ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update Kokkos library in LAMMPS to v2.9.00

Browse Source
This commit is contained in:
Stan Moore
2019-06-28 11:23:24 -06:00
parent 7f342b1cd0
commit ea2e73119d
506 changed files with 37043 additions and 6851 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
lib/kokkos/core/src/Kokkos_Cuda.hpp
View File

@ -52,6 +52,7 @@
#include <iosfwd>
#include <vector>
#include <impl/Kokkos_AnalyzePolicy.hpp>
#include <Kokkos_CudaSpace.hpp>
#include <Kokkos_Parallel.hpp>
@ -67,6 +68,7 @@
namespace Kokkos {
namespace Impl {
class CudaExec ;
class CudaInternal ;
} // namespace Impl
} // namespace Kokkos
@ -74,6 +76,23 @@ class CudaExec ;
namespace Kokkos {
namespace Impl {
namespace Experimental {
enum class CudaLaunchMechanism:unsigned{Default=0,ConstantMemory=1,GlobalMemory=2,LocalMemory=4};
constexpr inline CudaLaunchMechanism operator | (CudaLaunchMechanism p1, CudaLaunchMechanism p2) {
return static_cast<CudaLaunchMechanism>(static_cast<unsigned>(p1) | static_cast<unsigned>(p2));
}
constexpr inline CudaLaunchMechanism operator & (CudaLaunchMechanism p1, CudaLaunchMechanism p2) {
return static_cast<CudaLaunchMechanism>(static_cast<unsigned>(p1) & static_cast<unsigned>(p2));
}
template<CudaLaunchMechanism l>
struct CudaDispatchProperties {
CudaLaunchMechanism launch_mechanism = l;
};
}
}
/// \class Cuda
/// \brief Kokkos Execution Space that uses CUDA to run on GPUs.
///
@ -153,7 +172,13 @@ public:
/// return asynchronously, before the functor completes. This
/// method does not return until all dispatched functors on this
/// device have completed.
static void impl_static_fence();
#ifdef KOKKOS_ENABLE_DEPRECATED_CODE
static void fence();
#else
void fence() const;
#endif
/** \brief Return the maximum amount of concurrency. */
static int concurrency();
@ -165,15 +190,18 @@ public:
//--------------------------------------------------
//! \name Cuda space instances
KOKKOS_INLINE_FUNCTION
~Cuda() {}
Cuda();
explicit Cuda( const int instance_id );
Cuda( Cuda && ) = default ;
Cuda( const Cuda & ) = default ;
Cuda & operator = ( Cuda && ) = default ;
Cuda & operator = ( const Cuda & ) = default ;
Cuda(cudaStream_t stream);
//--------------------------------------------------------------------------
//! \name Device-specific functions
//@{
@ -219,18 +247,18 @@ public:
*/
static std::vector<unsigned> detect_device_arch();
cudaStream_t cuda_stream() const { return m_stream ; }
int cuda_device() const { return m_device ; }
cudaStream_t cuda_stream() const;
int cuda_device() const;
//@}
//--------------------------------------------------------------------------
static const char* name();
inline Impl::CudaInternal* impl_internal_space_instance() const { return m_space_instance; }
private:
int m_device ;
cudaStream_t m_stream ;
Impl::CudaInternal* m_space_instance;
};
} // namespace Kokkos
@ -302,7 +330,8 @@ struct VerifyExecutionCanAccessMemorySpace
/*--------------------------------------------------------------------------*/
/*--------------------------------------------------------------------------*/
#include <Cuda/Kokkos_CudaExec.hpp>
#include <Cuda/Kokkos_Cuda_KernelLaunch.hpp>
#include <Cuda/Kokkos_Cuda_Instance.hpp>
#include <Cuda/Kokkos_Cuda_View.hpp>
#include <Cuda/Kokkos_Cuda_Team.hpp>
#include <Cuda/Kokkos_Cuda_Parallel.hpp>