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/SmallSet.h"
16 #include "llvm/ADT/SmallVector.h"
17 #include "llvm/ADT/StringRef.h"
18 #include "llvm/ADT/StringSwitch.h"
19 #include "llvm/ADT/Triple.h"
20 #include "llvm/Config/config.h"
21 #include "llvm/Support/Debug.h"
22 #include "llvm/Support/FileSystem.h"
23 #include "llvm/Support/MemoryBuffer.h"
24 #include "llvm/Support/raw_ostream.h"
25 #include <assert.h>
26 #include <string.h>
27 
28 // Include the platform-specific parts of this class.
29 #ifdef LLVM_ON_UNIX
30 #include "Unix/Host.inc"
31 #endif
32 #ifdef LLVM_ON_WIN32
33 #include "Windows/Host.inc"
34 #endif
35 #ifdef _MSC_VER
36 #include <intrin.h>
37 #endif
38 #if defined(__APPLE__) && (defined(__ppc__) || defined(__powerpc__))
39 #include <mach/host_info.h>
40 #include <mach/mach.h>
41 #include <mach/mach_host.h>
42 #include <mach/machine.h>
43 #endif
44 
45 #define DEBUG_TYPE "host-detection"
46 
47 //===----------------------------------------------------------------------===//
48 //
49 //  Implementations of the CPU detection routines
50 //
51 //===----------------------------------------------------------------------===//
52 
53 using namespace llvm;
54 
55 static std::unique_ptr<llvm::MemoryBuffer>
56     LLVM_ATTRIBUTE_UNUSED getProcCpuinfoContent() {
57   llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> Text =
58       llvm::MemoryBuffer::getFileAsStream("/proc/cpuinfo");
59   if (std::error_code EC = Text.getError()) {
60     llvm::errs() << "Can't read "
61                  << "/proc/cpuinfo: " << EC.message() << "\n";
62     return nullptr;
63   }
64   return std::move(*Text);
65 }
66 
67 StringRef sys::detail::getHostCPUNameForPowerPC(
68     const StringRef &ProcCpuinfoContent) {
69   // Access to the Processor Version Register (PVR) on PowerPC is privileged,
70   // and so we must use an operating-system interface to determine the current
71   // processor type. On Linux, this is exposed through the /proc/cpuinfo file.
72   const char *generic = "generic";
73 
74   // The cpu line is second (after the 'processor: 0' line), so if this
75   // buffer is too small then something has changed (or is wrong).
76   StringRef::const_iterator CPUInfoStart = ProcCpuinfoContent.begin();
77   StringRef::const_iterator CPUInfoEnd = ProcCpuinfoContent.end();
78 
79   StringRef::const_iterator CIP = CPUInfoStart;
80 
81   StringRef::const_iterator CPUStart = 0;
82   size_t CPULen = 0;
83 
84   // We need to find the first line which starts with cpu, spaces, and a colon.
85   // After the colon, there may be some additional spaces and then the cpu type.
86   while (CIP < CPUInfoEnd && CPUStart == 0) {
87     if (CIP < CPUInfoEnd && *CIP == '\n')
88       ++CIP;
89 
90     if (CIP < CPUInfoEnd && *CIP == 'c') {
91       ++CIP;
92       if (CIP < CPUInfoEnd && *CIP == 'p') {
93         ++CIP;
94         if (CIP < CPUInfoEnd && *CIP == 'u') {
95           ++CIP;
96           while (CIP < CPUInfoEnd && (*CIP == ' ' || *CIP == '\t'))
97             ++CIP;
98 
99           if (CIP < CPUInfoEnd && *CIP == ':') {
100             ++CIP;
101             while (CIP < CPUInfoEnd && (*CIP == ' ' || *CIP == '\t'))
102               ++CIP;
103 
104             if (CIP < CPUInfoEnd) {
105               CPUStart = CIP;
106               while (CIP < CPUInfoEnd && (*CIP != ' ' && *CIP != '\t' &&
107                                           *CIP != ',' && *CIP != '\n'))
108                 ++CIP;
109               CPULen = CIP - CPUStart;
110             }
111           }
112         }
113       }
114     }
115 
116     if (CPUStart == 0)
117       while (CIP < CPUInfoEnd && *CIP != '\n')
118         ++CIP;
119   }
120 
121   if (CPUStart == 0)
122     return generic;
123 
124   return StringSwitch<const char *>(StringRef(CPUStart, CPULen))
125       .Case("604e", "604e")
126       .Case("604", "604")
127       .Case("7400", "7400")
128       .Case("7410", "7400")
129       .Case("7447", "7400")
130       .Case("7455", "7450")
131       .Case("G4", "g4")
132       .Case("POWER4", "970")
133       .Case("PPC970FX", "970")
134       .Case("PPC970MP", "970")
135       .Case("G5", "g5")
136       .Case("POWER5", "g5")
137       .Case("A2", "a2")
138       .Case("POWER6", "pwr6")
139       .Case("POWER7", "pwr7")
140       .Case("POWER8", "pwr8")
141       .Case("POWER8E", "pwr8")
142       .Case("POWER8NVL", "pwr8")
143       .Case("POWER9", "pwr9")
144       .Default(generic);
145 }
146 
147 StringRef sys::detail::getHostCPUNameForARM(
148     const StringRef &ProcCpuinfoContent) {
149   // The cpuid register on arm is not accessible from user space. On Linux,
150   // it is exposed through the /proc/cpuinfo file.
