gmsh: patch for read_parallel and write_parallel with CUDA backend

[cwsmith]

@tristan0x

We were merging the snl repo into the scorec fork and some code in Omega_h_gmsh.cpp caught my attention. It appears that in a few places device arrays are being used in host code. E.g.; vert_globals_w in this block:

omega_h/src/Omega_h_gmsh.cpp

Lines 557 to 563 in dc23545

	Write<GO> vert_globals_w(nnodes);
	for (LO local_index = 0; local_index < nnodes; ++local_index) {
	const auto global_index =
	static_cast<GO>(node_tags[static_cast<std::size_t>(local_index)]);
	node_number_map[global_index] = local_index;
	vert_globals_w[local_index] = global_index;
	}

This branch main...SCOREC:omega_h:cws/gmshFix has a partial fix but fails during the serial vs parallel mesh comparison test here:

omega_h/src/unit_io.cpp

Line 1358 in dc23545

OMEGA_H_CHECK(light_compare_meshes(mesh, pmesh) == OMEGA_H_SAME);

Details on the build and failure are below.

Any help would be appreciated.

---

GCC version: 7.4.0
CUDA version: 11.4
MPICH version: 3.3.1

Gmsh version: 4.11.0 (tag=gmsh_4_11_0) from https://gitlab.onelab.info/gmsh/gmsh.git
Gmsh cmake build command:

d=buildGmsh
cmake -S gmsh -B $d \
  -DCMAKE_INSTALL_PREFIX=$d/install \
  -DENABLE_BUILD_DYNAMIC=on \
  -DBUILD_SHARED_LIBS=ON 
cmake --build $d --target install -j8

Omegah cmake build command:

d=buildOmegah
cmake -S omega_h -B $d \
  -DGmsh_INCLUDE_DIRS=$PWD/buildGmsh/install/include \
  -DGmsh_LIBRARIES=$PWD/buildGmsh/install/lib64/libgmsh.so \
  -DGmsh_VERSION_STRING=4.11.0 \
  -DCMAKE_INSTALL_PREFIX=$d/install \
  -DBUILD_TESTING=on  \
  -DOmega_h_USE_CUDA=on \
  -DOmega_h_CUDA_ARCH=75 \
  -DOmega_h_USE_MPI=on  \
  -DOmega_h_USE_Gmsh=on  \
  -DBUILD_SHARED_LIBS=ON \
  -DBUILD_TESTING=on
cmake --build $d --target install -j8

GDB stack trace

Note, I'm not sure why, but when I run gmsh --version on the command line the output Error : Unknown string option 'General.WebBrowser' appears. This output also appears when using the API (see below). I didn't find anything referencing this output in the gitlab issues page. It appears to be non-fatal as I was able to load the gmsh file, creating the omegah mesh, and writing VTK files (before the failing mesh comparison).

gdb ./src/unit_io
─── Output/messages ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
Error   : Unknown string option 'General.WebBrowser'

Thread 1 "unit_io" received signal SIGSEGV, Segmentation fault.
0x00007ffff3790220 in __memcmp_sse4_1 () from /lib64/libc.so.6
─── Source ───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
─── Stack ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
[0] from 0x00007ffff3790220 in __memcmp_sse4_1
[1] from 0x00000000004201a7 in std::__equal<true>::equal<long>(long const*, long const*, long const*)+80 at /opt/scorec/spack/dev/install/linux-rhel7-x86_64/gcc-6.5.0/gcc-7.4.0-c5aaloybb5jqiolgakbf5sxir4axkly4/include/c++/7.4.0/bits/stl_algobase.h:814
[2] from 0x000000000041ee73 in std::__equal_aux<long const*, long const*>(long const*, long const*, long const*)+47 at /opt/scorec/spack/dev/install/linux-rhel7-x86_64/gcc-6.5.0/gcc-7.4.0-c5aaloybb5jqiolgakbf5sxir4axkly4/include/c++/7.4.0/bits/stl_algobase.h:831
[3] from 0x000000000041de87 in std::equal<long const*, long const*>(long const*, long const*, long const*)+79 at /opt/scorec/spack/dev/install/linux-rhel7-x86_64/gcc-6.5.0/gcc-7.4.0-c5aaloybb5jqiolgakbf5sxir4axkly4/include/c++/7.4.0/bits/stl_algobase.h:1051
[4] from 0x000000000041a0fd in light_compare_meshes(Omega_h::Mesh&, Omega_h::Mesh&)+978 at /space/cwsmith/omegahGmsh/omega_h/src/unit_io.cpp:1271
[5] from 0x000000000041ae46 in test_gmsh_parallel(Omega_h::Library*)+675 at /space/cwsmith/omegahGmsh/omega_h/src/unit_io.cpp:1358
[6] from 0x000000000041cb0d in main(int, char**)+317 at /space/cwsmith/omegahGmsh/omega_h/src/unit_io.cpp:1485
──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
>>> where
#0  0x00007ffff3790220 in __memcmp_sse4_1 () from /lib64/libc.so.6
#1  0x00000000004201a7 in std::__equal<true>::equal<long> (__first1=0x7fffc4410a00, __last1=0x7fffc4413708, __first2=0x7fffc4424200) at /opt/scorec/spack/dev/install/linux-rhel7-x86_64/gcc-6.5.0/gcc-7.4.0-c5aaloybb5jqiolgakbf5sxir4axkly4/include/c++/7.4.0/bits/stl_algobase.h:814
#2  0x000000000041ee73 in std::__equal_aux<long const*, long const*> (__first1=0x7fffc4410a00, __last1=0x7fffc4413708, __first2=0x7fffc4424200) at /opt/scorec/spack/dev/install/linux-rhel7-x86_64/gcc-6.5.0/gcc-7.4.0-c5aaloybb5jqiolgakbf5sxir4axkly4/include/c++/7.4.0/bits/stl_algobase.h:831
#3  0x000000000041de87 in std::equal<long const*, long const*> (__first1=0x7fffc4410a00, __last1=0x7fffc4413708, __first2=0x7fffc4424200) at /opt/scorec/spack/dev/install/linux-rhel7-x86_64/gcc-6.5.0/gcc-7.4.0-c5aaloybb5jqiolgakbf5sxir4axkly4/include/c++/7.4.0/bits/stl_algobase.h:1051
#4  0x000000000041a0fd in light_compare_meshes (a=..., b=...) at /space/cwsmith/omegahGmsh/omega_h/src/unit_io.cpp:1271
#5  0x000000000041ae46 in test_gmsh_parallel (lib=0x7fffffffac30) at /space/cwsmith/omegahGmsh/omega_h/src/unit_io.cpp:1358
#6  0x000000000041cb0d in main (argc=1, argv=0x7fffffffade8) at /space/cwsmith/omegahGmsh/omega_h/src/unit_io.cpp:1485

[cwsmith]

Note, the General.WebBrowser output was caused by the contents of the ~/.gmsh-options file.