https://sourceware.org/bugzilla/show_bug.cgi?id=34237

--- Comment #4 from Rainer Orth <ro at CeBiTec dot Uni-Bielefeld.DE> ---
> If the addition is benign, one could pass $target to the gprof tests scripts
> and expect the additional line.  It would be best to have a single *.expected
> file per target, however I cannot tell which targets need the powerpc special
> version.

I've looked into the powerpc situation some more:

* On AIX (powerpc-ibm-aix7.3.1.0), a trivial program has ".main" in its
  symbol table, so this is where the powerpc special cases apply.
  However, even a trivial program fails to link with -pg with both gcc
  and the bundled ibm-clang for lack of gcrt0.o or gcrt0_64.o.  Besides,
  gprof is disabled in the toplevel configure.ac.

* Debian/powerpc64le (powerpc64le-unknown-linux-gnu) has no '.' prefix
  for symbols and works as other ELF targets.

* On FreeBSD/powerpc64le (powerpc64le-unknown-freebsd16.0), gcc >= 15 is
  required to link with -pg, otherwise it tries to link with -lc_p which
  no longer exists since FreeBSD 14.  However, with both gcc and the
  bundled clang a trivial program SEGVs in _mcount if built/linked with -pg.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

Reply via email to