This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: PATCH: RFC libgomp/testsuite/libgomp.fortran


On Wed, Sep 02, 2009 at 05:31:16PM -0500, Loren James Rittle wrote:
> (CCing fortran mailing list since perhaps related to libgomp.fortran.)
> (CCing Steve and John since their Aug. reports also display this problem.)
> 
> Inside gcc mainline, I noticed that various tests in
> libgomp/testsuite/libgomp.fortran pass on i386-*-freebsd7 when
> -pthread was globally applied (i.e. as by setting RUNTESTFLAGS) but
> fail when run with defaults due to missing pthread_create symbol.
> 
> Each failing test has: { dg-options "-fno-openmp" }
> 
> Since I don't see this in reports for many other ports, I ponder: Is
> this particular port slightly misconfigured somewhere?  Is the test
> code configuration for this port missing something?
> 
> If neither, then the following patch allows the 4 failing tests to
> produce executables and then pass.  I obviously don't want to paper
> over a possible misconfiguration thus I post...
> 
> Regards,
> Loren
> 
> 	* testsuite/libgomp.fortran/omp_cond4.F90: Add -pthread to
> 	dg-options for *-*-freebsd*.
> 	* testsuite/libgomp.fortran/condinc2.f: Likewise.
> 	* testsuite/libgomp.fortran/condinc4.f90: Likewise.
> 	* testsuite/libgomp.fortran/omp_cond2.f: Likewise.
> 

As a gfortran maintainer and FreeBSD user, I've always considered these
failures to be bogus.  The inclusion of -lgomp and -pthread are tied
to -fopenmp.  If you look at the failures, you'll see an ordering of
'-fopenmp -O -fno-openmp', which correctly disables the inclusion
of -lgomp and -pthread.  The only problem is that 

troutmask:sgk[251] find . -name \*.exp | xargs grep lgomp
./lib/libgomp.exp:    lappend ALWAYS_CFLAGS "ldflags=-lgomp"

so dejagnu overrides -fno-openmp and adds -lgomp without adding
-pthread.

Your changes look correct to me, but I can't approve patches
in the libgomp hierarchy.

-- 
Steve


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]