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 <sstream> 20 #include <stdarg.h> 21 #include <string.h> 22 #include <hip/hiprtc.h> 23 #include "ceed-hip-common.h" 24 #include "ceed-hip-compile.h" 25 26 #define CeedChk_hiprtc(ceed, x) \ 27 do { \ 28 hiprtcResult result = static_cast<hiprtcResult>(x); \ 29 if (result != HIPRTC_SUCCESS) \ 30 return CeedError((ceed), CEED_ERROR_BACKEND, hiprtcGetErrorString(result)); \ 31 } while (0) 32 33 //------------------------------------------------------------------------------ 34 // Compile HIP kernel 35 //------------------------------------------------------------------------------ 36 int CeedCompileHip(Ceed ceed, const char *source, hipModule_t *module, 37 const CeedInt numopts, ...) { 38 int ierr; 39 hipFree(0); // Make sure a Context exists for hiprtc 40 hiprtcProgram prog; 41 42 // Add hip runtime include to string for generation 43 std::ostringstream code; 44 code << "\n#include <hip/hip_runtime.h>\n"; 45 46 // Macro definitions 47 // Get kernel specific options, such as kernel constants 48 const int optssize = 2; 49 const char *opts[optssize]; 50 if (numopts > 0) { 51 va_list args; 52 va_start(args, numopts); 53 char *name; 54 int val; 55 for (int i = 0; i < numopts; i++) { 56 name = va_arg(args, char *); 57 val = va_arg(args, int); 58 code << "#define " << name << " " << val << "\n"; 59 } 60 va_end(args); 61 } 62 63 // Standard backend options 64 if (CEED_SCALAR_TYPE == CEED_SCALAR_FP32) { 65 code << "#define CeedScalar float\n"; 66 } 67 else { 68 code << "#define CeedScalar double\n"; 69 } 70 code << "#define CeedInt int\n"; 71 code << "#define CEED_ERROR_SUCCESS 0\n\n"; 72 73 // Non-macro options 74 opts[0] = "-default-device"; 75 struct hipDeviceProp_t prop; 76 Ceed_Hip *ceed_data; 77 ierr = CeedGetData(ceed, (void **)&ceed_data); CeedChkBackend(ierr); 78 CeedChk_Hip(ceed, hipGetDeviceProperties(&prop, ceed_data->deviceId)); 79 std::string archArg = "--gpu-architecture=" + std::string(prop.gcnArchName); 80 opts[1] = archArg.c_str(); 81 82 // Add string source argument provided in call 83 code << source; 84 85 // Create Program 86 CeedChk_hiprtc(ceed, hiprtcCreateProgram(&prog, code.str().c_str(), NULL, 0, NULL, NULL)); 87 88 // Compile kernel 89 hiprtcResult result = hiprtcCompileProgram(prog, optssize, opts); 90 if (result != HIPRTC_SUCCESS) { 91 size_t logsize; 92 CeedChk_hiprtc(ceed, hiprtcGetProgramLogSize(prog, &logsize)); 93 char *log; 94 ierr = CeedMalloc(logsize, &log); CeedChkBackend(ierr); 95 CeedChk_hiprtc(ceed, hiprtcGetProgramLog(prog, log)); 96 return CeedError(ceed, CEED_ERROR_BACKEND, "%s\n%s", hiprtcGetErrorString(result), log); 97 } 98 99 size_t ptxsize; 100 CeedChk_hiprtc(ceed, hiprtcGetCodeSize(prog, &ptxsize)); 101 char *ptx; 102 ierr = CeedMalloc(ptxsize, &ptx); CeedChkBackend(ierr); 103 CeedChk_hiprtc(ceed, hiprtcGetCode(prog, ptx)); 104 CeedChk_hiprtc(ceed, hiprtcDestroyProgram(&prog)); 105 106 CeedChk_Hip(ceed, hipModuleLoadData(module, ptx)); 107 ierr = CeedFree(&ptx); CeedChkBackend(ierr); 108 109 return CEED_ERROR_SUCCESS; 110 } 111 112 //------------------------------------------------------------------------------ 113 // Get HIP kernel 114 //------------------------------------------------------------------------------ 115 int CeedGetKernelHip(Ceed ceed, hipModule_t module, const char *name, 116 hipFunction_t *kernel) { 117 118 CeedChk_Hip(ceed, hipModuleGetFunction(kernel, module, name)); 119 return CEED_ERROR_SUCCESS; 120 } 121 122 //------------------------------------------------------------------------------ 123 // Run HIP kernel 124 //------------------------------------------------------------------------------ 125 int CeedRunKernelHip(Ceed ceed, hipFunction_t kernel, const int gridSize, 126 const int blockSize, void **args) { 127 CeedChk_Hip(ceed, hipModuleLaunchKernel(kernel, gridSize, 1, 1, blockSize, 1, 128 1, 0, NULL, args, NULL)); 129 return CEED_ERROR_SUCCESS; 130 } 131 132 //------------------------------------------------------------------------------ 133 // Run HIP kernel for spatial dimension 134 //------------------------------------------------------------------------------ 135 int CeedRunKernelDimHip(Ceed ceed, hipFunction_t kernel, const int gridSize, 136 const int blockSizeX, const int blockSizeY, 137 const int blockSizeZ, void **args) { 138 CeedChk_Hip(ceed, hipModuleLaunchKernel(kernel, gridSize, 1, 1, 139 blockSizeX, blockSizeY, blockSizeZ, 140 0, NULL, args, NULL)); 141 return CEED_ERROR_SUCCESS; 142 } 143 144 //------------------------------------------------------------------------------ 145 // Run HIP kernel for spatial dimension with shared memory 146 //------------------------------------------------------------------------------ 147 int CeedRunKernelDimSharedHip(Ceed ceed, hipFunction_t kernel, const int gridSize, 148 const int blockSizeX, const int blockSizeY, 149 const int blockSizeZ, const int sharedMemSize, 150 void **args) { 151 CeedChk_Hip(ceed, hipModuleLaunchKernel(kernel, gridSize, 1, 1, 152 blockSizeX, blockSizeY, blockSizeZ, 153 sharedMemSize, NULL, args, NULL)); 154 return CEED_ERROR_SUCCESS; 155 } 156