SuperBuild fails at eb_spherecube.cpp(21) (Cuda 11, GCC 9)

[markstock]

This spun off of #56

Using most recent pulls of AMReX, AMReX-Hydro, and incflo, a "superbuild" on Fedora 31, GCC 9.3.1, and CUDA 11.2.0 using cmake fails on the following error.

[ 77%] Building CUDA object CMakeFiles/incflo.dir/src/embedded_boundaries/eb_spherecube.cpp.o
/home/mstock/opt/incflo/src/embedded_boundaries/eb_spherecube.cpp(21): error: no instance of constructor "amrex::EB2::SphereIF::SphereIF" matches the argument list
            argument types are: (double, {...}, __nv_bool)

/home/mstock/opt/incflo/src/embedded_boundaries/eb_spherecube.cpp(22): error: no instance of constructor "amrex::EB2::BoxIF::BoxIF" matches the argument list
            argument types are: ({...}, {...}, __nv_bool)

This is a 2D-DOUBLE build with OpenMP and CUDA but no MPI.

It looks like src/embedded_boundaries/eb_spherecube.cpp assumes 3D in it's constructor (code below), while RealArray is defined in amrex/Src/Base/AMReX_Array.H as Array<Real, AMREX_SPACEDIM>.

EB2::SphereIF sphere(0.5, {1.8, 1.8, 2.8}, false);
EB2::BoxIF cube({1.85, 1.85, 2.85}, {2.5, 2.5, 3.5}, false);

[WeiqunZhang]

This is a cmake build issue. In GNU Make, there is

ifeq ($(DIM), 3)
  CEXE_sources += eb_spherecube.cpp
  CEXE_sources += eb_tuscan.cpp
  CEXE_sources += eb_twocylinders.cpp
endif

We should have the same logic in cmake.

[WeiqunZhang]

@etpalmer63 Erik could you handle this?

[WeiqunZhang]

@markstock Meanwhile, you could also try to build with make USE_CUDA=TRUE in test_2d.

[etpalmer63]

@etpalmer63 Erik could you handle this?

Looking into it now.

[etpalmer63]

OK, I think I have it. The PR is up here #58. I will let the CI run.

[markstock]

Running make USE_CUDA=TRUE USE_MPI=FALSE CUDA_ARCH=86 CUDA_PATH=/opt/cuda/11.2 inside test_2d returns SUCCESS. Nice.