1 //===-- Host.cpp - Implement OS Host Concept --------------------*- C++ -*-===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file implements the operating system Host concept. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "llvm/Support/Host.h" 15 #include "llvm/ADT/SmallVector.h" 16 #include "llvm/ADT/StringRef.h" 17 #include "llvm/ADT/StringSwitch.h" 18 #include "llvm/ADT/Triple.h" 19 #include "llvm/Config/config.h" 20 #include "llvm/Support/Debug.h" 21 #include "llvm/Support/FileSystem.h" 22 #include "llvm/Support/raw_ostream.h" 23 #include <string.h> 24 25 // Include the platform-specific parts of this class. 26 #ifdef LLVM_ON_UNIX 27 #include "Unix/Host.inc" 28 #endif 29 #ifdef LLVM_ON_WIN32 30 #include "Windows/Host.inc" 31 #endif 32 #ifdef _MSC_VER 33 #include <intrin.h> 34 #endif 35 #if defined(__APPLE__) && (defined(__ppc__) || defined(__powerpc__)) 36 #include <mach/mach.h> 37 #include <mach/mach_host.h> 38 #include <mach/host_info.h> 39 #include <mach/machine.h> 40 #endif 41 42 #define DEBUG_TYPE "host-detection" 43 44 //===----------------------------------------------------------------------===// 45 // 46 // Implementations of the CPU detection routines 47 // 48 //===----------------------------------------------------------------------===// 49 50 using namespace llvm; 51 52 #if defined(__linux__) 53 static ssize_t LLVM_ATTRIBUTE_UNUSED readCpuInfo(void *Buf, size_t Size) { 54 // Note: We cannot mmap /proc/cpuinfo here and then process the resulting 55 // memory buffer because the 'file' has 0 size (it can be read from only 56 // as a stream). 57 58 int FD; 59 std::error_code EC = sys::fs::openFileForRead("/proc/cpuinfo", FD); 60 if (EC) { 61 DEBUG(dbgs() << "Unable to open /proc/cpuinfo: " << EC.message() << "\n"); 62 return -1; 63 } 64 int Ret = read(FD, Buf, Size); 65 int CloseStatus = close(FD); 66 if (CloseStatus) 67 return -1; 68 return Ret; 69 } 70 #endif 71 72 #if defined(i386) || defined(__i386__) || defined(__x86__) || defined(_M_IX86)\ 73 || defined(__x86_64__) || defined(_M_AMD64) || defined (_M_X64) 74 75 /// GetX86CpuIDAndInfo - Execute the specified cpuid and return the 4 values in the 76 /// specified arguments. If we can't run cpuid on the host, return true. 77 static bool GetX86CpuIDAndInfo(unsigned value, unsigned *rEAX, unsigned *rEBX, 78 unsigned *rECX, unsigned *rEDX) { 79 #if defined(__GNUC__) || defined(__clang__) 80 #if defined(__x86_64__) || defined(_M_AMD64) || defined (_M_X64) 81 // gcc doesn't know cpuid would clobber ebx/rbx. Preseve it manually. 82 asm ("movq\t%%rbx, %%rsi\n\t" 83 "cpuid\n\t" 84 "xchgq\t%%rbx, %%rsi\n\t" 85 : "=a" (*rEAX), 86 "=S" (*rEBX), 87 "=c" (*rECX), 88 "=d" (*rEDX) 89 : "a" (value)); 90 return false; 91 #elif defined(i386) || defined(__i386__) || defined(__x86__) || defined(_M_IX86) 92 asm ("movl\t%%ebx, %%esi\n\t" 93 "cpuid\n\t" 94 "xchgl\t%%ebx, %%esi\n\t" 95 : "=a" (*rEAX), 96 "=S" (*rEBX), 97 "=c" (*rECX), 98 "=d" (*rEDX) 99 : "a" (value)); 100 return false; 101 // pedantic #else returns to appease -Wunreachable-code (so we don't generate 102 // postprocessed code that looks like "return true; return false;") 103 #else 104 return true; 105 #endif 106 #elif defined(_MSC_VER) 107 // The MSVC intrinsic is portable across x86 and x64. 