151 
152   // Read 32 lines from /proc/cpuinfo, which should contain the CPU part line
153   // in all cases.
154   SmallVector<StringRef, 32> Lines;
155   ProcCpuinfoContent.split(Lines, "\n");
156 
157   // Look for the CPU implementer line.
158   StringRef Implementer;
159   StringRef Hardware;
160   for (unsigned I = 0, E = Lines.size(); I != E; ++I) {
161     if (Lines[I].startswith("CPU implementer"))
162       Implementer = Lines[I].substr(15).ltrim("\t :");
163     if (Lines[I].startswith("Hardware"))
164       Hardware = Lines[I].substr(8).ltrim("\t :");
165   }
166 
167   if (Implementer == "0x41") { // ARM Ltd.
168     // MSM8992/8994 may give cpu part for the core that the kernel is running on,
169     // which is undeterministic and wrong. Always return cortex-a53 for these SoC.
170     if (Hardware.endswith("MSM8994") || Hardware.endswith("MSM8996"))
171       return "cortex-a53";
172 
173 
174     // Look for the CPU part line.
175     for (unsigned I = 0, E = Lines.size(); I != E; ++I)
176       if (Lines[I].startswith("CPU part"))
177         // The CPU part is a 3 digit hexadecimal number with a 0x prefix. The
178         // values correspond to the "Part number" in the CP15/c0 register. The
179         // contents are specified in the various processor manuals.
180         return StringSwitch<const char *>(Lines[I].substr(8).ltrim("\t :"))
181             .Case("0x926", "arm926ej-s")
182             .Case("0xb02", "mpcore")
183             .Case("0xb36", "arm1136j-s")
184             .Case("0xb56", "arm1156t2-s")
185             .Case("0xb76", "arm1176jz-s")
186             .Case("0xc08", "cortex-a8")
187             .Case("0xc09", "cortex-a9")
188             .Case("0xc0f", "cortex-a15")
189             .Case("0xc20", "cortex-m0")
190             .Case("0xc23", "cortex-m3")
191             .Case("0xc24", "cortex-m4")
192             .Case("0xd04", "cortex-a35")
193             .Case("0xd03", "cortex-a53")
194             .Case("0xd07", "cortex-a57")
195             .Case("0xd08", "cortex-a72")
196             .Case("0xd09", "cortex-a73")
197             .Default("generic");
198   }
199 
200   if (Implementer == "0x51") // Qualcomm Technologies, Inc.
201     // Look for the CPU part line.
202     for (unsigned I = 0, E = Lines.size(); I != E; ++I)
203       if (Lines[I].startswith("CPU part"))
204         // The CPU part is a 3 digit hexadecimal number with a 0x prefix. The
205         // values correspond to the "Part number" in the CP15/c0 register. The
206         // contents are specified in the various processor manuals.
207         return StringSwitch<const char *>(Lines[I].substr(8).ltrim("\t :"))
208             .Case("0x06f", "krait") // APQ8064
209             .Case("0x201", "kryo")
210             .Case("0x205", "kryo")
211             .Default("generic");
212 
213   return "generic";
214 }
215 
216 StringRef sys::detail::getHostCPUNameForS390x(
217     const StringRef &ProcCpuinfoContent) {
218   // STIDP is a privileged operation, so use /proc/cpuinfo instead.
219 
220   // The "processor 0:" line comes after a fair amount of other information,
221   // including a cache breakdown, but this should be plenty.
222   SmallVector<StringRef, 32> Lines;
223   ProcCpuinfoContent.split(Lines, "\n");
224 
225   // Look for the CPU features.
226   SmallVector<StringRef, 32> CPUFeatures;
227   for (unsigned I = 0, E = Lines.size(); I != E; ++I)
228     if (Lines[I].startswith("features")) {
229       size_t Pos = Lines[I].find(":");
230       if (Pos != StringRef::npos) {
231         Lines[I].drop_front(Pos + 1).split(CPUFeatures, ' ');
232         break;
233       }
234     }
235 
236   // We need to check for the presence of vector support independently of
237   // the machine type, since we may only use the vector register set when
238   // supported by the kernel (and hypervisor).
