nmm0
commented
3 years ago TestResilience.cpp should currently be disabled, since it relies on code that is not implemented. Are there additional tests giving problems?
Open nphtan opened 3 years ago
nmm0
commented
3 years ago TestResilience.cpp should currently be disabled, since it relies on code that is not implemented. Are there additional tests giving problems?
nphtan
commented
3 years ago There's a syntax bug in ResCudaSpace.hpp
diff --git a/src/resilience/cuda/ResCudaSpace.hpp b/src/resilience/cuda/ResCudaSpace.hpp index 970151e..8fc3209 100644 --- a/src/resilience/cuda/ResCudaSpace.hpp +++ b/src/resilience/cuda/ResCudaSpace.hpp @@ -270,7 +270,7 @@ struct VerifyExecutionCanAccessMemorySpace< KokkosResilience::ResCudaSpace , Kok /* Running in CudaSpace attempting to access an unknown space: error / template< class OtherSpace > struct VerifyExecutionCanAccessMemorySpace<
With removed TestResilience.cpp and the syntax error fix the build fails while trying to make TestVelocMemoryBackend.cpp with the following errors.
/home/ntan1/KokkosResilience/kokkos-resilience/src/resilience/util/Trace.hpp(288): error: expression must have class type
detected during:
instantiation of "auto KokkosResilience::Util::begin_trace<TraceType,Context,Args...>(Context &, Args &&...) [with TraceType=KokkosResilience::Util::TimingTrace
/home/ntan1/KokkosResilience/kokkos-resilience/src/resilience/util/Trace.hpp(290): error: expression must have class type
detected during:
instantiation of "auto KokkosResilience::Util::begin_trace<TraceType,Context,Args...>(Context &, Args &&...) [with TraceType=KokkosResilience::Util::TimingTrace
2 errors detected in the compilation of "/tmp/tmpxft_00010e8b_00000000-6_TestVelocMemoryBackend.cpp1.ii". make[2]: [tests/CMakeFiles/resilience_tests.dir/TestVelocMemoryBackend.cpp.o] Error 1 make[1]: [tests/CMakeFiles/resilience_tests.dir/all] Error 2 make: *** [all] Error 2
I'm running into issues building with CUDA support on Power9. The platform is a dual socket Power9 node with 32 cores and 2 V100 GPUs per node. Building with CUDA support has 2 issues I've seen so far. The first is a simple mistake in ResCudaSpace.hpp(273) that generates a bunch of syntax errors. ... [ 25%] Building CXX object CMakeFiles/resilience.dir/src/resilience/cuda/ResCuda.cpp.o nvcc_wrapper has been given GNU extension standard flag -std=gnu++14 - reverting flag to -std=c++14 /home/ntan1/KokkosResilience/kokkos-resilience/src/resilience/cuda/ResCudaSpace.hpp(273): error: enable_if is not a template
The fix is to add std:: to both the enable_if and is_same template functions on line 273.
The second error comes further along when building the tests.
[ 50%] Building CXX object tests/CMakeFiles/resilience_tests.dir/TestResilience.cpp.o /home/ntan1/KokkosResilience/kokkos/build/install/include/impl/Kokkos_Profiling_Interface.hpp(79): error: incomplete type is not allowed detected during: instantiation of "uint32_t Kokkos::Profiling::Experimental::device_id(const ExecutionSpace &) [with ExecutionSpace=KokkosResilience::ResCuda]" /home/ntan1/KokkosResilience/kokkos/build/install/include/Kokkos_Parallel.hpp(171): here instantiation of "void Kokkos::parallel_for(const ExecPolicy &, const FunctorType &, const std::cxx11::string &, std::enable_if<Kokkos::is_execution_policy::value, void>::type *) [with ExecPolicy=Kokkos::RangePolicy, FunctorType=lambda ->void]"
/home/ntan1/KokkosResilience/kokkos-resilience/tests/TestResilience.cpp(93): here
instantiation of "void TestResilientRange<ExecSpace, ScheduleType, DataType>::test_for() [with ExecSpace=Kokkos::Serial, ScheduleType=Kokkos::Schedule, DataType=int]"
/home/ntan1/KokkosResilience/kokkos-resilience/tests/TestResilience.cpp(117): here
instantiation of "void TestResilience_range_Test::TestBody() [with gtestTypeParam=Kokkos::Serial]"
/home/ntan1/KokkosResilience/kokkos-resilience/build/_deps/googletest-src/googletest/include/gtest/internal/gtest-internal.h(470): here
implicit generation of "TestResilience_range_Test::~TestResilience_range_Test() [with gtestTypeParam=Kokkos::Serial]"
/home/ntan1/KokkosResilience/kokkos-resilience/build/_deps/googletest-src/googletest/include/gtest/internal/gtest-internal.h(470): here
[ 4 instantiation contexts not shown ]
implicit generation of "testing::internal::TestFactoryImpl::~TestFactoryImpl() [with TestClass=TestResilience_range_Test]"
/home/ntan1/KokkosResilience/kokkos-resilience/build/_deps/googletest-src/googletest/include/gtest/internal/gtest-internal.h(728): here
instantiation of class "testing::internal::TestFactoryImpl [with TestClass=TestResilience_range_Test]"
/home/ntan1/KokkosResilience/kokkos-resilience/build/_deps/googletest-src/googletest/include/gtest/internal/gtest-internal.h(728): here
implicit generation of "testing::internal::TestFactoryImpl::TestFactoryImpl() [with TestClass=TestResilience_range_Test]"
/home/ntan1/KokkosResilience/kokkos-resilience/build/_deps/googletest-src/googletest/include/gtest/internal/gtest-internal.h(728): here
instantiation of class "testing::internal::TestFactoryImpl [with TestClass=TestResilience_range_Test]"
/home/ntan1/KokkosResilience/kokkos-resilience/build/_deps/googletest-src/googletest/include/gtest/internal/gtest-internal.h(728): here
instantiation of " nv_bool testing::internal::TypeParameterizedTest<Fixture, TestSel, Types>::Register(const char , const testing::internal::CodeLocation &, const char , const char *, int, const std::vector<std::cxx11::string, std::allocator<std::cxx11::string>> &) [with Fixture=TestResilience, TestSel=testing::internal::TemplateSel, Types=gtest_type_paramsTestResilience]"
/home/ntan1/KokkosResilience/kokkos-resilience/tests/TestResilience.cpp(110): here
1 error detected in the compilation of "/tmp/tmpxft_00002401_00000000-6_TestResilience.cpp1.ii". make[2]: [tests/CMakeFiles/resilience_tests.dir/TestResilience.cpp.o] Error 1 make[1]: [tests/CMakeFiles/resilience_tests.dir/all] Error 2 make: *** [all] Error 2
I'm not sure how to fix this.