Christian Trott

Results 519 comments of Christian Trott

You can't use the same executable because then we would try to reinitialize. That is why we have like 20 initialization test executables.

Need to discuss what the actual semantics are we should support long term: A) More or less current behavior: finalize is a no op if Kokkos is not initialized makes...

That is actually not a bad idea. I mean my original interface design actually had the Kokkos::sort just as a helper interface not my main interface (I used the particular...

I think we should not: similar to extents(5,5) not deducing int in the C++ standard. I think its too easy to make a mistake.

looks like this fails in DEBUG build type with GCC 8: ``` cmake -DKokkos_ENABLE_SERIAL=ON -DKokkos_ARCH_SKX=ON -DKokkos_ENABLE_TESTS=ON -DCMAKE_BUILD_TYPE=DEBUG ../ ``` ``` [100%] Building CXX object simd/unit_tests/CMakeFiles/KokkosSimd_UnitTest_SIMD.dir/TestSIMD.cpp.o In file included from /ascldap/users/crtrott/Kokkos/kokkos/simd/src/Kokkos_SIMD.hpp:53,...

Yeah looks like the definition of those intrinsics is protected by `_OPTIMIZE_` ... ``` /home/projects/x86-64/gcc/8.2.0/lib/gcc/x86_64-pc-linux-gnu/8.2.0/include/avx512fintrin.h ```

And yeah: compiles with -O1 but fails with -O0 ... ``` /ascldap/users/projects/x86-64/gcc/8.2.0/bin/g++ -DKOKKOS_DEPENDENCE -I/ascldap/users/crtrott/Kokkos/kokkos/tpls/gtest -I/ascldap/users/crtrott/Kokkos/kokkos/build -I/ascldap/users/crtrott/Kokkos/kokkos/build/core/src -I/ascldap/users/crtrott/Kokkos/kokkos/core/src -I/ascldap/users/crtrott/Kokkos/kokkos/core/src/../../tpls/desul/include -I/ascldap/users/crtrott/Kokkos/kokkos/build/containers/src -I/ascldap/users/crtrott/Kokkos/kokkos/containers/src -I/ascldap/users/crtrott/Kokkos/kokkos/build/algorithms/src -I/ascldap/users/crtrott/Kokkos/kokkos/algorithms/src -I/ascldap/users/crtrott/Kokkos/kokkos/build/simd/src -I/ascldap/users/crtrott/Kokkos/kokkos/simd/src -g -march=skylake-avx512 -mtune=skylake-avx512 -std=gnu++17 -o CMakeFiles/KokkosSimd_UnitTest_SIMD.dir/TestSIMD.cpp.o...

I was testing this and it looks like some tests are hanging when trying to use this with compiling for 3.5 and running on Volta. I am doing more tests.

Non Kokkos reproducer, works with pure G++ (11.1) but not nvcc (11.6) main.cpp: ```c++ #include #include int main(int argc, char* argv[]) { std::vector abc(1000); #pragma omp parallel for for (auto...