offload_acc_omp.h

 
#ifndef OFFLOAD_ACC_OMP_H
#define OFFLOAD_ACC_OMP_H
 
#include "ascot5.h"
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_PARALLEL_LOOP_ALL_LEVELS \
    str_pragma(omp target teams distribute parallel for simd)
#elif defined(GPU) && defined(_OPENACC)
#define GPU_PARALLEL_LOOP_ALL_LEVELS str_pragma(acc parallel loop)
#else
#define GPU_PARALLEL_LOOP_ALL_LEVELS str_pragma(omp simd)
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_PARALLEL_LOOP_ALL_LEVELS_REDUCTION(...) \
    str_pragma(omp target teams distribute parallel for simd \
    reduction(+:__VA_ARGS__))
#elif defined(GPU) && defined(_OPENACC)
#define GPU_PARALLEL_LOOP_ALL_LEVELS_REDUCTION(...) \
    str_pragma(acc parallel loop reduction(+:__VA_ARGS__))
#else
#define GPU_PARALLEL_LOOP_ALL_LEVELS_REDUCTION(...)
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define DECLARE_TARGET str_pragma(omp declare target)
#elif defined(GPU) && defined(_OPENACC)
#define DECLARE_TARGET str_pragma(acc routine seq)
#else
#define DECLARE_TARGET
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define DECLARE_TARGET_END str_pragma(omp end declare target)
#elif defined(GPU) && defined(_OPENACC)
#define DECLARE_TARGET_END
#else
#define DECLARE_TARGET_END
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_MAP_TO_DEVICE(...) \
    str_pragma(omp target enter data map (to: __VA_ARGS__))
#elif defined(GPU) && defined(_OPENACC)
#define GPU_MAP_TO_DEVICE(...) str_pragma(acc enter data copyin (__VA_ARGS__))
#else
#define GPU_MAP_TO_DEVICE(...)
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_UPDATE_FROM_DEVICE(...) \
    str_pragma(omp target update from (__VA_ARGS__))
#elif defined(GPU) && defined(_OPENACC)
#define GPU_UPDATE_FROM_DEVICE(...) str_pragma(acc update host (__VA_ARGS__))
#else
#define GPU_UPDATE_FROM_DEVICE(...)
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_MAP_FROM_DEVICE(...) \
    str_pragma(omp target exit data map (from: __VA_ARGS__))
#elif defined(GPU) && defined(_OPENACC)
#define GPU_MAP_FROM_DEVICE(...) \
    str_pragma(acc exit data copyout (__VA_ARGS__))
#else
#define GPU_MAP_FROM_DEVICE(...)
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_MAP_DELETE_DEVICE(...) \
    str_pragma(omp target exit data (delete: __VA_ARGS__))
#elif defined(GPU) && defined(_OPENACC)
#define GPU_MAP_DELETE_DEVICE(...) \
    str_pragma(acc exit data delete (__VA_ARGS__))
#else
#define GPU_MAP_DELETE_DEVICE(...)
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_ATOMIC str_pragma(omp atomic)
#elif defined(GPU) && defined(_OPENACC)
#define GPU_ATOMIC str_pragma(omp atomic)
#else
#define GPU_ATOMIC str_pragma(omp atomic)
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define DECLARE_TARGET_SIMD
#elif defined(GPU) && defined(_OPENACC)
#define DECLARE_TARGET_SIMD
#else
#define DECLARE_TARGET_SIMD str_pragma(omp declare simd)
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define DECLARE_TARGET_SIMD_UNIFORM(...)
#elif defined(GPU) && defined(_OPENACC)
#define DECLARE_TARGET_SIMD_UNIFORM(...) \
    str_pragma(omp declare simd uniform (__VA_ARGS__))
#else
#define DECLARE_TARGET_SIMD_UNIFORM(...)
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_DECLARE_TARGET_SIMD str_pragma(omp declare target)
#elif defined(GPU) && defined(_OPENACC)
#define GPU_DECLARE_TARGET_SIMD str_pragma(acc routine seq)
#else
#define GPU_DECLARE_TARGET_SIMD str_pragma(omp declare simd)
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_DECLARE_TARGET_SIMD_END str_pragma(omp end declare target)
#elif defined(GPU) && defined(_OPENACC)
#define GPU_DECLARE_TARGET_SIMD_END
#else
#define GPU_DECLARE_TARGET_SIMD_END
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_DECLARE_TARGET_SIMD_UNIFORM(...) str_pragma(omp declare target)
#elif defined(GPU) && defined(_OPENACC)
#define GPU_DECLARE_TARGET_SIMD_UNIFORM(...) str_pragma(acc routine seq)
#else
#define GPU_DECLARE_TARGET_SIMD_UNIFORM(...) \
    str_pragma(omp declare simd uniform (__VA_ARGS__))
#endif
 
#if defined(GPU) && defined(_OPENMP)
#define GPU_DECLARE_TARGET_SIMD_UNIFORM_END str_pragma(omp end declare target)
#elif defined(GPU) && defined(_OPENACC)
#define GPU_DECLARE_TARGET_SIMD_UNIFORM_END
#else
#define GPU_DECLARE_TARGET_SIMD_UNIFORM_END
#endif
 
#if defined(GPU)
#define OMP_PARALLEL_CPU_ONLY
#else
#define OMP_PARALLEL_CPU_ONLY str_pragma(omp parallel)
#endif
 
#if defined(GPU) && defined(_OPENACC)
#define GPU_SEQUENTIAL_LOOP str_pragma(acc loop seq)
#else
#define GPU_SEQUENTIAL_LOOP
#endif
 
#if defined(GPU) && defined(_OPENACC)
#define GPU_DATA_IS_MAPPED(...) str_pragma(acc data present(__VA_ARGS__))
#else
#define GPU_DATA_IS_MAPPED(...)
#endif
 
#ifdef _OPENACC
#define NGANGS 1024
 
#define NWORKERS 8
 
#define NVECTORS 32
#endif // _OPENACC
 
#endif