search

NOAA-EMC / NCEPLIBS

Top level repo containing submodules for NCEPLIBS and associated dependencies for superproject builds
Other
42 stars 18 forks source link

Issue while building with Clang and Gfortran #91

Closed neekumar1 closed 4 years ago

neekumar1 commented 4 years ago

Hello All, I was trying to build NCEPLIBS with clang and gfortran on Linux Ubuntu 18.04

. NCEPLIBS-external got built successfully. But I am facing following errors while building NCEPLIBS. " [ 59%] Linking Fortran executable nhour /usr/bin/ld: ../../../NCEPLIBS-w3nco/libw3nco_v2.1.0_8.a(summary.c.o): relocation R_X86_64_32 against `.rodata.str1.1' can not be used when making a PIE o bject; recompile with -fPIC /usr/bin/ld: final link failed: Nonrepresentable section on output [ 59%] Linking Fortran executable mdate collect2: error: ld returned 1 exit status NCEPLIBS-grib_util/sorc/grib2grib.fd/CMakeFiles/grib2grib.dir/build.make:100: recipe for target 'NCEPLIBS-grib_util/sorc/grib2grib.fd/grib2grib' failed make[2]: [NCEPLIBS-grib_util/sorc/grib2grib.fd/grib2grib] Error 1 CMakeFiles/Makefile2:2253: recipe for target 'NCEPLIBS-grib_util/sorc/grib2grib.fd/CMakeFiles/grib2grib.dir/all' failed make[1]: [NCEPLIBS-grib_util/sorc/grib2grib.fd/CMakeFiles/grib2grib.dir/all] Error 2 " And then process ends with following error: “ [ 62%] Linking Fortran static library libcrtm_v2.3.0.a 9625 [ 62%] Built target crtm 9626 Makefile:129: recipe for target 'all' failed 9627 make: *** [all] Error 2 “

I am building public release v1 branch: git clone -b ufs-v1.0.0 --recursive https://github.com/NOAA-EMC/NCEPLIBS I have shared the log file. Please check and suggest.

kgerheiser commented 4 years ago

Could you run make VERBOSE=1 so I can see the link flags?

I don't why it's making PIE.

neekumar1 commented 4 years ago

Please find the attached log file with make VERBOSE=1. buildlog.NCEPLIBS.Thu_Jul_16_222542_IST_2020.txt

kgerheiser commented 4 years ago

I think it's because -fPIC is being added here. I don't know where that or -mcmodel=medium flags are coming from.

/usr/bin/gfortran  -fPIC -mcmodel=medium -g -O3 CMakeFiles/tocgrib2.dir/tocgrib2.f.o  -o tocgrib2   -L/usr/lib  -Wl,-rpath,/home/amd/neeraj/ufs_new/v1.0.0/install/NCEPLIBS-external/lib: ../../../NCEPLIBS-g2/libg2_v3.2.0_4.a ../../../NCEPLIBS-w3nco/libw3nco_v2.1.0_4.a ../../../NCEPLIBS-bacio/libbacio_v2.2.0_4.a /home/amd/neeraj/ufs_new/v1.0.0/install/NCEPLIBS-external/lib/libpng.so /home/amd/neeraj/ufs_new/v1.0.0/install/NCEPLIBS-external/lib/libz.so /home/amd/neeraj/ufs_new/v1.0.0/install/NCEPLIBS-external/lib/libjasper.so /home/amd/neeraj/ufs_new/v1.0.0/install/NCEPLIBS-external/lib/libjpeg.so /home/amd/neeraj/ufs_new/v1.0.0/install/NCEPLIBS-external/lib/libz.so /home/amd/neeraj/ufs_new/v1.0.0/install/NCEPLIBS-external/lib/libjasper.so /home/amd/neeraj/ufs_new/v1.0.0/install/NCEPLIBS-external/lib/libjpeg.so -lomp -lpthread 
/usr/bin/ld: ../../../NCEPLIBS-w3nco/libw3nco_v2.1.0_4.a(summary.c.o): relocation R_X86_64_32 against `.rodata.str1.1' can not be used when making a PIE object; recompile with -fPIC
/usr/bin/ld: final link failed: Nonrepresentable section on output
kgerheiser commented 4 years ago

I was able to reproduce your issue on a Linux machine I have, and I think I know what's going on.

Ubuntu compiles position independent executables by default and that doesn't mix with the way we use static and shared libraries since we don't compile with -fPIC. Or something along those lines.

The way I fixed it was to add set(CMAKE_EXE_LINKER_FLAGS -no-pie) in the top-level NCEPLIBS CMakeLists.txt. That will pass the -no-pie option to the linker and everything will build.

I also tried adding set(CMAKE_POSITION_INDEPENDENT_CODE ON) which will compile NCEPLIBS with -fPIC so you don't have to disable PIE. It gets farther, but then it fails linking chgres_cube in UFS_UTILS because wgrib2's internal libraries aren't compiled with -fPIC.

I'm not sure if this is a permanent fix, but it should resolve your issue. Thoughts @climbfuji?

neekumar1 commented 4 years ago

Thank you @kgerheiser for providing the suggestion. It worked for now. Is this error got anything to do with Clang ? I tried NCEPLIBS-external with combination of Clang and Flang but could not verify as the build failed with following error message: " checking for Fortran 90 integer kind for 8-byte integers... unavailable checking for Fortran 90 integer kind for 4-byte integers... unavailable configure: error: Unable to determine Fortran 90 KIND values for either address-sized integers or offset-sized integers. CMakeFiles/mpi.dir/build.make:106: recipe for target 'mpich/src/mpi-stamp/mpi-configure' failed make[2]: [mpich/src/mpi-stamp/mpi-configure] Error 1 CMakeFiles/Makefile2:338: recipe for target 'CMakeFiles/mpi.dir/all' failed make[1]: [CMakeFiles/mpi.dir/all] Error 2 Makefile:83: recipe for target 'all' failed make: *** [all] Error 2 " Any suggestions please ?

climbfuji commented 4 years ago

Thank you @kgerheiser for providing the suggestion. It worked for now. Is this error got anything to do with Clang ? I tried NCEPLIBS-external with combination of Clang and Flang but could not verify as the build failed with following error message: " checking for Fortran 90 integer kind for 8-byte integers... unavailable checking for Fortran 90 integer kind for 4-byte integers... unavailable configure: error: Unable to determine Fortran 90 KIND values for either address-sized integers or offset-sized integers. CMakeFiles/mpi.dir/build.make:106: recipe for target 'mpich/src/mpi-stamp/mpi-configure' failed make[2]: [mpich/src/mpi-stamp/mpi-configure] Error 1 CMakeFiles/Makefile2:338: recipe for target 'CMakeFiles/mpi.dir/all' failed make[1]: [CMakeFiles/mpi.dir/all] Error 2 Makefile:83: recipe for target 'all' failed make: *** [all] Error 2 " Any suggestions please ?

We have never tried flang and we don't support it - only gcc+gfortran and clang+gfortran.

aerorahul commented 4 years ago

@neekumar1 Is there still an issue with clang + gfortran as the issue topic mentions? If so, please indicate the versions you are using and the platform. As @climbfuji mentions, we have not tested with flang and don't support it at the moment.

neekumar1 commented 4 years ago

Hello @aerorahul and @climbfuji, the issue with clang+gfortran for NCEPLIBS is resolved now. Thanks for the support.