and see if it has an inferred range.
FOR_EACH_IMM_USE_FAST (use_p, iter, name)
{
gimple *s = USE_STMT (use_p);
gimple_infer_range infer (s, m_query);
for (unsigned x = 0; x < infer.num (); x++)
{
if (name == infer.name (x))
add_range (name, s, infer.range (x));
}
}
```
Which is inside the ranger code.
Which is because the ranger is called from this pattern:
```
(simplify
(cmp (nop_convert? @0) integer_zerop)
(if (tree_expr_nonzero_p (@0))
{ constant_boolean_node (cmp == NE_EXPR, type); }))
```
tree_expr_nonzero_p calls:
```
case SSA_NAME:
if (!INTEGRAL_TYPE_P (TREE_TYPE (t)))
break;
return expr_not_equal_to (t, wi::zero (TYPE_PRECISION (TREE_TYPE (t))));
```
expr_not_equal_to does:
get_range_query (cfun)->range_of_expr (vr, t);
on s4_31 .
I am not going to be able to solve this one by myself.
A few questions here which points to the best way of solving this:
Is calling fold_stmt inside a FOR_EACH_IMM_USE_STMT valid (or is it only valid
if no ranger is turned on; there are a few other places which does this and
might need to change too)?
Or is valid is match pattern ok to call into tree_expr_nonzero_p/ranger?
Or is what the ranger doing wrong and should instead use some other way to
iterate the uses?