108 int registers[4]; 109 __cpuid(registers, value); 110 *rEAX = registers[0]; 111 *rEBX = registers[1]; 112 *rECX = registers[2]; 113 *rEDX = registers[3]; 114 return false; 115 #else 116 return true; 117 #endif 118 } 119 120 /// GetX86CpuIDAndInfoEx - Execute the specified cpuid with subleaf and return the 121 /// 4 values in the specified arguments. If we can't run cpuid on the host, 122 /// return true. 123 static bool GetX86CpuIDAndInfoEx(unsigned value, unsigned subleaf, 124 unsigned *rEAX, unsigned *rEBX, unsigned *rECX, 125 unsigned *rEDX) { 126 #if defined(__x86_64__) || defined(_M_AMD64) || defined (_M_X64) 127 #if defined(__GNUC__) 128 // gcc doesn't know cpuid would clobber ebx/rbx. Preseve it manually. 129 asm ("movq\t%%rbx, %%rsi\n\t" 130 "cpuid\n\t" 131 "xchgq\t%%rbx, %%rsi\n\t" 132 : "=a" (*rEAX), 133 "=S" (*rEBX), 134 "=c" (*rECX), 135 "=d" (*rEDX) 136 : "a" (value), 137 "c" (subleaf)); 138 return false; 139 #elif defined(_MSC_VER) 140 // __cpuidex was added in MSVC++ 9.0 SP1 141 #if (_MSC_VER > 1500) || (_MSC_VER == 1500 && _MSC_FULL_VER >= 150030729) 142 int registers[4]; 143 __cpuidex(registers, value, subleaf); 144 *rEAX = registers[0]; 145 *rEBX = registers[1]; 146 *rECX = registers[2]; 147 *rEDX = registers[3]; 148 return false; 149 #else 150 return true; 151 #endif 152 #else 153 return true; 154 #endif 155 #elif defined(i386) || defined(__i386__) || defined(__x86__) || defined(_M_IX86) 156 #if defined(__GNUC__) 157 asm ("movl\t%%ebx, %%esi\n\t" 158 "cpuid\n\t" 159 "xchgl\t%%ebx, %%esi\n\t" 160 : "=a" (*rEAX), 161 "=S" (*rEBX), 162 "=c" (*rECX), 163 "=d" (*rEDX) 164 : "a" (value), 165 "c" (subleaf)); 166 return false; 167 #elif defined(_MSC_VER) 168 __asm { 169 mov eax,value 170 mov ecx,subleaf 171 cpuid 172 mov esi,rEAX 173 mov dword ptr [esi],eax 174 mov esi,rEBX 175 mov dword ptr [esi],ebx 176 mov esi,rECX 177 mov dword ptr [esi],ecx 178 mov esi,rEDX 179 mov dword ptr [esi],edx 180 } 181 return false; 182 #else 183 return true; 184 #endif 185 #else 186 return true; 187 #endif 188 } 189 190 static bool OSHasAVXSupport() { 191 #if defined(__GNUC__) 192 // Check xgetbv; this uses a .byte sequence instead of the instruction 193 // directly because older assemblers do not include support for xgetbv and 194 // there is no easy way to conditionally compile based on the assembler used. 195 int rEAX, rEDX; 196 __asm__ (".byte 0x0f, 0x01, 0xd0" : "=a" (rEAX), "=d" (rEDX) : "c" (0)); 197 #elif defined(_MSC_FULL_VER) && defined(_XCR_XFEATURE_ENABLED_MASK) 198 unsigned long long rEAX = _xgetbv(_XCR_XFEATURE_ENABLED_MASK); 199 #else 200 int rEAX = 0; // Ensures we return false 201 #endif 202 return (rEAX & 6) == 6; 203 } 204 205 static void DetectX86FamilyModel(unsigned EAX, unsigned &Family, 206 unsigned &Model) { 207 Family = (EAX >> 8) & 0xf; // Bits 8 - 11 208 Model = (EAX >> 4) & 0xf; // Bits 4 - 7 209 if (Family == 6 || Family == 0xf) { 210 if (Family == 0xf) 211 // Examine extended family ID if family ID is F. 212 Family += (EAX >> 20) & 0xff; // Bits 20 - 27 213 // Examine extended model ID if family ID is 6 or F. 214 Model += ((EAX >> 16) & 0xf) << 4; // Bits 16 - 19 215 } 216 } 217 218 StringRef sys::getHostCPUName() { 219 unsigned EAX = 0, EBX = 0, ECX = 0, EDX = 0; 220 if (GetX86CpuIDAndInfo(0x1, &EAX, &EBX, &ECX, &EDX)) 221 return "generic"; 222 unsigned Family = 0; 223 unsigned Model = 0; 224 DetectX86FamilyModel(EAX, Family, Model); 225 226 union { 227 unsigned u[3]; 228 char c[12]; 229 } text; 230 231 GetX86CpuIDAndInfo(0, &EAX, text.u+0, text.u+2, text.u+1); 232 233 unsigned MaxLeaf = EAX; 234 bool HasSSE3 = (ECX & 0x1); 235 bool HasSSE41 = (ECX & 0x80000); 236 // If CPUID indicates support for XSAVE, XRESTORE and AVX, and XGETBV 237 // indicates that the AVX registers will be saved and restored on context 238 // switch, then we have full AVX support. 239 const unsigned AVXBits = (1 << 27) | (1 << 28); 240 bool HasAVX = ((ECX & AVXBits) == AVXBits) && OSHasAVXSupport(); 241 bool HasAVX2 = HasAVX && MaxLeaf >= 0x7 && 242 !GetX86CpuIDAndInfoEx(0x7, 0x0, &EAX, &EBX, &ECX, &EDX) && 243 (EBX & 0x20); 244 GetX86CpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX); 245 bool Em64T = (EDX >> 29) & 0x1; 246 bool HasTBM = (ECX >> 21) & 0x1; 247 248 if (memcmp(text.c, "GenuineIntel", 12) == 0) { 249 switch (Family) { 250 case 3: 251 return "i386"; 252 case 4: 253 switch (Model) { 254 case 0: // Intel486 DX processors 255 case 1: // Intel486 DX processors 256 case 2: // Intel486 SX processors 257 case 3: // Intel487 processors, IntelDX2 OverDrive processors, 258 // IntelDX2 processors 259 case 4: // Intel486 SL processor 260 case 5: // IntelSX2 processors 261 case 7: // Write-Back Enhanced IntelDX2 processors 262 case 8: // IntelDX4 OverDrive processors, IntelDX4 processors 263 default: return "i486"; 264 } 265 case 5: 266 switch (Model) { 267 case 1: // Pentium OverDrive processor for Pentium processor (60, 66), 268 // Pentium processors (60, 66) 269 case 2: // Pentium OverDrive processor for Pentium processor (75, 90, 270 // 100, 120, 133), Pentium processors (75, 90, 100, 120, 133, 271 // 150, 166, 200) 272 case 3: // Pentium OverDrive processors for Intel486 processor-based 273 // systems 274 return "pentium"; 275 276 case 4: // Pentium OverDrive processor with MMX technology for Pentium 277 // processor (75, 90, 100, 120, 133), Pentium processor with 278 // MMX technology (166, 200) 279 return "pentium-mmx"; 280 281 default: return "pentium"; 282 } 283 case 6: 284 switch (Model) { 285 case 1: // Pentium Pro processor 286 return "pentiumpro"; 287 288 case 3: // Intel Pentium II OverDrive processor, Pentium II processor, 289 // model 03 290 case 5: // Pentium II processor, model 05, Pentium II Xeon processor, 291 // model 05, and Intel Celeron processor, model 05 292 case 6: // Celeron processor, model 06 293 return "pentium2"; 294 295 case 7: // Pentium III processor, model 07, and Pentium III Xeon 296 // processor, model 07 297 case 8: // Pentium III processor, model 08, Pentium III Xeon processor, 298 // model 08, and Celeron processor, model 08 299 case 10: // Pentium III Xeon processor, model 0Ah 300 case 11: // Pentium III processor, model 0Bh 301 return "pentium3"; 302 303 case 9: // Intel Pentium M processor, Intel Celeron M processor model 09. 