130f4f45fSnbeams // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 230f4f45fSnbeams // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 330f4f45fSnbeams // All Rights reserved. See files LICENSE and NOTICE for details. 430f4f45fSnbeams // 530f4f45fSnbeams // This file is part of CEED, a collection of benchmarks, miniapps, software 630f4f45fSnbeams // libraries and APIs for efficient high-order finite element and spectral 730f4f45fSnbeams // element discretizations for exascale applications. For more information and 830f4f45fSnbeams // source code availability see http://github.com/ceed. 930f4f45fSnbeams // 1030f4f45fSnbeams // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 1130f4f45fSnbeams // a collaborative effort of two U.S. Department of Energy organizations (Office 1230f4f45fSnbeams // of Science and the National Nuclear Security Administration) responsible for 1330f4f45fSnbeams // the planning and preparation of a capable exascale ecosystem, including 1430f4f45fSnbeams // software, applications, hardware, advanced system engineering and early 1530f4f45fSnbeams // testbed platforms, in support of the nation's exascale computing imperative. 163d576824SJeremy L Thompson 1730f4f45fSnbeams #ifndef _ceed_hip_compile_h 1830f4f45fSnbeams #define _ceed_hip_compile_h 1930f4f45fSnbeams 20ec3da8bcSJed Brown #include <ceed/ceed.h> 2130f4f45fSnbeams #include <hip/hip_runtime.h> 2230f4f45fSnbeams 23*0d0321e0SJeremy L Thompson static inline CeedInt CeedDivUpInt(CeedInt numerator, CeedInt denominator) { 24*0d0321e0SJeremy L Thompson return (numerator + denominator - 1) / denominator; 25*0d0321e0SJeremy L Thompson } 26*0d0321e0SJeremy L Thompson 27cf7477b8Snbeams CEED_INTERN int CeedCompileHip(Ceed ceed, const char *source, 28*0d0321e0SJeremy L Thompson hipModule_t *module, const CeedInt numopts, ...); 2930f4f45fSnbeams 30cf7477b8Snbeams CEED_INTERN int CeedGetKernelHip(Ceed ceed, hipModule_t module, 31*0d0321e0SJeremy L Thompson const char *name, hipFunction_t *kernel); 3230f4f45fSnbeams 3330f4f45fSnbeams CEED_INTERN int CeedRunKernelHip(Ceed ceed, hipFunction_t kernel, 34*0d0321e0SJeremy L Thompson const int grid_size, 35*0d0321e0SJeremy L Thompson const int block_size, void **args); 3630f4f45fSnbeams 3730f4f45fSnbeams CEED_INTERN int CeedRunKernelDimHip(Ceed ceed, hipFunction_t kernel, 38*0d0321e0SJeremy L Thompson const int grid_size, 39*0d0321e0SJeremy L Thompson const int block_size_x, const int block_size_y, 40*0d0321e0SJeremy L Thompson const int block_size_z, void **args); 4130f4f45fSnbeams 4230f4f45fSnbeams CEED_INTERN int CeedRunKernelDimSharedHip(Ceed ceed, hipFunction_t kernel, 43*0d0321e0SJeremy L Thompson const int grid_size, const int block_size_x, const int block_size_y, 44*0d0321e0SJeremy L Thompson const int block_size_z, const int shared_mem_size, void **args); 4530f4f45fSnbeams 463d576824SJeremy L Thompson #endif // _ceed_hip_compile_h 47