239   bool HaveVectorSupport = false;
240   for (unsigned I = 0, E = CPUFeatures.size(); I != E; ++I) {
241     if (CPUFeatures[I] == "vx")
242       HaveVectorSupport = true;
243   }
244 
245   // Now check the processor machine type.
246   for (unsigned I = 0, E = Lines.size(); I != E; ++I) {
247     if (Lines[I].startswith("processor ")) {
248       size_t Pos = Lines[I].find("machine = ");
249       if (Pos != StringRef::npos) {
250         Pos += sizeof("machine = ") - 1;
251         unsigned int Id;
252         if (!Lines[I].drop_front(Pos).getAsInteger(10, Id)) {
253           if (Id >= 3906 && HaveVectorSupport)
254             return "z14";
255           if (Id >= 2964 && HaveVectorSupport)
256             return "z13";
257           if (Id >= 2827)
258             return "zEC12";
259           if (Id >= 2817)
260             return "z196";
261         }
262       }
263       break;
264     }
265   }
266 
267   return "generic";
268 }
269 
270 #if defined(__i386__) || defined(_M_IX86) || \
271     defined(__x86_64__) || defined(_M_X64)
272 
273 enum VendorSignatures {
274   SIG_INTEL = 0x756e6547 /* Genu */,
275   SIG_AMD = 0x68747541 /* Auth */
276 };
277 
278 enum ProcessorVendors {
279   VENDOR_INTEL = 1,
280   VENDOR_AMD,
281   VENDOR_OTHER,
282   VENDOR_MAX
283 };
284 
285 enum ProcessorTypes {
286   INTEL_BONNELL = 1,
287   INTEL_CORE2,
288   INTEL_COREI7,
289   AMDFAM10H,
290   AMDFAM15H,
291   INTEL_SILVERMONT,
292   INTEL_KNL,
293   AMD_BTVER1,
294   AMD_BTVER2,
295   AMDFAM17H,
296   // Entries below this are not in libgcc/compiler-rt.
297   INTEL_i386,
298   INTEL_i486,
299   INTEL_PENTIUM,
300   INTEL_PENTIUM_PRO,
301   INTEL_PENTIUM_II,
302   INTEL_PENTIUM_III,
303   INTEL_PENTIUM_IV,
304   INTEL_PENTIUM_M,
305   INTEL_CORE_DUO,
306   INTEL_X86_64,
307   INTEL_NOCONA,
308   INTEL_PRESCOTT,
309   AMD_i486,
310   AMDPENTIUM,
311   AMDATHLON,
312   INTEL_GOLDMONT,
313   CPU_TYPE_MAX
314 };
315 
316 enum ProcessorSubtypes {
317   INTEL_COREI7_NEHALEM = 1,
318   INTEL_COREI7_WESTMERE,
319   INTEL_COREI7_SANDYBRIDGE,
320   AMDFAM10H_BARCELONA,
321   AMDFAM10H_SHANGHAI,
322   AMDFAM10H_ISTANBUL,
323   AMDFAM15H_BDVER1,
324   AMDFAM15H_BDVER2,
325   AMDFAM15H_BDVER3,
326   AMDFAM15H_BDVER4,
327   AMDFAM17H_ZNVER1,
328   INTEL_COREI7_IVYBRIDGE,
329   INTEL_COREI7_HASWELL,
330   INTEL_COREI7_BROADWELL,
331   INTEL_COREI7_SKYLAKE,
332   INTEL_COREI7_SKYLAKE_AVX512,
333   // Entries below this are not in libgcc/compiler-rt.
334   INTEL_PENTIUM_MMX,
335   INTEL_CORE2_65,
336   INTEL_CORE2_45,
337   AMDPENTIUM_K6,
338   AMDPENTIUM_K62,
339   AMDPENTIUM_K63,
340   AMDPENTIUM_GEODE,
341   AMDATHLON_CLASSIC,
342   AMDATHLON_XP,
343   AMDATHLON_K8,
344   AMDATHLON_K8SSE3,
345   CPU_SUBTYPE_MAX
346 };
347 
348 enum ProcessorFeatures {
349   FEATURE_CMOV = 0,
350   FEATURE_MMX,
351   FEATURE_POPCNT,
352   FEATURE_SSE,
353   FEATURE_SSE2,
354   FEATURE_SSE3,
355   FEATURE_SSSE3,
356   FEATURE_SSE4_1,
357   FEATURE_SSE4_2,
358   FEATURE_AVX,
359   FEATURE_AVX2,
360   FEATURE_SSE4_A,
361   FEATURE_FMA4,
362   FEATURE_XOP,
363   FEATURE_FMA,
364   FEATURE_AVX512F,
365   FEATURE_BMI,
366   FEATURE_BMI2,
367   FEATURE_AES,
368   FEATURE_PCLMUL,
369   FEATURE_AVX512VL,
370   FEATURE_AVX512BW,
371   FEATURE_AVX512DQ,
372   FEATURE_AVX512CD,
373   FEATURE_AVX512ER,
374   FEATURE_AVX512PF,
375   FEATURE_AVX512VBMI,
376   FEATURE_AVX512IFMA,
377   FEATURE_AVX5124VNNIW,
378   FEATURE_AVX5124FMAPS,
379   FEATURE_AVX512VPOPCNTDQ,
380   // Only one bit free left in the first 32 features.
