17d8d0e25Snbeams // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 27d8d0e25Snbeams // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 37d8d0e25Snbeams // All Rights reserved. See files LICENSE and NOTICE for details. 47d8d0e25Snbeams // 57d8d0e25Snbeams // This file is part of CEED, a collection of benchmarks, miniapps, software 67d8d0e25Snbeams // libraries and APIs for efficient high-order finite element and spectral 77d8d0e25Snbeams // element discretizations for exascale applications. For more information and 87d8d0e25Snbeams // source code availability see http://github.com/ceed. 97d8d0e25Snbeams // 107d8d0e25Snbeams // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 117d8d0e25Snbeams // a collaborative effort of two U.S. Department of Energy organizations (Office 127d8d0e25Snbeams // of Science and the National Nuclear Security Administration) responsible for 137d8d0e25Snbeams // the planning and preparation of a capable exascale ecosystem, including 147d8d0e25Snbeams // software, applications, hardware, advanced system engineering and early 157d8d0e25Snbeams // testbed platforms, in support of the nation's exascale computing imperative. 16*3d576824SJeremy L Thompson 17*3d576824SJeremy L Thompson #ifndef _ceed_hip_shared_h 18*3d576824SJeremy L Thompson #define _ceed_hip_shared_h 19*3d576824SJeremy L Thompson 20*3d576824SJeremy L Thompson #include <ceed.h> 21*3d576824SJeremy L Thompson #include <ceed-backend.h> 22*3d576824SJeremy L Thompson #include <hip/hip_runtime.h> 237d8d0e25Snbeams #include "../hip/ceed-hip.h" 247d8d0e25Snbeams 257d8d0e25Snbeams typedef struct { 267d8d0e25Snbeams hipModule_t module; 277d8d0e25Snbeams hipFunction_t interp; 287d8d0e25Snbeams hipFunction_t grad; 297d8d0e25Snbeams hipFunction_t weight; 309e31c45bSnbeams CeedInt blksizes[3]; // interp, grad, weight thread block sizes 317d8d0e25Snbeams CeedScalar *d_interp1d; 327d8d0e25Snbeams CeedScalar *d_grad1d; 337d8d0e25Snbeams CeedScalar *d_collograd1d; 347d8d0e25Snbeams CeedScalar *d_qweight1d; 357d8d0e25Snbeams CeedScalar *c_B; 367d8d0e25Snbeams CeedScalar *c_G; 377d8d0e25Snbeams } CeedBasis_Hip_shared; 387d8d0e25Snbeams 397d8d0e25Snbeams typedef struct { 407d8d0e25Snbeams Ceed_Hip base; 417d8d0e25Snbeams } Ceed_Hip_shared; 427d8d0e25Snbeams 437d8d0e25Snbeams CEED_INTERN int CeedBasisCreateTensorH1_Hip_shared(CeedInt dim, CeedInt P1d, 447d8d0e25Snbeams CeedInt Q1d, const CeedScalar *interp1d, const CeedScalar *grad1d, 457d8d0e25Snbeams const CeedScalar *qref1d, const CeedScalar *qweight1d, CeedBasis basis); 46526015bcSnbeams 47526015bcSnbeams CEED_INTERN int CeedHipInitInterp(CeedScalar *d_B, CeedInt P1d, CeedInt Q1d, 48526015bcSnbeams CeedScalar **c_B); 49526015bcSnbeams 507c2e65d5Snbeams CEED_INTERN int CeedHipInitInterpGrad(CeedScalar *d_B, CeedScalar *d_G, 517c2e65d5Snbeams CeedInt P1d, 52526015bcSnbeams CeedInt Q1d, CeedScalar **c_B_ptr, 53526015bcSnbeams CeedScalar **c_G_ptr); 54*3d576824SJeremy L Thompson 55*3d576824SJeremy L Thompson #endif // _ceed_hip_shared_h 56