| 1 | //===- AMDGPUArchByHIP.cpp - list AMDGPU installed ----------*- C++ -*-----===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // This file implements a tool for detecting name of AMDGPU installed in system |
| 10 | // using HIP runtime. This tool is used by AMDGPU OpenMP and HIP driver. |
| 11 | // |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
| 14 | #include "llvm/ADT/STLExtras.h" |
| 15 | #include "llvm/ADT/Sequence.h" |
| 16 | #include "llvm/Support/CommandLine.h" |
| 17 | #include "llvm/Support/ConvertUTF.h" |
| 18 | #include "llvm/Support/DynamicLibrary.h" |
| 19 | #include "llvm/Support/Error.h" |
| 20 | #include "llvm/Support/FileSystem.h" |
| 21 | #include "llvm/Support/Path.h" |
| 22 | #include "llvm/Support/Process.h" |
| 23 | #include "llvm/Support/Program.h" |
| 24 | #include "llvm/Support/VersionTuple.h" |
| 25 | #include "llvm/Support/raw_ostream.h" |
| 26 | #include <algorithm> |
| 27 | #include <string> |
| 28 | #include <vector> |
| 29 | |
| 30 | #ifdef _WIN32 |
| 31 | #include <windows.h> |
| 32 | #endif |
| 33 | |
| 34 | using namespace llvm; |
| 35 | |
| 36 | // R0600 struct layout (HIP 6.x+) |
| 37 | typedef struct alignas(8) { |
| 38 | char padding[1160]; |
| 39 | char gcnArchName[256]; |
| 40 | char padding2[56]; |
| 41 | } hipDeviceProp_tR0600; |
| 42 | |
| 43 | // R0000 struct layout (legacy) |
| 44 | typedef struct alignas(8) { |
| 45 | char padding[396]; |
| 46 | char gcnArchName[256]; |
| 47 | char padding2[1024]; |
| 48 | } hipDeviceProp_tR0000; |
| 49 | |
| 50 | typedef enum { |
| 51 | hipSuccess = 0, |
| 52 | } hipError_t; |
| 53 | |
| 54 | typedef hipError_t (*hipGetDeviceCount_t)(int *); |
| 55 | typedef hipError_t (*hipGetDevicePropertiesR0600_t)(hipDeviceProp_tR0600 *, |
| 56 | int); |
| 57 | typedef hipError_t (*hipGetDevicePropertiesR0000_t)(hipDeviceProp_tR0000 *, |
| 58 | int); |
| 59 | typedef hipError_t (*hipGetDeviceProperties_t)(hipDeviceProp_tR0000 *, int); |
| 60 | typedef hipError_t (*hipRuntimeGetVersion_t)(int *); |
| 61 | typedef const char *(*hipGetErrorString_t)(hipError_t); |
| 62 | |
| 63 | extern cl::opt<bool> Verbose; |
| 64 | |
| 65 | cl::OptionCategory AMDGPUArchByHIPCategory("amdgpu-arch (HIP) options" ); |
| 66 | |
| 67 | enum class HipApiVersion { |
| 68 | Auto, // Automatic fallback (R0600 -> R0000 -> unversioned) |
| 69 | R0600, // Force R0600 API (HIP 6.x+) |
| 70 | R0000, // Force R0000 API (legacy HIP) |
| 71 | Unversioned // Force unversioned API (very old HIP) |
| 72 | }; |
| 73 | |
| 74 | static cl::opt<HipApiVersion> HipApi( |
| 75 | "hip-api-version" , cl::desc("Select HIP API version for device properties" ), |
| 76 | cl::values(clEnumValN(HipApiVersion::Auto, "auto" , |
| 77 | "Auto-detect (R0600 -> R0000 -> unversioned)" ), |
| 78 | clEnumValN(HipApiVersion::R0600, "r0600" , "Force R0600 API" ), |