381   FEATURE_MOVBE = 32,
382   FEATURE_ADX,
383   FEATURE_EM64T,
384   FEATURE_CLFLUSHOPT,
385   FEATURE_SHA,
386 };
387 
388 // The check below for i386 was copied from clang's cpuid.h (__get_cpuid_max).
389 // Check motivated by bug reports for OpenSSL crashing on CPUs without CPUID
390 // support. Consequently, for i386, the presence of CPUID is checked first
391 // via the corresponding eflags bit.
392 // Removal of cpuid.h header motivated by PR30384
393 // Header cpuid.h and method __get_cpuid_max are not used in llvm, clang, openmp
394 // or test-suite, but are used in external projects e.g. libstdcxx
395 static bool isCpuIdSupported() {
396 #if defined(__GNUC__) || defined(__clang__)
397 #if defined(__i386__)
398   int __cpuid_supported;
399   __asm__("  pushfl\n"
400           "  popl   %%eax\n"
401           "  movl   %%eax,%%ecx\n"
402           "  xorl   $0x00200000,%%eax\n"
403           "  pushl  %%eax\n"
404           "  popfl\n"
405           "  pushfl\n"
406           "  popl   %%eax\n"
407           "  movl   $0,%0\n"
408           "  cmpl   %%eax,%%ecx\n"
409           "  je     1f\n"
410           "  movl   $1,%0\n"
411           "1:"
412           : "=r"(__cpuid_supported)
413           :
414           : "eax", "ecx");
415   if (!__cpuid_supported)
416     return false;
417 #endif
418   return true;
419 #endif
420   return true;
421 }
422 
423 /// getX86CpuIDAndInfo - Execute the specified cpuid and return the 4 values in
424 /// the specified arguments.  If we can't run cpuid on the host, return true.
425 static bool getX86CpuIDAndInfo(unsigned value, unsigned *rEAX, unsigned *rEBX,
426                                unsigned *rECX, unsigned *rEDX) {
427 #if defined(__GNUC__) || defined(__clang__)
428 #if defined(__x86_64__)
429   // gcc doesn't know cpuid would clobber ebx/rbx. Preserve it manually.
430   // FIXME: should we save this for Clang?
431   __asm__("movq\t%%rbx, %%rsi\n\t"
432           "cpuid\n\t"
433           "xchgq\t%%rbx, %%rsi\n\t"
434           : "=a"(*rEAX), "=S"(*rEBX), "=c"(*rECX), "=d"(*rEDX)
435           : "a"(value));
436   return false;
437 #elif defined(__i386__)
438   __asm__("movl\t%%ebx, %%esi\n\t"
439           "cpuid\n\t"
440           "xchgl\t%%ebx, %%esi\n\t"
441           : "=a"(*rEAX), "=S"(*rEBX), "=c"(*rECX), "=d"(*rEDX)
442           : "a"(value));
443   return false;
444 #else
445   return true;
446 #endif
447 #elif defined(_MSC_VER)
448   // The MSVC intrinsic is portable across x86 and x64.
449   int registers[4];
450   __cpuid(registers, value);
451   *rEAX = registers[0];
452   *rEBX = registers[1];
453   *rECX = registers[2];
454   *rEDX = registers[3];
455   return false;
456 #else
457   return true;
458 #endif
459 }
460 
461 /// getX86CpuIDAndInfoEx - Execute the specified cpuid with subleaf and return
462 /// the 4 values in the specified arguments.  If we can't run cpuid on the host,
463 /// return true.
464 static bool getX86CpuIDAndInfoEx(unsigned value, unsigned subleaf,
465                                  unsigned *rEAX, unsigned *rEBX, unsigned *rECX,
466                                  unsigned *rEDX) {
467 #if defined(__GNUC__) || defined(__clang__)
468 #if defined(__x86_64__)
469   // gcc doesn't know cpuid would clobber ebx/rbx. Preserve it manually.
470   // FIXME: should we save this for Clang?
