[cfe-dev] Alternative to -pg for clang?

Reid Kleckner reid.kleckner at gmail.com
Sun Sep 26 08:00:21 PDT 2010


Don't know about a plan for -pg, but since I've been evaluating Linux
profiling tools for a class recently, I'll give some recommendations.

If you're using Linux, I'd recommend "perf" (aka "Linux Perf Events"
or "Linux Performance Counters", names that are un-google-able).  The
command line interface is *much* easier to use than oprofile for
performance counters, which is what I was interested in.  I think the
callgraph support is a bit weaker than gprof.  It handled recursion
*very* poorly, which is pretty lame.  The main things to like about it
are the interface, low overhead, and the fact that you don't need to
rebuild.

There's also google-perftools cpuprofiler:
http://google-perftools.googlecode.com/svn/trunk/doc/cpuprofile.html
I've only used their heap profiler, which worked reasonably well.
Google uses this for everything apparently, so it's probably not half
bad.

Reid

On Sun, Sep 26, 2010 at 7:36 AM, Matt Joiner <anacrolix at gmail.com> wrote:
> I observed there is no implementation of -pg for clang. Is there any
> plan to do this, or an alternative framework in place? Can anyone
> recommend an alternative free and effective sampling profiler with
> minimal overhead?
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>



More information about the cfe-dev mailing list