gfortran/OpenMP/declare target link issue

Alin Marin Elena alinm.elena@gmail.com
Sat Sep 5 14:52:22 GMT 2020


Dear Ioannis,

this is my experience with offloading using omp4... it is slightly out
of date but i suspect syntax did not change.
in the module that contains the data (i suspect in reality is a
glorified common block)
you need things like this

```
  !$omp declare target(xxx,yyy,zzz,vxx,vyy,vzz,fxx,fyy,fzz)
```

then of course you will need to sync the data to device before the
call of the routine
```
!$omp target map(to: imcon) &
!$omp  map(to: xxx,yyy,zzz)
```

Regards,
Alin


Without Questions there are no Answers!
______________________________________________________________________
Dr. Alin Marin ELENA
http://alin.elena.space/
______________________________________________________________________

On Sat, 5 Sep 2020 at 15:35, Ioannis E. Venetis <venetis@ceid.upatras.gr> wrote:
>
> Any comment on this? Are global variables defined in a module and used
> in subroutines in the target region supported through the declare target
> link mechanism currently on gcc? If so, am I doing something wrong on
> how to use that mechanism? If not, is there any other approach I could
> follow without too many changes in the code? In the real code I have
> about 20 such variables and the call tree in the target region goes up
> to 3 subroutines deep.
>
> Ioannis E. Venetis
>
> On 3/9/2020 9:36 μ.μ., Ioannis E. Venetis wrote:
> > Hello everyone,
> >
> > I have a larger program that I am trying to convert so that the
> > computationally intensive part will run on an NVidia GPU using OpenMP.
> > However, I am running into trouble when compiling the program. The
> > part of the program to run on the GPU contains calls to subroutines,
> > where variables declared in a separate module are used. This seems to
> > be creating issues. I have reduced the problem to the attached files.
> >
> > I compile as follows:
> >
> > gfortran test_link.f90 common_vars.f90 parameters.f90 -O0 -fopenmp
> > -Wall -Wextra -o test_link
> >
> > With the file test_link.f90 as attached, the program compiles and runs
> > without a problem.
> >
> > If I remove the comments for the subroutine TEST() and comment out
> > line 31 in test_link.f90 (the line "I = NR") the compilation gives the
> > following error:
> >
> > ptxas /tmp/ccw3FqJD.o, line 52; error   : Illegal operand type to
> > instruction 'ld'
> > ptxas /tmp/ccw3FqJD.o, line 52; error   : Unknown symbol
> > '__common_vars_MOD_nr$linkptr'
> > ptxas fatal   : Ptx assembly aborted due to errors
> > nvptx-as: ptxas returned 255 exit status
> > mkoffload: fatal error:
> > /home/myself/apps/gcc-10.2.0/bin/x86_64-pc-linux-gnu-accel-nvptx-none-gcc
> > returned 1 exit status
> > compilation terminated.
> > lto-wrapper: fatal error:
> > /home/myself/apps/gcc-10.2.0/libexec/gcc/x86_64-pc-linux-gnu/10.2.0//accel/nvptx-none/mkoffload
> > returned 1 exit status
> > compilation terminated.
> > /usr/bin/ld: error: lto-wrapper failed
> > collect2: error: ld returned 1 exit status
> >
> > Is this a bug in gfortran or have I misunderstood how DECLARE TARGET
> > LINK works in combination with using subroutines?
> >
> > And a last point: If I use -O3 during compilation the program compiles
> > and runs fine in both cases. I assume that TEST() is inlined in this
> > case and the error disappears?
> >
> > Any help to overcome this issue is more than welcome.
> >
> > Best regards,
> >
> > Ioannis E. Venetis
> >
> > PS1: The problem happens with gcc 10.2 that I compiled myself:
> >
> > $ ~/apps/gcc-10.2.0/bin/gfortran -v
> > Using built-in specs.
> > COLLECT_GCC=/home/myself/apps/gcc-10.2.0/bin/gfortran
> > COLLECT_LTO_WRAPPER=/home/myself/apps/gcc-10.2.0/libexec/gcc/x86_64-pc-linux-gnu/10.2.0/lto-wrapper
> >
> > OFFLOAD_TARGET_NAMES=nvptx-none
> > Target: x86_64-pc-linux-gnu
> > Configured with: ../gcc-10.2.0/configure
> > --enable-offload-targets=nvptx-none
> > --with-cuda-driver-include=/usr/local/cuda/include
> > --with-cuda-driver-lib=/usr/local/cuda/lib64 --disable-bootstrap
> > --disable-multilib --enable-languages=c,c++,fortran,lto
> > --prefix=/home/myself/apps/gcc-10.2.0
> > Thread model: posix
> > Supported LTO compression algorithms: zlib
> > gcc version 10.2.0 (GCC)
> >
> >
> > PS2: Same problem also happens with gcc 9.3 as installed on Ubuntu
> > 16.04 from apt:
> >
> > $ gfortran -v
> > Using built-in specs.
> > COLLECT_GCC=gfortran
> > COLLECT_LTO_WRAPPER=/usr/lib/gcc/x86_64-linux-gnu/9/lto-wrapper
> > OFFLOAD_TARGET_NAMES=nvptx-none:hsa
> > OFFLOAD_TARGET_DEFAULT=1
> > Target: x86_64-linux-gnu
> > Configured with: ../src/configure -v --with-pkgversion='Ubuntu
> > 9.3.0-10ubuntu2~16.04'
> > --with-bugurl=file:///usr/share/doc/gcc-9/README.Bugs
> > --enable-languages=c,ada,c++,go,brig,d,fortran,objc,obj-c++,gm2
> > --prefix=/usr --with-gcc-major-version-only --program-suffix=-9
> > --program-prefix=x86_64-linux-gnu- --enable-shared
> > --enable-linker-build-id --libexecdir=/usr/lib
> > --without-included-gettext --enable-threads=posix --libdir=/usr/lib
> > --enable-nls --enable-clocale=gnu --enable-libstdcxx-debug
> > --enable-libstdcxx-time=yes --with-default-libstdcxx-abi=new
> > --enable-gnu-unique-object --disable-vtable-verify --enable-plugin
> > --with-system-zlib --with-target-system-zlib=auto
> > --enable-objc-gc=auto --enable-multiarch --disable-werror
> > --with-arch-32=i686 --with-abi=m64 --with-multilib-list=m32,m64,mx32
> > --enable-multilib --with-tune=generic
> > --enable-offload-targets=nvptx-none,hsa --without-cuda-driver
> > --enable-checking=release --build=x86_64-linux-gnu
> > --host=x86_64-linux-gnu --target=x86_64-linux-gnu
> > Thread model: posix
> > gcc version 9.3.0 (Ubuntu 9.3.0-10ubuntu2~16.04)
> >


More information about the Gcc-help mailing list