| 79 | clEnumValN(HipApiVersion::R0000, "r0000" , "Force R0000 API" ), |
| 80 | clEnumValN(HipApiVersion::Unversioned, "unversioned" , |
| 81 | "Force unversioned API" )), |
| 82 | cl::init(Val: HipApiVersion::Auto), cl::cat(AMDGPUArchByHIPCategory)); |
| 83 | |
| 84 | #ifdef _WIN32 |
| 85 | static std::vector<std::string> getSearchPaths() { |
| 86 | std::vector<std::string> Paths; |
| 87 | |
| 88 | // Get the directory of the current executable |
| 89 | if (auto MainExe = sys::fs::getMainExecutable(nullptr, nullptr); |
| 90 | !MainExe.empty()) |
| 91 | Paths.push_back(sys::path::parent_path(MainExe).str()); |
| 92 | |
| 93 | // Get the system directory |
| 94 | wchar_t SystemDirectory[MAX_PATH]; |
| 95 | if (GetSystemDirectoryW(SystemDirectory, MAX_PATH) > 0) { |
| 96 | std::string Utf8SystemDir; |
| 97 | if (convertUTF16ToUTF8String( |
| 98 | ArrayRef<UTF16>(reinterpret_cast<const UTF16 *>(SystemDirectory), |
| 99 | wcslen(SystemDirectory)), |
| 100 | Utf8SystemDir)) |
| 101 | Paths.push_back(Utf8SystemDir); |
| 102 | } |
| 103 | |
| 104 | // Get the Windows directory |
| 105 | wchar_t WindowsDirectory[MAX_PATH]; |
| 106 | if (GetWindowsDirectoryW(WindowsDirectory, MAX_PATH) > 0) { |
| 107 | std::string Utf8WindowsDir; |
| 108 | if (convertUTF16ToUTF8String( |
| 109 | ArrayRef<UTF16>(reinterpret_cast<const UTF16 *>(WindowsDirectory), |
| 110 | wcslen(WindowsDirectory)), |
| 111 | Utf8WindowsDir)) |
| 112 | Paths.push_back(Utf8WindowsDir); |
| 113 | } |
| 114 | |
| 115 | // Get the current working directory |
| 116 | SmallVector<char, 256> CWD; |
| 117 | if (sys::fs::current_path(CWD)) |
| 118 | Paths.push_back(std::string(CWD.begin(), CWD.end())); |
| 119 | |
| 120 | // Get the PATH environment variable |
| 121 | if (std::optional<std::string> PathEnv = sys::Process::GetEnv("PATH" )) { |
| 122 | SmallVector<StringRef, 16> PathList; |
| 123 | StringRef(*PathEnv).split(PathList, sys::EnvPathSeparator); |
| 124 | for (auto &Path : PathList) |
| 125 | Paths.push_back(Path.str()); |
| 126 | } |
| 127 | |
| 128 | return Paths; |
| 129 | } |
| 130 | |
| 131 | // Custom comparison function for dll name |
| 132 | static bool compareVersions(StringRef A, StringRef B) { |
| 133 | auto ParseVersion = [](StringRef S) -> VersionTuple { |
| 134 | StringRef Filename = sys::path::filename(S); |
| 135 | size_t Pos = Filename.find_last_of('_'); |
| 136 | if (Pos == StringRef::npos) |
| 137 | return VersionTuple(); |
| 138 | |
| 139 | StringRef VerStr = Filename.substr(Pos + 1); |
| 140 | size_t DotPos = VerStr.find('.'); |
| 141 | if (DotPos != StringRef::npos) |
| 142 | VerStr = VerStr.substr(0, DotPos); |
| 143 | |
| 144 | VersionTuple Vt; |
| 145 | (void)Vt.tryParse(VerStr); |
| 146 | return Vt; |
| 147 | }; |
| 148 | |
