https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119293
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|pinskia at gcc dot gnu.org |unassigned at gcc dot
gnu.org
Status|ASSIGNED |NEW
CC| |hubicka at gcc dot gnu.org
--- Comment #10 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I suspect if we fix this, it will fix a lot more of the "Invalid sum"s that
show up.
I think we need to prop the costs on both sides for the successor until we end
up in the same bb but how to handle loops. I don't know.
Also vect produces some odd code which it could prove as false:
We have:
```
_28 = (unsigned int) end_11(D);
_27 = (unsigned int) start_9(D);
_26 = _28 - _27;
_25 = _26 + 4294967295;
_24 = _25 > 2;
if (_24 != 0)
````
And:
```
_63 = (unsigned int) end_11(D);
_64 = (unsigned int) start_9(D);
_65 = _63 - _64;
_66 = _65 + 4294967295;
if (_66 <= 2)
```
This is how fre4 is able to figure out `_66 <= 2` is false.
Honza,
So the problem is when we have say `if (a) goto b[75%]; else goto c[25%];`
and there are a lot of blocks between the common parts until sucessors of b and
sucessors of c meet. and we have found that a is either true or false. There
seems like we have no way to update the counts on those 2 pathes.
Which is why we end up with "Invalid sum" in this case.
