CUDA.h

Go to the documentation of this file.

#ifndef VECCORE_CUDA_H
#define VECCORE_CUDA_H
 
#if !defined(VECCORE_ENABLE_CUDA)
 
#define VECCORE_IMPL_NAMESPACE cxx
 
#define VECCORE_ATT_HOST               /* empty */
#define VECCORE_ATT_DEVICE             /* empty */
#define VECCORE_ATT_HOST_DEVICE        /* empty */
#define VECCORE_ATT_GLOBAL             /* empty */
#define VECCORE_CUDA_ALIGN             /* empty */
#define VECCORE_DECLARE_CUDA(x)        /* empty */
#define VECCORE_DECLARE_CUDA_TYPE(T)   /* empty */
#define VECCORE_DECLARE_CUDA_CLASS(x)  /* empty */
#define VECCORE_DECLARE_CUDA_STRUCT(x) /* empty */
#define VECCORE_DEVICE_CONSTANT
 
#elif (defined(__CUDACC__) || defined(__NVCC__))
 
#define VECCORE_IMPL_NAMESPACE cuda
 
#define VECCORE_CUDA
#ifdef __CUDA_ARCH__
#define VECCORE_CUDA_DEVICE_COMPILATION
#endif
 
#define VECCORE_ATT_HOST __host__
#define VECCORE_ATT_DEVICE __device__
#define VECCORE_ATT_HOST_DEVICE __host__ __device__
#define VECCORE_ATT_GLOBAL __global__
#define VECCORE_CUDA_ALIGN __align__((64))
#define VECCORE_DECLARE_CUDA(x)          /* empty */
#define VECCORE_DECLARE_CUDA_TYPE(T)     /* empty */
#define VECCORE_DECLARE_CUDA_CLASS(x)    /* empty */
#define VECCORE_DECLARE_CUDA_STRUCT(x)   /* empty */
#define VECCORE_DECLARE_CUDA_TEMPLATE(x) /* empty */
#define VECCORE_DEVICE_CONSTANT __constant__
 
#else // CUDA enabled, but compiling regular C++ code
 
#define VECCORE_CUDA_INTERFACE
#define VECCORE_ATT_HOST        /* empty */
#define VECCORE_ATT_DEVICE      /* empty */
#define VECCORE_ATT_HOST_DEVICE /* empty */
#define VECCORE_ATT_GLOBAL      /* empty */
#define VECCORE_CUDA_ALIGN      /* empty */
 
// Keep the macro compact
// clang-format off
 
#define VECCORE_CUDA_BASIC_TYPES                                               \
template <typename T> struct CudaTypeTraits;                                   \
template <> struct CudaTypeTraits<float>    { using Type = float;    };        \
template <> struct CudaTypeTraits<double>   { using Type = double;   };        \
template <> struct CudaTypeTraits<int8_t>   { using Type = int8_t;   };        \
template <> struct CudaTypeTraits<int16_t>  { using Type = int16_t;  };        \
template <> struct CudaTypeTraits<int32_t>  { using Type = int32_t;  };        \
template <> struct CudaTypeTraits<int64_t>  { using Type = int64_t;  };        \
template <> struct CudaTypeTraits<uint8_t>  { using Type = uint8_t;  };        \
template <> struct CudaTypeTraits<uint16_t> { using Type = uint16_t; };        \
template <> struct CudaTypeTraits<uint32_t> { using Type = uint32_t; };        \
template <> struct CudaTypeTraits<uint64_t> { using Type = uint64_t; };        \
template <typename T> using CudaType = typename CudaTypeTraits<T>::Type;
 
#define VECCORE_DECLARE_CUDA(T) T; namespace cuda { T; }
 
#define VECCORE_DECLARE_CUDA_TYPE(T)                                           \
  template <> struct CudaTypeTraits<T> { using Type = cuda::T; };              \
 
#define VECCORE_DECLARE_CUDA_CLASS(x)                                          \
  VECCORE_DECLARE_CUDA(class x) VECCORE_DECLARE_CUDA_TYPE(x)
 
#define VECCORE_DECLARE_CUDA_STRUCT(x)                                         \
  VECCORE_DECLARE_CUDA(struct x) VECCORE_DECLARE_CUDA_TYPE(x)
 
// clang-format on
 
#endif // defined (VECCORE_ENABLE_CUDA)
 
#endif