| 149 | VersionTuple VtA = ParseVersion(A); |
| 150 | VersionTuple VtB = ParseVersion(B); |
| 151 | return VtA > VtB; |
| 152 | } |
| 153 | #endif |
| 154 | |
| 155 | // On Windows, prefer amdhip64_n.dll where n is ROCm major version and greater |
| 156 | // value of n takes precedence. If amdhip64_n.dll is not found, fall back to |
| 157 | // amdhip64.dll. The reason is that a normal driver installation only has |
| 158 | // amdhip64_n.dll but we do not know what n is since this program may be used |
| 159 | // with a future version of HIP runtime. |
| 160 | // |
| 161 | // On Linux, always use default libamdhip64.so. |
| 162 | static std::pair<std::string, bool> findNewestHIPDLL() { |
| 163 | #ifdef _WIN32 |
| 164 | StringRef HipDLLPrefix = "amdhip64_" ; |
| 165 | StringRef HipDLLSuffix = ".dll" ; |
| 166 | |
| 167 | std::vector<std::string> SearchPaths = getSearchPaths(); |
| 168 | std::vector<std::string> DLLNames; |
| 169 | |
| 170 | for (const auto &Dir : SearchPaths) { |
| 171 | std::error_code EC; |
| 172 | for (sys::fs::directory_iterator DirIt(Dir, EC), DirEnd; |
| 173 | DirIt != DirEnd && !EC; DirIt.increment(EC)) { |
| 174 | StringRef Filename = sys::path::filename(DirIt->path()); |
| 175 | if (Filename.starts_with(HipDLLPrefix) && |
| 176 | Filename.ends_with(HipDLLSuffix)) |
| 177 | DLLNames.push_back(sys::path::convert_to_slash(DirIt->path())); |
| 178 | } |
| 179 | } |
| 180 | |
| 181 | if (DLLNames.empty()) |
| 182 | return {"amdhip64.dll" , true}; |
| 183 | |
| 184 | llvm::sort(DLLNames, compareVersions); |
| 185 | return {DLLNames[0], false}; |
| 186 | #else |
| 187 | // On Linux, fallback to default shared object |
| 188 | return {"libamdhip64.so" , true}; |
| 189 | #endif |
| 190 | } |
| 191 | |
| 192 | int printGPUsByHIP() { |
| 193 | auto [DynamicHIPPath, IsFallback] = findNewestHIPDLL(); |
| 194 | |
| 195 | if (Verbose) { |
| 196 | if (IsFallback) |
| 197 | outs() << "Using default HIP runtime: " << DynamicHIPPath << '\n'; |
| 198 | else |
| 199 | outs() << "Found HIP runtime: " << DynamicHIPPath << '\n'; |
| 200 | } |
| 201 | |
| 202 | std::string ErrMsg; |
| 203 | auto DynlibHandle = std::make_unique<llvm::sys::DynamicLibrary>( |
| 204 | args: llvm::sys::DynamicLibrary::getPermanentLibrary(filename: DynamicHIPPath.c_str(), |
| 205 | errMsg: &ErrMsg)); |
| 206 | if (!DynlibHandle->isValid()) { |
| 207 | if (Verbose) |
| 208 | llvm::errs() << "Failed to load " << DynamicHIPPath << ": " << ErrMsg |
| 209 | << '\n'; |
| 210 | return 1; |
| 211 | } |
| 212 | |
| 213 | if (Verbose) |
| 214 | outs() << "Successfully loaded HIP runtime library\n" ; |
| 215 | |
| 216 | #define DYNAMIC_INIT_HIP(SYMBOL) \ |
| 217 | { \ |
| 218 | void *SymbolPtr = DynlibHandle->getAddressOfSymbol(#SYMBOL); \ |
| 219 | if (!SymbolPtr) { \ |
| 220 | llvm::errs() << "Failed to find symbol " << #SYMBOL << '\n'; \ |