304 case 13: // Intel Pentium M processor, Intel Celeron M processor, model 305 // 0Dh. All processors are manufactured using the 90 nm process. 306 return "pentium-m"; 307 308 case 14: // Intel Core Duo processor, Intel Core Solo processor, model 309 // 0Eh. All processors are manufactured using the 65 nm process. 310 return "yonah"; 311 312 case 15: // Intel Core 2 Duo processor, Intel Core 2 Duo mobile 313 // processor, Intel Core 2 Quad processor, Intel Core 2 Quad 314 // mobile processor, Intel Core 2 Extreme processor, Intel 315 // Pentium Dual-Core processor, Intel Xeon processor, model 316 // 0Fh. All processors are manufactured using the 65 nm process. 317 case 22: // Intel Celeron processor model 16h. All processors are 318 // manufactured using the 65 nm process 319 return "core2"; 320 321 case 21: // Intel EP80579 Integrated Processor and Intel EP80579 322 // Integrated Processor with Intel QuickAssist Technology 323 return "i686"; // FIXME: ??? 324 325 case 23: // Intel Core 2 Extreme processor, Intel Xeon processor, model 326 // 17h. All processors are manufactured using the 45 nm process. 327 // 328 // 45nm: Penryn , Wolfdale, Yorkfield (XE) 329 // Not all Penryn processors support SSE 4.1 (such as the Pentium brand) 330 return HasSSE41 ? "penryn" : "core2"; 331 332 case 26: // Intel Core i7 processor and Intel Xeon processor. All 333 // processors are manufactured using the 45 nm process. 334 case 29: // Intel Xeon processor MP. All processors are manufactured using 335 // the 45 nm process. 336 case 30: // Intel(R) Core(TM) i7 CPU 870 @ 2.93GHz. 337 // As found in a Summer 2010 model iMac. 338 case 37: // Intel Core i7, laptop version. 339 case 44: // Intel Core i7 processor and Intel Xeon processor. All 340 // processors are manufactured using the 32 nm process. 341 case 46: // Nehalem EX 342 case 47: // Westmere EX 343 return "corei7"; 344 345 // SandyBridge: 346 case 42: // Intel Core i7 processor. All processors are manufactured 347 // using the 32 nm process. 348 case 45: 349 // Not all Sandy Bridge processors support AVX (such as the Pentium 350 // versions instead of the i7 versions). 351 return HasAVX ? "corei7-avx" : "corei7"; 352 353 // Ivy Bridge: 354 case 58: 355 case 62: // Ivy Bridge EP 356 // Not all Ivy Bridge processors support AVX (such as the Pentium 357 // versions instead of the i7 versions). 358 return HasAVX ? "core-avx-i" : "corei7"; 359 360 // Haswell: 361 case 60: 362 case 63: 363 case 69: 364 case 70: 365 // Not all Haswell processors support AVX too (such as the Pentium 366 // versions instead of the i7 versions). 367 return HasAVX2 ? "core-avx2" : "corei7"; 368 369 case 28: // Most 45 nm Intel Atom processors 370 case 38: // 45 nm Atom Lincroft 371 case 39: // 32 nm Atom Medfield 372 case 53: // 32 nm Atom Midview 373 case 54: // 32 nm Atom Midview 374 return "atom"; 375 376 // Atom Silvermont codes from the Intel software optimization guide. 