[cfe-dev] "You don't pay for what you don't use" principle violation?

[Eric Christopher]

On Sun, May 4, 2014 at 3:59 PM, David Blaikie <dblaikie at gmail.com> wrote:
> On Sun, May 4, 2014 at 3:56 PM, Fernando Pelliccioni
> <fpelliccioni at gmail.com> wrote:
>>
>>
>>
>> On Sun, May 4, 2014 at 7:27 PM, David Blaikie <dblaikie at gmail.com> wrote:
>>>
>>> On Sun, May 4, 2014 at 3:08 PM, Fernando Pelliccioni
>>> <fpelliccioni at gmail.com> wrote:
>>> > Hi all,
>>> >
>>> > Is there any reason why the following code produces the attached
>>> > assembly
>>> > code?
>>> >
>>> >     struct X { int v; };
>>> >
>>> >     int func( X** x, int n )
>>> >     {
>>> >         int sum = 0;
>>> >
>>> >         while ( n > 0 )
>>> >         {
>>> >             sum += (*x)->v;
>>> >             ++x;
>>> >             --n;
>>> >         }
>>> >         return sum;
>>> >     }
>>>
>>> This is an externally visible function - it can't be omitted from the
>>> object file as it may be used by some other object file that will be
>>> linked in. (it could be a global ctor, so the absence of code in
>>> 'main' doesn't indicate that it will never be used)
>>>
>>> LTO can remove functions like this.
>>>
>>> >     int main() {}
>>> >
>>> > The "func" function is not used, I think it should be omitted by the
>>> > compiler.
>>> >
>>> > If I change the function to a template function
>>> >
>>> >     template <typename T>
>>> >     int func( T** x, int n )
>>> >     {
>>> >         int sum = 0;
>>> >
>>> >         while ( n > 0 )
>>> >         {
>>> >             sum += (*x)->v;
>>> >             ++x;
>>> >             --n;
>>> >         }
>>> >         return sum;
>>> >     }
>>> >
>>> > Then, no code is generated.
>>>
>>> Since there's no call, there's no template instantiation, and no code.
>>> The C++ language requires callers in other translation units to have
>>> an instantiation of the template somewhere, it just doesn't have to be
>>> here.
>>>
>>> >
>>> >
>>> > Compiled using:
>>> >     clang++ -S -O3 -std=c++11 -S code.cpp
>>> >
>>> > clang++ --version
>>> >     clang version 3.5 (trunk 200620)
>>> >     Target: x86_64-apple-darwin13.1.0
>>> >     Thread model: posix
>>> >
>>> >
>>> > Best,
>>> > Fernando.
>>> >
>>> > _______________________________________________
>>> > cfe-dev mailing list
>>> > cfe-dev at cs.uiuc.edu
>>> > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>> >
>>
>>
>>
>> Thanks guys, sorry for the lack of precision in my words. You are right, the
>> template function example is useless.
>> Actually I was thinking in both, compile and linking time optimizations.
>> I rephrase my question. Why the function is not omitted by the linker? (Is
>> the -S flag the culprit?)
>
> Could be - I'm not particularly familiar with linker flags, though
> that's a bit out of scope for this list anyway.
>

You'll need to tell the linker to dead strip the binary.

-Wl,-dead_strip

since you appear to be on darwin. Otherwise a linker will just assume
that you have that extra code in there for some reason. (This will
slow down the linking process a bit, but not a lot.)

-eric