[tree-ssa] where to fix this?

[Andrew Pinski]

On Dec 19, 2003, at 10:08, Dale Johannesen wrote:

> On Dec 19, 2003, at 8:17 AM, law@redhat.com wrote:
>> In message <978715DA-31C9-11D8-B11A-000A95D7CD40@apple.com>, Dale 
>> Johannesen wr
>> ites:
>>> I have a case where tree-ssa is calling fold() on this:
>>>
>>> <plus_expr 0x9c80a0
>>>     type <integer_type 0x40d42528 int SI
>>>         size <integer_cst 0x8e4780 constant invariant 32>
>>>         unit size <integer_cst 0x8e4798 constant invariant 4>...
>>>     invariant
>>>     arg 0 <integer_cst 0x40d40570 type <boolean_type 0x40d427f8 
>>> _Bool>
>>> constant invariant 1>
>>>     arg 1 <integer_cst 0x97eed0 type <integer_type 0x40d42528 int>
>>> constant invariant 1>>
>>>
>>> which gets into int_const_binop, which thinks the result is _Bool and
>>> truncates it to 0.
>>> Not what we want.  Would the preference be to fix this in fold(), or
>>> make tree_ssa
>>> produce trees that incorporate the C promotions?
>> We'd need more context to know what's happening.   testcases help a 
>> lot :-)
>
> I'm not asking anybody to fix the bug, I will do that, I just want to 
> know what the
> defined semantics of fold() are.  Is is supposed to handle the tree 
> above or not?
>
> This arises from perlbmk in SPEC compiled -O3.  It hangs at runtime 
> and I'm surprised
> nobody else has seen it.  The original code was
>
>   x = (p !=0) + (q!=0)
>
> and after a few rounds of optimization it transforms into the above.
> I haven't tried to construct a small test yet, but it should certainly 
> be possible.

Fold should not allow this to happen, it should abort if it sees 
something like this.
The reason why no one has seen this before is because PPC Darwin is the 
only
target which has a different size than 1 byte for _Bool (4 bytes).
Most likely what is happening is someone somewhere thinks a cast is not 
need and removes the cast which
causes the tree above.
Actually I did see a problem like this before, see 
<http://gcc.gnu.org/ml/gcc/2003-08/msg01290.html>
and the references.

PPC Darwin is still referenced in that code segment.

Thanks,
Andrew Pinski