471   __asm__("movq\t%%rbx, %%rsi\n\t"
472           "cpuid\n\t"
473           "xchgq\t%%rbx, %%rsi\n\t"
474           : "=a"(*rEAX), "=S"(*rEBX), "=c"(*rECX), "=d"(*rEDX)
475           : "a"(value), "c"(subleaf));
476   return false;
477 #elif defined(__i386__)
478   __asm__("movl\t%%ebx, %%esi\n\t"
479           "cpuid\n\t"
480           "xchgl\t%%ebx, %%esi\n\t"
481           : "=a"(*rEAX), "=S"(*rEBX), "=c"(*rECX), "=d"(*rEDX)
482           : "a"(value), "c"(subleaf));
483   return false;
484 #else
485   return true;
486 #endif
487 #elif defined(_MSC_VER)
488   int registers[4];
489   __cpuidex(registers, value, subleaf);
490   *rEAX = registers[0];
491   *rEBX = registers[1];
492   *rECX = registers[2];
493   *rEDX = registers[3];
494   return false;
495 #else
496   return true;
497 #endif
498 }
499 
500 // Read control register 0 (XCR0). Used to detect features such as AVX.
501 static bool getX86XCR0(unsigned *rEAX, unsigned *rEDX) {
502 #if defined(__GNUC__) || defined(__clang__)
503   // Check xgetbv; this uses a .byte sequence instead of the instruction
504   // directly because older assemblers do not include support for xgetbv and
505   // there is no easy way to conditionally compile based on the assembler used.
506   __asm__(".byte 0x0f, 0x01, 0xd0" : "=a"(*rEAX), "=d"(*rEDX) : "c"(0));
507   return false;
508 #elif defined(_MSC_FULL_VER) && defined(_XCR_XFEATURE_ENABLED_MASK)
509   unsigned long long Result = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
510   *rEAX = Result;
511   *rEDX = Result >> 32;
512   return false;
513 #else
514   return true;
515 #endif
516 }
517 
518 static void detectX86FamilyModel(unsigned EAX, unsigned *Family,
519                                  unsigned *Model) {
520   *Family = (EAX >> 8) & 0xf; // Bits 8 - 11
521   *Model = (EAX >> 4) & 0xf;  // Bits 4 - 7
522   if (*Family == 6 || *Family == 0xf) {
523     if (*Family == 0xf)
524       // Examine extended family ID if family ID is F.
525       *Family += (EAX >> 20) & 0xff; // Bits 20 - 27
526     // Examine extended model ID if family ID is 6 or F.
527     *Model += ((EAX >> 16) & 0xf) << 4; // Bits 16 - 19
528   }
529 }
530 
531 static void
532 getIntelProcessorTypeAndSubtype(unsigned Family, unsigned Model,
533                                 unsigned Brand_id, unsigned Features,
534                                 unsigned Features2, unsigned *Type,
535                                 unsigned *Subtype) {
536   if (Brand_id != 0)
537     return;
538   switch (Family) {
539   case 3:
540     *Type = INTEL_i386;
541     break;
542   case 4:
543     switch (Model) {
544     case 0: // Intel486 DX processors
545     case 1: // Intel486 DX processors
546     case 2: // Intel486 SX processors
547     case 3: // Intel487 processors, IntelDX2 OverDrive processors,
548             // IntelDX2 processors
549     case 4: // Intel486 SL processor
550     case 5: // IntelSX2 processors
551     case 7: // Write-Back Enhanced IntelDX2 processors
552     case 8: // IntelDX4 OverDrive processors, IntelDX4 processors
553     default:
554       *Type = INTEL_i486;
555       break;
556     }
557     break;
558   case 5:
559     switch (Model) {
560     case 1: // Pentium OverDrive processor for Pentium processor (60, 66),
561             // Pentium processors (60, 66)
562     case 2: // Pentium OverDrive processor for Pentium processor (75, 90,
563             // 100, 120, 133), Pentium processors (75, 90, 100, 120, 133,
564             // 150, 166, 200)
565     case 3: // Pentium OverDrive processors for Intel486 processor-based
566             // systems
567       *Type = INTEL_PENTIUM;
568       break;
569     case 4: // Pentium OverDrive processor with MMX technology for Pentium
570             // processor (75, 90, 100, 120, 133), Pentium processor with
571             // MMX technology (166, 200)
572       *Type = INTEL_PENTIUM;
573       *Subtype = INTEL_PENTIUM_MMX;
574       break;
575     default:
576       *Type = INTEL_PENTIUM;
577       break;
578     }
579     break;
580   case 6:
581     switch (Model) {
582     case 0x01: // Pentium Pro processor
583       *Type = INTEL_PENTIUM_PRO;
584       break;
585     case 0x03: // Intel Pentium II OverDrive processor, Pentium II processor,
586                // model 03
587     case 0x05: // Pentium II processor, model 05, Pentium II Xeon processor,
588                // model 05, and Intel Celeron processor, model 05
589     case 0x06: // Celeron processor, model 06
590       *Type = INTEL_PENTIUM_II;
591       break;
592     case 0x07: // Pentium III processor, model 07, and Pentium III Xeon
593                // processor, model 07
594     case 0x08: // Pentium III processor, model 08, Pentium III Xeon processor,
595                // model 08, and Celeron processor, model 08
596     case 0x0a: // Pentium III Xeon processor, model 0Ah
597     case 0x0b: // Pentium III processor, model 0Bh
598       *Type = INTEL_PENTIUM_III;
599       break;
600     case 0x09: // Intel Pentium M processor, Intel Celeron M processor model 09.
