[cfe-dev] missing optimization opportunity for const std::vector compared to std::array

[Dennis Luehring]

Am 04.06.2014 21:03, schrieb Richard Smith:
> > #include <string>
> > int main()
> > {
> >   return std::string("hello").size();
> > }
> >
> > results in:
>
>
> main:                                   # @main
>          movl    $5, %eax
>          retq
>
> > and
> >
> > #include <vector>
> > #include <numeric>
> >
> >
> > typedef std::vector<int> container_t;
> >
> > int main()
> > {
> >
> >   const container_t a{1,2};
> >   const container_t b{4,5};
> >   const container_t ints
> >   {
> >     std::accumulate(a.begin(),a.end(),1),
> >     std::accumulate(b.begin(),b.end(),2),
> >   };
> >   return std::accumulate(ints.begin(),ints.end(),100);
> > }
> >
> > results in:
> >
>
> main:                                   # @main
>          movl    $115, %eax
>          retq
>

so the introduction of

void *__builtin_operator_new(size_t)
void __builtin_operator_delete(void*)

in clang and your patching of libc++ reduces my testszenarios results
down to the absolutely bare minimum
it can't get better - thank you very much

its a ~small~ optimization - but i think the impact could be very huge 
for libc++ using code bases

can you post the patch/review link?

are there still known corner cases / similar examples that do not profit 
from your patch?

dennis