[Bug ipa/60315] [4.8/4.9 Regression] template constructor switch optimization

[rguenther at suse dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60315

--- Comment #17 from rguenther at suse dot de <rguenther at suse dot de> ---
On March 26, 2014 10:58:18 PM CET, hubicka at ucw dot cz
<gcc-bugzilla@gcc.gnu.org> wrote:
>http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60315
>
>--- Comment #16 from Jan Hubicka <hubicka at ucw dot cz> ---
>> forwprop would do that, but the enum is unsigned int while the
>> switch value is int and thus simplify_gimple_switch bails out
>> because the conversion is not value-preserving.
>> 
>> So the frontend would need to be changed here or we need to
>> "complicate" the transform by not looking at the type of
>> the existing switch argument but instead by looking at the
>> actual switch label values to see if their value would be
>> preserved.  But yes, that enum -> int conversion asked for
>> by the C++ standard seems to be common that this should be
>> worth the trouble.
>
>Yep, it seems that the "complicate" transform is actually the most
>generic
>thing to do. (we won't need to modify all FE's and we will likely get
>more
>simplifications done) Shall I try to dig into it or you know how to do
>that
>better?

I've posted a patch but it causes some regressions I need to investigate.