Re: g++ and library optimisations

[pemo]

On May 13, 4:34 pm, Paul Pluzhnikov <ppluzhnikov-...@charter.net>
wrote:
> "pemo" <usenetmeis...@gmail.com> writes:
> > Where is the STL source normally located - can find the vector.h but not the
> > cpp?
>
> You aren't looking for either vector.h or vector.cpp.
>
> You are looking for 'vector' (and, as I already explained, there
> *isn't* a .cpp file, because STL is all templates, which are *not*
> compiled into a library).
>
> Assuming gcc 4.1.1 in default RH installation, the 'vector' would
> be found in:
>
> $ rpm -ql libstdc++-devel | grep 'vector$'
> /usr/include/c++/4.1.1/debug/vector     # this is internal debug version
> /usr/include/c++/4.1.1/vector           # this is what client code uses.

Thanks.

Could you clarify it's neither a .h/.cpp?  The source code behind the
vector template has to be compiled, and thus found.  Intuitively, I
would expect this be be in some include path, i.e., as vector is
included in her source, the template boiler plate has also to be
'included' somehow.  This seems *not* to be the case however, so I'm
curious as to how the code 'arrives' for compilation.

> >> Sure: just add '-Wl,-verbose' and it will spit lines like this:
>
> > Her g++ [4.1.1 20070105 (Red Hat 4.1.1-51)] didn't understand either of
> > those switches - not did gcc [should one use gcc rather than g++?]
>
> There are no "those switches". It's a *single* switch, spelled
> *exactly* this way:
>
>    g++ ... -Wl,-verbose

Ok, will try this.

> >> Note that correct profiling is a bit of "black art", best performed
> > -pg seemed to work, but nothing seemed to be generated. running gprof, it
> > couldn't find gmon sometimes, or a.out when the app was compiled via gcc.
>
> You seem to be lacking the necessary skills to help her with
> profiling. As I said, it may require a bit of wizardry.

Yes I am - used to selecting a simple checkbox to say I'd like to
profile some code, and being shown VTune output [I have the Intel
compiler for Windows [not that VTune requires that of course]].

> Given that you didn't supply any details, other than "nothing seemed
> to be generated", it's hard to tell what may have gone wrong.
>
> Suggestion: compile a simple application (main with a couple of
> functions with time-consuming loops) with '-pg'. Run it to learn
> how gprof works, what output to expect, etc. Only then try the real
> application.

Ok, but can you say whether after doing this [away from her machine at
the mo] whether gprof requires some explicit input, e.g. gprof a.out?