Tree tail merging breaks __builtin_unreachable optimization

[Tom de Vries]

On 05/07/12 15:30, Michael Matz wrote:
> Hi,
> 
> On Thu, 5 Jul 2012, Tom de Vries wrote:
> 
>> The asserts allow the return result to be optimized, but not the cfg 
>> conditions.
>>
>> AFAIU, we can insert the asserts earlier. F.i., we can insert
>>   aD.1711_6 = ASSERT_EXPR <aD.1711_1(D), aD.1711_1(D) > 0>
>> before the GIMPLE_COND in bb2.
> 
> Nope.  That would require some more checks, in particular that the BB 
> containing builtin_unreachable doesn't contain any other side-effects.  
> Given this:
> 
> if (i < 0)
>   { do_something_interesting();
>     __builtin_unreachable();
>   }
> 
> moving the assert before the if would remove the if condition, hence 
> the call to do_something_interesting.  You need to retain side-effects if 
> there are any.
> 

Michael,

Thanks for pointing that out.

I tried a first stab at your suggestion of implementing the optimization in
pass_fold_builtins, it works for the test-case.

Thanks,
- Tom

> 
> Ciao,
> Michael.
> 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: ccp-unreachable.patch
Type: text/x-patch
Size: 1775 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20120705/1e918874/attachment.bin>