-# B) Define the common options for Kokkos first so they can be used by
-# subpackages as well.
-#
-
-TRIBITS_ADD_DEBUG_OPTION()
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_SIERRA_BUILD
- KOKKOS_FOR_SIERRA
- "Configure Kokkos for building within the Sierra build system."
- OFF
- )
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_Cuda
- KOKKOS_HAVE_CUDA
- "Enable CUDA support in Kokkos."
- "${TPL_ENABLE_CUDA}"
- )
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_Cuda_UVM
- KOKKOS_USE_CUDA_UVM
- "Enable CUDA Unified Virtual Memory support in Kokkos."
- OFF
- )
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_Pthread
- KOKKOS_HAVE_PTHREAD
- "Enable Pthread support in Kokkos."
- OFF
- )
-
-ASSERT_DEFINED(TPL_ENABLE_Pthread)
-IF (Kokkos_ENABLE_Pthread AND NOT TPL_ENABLE_Pthread)
- MESSAGE(FATAL_ERROR "You set Kokkos_ENABLE_Pthread=ON, but Trilinos' support for Pthread(s) is not enabled (TPL_ENABLE_Pthread=OFF). This is not allowed. Please enable Pthreads in Trilinos before attempting to enable Kokkos' support for Pthreads.")
-ENDIF ()
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_OpenMP
- KOKKOS_HAVE_OPENMP
- "Enable OpenMP support in Kokkos."
- "${${PROJECT_NAME}_ENABLE_OpenMP}"
- )
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_QTHREAD
- KOKKOS_HAVE_QTHREAD
- "Enable QTHREAD support in Kokkos."
- "${TPL_ENABLE_QTHREAD}"
- )
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_CXX11
- KOKKOS_HAVE_CXX11
- "Enable C++11 support in Kokkos."
- "${${PROJECT_NAME}_ENABLE_CXX11}"
- )
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_HWLOC
- KOKKOS_HAVE_HWLOC
- "Enable HWLOC support in Kokkos."
- "${TPL_ENABLE_HWLOC}"
- )
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_MPI
- KOKKOS_HAVE_MPI
- "Enable MPI support in Kokkos."
- "${TPL_ENABLE_MPI}"
- )
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_Debug_Bounds_Check
- KOKKOS_ENABLE_DEBUG_BOUNDS_CHECK
- "Enable bounds checking support in Kokkos."
- OFF
- )
-
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_Profiling
- KOKKOS_ENABLE_PROFILING_INTERNAL
- "Enable KokkosP profiling support for kernel data collections."
- "${TPL_ENABLE_DLlib}"
- )
-
-# placeholder for future device...
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_ENABLE_Winthread
- KOKKOS_HAVE_WINTHREAD
- "Enable Winthread support in Kokkos."
- "${TPL_ENABLE_Winthread}"
- )
-
-# use new/old View
-TRIBITS_ADD_OPTION_AND_DEFINE(
- Kokkos_USING_DEPRECATED_VIEW
- KOKKOS_USING_DEPRECATED_VIEW
- "Choose whether to use the old, deprecated Kokkos::View"
-# Skip NVCC compilation and use host compiler directly
-host_only=0
-
-# Enable workaround for CUDA 6.5 for pragma ident
-replace_pragma_ident=0
-
-# Mark first host compiler argument
-first_xcompiler_arg=1
-
-temp_dir=${TMPDIR:-/tmp}
-
-#echo "Arguments: $# $@"
-
-while [ $# -gt 0 ]
-do
- case $1 in
- #show the executed command
- --show|--nvcc-wrapper-show)
- dry_run=1
- ;;
- #run host compilation only
- --host-only)
- host_only=1
- ;;
- #replace '#pragma ident' with '#ident' this is needed to compile OpenMPI due to a configure script bug and a non standardized behaviour of pragma with macros
- --replace-pragma-ident)
- replace_pragma_ident=1
- ;;
- #handle source files to be compiled as cuda files
- *.cpp|*.cxx|*.cc|*.C|*.c++|*.cu)
- cpp_files="$cpp_files $1"
- ;;
- #Handle shared args (valid for both nvcc and the host compiler)
- static_assert( !std::is_array<DataType>::value && !std::is_pointer<DataType>::value , "Cannot template DynRankView with array or pointer datatype - must be pod" );
-
-public:
- using view_type = View< DataType******* , Properties...>;
- using reference_type = typename view_type::reference_type;
-
-private:
- template < class , class ... > friend class DynRankView ;
- template< class , class ... > friend class Impl::ViewMapping ;
- , m_next_index(ViewAllocateWithoutInitializing("UnorderedMap next index"), capacity()+1) // +1 so that the *_at functions can always return a valid reference
- typedef typename Kokkos::Experimental::DynRankView< const T , Kokkos::LayoutStride , host > const_svector_right_type ; //LayoutStride, not right; setup to match original ViewAPI calls... update
-# error "It doesn't make sense to build this file unless the Kokkos::Serial device is enabled. If you see this message, it probably means that there is an error in Kokkos' CMake build infrastructure."