601     case 0x0d: // Intel Pentium M processor, Intel Celeron M processor, model
602                // 0Dh. All processors are manufactured using the 90 nm process.
603     case 0x15: // Intel EP80579 Integrated Processor and Intel EP80579
604                // Integrated Processor with Intel QuickAssist Technology
605       *Type = INTEL_PENTIUM_M;
606       break;
607     case 0x0e: // Intel Core Duo processor, Intel Core Solo processor, model
608                // 0Eh. All processors are manufactured using the 65 nm process.
609       *Type = INTEL_CORE_DUO;
610       break;   // yonah
611     case 0x0f: // Intel Core 2 Duo processor, Intel Core 2 Duo mobile
612                // processor, Intel Core 2 Quad processor, Intel Core 2 Quad
613                // mobile processor, Intel Core 2 Extreme processor, Intel
614                // Pentium Dual-Core processor, Intel Xeon processor, model
615                // 0Fh. All processors are manufactured using the 65 nm process.
616     case 0x16: // Intel Celeron processor model 16h. All processors are
617                // manufactured using the 65 nm process
618       *Type = INTEL_CORE2; // "core2"
619       *Subtype = INTEL_CORE2_65;
620       break;
621     case 0x17: // Intel Core 2 Extreme processor, Intel Xeon processor, model
622                // 17h. All processors are manufactured using the 45 nm process.
623                //
624                // 45nm: Penryn , Wolfdale, Yorkfield (XE)
625     case 0x1d: // Intel Xeon processor MP. All processors are manufactured using
626                // the 45 nm process.
627       *Type = INTEL_CORE2; // "penryn"
628       *Subtype = INTEL_CORE2_45;
629       break;
630     case 0x1a: // Intel Core i7 processor and Intel Xeon processor. All
631                // processors are manufactured using the 45 nm process.
632     case 0x1e: // Intel(R) Core(TM) i7 CPU         870  @ 2.93GHz.
633                // As found in a Summer 2010 model iMac.
634     case 0x1f:
635     case 0x2e:             // Nehalem EX
636       *Type = INTEL_COREI7; // "nehalem"
637       *Subtype = INTEL_COREI7_NEHALEM;
638       break;
639     case 0x25: // Intel Core i7, laptop version.
640     case 0x2c: // Intel Core i7 processor and Intel Xeon processor. All
641                // processors are manufactured using the 32 nm process.
642     case 0x2f: // Westmere EX
643       *Type = INTEL_COREI7; // "westmere"
644       *Subtype = INTEL_COREI7_WESTMERE;
645       break;
646     case 0x2a: // Intel Core i7 processor. All processors are manufactured
647                // using the 32 nm process.
648     case 0x2d:
649       *Type = INTEL_COREI7; //"sandybridge"
650       *Subtype = INTEL_COREI7_SANDYBRIDGE;
651       break;
652     case 0x3a:
653     case 0x3e:             // Ivy Bridge EP
654       *Type = INTEL_COREI7; // "ivybridge"
655       *Subtype = INTEL_COREI7_IVYBRIDGE;
656       break;
657 
658     // Haswell:
659     case 0x3c:
660     case 0x3f:
661     case 0x45:
662     case 0x46:
663       *Type = INTEL_COREI7; // "haswell"
664       *Subtype = INTEL_COREI7_HASWELL;
665       break;
666 
667     // Broadwell:
668     case 0x3d:
669     case 0x47:
670     case 0x4f:
671     case 0x56:
672       *Type = INTEL_COREI7; // "broadwell"
673       *Subtype = INTEL_COREI7_BROADWELL;
674       break;
675 
676     // Skylake:
677     case 0x4e: // Skylake mobile
678     case 0x5e: // Skylake desktop
679     case 0x8e: // Kaby Lake mobile
680     case 0x9e: // Kaby Lake desktop
681       *Type = INTEL_COREI7; // "skylake"
682       *Subtype = INTEL_COREI7_SKYLAKE;
683       break;
684 
685     // Skylake Xeon:
686     case 0x55:
687       *Type = INTEL_COREI7;
688       *Subtype = INTEL_COREI7_SKYLAKE_AVX512; // "skylake-avx512"
689       break;
690 
691     case 0x1c: // Most 45 nm Intel Atom processors
692     case 0x26: // 45 nm Atom Lincroft
693     case 0x27: // 32 nm Atom Medfield
694     case 0x35: // 32 nm Atom Midview
695     case 0x36: // 32 nm Atom Midview
696       *Type = INTEL_BONNELL;
697       break; // "bonnell"
698 
699     // Atom Silvermont codes from the Intel software optimization guide.
