https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118804
--- Comment #4 from Andrew Macleod <amacleod at redhat dot com> ---
Range isn't getting it because its not an equivalency, but rather an
unquivalence.
=========== BB 4 ============
x_3(D) [irange] int VARYING
Relational : (x_3(D) != y_4(D))
<bb 4> :
if (x_3(D) == 999)
goto <bb 5>; [INV]
else
goto <bb 7>; [INV]
=========== BB 5 ============
Imports: y_4(D)
Exports: y_4(D)
y_4(D) [irange] int VARYING
<bb 5> :
if (y_4(D) == 999)
goto <bb 6>; [INV]
else
goto <bb 7>; [INV]
It knows that x_3 != y_4, and we also know that x_4 is [999, 999] in BB5.. but
there is nothing to indicate that when evaluating y_4 == 999, we should look at
the value of x_3.
Its too expensive to check all relations every time. Well, it use to be, I
will run an experiment to see what the impact is there days. I had also
experimented with querying the most recent relation for info.
Bottom line is a relation query needs to be made to see if there is an
equivalence or other relation with X, and then apply that relation to the range
of X and
see if it has any impact on the range of Y
.
The question is how to do it efficiently as functions gt larger.
A query needs to be made if there is a relation (find x_3 != y_4)
Then a query to be made for the range of the other part of the relation (x_3)
The apply that relation& range to y_4 and see if it makes any difference.