-# error "You must enable at least one of the following execution spaces in order to build this test: Kokkos::Threads, Kokkos::OpenMP, or Kokkos::Serial."
- "Whether to enable the Kokkos::Serial device. This device executes \"parallel\" kernels sequentially on a single CPU thread. It is enabled by default. If you disable this device, please enable at least one other CPU device, such as Kokkos::OpenMP or Kokkos::Threads."
- "Whether Kokkos allows use of lambdas at the outer level of parallel dispatch (that is, as the argument to an outer parallel_for, parallel_reduce, or parallel_scan). This requires C++11. It also does not currently work with public releases of CUDA. As a result, even if C++11 is enabled, this will be OFF by default if CUDA is enabled. If this option is ON, the macro KOKKOS_HAVE_CXX11_DISPATCH_LAMBDA will be defined. For compatibility with Kokkos' Makefile build system, it is also possible to define that macro on the command line."
- Impl::throw_runtime_exception( "Requested too large league_size for TeamPolicy on Cuda execution space.");
-
- // Make sure total block size is permissable
- if ( m_team_size * m_vector_length > 1024 ) {
- Impl::throw_runtime_exception(std::string("Kokkos::TeamPolicy< Cuda > the team size is too large. Team size x vector length must be smaller than 1024."));
- }
- }
-
- /** \brief Specify league size, request team size */
- Impl::throw_runtime_exception( "Requested too large league_size for TeamPolicy on Cuda execution space.");
-
- // Make sure total block size is permissable
- if ( m_team_size * m_vector_length > 1024 ) {
- Impl::throw_runtime_exception(std::string("Kokkos::TeamPolicy< Cuda > the team size is too large. Team size x vector length must be smaller than 1024."));
- // The global parallel_reduce does not support vector_length other than 1 at the moment
- if( arg_policy.vector_length() > 1)
- Impl::throw_runtime_exception( "Kokkos::parallel_reduce with a TeamPolicy using a vector length of greater than 1 is not currently supported for CUDA.");
-
- if( m_team_size < 32)
- Impl::throw_runtime_exception( "Kokkos::parallel_reduce with a TeamPolicy using a team_size smaller than 32 is not currently supported with CUDA.");
-
- // Functor's reduce memory, team scan memory, and team shared memory depend upon team size.
-#if defined ( KOKKOS_HAVE_DEFAULT_DEVICE_TYPE_CUDA )
- typedef Cuda DefaultExecutionSpace ;
-#elif defined ( KOKKOS_HAVE_DEFAULT_DEVICE_TYPE_OPENMP )
- typedef OpenMP DefaultExecutionSpace ;
-#elif defined ( KOKKOS_HAVE_DEFAULT_DEVICE_TYPE_THREADS )
- typedef Threads DefaultExecutionSpace ;
-#elif defined ( KOKKOS_HAVE_DEFAULT_DEVICE_TYPE_SERIAL )
- typedef Serial DefaultExecutionSpace ;
-#else
-# error "At least one of the following execution spaces must be defined in order to use Kokkos: Kokkos::Cuda, Kokkos::OpenMP, Kokkos::Serial, or Kokkos::Threads."