700     case 0x37:
701     case 0x4a:
702     case 0x4d:
703     case 0x5a:
704     case 0x5d:
705     case 0x4c: // really airmont
706       *Type = INTEL_SILVERMONT;
707       break; // "silvermont"
708     // Goldmont:
709     case 0x5c:
710     case 0x5f:
711       *Type = INTEL_GOLDMONT;
712       break; // "goldmont"
713     case 0x57:
714       *Type = INTEL_KNL; // knl
715       break;
716 
717     default: // Unknown family 6 CPU, try to guess.
718       if (Features & (1 << FEATURE_AVX512F)) {
719         if (Features & (1 << FEATURE_AVX512VL)) {
720           *Type = INTEL_COREI7;
721           *Subtype = INTEL_COREI7_SKYLAKE_AVX512;
722         } else {
723           *Type = INTEL_KNL; // knl
724         }
725         break;
726       }
727       if (Features2 & (1 << (FEATURE_CLFLUSHOPT - 32))) {
728         if (Features2 & (1 << (FEATURE_SHA - 32))) {
729           *Type = INTEL_GOLDMONT;
730         } else {
731           *Type = INTEL_COREI7;
732           *Subtype = INTEL_COREI7_SKYLAKE;
733         }
734         break;
735       }
736       if (Features2 & (1 << (FEATURE_ADX - 32))) {
737         *Type = INTEL_COREI7;
738         *Subtype = INTEL_COREI7_BROADWELL;
739         break;
740       }
741       if (Features & (1 << FEATURE_AVX2)) {
742         *Type = INTEL_COREI7;
743         *Subtype = INTEL_COREI7_HASWELL;
744         break;
745       }
746       if (Features & (1 << FEATURE_AVX)) {
747         *Type = INTEL_COREI7;
748         *Subtype = INTEL_COREI7_SANDYBRIDGE;
749         break;
750       }
751       if (Features & (1 << FEATURE_SSE4_2)) {
752         if (Features2 & (1 << (FEATURE_MOVBE - 32))) {
753           *Type = INTEL_SILVERMONT;
754         } else {
755           *Type = INTEL_COREI7;
756           *Subtype = INTEL_COREI7_NEHALEM;
757         }
758         break;
759       }
760       if (Features & (1 << FEATURE_SSE4_1)) {
761         *Type = INTEL_CORE2; // "penryn"
762         *Subtype = INTEL_CORE2_45;
763         break;
764       }
765       if (Features & (1 << FEATURE_SSSE3)) {
766         if (Features2 & (1 << (FEATURE_MOVBE - 32))) {
767           *Type = INTEL_BONNELL; // "bonnell"
768         } else {
769           *Type = INTEL_CORE2; // "core2"
770           *Subtype = INTEL_CORE2_65;
771         }
772         break;
773       }
774       if (Features2 & (1 << (FEATURE_EM64T - 32))) {
775         *Type = INTEL_X86_64;
776         break; // x86-64
777       }
778       if (Features & (1 << FEATURE_SSE2)) {
779         *Type = INTEL_PENTIUM_M;
780         break;
781       }
782       if (Features & (1 << FEATURE_SSE)) {
783         *Type = INTEL_PENTIUM_III;
784         break;
785       }
786       if (Features & (1 << FEATURE_MMX)) {
787         *Type = INTEL_PENTIUM_II;
788         break;
789       }
790       *Type = INTEL_PENTIUM_PRO;
791       break;
792     }
793     break;
794   case 15: {
795     switch (Model) {
796     case 0: // Pentium 4 processor, Intel Xeon processor. All processors are
797             // model 00h and manufactured using the 0.18 micron process.
798     case 1: // Pentium 4 processor, Intel Xeon processor, Intel Xeon
799             // processor MP, and Intel Celeron processor. All processors are
800             // model 01h and manufactured using the 0.18 micron process.
801     case 2: // Pentium 4 processor, Mobile Intel Pentium 4 processor - M,
802             // Intel Xeon processor, Intel Xeon processor MP, Intel Celeron
803             // processor, and Mobile Intel Celeron processor. All processors
804             // are model 02h and manufactured using the 0.13 micron process.
805       *Type = ((Features2 & (1 << (FEATURE_EM64T - 32))) ? INTEL_X86_64
806                                                          : INTEL_PENTIUM_IV);
807       break;
808 
809     case 3: // Pentium 4 processor, Intel Xeon processor, Intel Celeron D
810             // processor. All processors are model 03h and manufactured using
811             // the 90 nm process.
812     case 4: // Pentium 4 processor, Pentium 4 processor Extreme Edition,
813             // Pentium D processor, Intel Xeon processor, Intel Xeon
814             // processor MP, Intel Celeron D processor. All processors are
815             // model 04h and manufactured using the 90 nm process.
