1 // Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 2 // Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 3 // All Rights reserved. See files LICENSE and NOTICE for details. 4 // 5 // This file is part of CEED, a collection of benchmarks, miniapps, software 6 // libraries and APIs for efficient high-order finite element and spectral 7 // element discretizations for exascale applications. For more information and 8 // source code availability see http://github.com/ceed. 9 // 10 // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 11 // a collaborative effort of two U.S. Department of Energy organizations (Office 12 // of Science and the National Nuclear Security Administration) responsible for 13 // the planning and preparation of a capable exascale ecosystem, including 14 // software, applications, hardware, advanced system engineering and early 15 // testbed platforms, in support of the nation's exascale computing imperative. 16 17 #include <ceed/ceed.h> 18 #include <ceed/backend.h> 19 #include <iostream> 20 #include <sstream> 21 #include <string.h> 22 #include "ceed-hip-ref.h" 23 #include "../hip/ceed-hip-compile.h" 24 25 static const char *qReadWrite = QUOTE( 26 template <int SIZE> 27 //------------------------------------------------------------------------------ 28 // Read from quadrature points 29 //------------------------------------------------------------------------------ 30 inline __device__ void readQuads(const CeedInt quad, const CeedInt nquads, const CeedScalar* d_u, CeedScalar* r_u) { 31 for(CeedInt comp = 0; comp < SIZE; ++comp) { 32 r_u[comp] = d_u[quad + nquads * comp]; 33 } 34 } 35 36 //------------------------------------------------------------------------------ 37 // Write at quadrature points 38 //------------------------------------------------------------------------------ 39 template <int SIZE> 40 inline __device__ void writeQuads(const CeedInt quad, const CeedInt nquads, const CeedScalar* r_v, CeedScalar* d_v) { 41 for(CeedInt comp = 0; comp < SIZE; ++comp) { 42 d_v[quad + nquads * comp] = r_v[comp]; 43 } 44 } 45 ); 46 47 //------------------------------------------------------------------------------ 48 // Build QFunction kernel 49 //------------------------------------------------------------------------------ 50 extern "C" int CeedHipBuildQFunction(CeedQFunction qf) { 51 CeedInt ierr; 52 using std::ostringstream; 53 using std::string; 54 CeedQFunction_Hip *data; 55 ierr = CeedQFunctionGetData(qf, (void **)&data); CeedChkBackend(ierr); 56 // QFunction is built 57 if (!data->qFunctionSource) 58 return CEED_ERROR_SUCCESS; 59 60 // QFunction kernel generation 61 CeedInt numinputfields, numoutputfields, size; 62 CeedQFunctionField *qfinputfields, *qfoutputfields; 63 ierr = CeedQFunctionGetFields(qf, &numinputfields, &qfinputfields, &numoutputfields, &qfoutputfields); 64 CeedChkBackend(ierr); 65 66 // Build strings for final kernel 67 string qFunction(data->qFunctionSource); 68 string qReadWriteS(qReadWrite); 69 ostringstream code; 70 string qFunctionName(data->qFunctionName); 71 string kernelName; 72 kernelName = "CeedKernel_Hip_ref_" + qFunctionName; 73 74 // Defintions 75 code << "\n#define CEED_QFUNCTION(name) inline __device__ int name\n"; 76 code << "#define CEED_QFUNCTION_HELPER inline __device__ __forceinline__\n"; 77 code << "#define CeedPragmaSIMD\n"; 78 code << "#define CEED_ERROR_SUCCESS 0\n"; 79 code << "#define CEED_Q_VLA 1\n\n"; 80 code << "typedef struct { const CeedScalar* inputs[16]; CeedScalar* outputs[16]; } Fields_Hip;\n"; 81 code << qReadWriteS; 82 code << qFunction; 83 code << "extern \"C\" __global__ void " << kernelName << "(void *ctx, CeedInt Q, Fields_Hip fields) {\n"; 84 85 // Inputs 86 for (CeedInt i = 0; i < numinputfields; i++) { 87 code << "// Input field "<<i<<"\n"; 88 ierr = CeedQFunctionFieldGetSize(qfinputfields[i], &size); CeedChkBackend(ierr); 89 code << " const CeedInt size_in_"<<i<<" = "<<size<<";\n"; 90 code << " CeedScalar r_q"<<i<<"[size_in_"<<i<<"];\n"; 91 } 92 93 // Outputs 94 for (CeedInt i = 0; i < numoutputfields; i++) { 95 code << "// Output field "<<i<<"\n"; 96 ierr = CeedQFunctionFieldGetSize(qfoutputfields[i], &size); CeedChkBackend(ierr); 97 code << " const CeedInt size_out_"<<i<<" = "<<size<<";\n"; 98 code << " CeedScalar r_qq"<<i<<"[size_out_"<<i<<"];\n"; 99 } 100 101 // Setup input/output arrays 102 code << " const CeedScalar* in["<<numinputfields<<"];\n"; 103 for (CeedInt i = 0; i < numinputfields; i++) { 104 code << " in["<<i<<"] = r_q"<<i<<";\n"; 105 } 106 code << " CeedScalar* out["<<numoutputfields<<"];\n"; 107 for (CeedInt i = 0; i < numoutputfields; i++) { 108 code << " out["<<i<<"] = r_qq"<<i<<";\n"; 109 } 110 111 // Loop over quadrature points 112 code << " for (CeedInt q = blockIdx.x * blockDim.x + threadIdx.x; q < Q; q += blockDim.x * gridDim.x) {\n"; 113 114 // Load inputs 115 for (CeedInt i = 0; i < numinputfields; i++) { 116 code << "// Input field "<<i<<"\n"; 117 code << " readQuads<size_in_"<<i<<">(q, Q, fields.inputs["<<i<<"], r_q"<<i<<");\n"; 118 } 119 // QFunction 120 code << "// QFunction\n"; 121 code << " "<<qFunctionName<<"(ctx, 1, in, out);\n"; 122 123 // Write outputs 124 for (CeedInt i = 0; i < numoutputfields; i++) { 125 code << "// Output field "<<i<<"\n"; 126 code << " writeQuads<size_out_"<<i<<">(q, Q, r_qq"<<i<<", fields.outputs["<<i<<"]);\n"; 127 } 128 code << " }\n"; 129 code << "}\n"; 130 131 // View kernel for debugging 132 Ceed ceed; 133 CeedQFunctionGetCeed(qf, &ceed); 134 CeedDebug(ceed, code.str().c_str()); 135 136 // Compile kernel 137 ierr = CeedCompileHip(ceed, code.str().c_str(), &data->module, 0); 138 CeedChkBackend(ierr); 139 ierr = CeedGetKernelHip(ceed, data->module, kernelName.c_str(), &data->qFunction); 140 CeedChkBackend(ierr); 141 142 // Cleanup 143 ierr = CeedFree(&data->qFunctionSource); CeedChkBackend(ierr); 144 return CEED_ERROR_SUCCESS; 145 } 146 //------------------------------------------------------------------------------ 147