On Wed, 29 Jan 2020, dbroemmel wrote: > Where could I have come across this explanation and the hint that > --disable-multilib will result in an incomplete build? I believe it's not documented. Ideally though mkoffload would be able to see that offload compiler lacks mgomp multilibs and issue a human-readable diagnostic. You can file a bug in the GCC Bugzilla to draw more attention to this issue. > We did configure with that switch (as we always do) and I was struggling > for a while to get it working without (building with 32-bit support > doesn't work, so eventually I've tried with --with-multilib-list=m64 > instead - not sure this breaks things again). For offload compiler you don't need to touch multilib options at all: it will build as if with --enable-multilib, which is correct for OpenMP. For host compiler you can pass --disable-multilib if you don't need libraries for 32-bit x86. > For a somewhat longer example, even when compiled with > -foffload=-lgfortran > I now get > unresolved symbol _gfortran_os_error > collect2: error: ld returned 1 exit status > I suspect this is a similar problem? I think that's a different problem, but I'm not so familiar with libgfortran. I think the function should be implemented in the so-called "minimal" libgfortran runtime. Alexander
