Fixed block row formula in block GS test (Fix #648)

[brian-kelley]

(an example of #478, which is incorrect integer division with rounding up)

This caused the X vector to have the wrong length. I replicated #648 locally and checked that this fixed it.

RIDE:
#######################################################
PASSED TESTS
#######################################################
cuda-10.1.105-Cuda_OpenMP-release build_time=541 run_time=398
cuda-10.1.105-Cuda_Serial-release build_time=551 run_time=508
cuda-9.2.88-Cuda_OpenMP-release build_time=531 run_time=545
cuda-9.2.88-Cuda_Serial-release build_time=571 run_time=647
gcc-6.4.0-OpenMP_Serial-release build_time=207 run_time=382
gcc-7.2.0-OpenMP-release build_time=130 run_time=126
gcc-7.2.0-OpenMP_Serial-release build_time=183 run_time=356
gcc-7.2.0-Serial-release build_time=122 run_time=226
ibm-16.1.0-Serial-release build_time=504 run_time=392

Kokkos-dev2 (only failure is due to #645 )
#######################################################
PASSED TESTS
#######################################################
clang-8.0-Pthread_Serial-release build_time=86 run_time=370
cuda-10.1-Cuda_OpenMP-release build_time=272 run_time=318
cuda-9.2-Cuda_Serial-release build_time=264 run_time=428
gcc-7.3.0-OpenMP-release build_time=60 run_time=117
gcc-7.3.0-Pthread-release build_time=58 run_time=186
gcc-8.3.0-Serial-release build_time=59 run_time=177
gcc-9.1-OpenMP-release build_time=72 run_time=109
gcc-9.1-Serial-release build_time=65 run_time=175
intel-18.0.5-OpenMP-release build_time=169 run_time=121
#######################################################
FAILED TESTS
#######################################################
clang-8.0-Cuda_OpenMP-release (test failed)

[ndellingwood]

Thanks @brian-kelley ! I like that you added the non-power-of-two block size test as well :)

[brian-kelley]

@ndellingwood Argh, I just noticed that I changed the block size in the rank-2 test to divide evenly into num rows, but that was not intentional. Block GS is intended to work with block size not being a multiple of num rows, the issue here was the way ceiling integer division was done.

So I'll change that back and retest.

[brian-kelley]

Still good, so I'm merging:
#######################################################
PASSED TESTS
#######################################################
cuda-10.1.105-Cuda_OpenMP-release build_time=500 run_time=405
cuda-10.1.105-Cuda_Serial-release build_time=556 run_time=516
cuda-9.2.88-Cuda_OpenMP-release build_time=551 run_time=546
cuda-9.2.88-Cuda_Serial-release build_time=563 run_time=676
gcc-6.4.0-OpenMP_Serial-release build_time=221 run_time=386
gcc-7.2.0-OpenMP-release build_time=121 run_time=129
gcc-7.2.0-OpenMP_Serial-release build_time=196 run_time=361
gcc-7.2.0-Serial-release build_time=116 run_time=226
ibm-16.1.0-Serial-release build_time=521 run_time=392

#######################################################
PASSED TESTS
#######################################################
clang-8.0-Pthread_Serial-release build_time=92 run_time=423
cuda-10.1-Cuda_OpenMP-release build_time=272 run_time=327
cuda-9.2-Cuda_Serial-release build_time=274 run_time=441
gcc-7.3.0-OpenMP-release build_time=58 run_time=119
gcc-7.3.0-Pthread-release build_time=57 run_time=185
gcc-8.3.0-Serial-release build_time=59 run_time=179
gcc-9.1-OpenMP-release build_time=71 run_time=112
gcc-9.1-Serial-release build_time=66 run_time=181
intel-18.0.5-OpenMP-release build_time=186 run_time=121
#######################################################
FAILED TESTS
#######################################################
clang-8.0-Cuda_OpenMP-release (test failed)