377 case 55: 378 case 74: 379 case 77: 380 return "slm"; 381 382 default: return (Em64T) ? "x86-64" : "i686"; 383 } 384 case 15: { 385 switch (Model) { 386 case 0: // Pentium 4 processor, Intel Xeon processor. All processors are 387 // model 00h and manufactured using the 0.18 micron process. 388 case 1: // Pentium 4 processor, Intel Xeon processor, Intel Xeon 389 // processor MP, and Intel Celeron processor. All processors are 390 // model 01h and manufactured using the 0.18 micron process. 391 case 2: // Pentium 4 processor, Mobile Intel Pentium 4 processor - M, 392 // Intel Xeon processor, Intel Xeon processor MP, Intel Celeron 393 // processor, and Mobile Intel Celeron processor. All processors 394 // are model 02h and manufactured using the 0.13 micron process. 395 return (Em64T) ? "x86-64" : "pentium4"; 396 397 case 3: // Pentium 4 processor, Intel Xeon processor, Intel Celeron D 398 // processor. All processors are model 03h and manufactured using 399 // the 90 nm process. 400 case 4: // Pentium 4 processor, Pentium 4 processor Extreme Edition, 401 // Pentium D processor, Intel Xeon processor, Intel Xeon 402 // processor MP, Intel Celeron D processor. All processors are 403 // model 04h and manufactured using the 90 nm process. 404 case 6: // Pentium 4 processor, Pentium D processor, Pentium processor 405 // Extreme Edition, Intel Xeon processor, Intel Xeon processor 406 // MP, Intel Celeron D processor. All processors are model 06h 407 // and manufactured using the 65 nm process. 408 return (Em64T) ? "nocona" : "prescott"; 409 410 default: 411 return (Em64T) ? "x86-64" : "pentium4"; 412 } 413 } 414 415 default: 416 return "generic"; 417 } 418 } else if (memcmp(text.c, "AuthenticAMD", 12) == 0) { 419 // FIXME: this poorly matches the generated SubtargetFeatureKV table. There 420 // appears to be no way to generate the wide variety of AMD-specific targets 421 // from the information returned from CPUID. 422 switch (Family) { 423 case 4: 424 return "i486"; 425 case 5: 426 switch (Model) { 427 case 6: 428 case 7: return "k6"; 429 case 8: return "k6-2"; 430 case 9: 431 case 13: return "k6-3"; 432 case 10: return "geode"; 433 default: return "pentium"; 434 } 435 case 6: 436 switch (Model) { 437 case 4: return "athlon-tbird"; 438 case 6: 439 case 7: 440 case 8: return "athlon-mp"; 441 case 10: return "athlon-xp"; 442 default: return "athlon"; 443 } 444 case 15: 445 if (HasSSE3) 446 return "k8-sse3"; 447 switch (Model) { 448 case 1: return "opteron"; 449 case 5: return "athlon-fx"; // also opteron 450 default: return "athlon64"; 451 } 452 case 16: 453 return "amdfam10"; 454 case 20: 455 return "btver1"; 456 case 21: 457 if (!HasAVX) // If the OS doesn't support AVX provide a sane fallback. 458 return "btver1"; 459 if (Model >= 0x50) 460 return "bdver4"; // 50h-6Fh: Excavator 461 if (Model >= 0x30) 462 return "bdver3"; // 30h-3Fh: Steamroller 463 if (Model >= 0x10 || HasTBM) 464 return "bdver2"; // 10h-1Fh: Piledriver 465 return "bdver1"; // 00h-0Fh: Bulldozer 466 case 22: 467 if (!HasAVX) // If the OS doesn't support AVX provide a sane fallback. 