openmp.batched_vector_arithmatic_simd_dcomplex2 failure with gcc/4.8.4 + OpenMP

[ndellingwood]

@kyungjoo-kim A spot-check on Apollo exposed this test failure below, hopefully just a small touch-up needed for your SIMD fixes?

[ RUN      ] openmp.batched_vector_arithmatic_simd_dcomplex2
/home/ndellin/kokkos-kernels/unit_test/../test_common/KokkosKernels_TestUtils.hpp:85: Failure
The difference between double(AT1::abs(val1)) and double(AT2::abs(val2)) is 1.3067744696188346, which exceeds double(AT3::abs(tol)), where
double(AT1::abs(val1)) evaluates to 0,
double(AT2::abs(val2)) evaluates to 1.3067744696188346, and
double(AT3::abs(tol)) evaluates to 0.
/home/ndellin/kokkos-kernels/unit_test/../test_common/KokkosKernels_TestUtils.hpp:85: Failure
The difference between double(AT1::abs(val1)) and double(AT2::abs(val2)) is 0.49846081263005498, which exceeds double(AT3::abs(tol)), where
double(AT1::abs(val1)) evaluates to 0,
double(AT2::abs(val2)) evaluates to 0.49846081263005498, and
double(AT3::abs(tol)) evaluates to 0.
//many more repeats like this...

[ndellingwood]

gcc/5.3.0 + Serial backend also has this test failure on Apollo

[  FAILED  ] 1 test, listed below:
[  FAILED  ] serial.batched_vector_arithmatic_simd_dcomplex2

As an extra data point, clang/3.9.0 + Serial passed.

[kyungjoo-kim]

thank you. I think that I know the problem. will add a fix .

[ndellingwood]

@kyungjoo-kim the test_all_sandia script for Apollo explicitly sets SNB as the arch, in case this is relevant info.