-#endif
-
-#if defined ( KOKKOS_HAVE_DEFAULT_DEVICE_TYPE_OPENMP )
- typedef OpenMP DefaultHostExecutionSpace ;
-#elif defined ( KOKKOS_HAVE_DEFAULT_DEVICE_TYPE_THREADS )
- typedef Threads DefaultHostExecutionSpace ;
-#elif defined ( KOKKOS_HAVE_DEFAULT_DEVICE_TYPE_SERIAL )
- typedef Serial DefaultHostExecutionSpace ;
-#elif defined ( KOKKOS_HAVE_OPENMP )
- typedef OpenMP DefaultHostExecutionSpace ;
-#elif defined ( KOKKOS_HAVE_PTHREAD )
- typedef Threads DefaultHostExecutionSpace ;
-#elif defined ( KOKKOS_HAVE_SERIAL )
- typedef Serial DefaultHostExecutionSpace ;
-#else
-# error "At least one of the following execution spaces must be defined in order to use Kokkos: Kokkos::OpenMP, Kokkos::Serial, or Kokkos::Threads."
-# error "At least one of the following host execution spaces must be defined: Kokkos::OpenMP, Kokkos::Serial, or Kokkos::Threads. You might be seeing this message if you disabled the Kokkos::Serial device explicitly using the Kokkos_ENABLE_Serial:BOOL=OFF CMake option, but did not enable any of the other host execution space devices."
-# error "At least one of the following host execution spaces must be defined: Kokkos::OpenMP, Kokkos::Serial, or Kokkos::Threads. You might be seeing this message if you disabled the Kokkos::Serial device explicitly using the Kokkos_ENABLE_Serial:BOOL=OFF CMake option, but did not enable any of the other host execution space devices."
- Kokkos::Impl::throw_runtime_exception("Kokkos::Experimental::Impl::SharedAllocationRecord<>::tracking_release_and_enable FAILED, this host process thread did not hold the lock" );
- if ((strncmp(arg[iarg],"--kokkos-threads",16) == 0) || (strncmp(arg[iarg],"--threads",9) == 0)) {
- //Find the number of threads (expecting --threads=XX)
- if (!((strncmp(arg[iarg],"--kokkos-threads=",17) == 0) || (strncmp(arg[iarg],"--threads=",10) == 0)))
- Impl::throw_runtime_exception("Error: expecting an '=INT' after command line argument '--threads/--kokkos-threads'. Raised by Kokkos::initialize(int narg, char* argc[]).");
- Impl::throw_runtime_exception("Error: expecting an '=INT' after command line argument '--threads/--kokkos-threads'. Raised by Kokkos::initialize(int narg, char* argc[]).");
- if (!((strncmp(arg[iarg],"--kokkos-numa=",14) == 0) || (strncmp(arg[iarg],"--numa=",7) == 0)))
- Impl::throw_runtime_exception("Error: expecting an '=INT' after command line argument '--numa/--kokkos-numa'. Raised by Kokkos::initialize(int narg, char* argc[]).");
- Impl::throw_runtime_exception("Error: expecting an '=INT' after command line argument '--numa/--kokkos-numa'. Raised by Kokkos::initialize(int narg, char* argc[]).");
- //Find the number of device (expecting --device=XX)
- if (!((strncmp(arg[iarg],"--kokkos-device=",16) == 0) || (strncmp(arg[iarg],"--device=",9) == 0)))
- Impl::throw_runtime_exception("Error: expecting an '=INT' after command line argument '--device/--kokkos-device'. Raised by Kokkos::initialize(int narg, char* argc[]).");
- Impl::throw_runtime_exception("Error: expecting an '=INT' after command line argument '--device/--kokkos-device'. Raised by Kokkos::initialize(int narg, char* argc[]).");
- //Find the number of device (expecting --device=XX)
- if (!((strncmp(arg[iarg],"--kokkos-ndevices=",18) == 0) || (strncmp(arg[iarg],"--ndevices=",11) == 0)))
- Impl::throw_runtime_exception("Error: expecting an '=INT[,INT]' after command line argument '--ndevices/--kokkos-ndevices'. Raised by Kokkos::initialize(int narg, char* argc[]).");
-
- int ndevices=-1;
- int skip_device = 9999;
-
- char* num1 = strchr(arg[iarg],'=')+1;
- char* num2 = strpbrk(num1,",");
- int num1_len = num2==NULL?strlen(num1):num2-num1;
- Impl::throw_runtime_exception("Error: expecting an integer number after command line argument '--kokkos-ndevices'. Raised by Kokkos::initialize(int narg, char* argc[]).");
- Impl::throw_runtime_exception("Error: expecting an integer number after command line argument '--kokkos-ndevices=XX,'. Raised by Kokkos::initialize(int narg, char* argc[]).");