468 return "btver1"; 469 return "btver2"; 470 default: 471 return "generic"; 472 } 473 } 474 return "generic"; 475 } 476 #elif defined(__APPLE__) && (defined(__ppc__) || defined(__powerpc__)) 477 StringRef sys::getHostCPUName() { 478 host_basic_info_data_t hostInfo; 479 mach_msg_type_number_t infoCount; 480 481 infoCount = HOST_BASIC_INFO_COUNT; 482 host_info(mach_host_self(), HOST_BASIC_INFO, (host_info_t)&hostInfo, 483 &infoCount); 484 485 if (hostInfo.cpu_type != CPU_TYPE_POWERPC) return "generic"; 486 487 switch(hostInfo.cpu_subtype) { 488 case CPU_SUBTYPE_POWERPC_601: return "601"; 489 case CPU_SUBTYPE_POWERPC_602: return "602"; 490 case CPU_SUBTYPE_POWERPC_603: return "603"; 491 case CPU_SUBTYPE_POWERPC_603e: return "603e"; 492 case CPU_SUBTYPE_POWERPC_603ev: return "603ev"; 493 case CPU_SUBTYPE_POWERPC_604: return "604"; 494 case CPU_SUBTYPE_POWERPC_604e: return "604e"; 495 case CPU_SUBTYPE_POWERPC_620: return "620"; 496 case CPU_SUBTYPE_POWERPC_750: return "750"; 497 case CPU_SUBTYPE_POWERPC_7400: return "7400"; 498 case CPU_SUBTYPE_POWERPC_7450: return "7450"; 499 case CPU_SUBTYPE_POWERPC_970: return "970"; 500 default: ; 501 } 502 503 return "generic"; 504 } 505 #elif defined(__linux__) && (defined(__ppc__) || defined(__powerpc__)) 506 StringRef sys::getHostCPUName() { 507 // Access to the Processor Version Register (PVR) on PowerPC is privileged, 508 // and so we must use an operating-system interface to determine the current 509 // processor type. On Linux, this is exposed through the /proc/cpuinfo file. 510 const char *generic = "generic"; 511 512 // The cpu line is second (after the 'processor: 0' line), so if this 513 // buffer is too small then something has changed (or is wrong). 514 char buffer[1024]; 515 ssize_t CPUInfoSize = readCpuInfo(buffer, sizeof(buffer)); 516 if (CPUInfoSize == -1) 517 return generic; 518 519 const char *CPUInfoStart = buffer; 520 const char *CPUInfoEnd = buffer + CPUInfoSize; 521 522 const char *CIP = CPUInfoStart; 523 524 const char *CPUStart = 0; 525 size_t CPULen = 0; 526 527 // We need to find the first line which starts with cpu, spaces, and a colon. 528 // After the colon, there may be some additional spaces and then the cpu type. 529 while (CIP < CPUInfoEnd && CPUStart == 0) { 530 if (CIP < CPUInfoEnd && *CIP == '\n') 531 ++CIP; 532 533 if (CIP < CPUInfoEnd && *CIP == 'c') { 534 ++CIP; 535 if (CIP < CPUInfoEnd && *CIP == 'p') { 536 ++CIP; 537 if (CIP < CPUInfoEnd && *CIP == 'u') { 538 ++CIP; 539 while (CIP < CPUInfoEnd && (*CIP == ' ' || *CIP == '\t')) 540 ++CIP; 541 542 if (CIP < CPUInfoEnd && *CIP == ':') { 543 ++CIP; 544 while (CIP < CPUInfoEnd && (*CIP == ' ' || *CIP == '\t')) 545 ++CIP; 546 547 if (CIP < CPUInfoEnd) { 548 CPUStart = CIP; 549 while (CIP < CPUInfoEnd && (*CIP != ' ' && *CIP != '\t' && 550 *CIP != ',' && *CIP != '\n')) 551 ++CIP; 552 CPULen = CIP - CPUStart; 553 } 554 } 555 } 556 } 557 } 558 559 if (CPUStart == 0) 560 while (CIP < CPUInfoEnd && *CIP != '\n') 561 ++CIP; 562 } 563 564 if (CPUStart == 0) 565 return generic; 566 567 return StringSwitch<const char *>(StringRef(CPUStart, CPULen)) 568 .Case("604e", "604e") 569 .Case("604", "604") 570 .Case("7400", "7400") 