[Ecls-list] ECL shootout
Gabriel Dos Reis
gdr at integrable-solutions.net
Fri Jul 3 03:00:09 UTC 2009
On Thu, Jul 2, 2009 at 4:10 AM, Juan Jose
Garcia-Ripoll<juanjose.garciaripoll at googlemail.com> wrote:
> On Thu, Jul 2, 2009 at 1:51 AM, Gabriel Dos
> Reis<gdr at integrable-solutions.net> wrote:
>> I agree that debugging flags vary from compiler brands to compiler
>> brands. However, having the user (builder) to explicitly specify
>> the profiling flags needed has to be coordinated by the user who does not
>> always know the properties of the C compiler detected by configure.
>> I think we need a flag that most of the time conservatively *infers*
>> the right flags to build ECL to support profiling. We may however
>> need a way to overrule configure, along the line you suggest.
>
> I added --with-profile-cflags which can be yes, no or the actual
> flags. By default "yes" means "-pg" for compiler and linker. I have
> not tested this, though, but it should just work,
Thanks! I've tested this with OpenAxiom trunk[1] on an openSUSE
11.1 system. I was able to get profile data on some runs that
traditionally takes a lot of time. As I anticipated, gprof does not
give data on functions in shared libraries, and I got nothing about
the ECL library. All information are about OpenAxiom internal
functions. Absolutely nothing about functions in libecl.
I think we need --with-profile-cflags to also imply a static link for
any ECL-based program -- note that does not imply that dynamic
linking (especially load of FASLs) should be suppressed. All
it implies is that an ECL-base standalone program (including
the ecl binary itself) be statically linked. Is that workable?
[1] OpenAxiom trunk now supports --enable-profiling for ECL-based
builds. The practical effect is that it passes the flags
:c-file t :h-file t
to compile-file when producing a .o object.
And the build takes a looonng time (as expected).
-- Gaby
>
> Juanjo
>
> --
> Instituto de Física Fundamental, CSIC
> c/ Serrano, 113b, Madrid 28006 (Spain)
> http://juanjose.garciaripoll.googlepages.com
>
More information about the ecl-devel
mailing list