-//change this if you want to allow oversubscription of the system, by default only the range {1-(system size)} is tested
-#define FOR_GAUNTLET(x) for(unsigned x = (std::min)(std::thread::hardware_concurrency()*8,unsigned(sizeof(next_table)/sizeof(unsigned))); x; x = next_table[x-1])
-
-//set this to override the benchmark of barriers to use OMP barriers instead of n3998 std::barrier
-//#define USEOMP
-
-#if defined(__SYNCHRONIC_COMPATIBLE)
- #define PREFIX "futex-"
-#else
- #define PREFIX "backoff-"
-#endif
-
-//this test uses a custom Mersenne twister to eliminate implementation variation
-\mainpage Trilinos/Kokkos: Shared-memory programming interface and computational kernels
-
-\section Kokkos_Intro Introduction
-
-The %Kokkos package has two main components. The first, sometimes
-called "%Kokkos Array" or just "%Kokkos," implements a
-performance-portable shared-memory parallel programming model and data
-containers. The second, called "%Kokkos Classic," consists of
-computational kernels that support the %Tpetra package.
-
-\section Kokkos_Kokkos The %Kokkos programming model
-
-%Kokkos implements a performance-portable shared-memory parallel
-programming model and data containers. It lets you write an algorithm
-once, and just change a template parameter to get the optimal data
-layout for your hardware. %Kokkos has back-ends for the following
-parallel programming models:
-
-- Kokkos::Threads: POSIX Threads (Pthreads)
-- Kokkos::OpenMP: OpenMP
-- Kokkos::Cuda: NVIDIA's CUDA programming model for graphics
- processing units (GPUs)
-- Kokkos::Serial: No thread parallelism
-
-%Kokkos also has optimizations for shared-memory parallel systems with
-nonuniform memory access (NUMA). Its containers can hold data of any
-primitive ("plain old") data type (and some aggregate types). %Kokkos
-Array may be used as a stand-alone programming model.
-
-%Kokkos' parallel operations include the following:
-
-- parallel_for: a thread-parallel "for loop"
-- parallel_reduce: a thread-parallel reduction
-- parallel_scan: a thread-parallel prefix scan operation
-
-as well as expert-level platform-independent interfaces to thread
-"teams," per-team "shared memory," synchronization, and atomic update
-operations.
-
-%Kokkos' data containers include the following:
-
-- Kokkos::View: A multidimensional array suitable for thread-parallel
- operations. Its layout (e.g., row-major or column-major) is
- optimized by default for the particular thread-parallel device.
-- Kokkos::Vector: A drop-in replacement for std::vector that eases
- porting from standard sequential C++ data structures to %Kokkos'
- parallel data structures.
-- Kokkos::UnorderedMap: A parallel lookup table comparable in
- functionality to std::unordered_map.
-
-%Kokkos also uses the above basic containers to implement higher-level
-data structures, like sparse graphs and matrices.
-
-A good place to start learning about %Kokkos would be <a href="http://trilinos.sandia.gov/events/trilinos_user_group_2013/presentations/2013-11-TUG-Kokkos-Tutorial.pdf">these tutorial slides</a> from the 2013 Trilinos Users' Group meeting.
-
-\section Kokkos_Classic %Kokkos Classic
-
-"%Kokkos Classic" consists of computational kernels that support the
-%Tpetra package. These kernels include sparse matrix-vector multiply,
-sparse triangular solve, Gauss-Seidel, and dense vector operations.
-They are templated on the type of objects (\c Scalar) on which they
-operate. This component was not meant to be visible to users; it is
-an implementation detail of the %Tpetra distributed linear algebra
-package.
-
-%Kokkos Classic also implements a shared-memory parallel programming
-model. This inspired and preceded the %Kokkos programming model
-described in the previous section. Users should consider the %Kokkos
-Classic programming model deprecated, and prefer the new %Kokkos