Full MatrixMarket support for read_mtx()

[brian-kelley]

• read_mtx now supports pattern, complex, array, Hermitian, symmetric,
  and skew-symmetric (the full MatrixMarket specification). Before, it just
  supported "matrix coordinate real general".
• support Kokkos::complex scalar type in write_graph_mtx()
• support non-host accessible views in write_kokkos_crst_matrix, by
  getting a host mirror

There are no tests that use the matrix IO, so the spot check doesn't help evaluate these changes. However, once we migrate to CMake for testing, a "copy files to binary dir" will make it easy to test each possible format/field/symmetry combination for read_mtx.

For now, I tested the I/O manually by running the following matrices from the SuiteSparse collection through the SPGEMM testing driver:

• 130bit (pattern, general)
• aa01 (pattern, general)
• can_73 (pattern, symmetric)
• mhd1280b (complex, Hermitian)
• plsk1919 (real, skew-symmetric)

[srajama1]

@brian-kelley : Thanks for fixing this quickly. One comment, see below.

[srajama1]

Thanks, @brian-kelley : We can merge if spot check passes.

[brian-kelley]

@srajama1 Running them now.

[brian-kelley]

@srajama1 It's a good thing I ran the spot check because there was a build error. But it works now :)

kokkos-dev spot-check:
#######################################################
PASSED TESTS
#######################################################
clang-4.0.1-Pthread_Serial-hwloc-release build_time=717 run_time=368
clang-4.0.1-Pthread_Serial-release build_time=737 run_time=588
cuda-8.0.44-Cuda_OpenMP-release build_time=1209 run_time=337
gcc-5.3.0-Serial-hwloc-release build_time=499 run_time=192
gcc-5.3.0-Serial-release build_time=497 run_time=199
gcc-7.2.0-Serial-hwloc-release build_time=439 run_time=179
gcc-7.2.0-Serial-release build_time=430 run_time=168
intel-17.0.1-OpenMP-hwloc-release build_time=963 run_time=164
intel-17.0.1-OpenMP-release build_time=937 run_time=118
#######################################################
FAILED TESTS
#######################################################

White spot-check:
#######################################################
PASSED TESTS
#######################################################
cuda-10.0.130-Cuda_Serial-release build_time=1132 run_time=367
cuda-9.2.88-Cuda_OpenMP-release build_time=1102 run_time=306
gcc-6.4.0-OpenMP_Serial-release build_time=618 run_time=306
gcc-7.2.0-OpenMP-release build_time=368 run_time=150
gcc-7.2.0-OpenMP_Serial-release build_time=753 run_time=372
gcc-7.2.0-Serial-release build_time=271 run_time=174
ibm-16.1.0-Serial-release build_time=1438 run_time=270
#######################################################
FAILED TESTS
#######################################################

[srajama1]

I assume bowman is fine as well ?

[brian-kelley]

@srajama1 Yes:
#######################################################
PASSED TESTS
#######################################################
intel-16.4.258-Pthread-release build_time=1520 run_time=583
intel-16.4.258-Pthread_Serial-release build_time=2488 run_time=1262
intel-16.4.258-Serial-release build_time=1476 run_time=617
intel-17.2.174-OpenMP-release build_time=1979 run_time=386
intel-17.2.174-OpenMP_Serial-release build_time=2881 run_time=1169
intel-17.2.174-Pthread-release build_time=1331 run_time=629
intel-17.2.174-Pthread_Serial-release build_time=2425 run_time=1265
intel-17.2.174-Serial-release build_time=1391 run_time=654
intel-18.2.199-OpenMP-release build_time=1599 run_time=433
intel-18.2.199-OpenMP_Serial-release build_time=2747 run_time=1055
intel-18.2.199-Pthread-release build_time=1208 run_time=632
intel-18.2.199-Pthread_Serial-release build_time=2364 run_time=1218
intel-18.2.199-Serial-release build_time=1212 run_time=667
#######################################################
FAILED TESTS
#######################################################

[srajama1]

@brian-kelley Thanks a lot ! Merging it.

[brian-kelley]

@srajama1 Thanks! Bowman takes way longer than the other two so I get impatient and post them first :)

[srajama1]

Thank @brian-kelley, understood ! I hope I am not annoying you by asking for these in multiple PRs. We have to move to an automated PR testing next year and hopefully that will resolve this.