571 .Case("7410", "7400") 572 .Case("7447", "7400") 573 .Case("7455", "7450") 574 .Case("G4", "g4") 575 .Case("POWER4", "970") 576 .Case("PPC970FX", "970") 577 .Case("PPC970MP", "970") 578 .Case("G5", "g5") 579 .Case("POWER5", "g5") 580 .Case("A2", "a2") 581 .Case("POWER6", "pwr6") 582 .Case("POWER7", "pwr7") 583 .Case("POWER8", "pwr8") 584 .Case("POWER8E", "pwr8") 585 .Default(generic); 586 } 587 #elif defined(__linux__) && defined(__arm__) 588 StringRef sys::getHostCPUName() { 589 // The cpuid register on arm is not accessible from user space. On Linux, 590 // it is exposed through the /proc/cpuinfo file. 591 592 // Read 1024 bytes from /proc/cpuinfo, which should contain the CPU part line 593 // in all cases. 594 char buffer[1024]; 595 ssize_t CPUInfoSize = readCpuInfo(buffer, sizeof(buffer)); 596 if (CPUInfoSize == -1) 597 return "generic"; 598 599 StringRef Str(buffer, CPUInfoSize); 600 601 SmallVector<StringRef, 32> Lines; 602 Str.split(Lines, "\n"); 603 604 // Look for the CPU implementer line. 605 StringRef Implementer; 606 for (unsigned I = 0, E = Lines.size(); I != E; ++I) 607 if (Lines[I].startswith("CPU implementer")) 608 Implementer = Lines[I].substr(15).ltrim("\t :"); 609 610 if (Implementer == "0x41") // ARM Ltd. 611 // Look for the CPU part line. 612 for (unsigned I = 0, E = Lines.size(); I != E; ++I) 613 if (Lines[I].startswith("CPU part")) 614 // The CPU part is a 3 digit hexadecimal number with a 0x prefix. The 615 // values correspond to the "Part number" in the CP15/c0 register. The 616 // contents are specified in the various processor manuals. 617 return StringSwitch<const char *>(Lines[I].substr(8).ltrim("\t :")) 618 .Case("0x926", "arm926ej-s") 619 .Case("0xb02", "mpcore") 620 .Case("0xb36", "arm1136j-s") 621 .Case("0xb56", "arm1156t2-s") 622 .Case("0xb76", "arm1176jz-s") 623 .Case("0xc08", "cortex-a8") 624 .Case("0xc09", "cortex-a9") 625 .Case("0xc0f", "cortex-a15") 626 .Case("0xc20", "cortex-m0") 627 .Case("0xc23", "cortex-m3") 628 .Case("0xc24", "cortex-m4") 629 .Default("generic"); 630 631 if (Implementer == "0x51") // Qualcomm Technologies, Inc. 632 // Look for the CPU part line. 633 for (unsigned I = 0, E = Lines.size(); I != E; ++I) 634 if (Lines[I].startswith("CPU part")) 635 // The CPU part is a 3 digit hexadecimal number with a 0x prefix. The 636 // values correspond to the "Part number" in the CP15/c0 register. The 637 // contents are specified in the various processor manuals. 638 return StringSwitch<const char *>(Lines[I].substr(8).ltrim("\t :")) 639 .Case("0x06f", "krait") // APQ8064 640 .Default("generic"); 641 642 return "generic"; 643 } 644 #elif defined(__linux__) && defined(__s390x__) 645 StringRef sys::getHostCPUName() { 646 // STIDP is a privileged operation, so use /proc/cpuinfo instead. 647 648 // The "processor 0:" line comes after a fair amount of other information, 649 // including a cache breakdown, but this should be plenty. 