| 221 | return 1; \ |
| 222 | } \ |
| 223 | if (Verbose) \ |
| 224 | outs() << "Found symbol: " << #SYMBOL << '\n'; \ |
| 225 | SYMBOL = reinterpret_cast<decltype(SYMBOL)>(SymbolPtr); \ |
| 226 | } |
| 227 | |
| 228 | hipGetDeviceCount_t hipGetDeviceCount; |
| 229 | hipRuntimeGetVersion_t hipRuntimeGetVersion = nullptr; |
| 230 | hipGetDevicePropertiesR0600_t hipGetDevicePropertiesR0600 = nullptr; |
| 231 | hipGetDevicePropertiesR0000_t hipGetDevicePropertiesR0000 = nullptr; |
| 232 | hipGetDeviceProperties_t hipGetDeviceProperties = nullptr; |
| 233 | hipGetErrorString_t hipGetErrorString = nullptr; |
| 234 | |
| 235 | DYNAMIC_INIT_HIP(hipGetDeviceCount); |
| 236 | |
| 237 | #undef DYNAMIC_INIT_HIP |
| 238 | |
| 239 | auto LoadSymbol = [&](const char *Name, auto &FuncPtr, |
| 240 | const char *Desc = "" ) { |
| 241 | void *Sym = DynlibHandle->getAddressOfSymbol(symbolName: Name); |
| 242 | if (Sym) { |
| 243 | FuncPtr = reinterpret_cast<decltype(FuncPtr)>(Sym); |
| 244 | if (Verbose) |
| 245 | outs() << "Found symbol: " << Name << (Desc[0] ? " " : "" ) << Desc |
| 246 | << '\n'; |
| 247 | return true; |
| 248 | } |
| 249 | return false; |
| 250 | }; |
| 251 | |
| 252 | LoadSymbol("hipGetErrorString" , hipGetErrorString); |
| 253 | |
| 254 | if (LoadSymbol("hipRuntimeGetVersion" , hipRuntimeGetVersion)) { |
| 255 | int RuntimeVersion = 0; |
| 256 | if (hipRuntimeGetVersion(&RuntimeVersion) == hipSuccess) { |
| 257 | int Major = RuntimeVersion / 10000000; |
| 258 | int Minor = (RuntimeVersion / 100000) % 100; |
| 259 | int Patch = RuntimeVersion % 100000; |
| 260 | if (Verbose) |
| 261 | outs() << "HIP Runtime Version: " << Major << "." << Minor << "." |
| 262 | << Patch << '\n'; |
| 263 | } |
| 264 | } |
| 265 | |
| 266 | LoadSymbol("hipGetDevicePropertiesR0600" , hipGetDevicePropertiesR0600, |
| 267 | "(HIP 6.x+ API)" ); |
| 268 | LoadSymbol("hipGetDevicePropertiesR0000" , hipGetDevicePropertiesR0000, |
| 269 | "(legacy API)" ); |
| 270 | if (!hipGetDevicePropertiesR0600 && !hipGetDevicePropertiesR0000) |
| 271 | LoadSymbol("hipGetDeviceProperties" , hipGetDeviceProperties, |
| 272 | "(unversioned legacy API)" ); |
| 273 | |
| 274 | int DeviceCount; |
| 275 | if (Verbose) |
| 276 | outs() << "Calling hipGetDeviceCount...\n" ; |
| 277 | hipError_t Err = hipGetDeviceCount(&DeviceCount); |
| 278 | if (Err != hipSuccess) { |
| 279 | llvm::errs() << "Failed to get device count" ; |
| 280 | if (hipGetErrorString) { |
| 281 | llvm::errs() << ": " << hipGetErrorString(Err); |
| 282 | } |
| 283 | llvm::errs() << " (error code: " << Err << ")\n" ; |
| 284 | return 1; |
| 285 | } |
| 286 | |
| 287 | if (Verbose) |
| 288 | outs() << "Found " << DeviceCount << " device(s)\n" ; |
| 289 | |
| 290 | auto TryGetProperties = [&](auto *ApiFunc, auto *DummyProp, const char *Name, |