650 char buffer[2048]; 651 ssize_t CPUInfoSize = readCpuInfo(buffer, sizeof(buffer)); 652 if (CPUInfoSize == -1) 653 return "generic"; 654 655 StringRef Str(buffer, CPUInfoSize); 656 SmallVector<StringRef, 32> Lines; 657 Str.split(Lines, "\n"); 658 for (unsigned I = 0, E = Lines.size(); I != E; ++I) { 659 if (Lines[I].startswith("processor ")) { 660 size_t Pos = Lines[I].find("machine = "); 661 if (Pos != StringRef::npos) { 662 Pos += sizeof("machine = ") - 1; 663 unsigned int Id; 664 if (!Lines[I].drop_front(Pos).getAsInteger(10, Id)) { 665 if (Id >= 2827) 666 return "zEC12"; 667 if (Id >= 2817) 668 return "z196"; 669 } 670 } 671 break; 672 } 673 } 674 675 return "generic"; 676 } 677 #else 678 StringRef sys::getHostCPUName() { 679 return "generic"; 680 } 681 #endif 682 683 #if defined(__linux__) && (defined(__arm__) || defined(__aarch64__)) 684 bool sys::getHostCPUFeatures(StringMap<bool> &Features) { 685 // Read 1024 bytes from /proc/cpuinfo, which should contain the Features line 686 // in all cases. 687 char buffer[1024]; 688 ssize_t CPUInfoSize = readCpuInfo(buffer, sizeof(buffer)); 689 if (CPUInfoSize == -1) 690 return false; 691 692 StringRef Str(buffer, CPUInfoSize); 693 694 SmallVector<StringRef, 32> Lines; 695 Str.split(Lines, "\n"); 696 697 SmallVector<StringRef, 32> CPUFeatures; 698 699 // Look for the CPU features. 700 for (unsigned I = 0, E = Lines.size(); I != E; ++I) 701 if (Lines[I].startswith("Features")) { 702 Lines[I].split(CPUFeatures, " "); 703 break; 704 } 705 706 #if defined(__aarch64__) 707 // Keep track of which crypto features we have seen 708 enum { 709 CAP_AES = 0x1, 710 CAP_PMULL = 0x2, 711 CAP_SHA1 = 0x4, 712 CAP_SHA2 = 0x8 713 }; 714 uint32_t crypto = 0; 715 #endif 716 717 for (unsigned I = 0, E = CPUFeatures.size(); I != E; ++I) { 718 StringRef LLVMFeatureStr = StringSwitch<StringRef>(CPUFeatures[I]) 719 #if defined(__aarch64__) 720 .Case("asimd", "neon") 721 .Case("fp", "fp-armv8") 722 .Case("crc32", "crc") 723 #else 724 .Case("half", "fp16") 725 .Case("neon", "neon") 726 .Case("vfpv3", "vfp3") 727 .Case("vfpv3d16", "d16") 728 .Case("vfpv4", "vfp4") 729 .Case("idiva", "hwdiv-arm") 730 .Case("idivt", "hwdiv") 731 #endif 732 .Default(""); 733 734 #if defined(__aarch64__) 735 // We need to check crypto separately since we need all of the crypto 736 // extensions to enable the subtarget feature 737 if (CPUFeatures[I] == "aes") 738 crypto |= CAP_AES; 739 else if (CPUFeatures[I] == "pmull") 740 crypto |= CAP_PMULL; 741 else if (CPUFeatures[I] == "sha1") 742 crypto |= CAP_SHA1; 743 else if (CPUFeatures[I] == "sha2") 744 crypto |= CAP_SHA2; 745 #endif 746 747 if (LLVMFeatureStr != "") 748 Features[LLVMFeatureStr] = true; 749 } 750 751 #if defined(__aarch64__) 752 // If we have all crypto bits we can add the feature 753 if (crypto == (CAP_AES | CAP_PMULL | CAP_SHA1 | CAP_SHA2)) 754 Features["crypto"] = true; 755 #endif 756 757 return true; 758 } 759 #else 760 bool sys::getHostCPUFeatures(StringMap<bool> &Features){ 761 return false; 762 } 763 #endif 764 765 std::string sys::getProcessTriple() { 766 Triple PT(Triple::normalize(LLVM_HOST_TRIPLE)); 767 768 if (sizeof(void *) == 8 && PT.isArch32Bit()) 769 PT = PT.get64BitArchVariant(); 770 if (sizeof(void *) == 4 && PT.isArch64Bit()) 771 PT = PT.get32BitArchVariant(); 772 773 return PT.str(); 774 } 775