| 291 | int DeviceId) -> std::string { |
| 292 | if (!ApiFunc) |
| 293 | return "" ; |
| 294 | |
| 295 | if (Verbose) |
| 296 | outs() << "Using " << Name << "...\n" ; |
| 297 | |
| 298 | using PropType = std::remove_pointer_t<decltype(DummyProp)>; |
| 299 | PropType Prop; |
| 300 | hipError_t Err = ApiFunc(&Prop, DeviceId); |
| 301 | |
| 302 | if (Err == hipSuccess) { |
| 303 | if (Verbose) { |
| 304 | outs() << Name << " struct: sizeof = " << sizeof(PropType) |
| 305 | << " bytes, offsetof(gcnArchName) = " |
| 306 | << offsetof(PropType, gcnArchName) << " bytes\n" ; |
| 307 | } |
| 308 | return Prop.gcnArchName; |
| 309 | } |
| 310 | |
| 311 | if (Verbose) |
| 312 | llvm::errs() << Name << " failed (error code: " << Err << ")\n" ; |
| 313 | return "" ; |
| 314 | }; |
| 315 | |
| 316 | for (auto I : llvm::seq(Size: DeviceCount)) { |
| 317 | if (Verbose) |
| 318 | outs() << "Processing device " << I << "...\n" ; |
| 319 | |
| 320 | std::string ArchName; |
| 321 | auto TryR0600 = [&](int Dev) -> bool { |
| 322 | if (!hipGetDevicePropertiesR0600) |
| 323 | return false; |
| 324 | ArchName = TryGetProperties(hipGetDevicePropertiesR0600, |
| 325 | (hipDeviceProp_tR0600 *)nullptr, |
| 326 | "R0600 API (HIP 6.x+)" , Dev); |
| 327 | return !ArchName.empty(); |
| 328 | }; |
| 329 | auto TryR0000 = [&](int Dev) -> bool { |
| 330 | if (!hipGetDevicePropertiesR0000) |
| 331 | return false; |
| 332 | ArchName = TryGetProperties(hipGetDevicePropertiesR0000, |
| 333 | (hipDeviceProp_tR0000 *)nullptr, |
| 334 | "R0000 API (legacy HIP)" , Dev); |
| 335 | return !ArchName.empty(); |
| 336 | }; |
| 337 | auto TryUnversioned = [&](int Dev) -> bool { |
| 338 | if (!hipGetDeviceProperties) |
| 339 | return false; |
| 340 | ArchName = TryGetProperties(hipGetDeviceProperties, |
| 341 | (hipDeviceProp_tR0000 *)nullptr, |
| 342 | "unversioned API (very old HIP)" , Dev); |
| 343 | return !ArchName.empty(); |
| 344 | }; |
| 345 | |
| 346 | [[maybe_unused]] bool OK; |
| 347 | switch (HipApi) { |
| 348 | case HipApiVersion::Auto: |
| 349 | OK = TryR0600(I) || TryR0000(I) || TryUnversioned(I); |
| 350 | break; |
| 351 | case HipApiVersion::R0600: |
| 352 | OK = TryR0600(I); |
| 353 | break; |
| 354 | case HipApiVersion::R0000: |
| 355 | OK = TryR0000(I); |
| 356 | break; |
| 357 | case HipApiVersion::Unversioned: |
| 358 | OK = TryUnversioned(I); |
| 359 | } |
| 360 | |
| 361 | if (ArchName.empty()) { |
| 362 | llvm::errs() << "Failed to get device properties for device " << I |
| 363 | << " - no APIs available or all failed\n" ; |
| 364 | return 1; |
| 365 | } |
| 366 | |
| 367 | if (Verbose) |
| 368 | outs() << "Device " << I << " arch name: " ; |
| 369 | llvm::outs() << ArchName << '\n'; |
| 370 | } |
| 371 | |
| 372 | return 0; |
| 373 | } |
| 374 | |