816     case 6: // Pentium 4 processor, Pentium D processor, Pentium processor
817             // Extreme Edition, Intel Xeon processor, Intel Xeon processor
818             // MP, Intel Celeron D processor. All processors are model 06h
819             // and manufactured using the 65 nm process.
820       *Type = ((Features2 & (1 << (FEATURE_EM64T - 32))) ? INTEL_NOCONA
821                                                          : INTEL_PRESCOTT);
822       break;
823 
824     default:
825       *Type = ((Features2 & (1 << (FEATURE_EM64T - 32))) ? INTEL_X86_64
826                                                          : INTEL_PENTIUM_IV);
827       break;
828     }
829     break;
830   }
831   default:
832     break; /*"generic"*/
833   }
834 }
835 
836 static void getAMDProcessorTypeAndSubtype(unsigned Family, unsigned Model,
837                                           unsigned Features, unsigned *Type,
838                                           unsigned *Subtype) {
839   // FIXME: this poorly matches the generated SubtargetFeatureKV table.  There
840   // appears to be no way to generate the wide variety of AMD-specific targets
841   // from the information returned from CPUID.
842   switch (Family) {
843   case 4:
844     *Type = AMD_i486;
845     break;
846   case 5:
847     *Type = AMDPENTIUM;
848     switch (Model) {
849     case 6:
850     case 7:
851       *Subtype = AMDPENTIUM_K6;
852       break; // "k6"
853     case 8:
854       *Subtype = AMDPENTIUM_K62;
855       break; // "k6-2"
856     case 9:
857     case 13:
858       *Subtype = AMDPENTIUM_K63;
859       break; // "k6-3"
860     case 10:
861       *Subtype = AMDPENTIUM_GEODE;
862       break; // "geode"
863     }
864     break;
865   case 6:
866     *Type = AMDATHLON;
867     if (Features & (1 << FEATURE_SSE)) {
868       *Subtype = AMDATHLON_XP;
869       break; // "athlon-xp"
870     }
871     *Subtype = AMDATHLON_CLASSIC;
872     break; // "athlon"
873   case 15:
874     *Type = AMDATHLON;
875     if (Features & (1 << FEATURE_SSE3)) {
876       *Subtype = AMDATHLON_K8SSE3;
877       break; // "k8-sse3"
878     }
879     *Subtype = AMDATHLON_K8;
880     break; // "k8"
881   case 16:
882     *Type = AMDFAM10H; // "amdfam10"
883     switch (Model) {
884     case 2:
885       *Subtype = AMDFAM10H_BARCELONA;
886       break;
887     case 4:
888       *Subtype = AMDFAM10H_SHANGHAI;
889       break;
890     case 8:
891       *Subtype = AMDFAM10H_ISTANBUL;
892       break;
893     }
894     break;
895   case 20:
896     *Type = AMD_BTVER1;
897     break; // "btver1";
898   case 21:
899     *Type = AMDFAM15H;
900     if (Model >= 0x60 && Model <= 0x7f) {
901       *Subtype = AMDFAM15H_BDVER4;
902       break; // "bdver4"; 60h-7Fh: Excavator
903     }
904     if (Model >= 0x30 && Model <= 0x3f) {
905       *Subtype = AMDFAM15H_BDVER3;
906       break; // "bdver3"; 30h-3Fh: Steamroller
907     }
908     if (Model >= 0x10 && Model <= 0x1f) {
909       *Subtype = AMDFAM15H_BDVER2;
910       break; // "bdver2"; 10h-1Fh: Piledriver
911     }
912     if (Model <= 0x0f) {
913       *Subtype = AMDFAM15H_BDVER1;
914       break; // "bdver1"; 00h-0Fh: Bulldozer
915     }
916     break;
917   case 22:
918     *Type = AMD_BTVER2;
919     break; // "btver2"
920   case 23:
921     *Type = AMDFAM17H;
922     *Subtype = AMDFAM17H_ZNVER1;
923     break;
924   default:
925     break; // "generic"
926   }
927 }
928 
929 static void getAvailableFeatures(unsigned ECX, unsigned EDX, unsigned MaxLeaf,
930                                  unsigned *FeaturesOut,
931                                  unsigned *Features2Out) {
932   unsigned Features = 0;
933   unsigned Features2 = 0;
934   unsigned EAX, EBX;
935 
936   if ((EDX >> 15) & 1)
937     Features |= 1 << FEATURE_CMOV;
938   if ((EDX >> 23) & 1)
939     Features |= 1 << FEATURE_MMX;
940   if ((EDX >> 25) & 1)
941     Features |= 1 << FEATURE_SSE;
942   if ((EDX >> 26) & 1)
943     Features |= 1 << FEATURE_SSE2;
944 
945   if ((ECX >> 0) & 1)
946     Features |= 1 << FEATURE_SSE3;
947   if ((ECX >> 1) & 1)
948     Features |= 1 << FEATURE_PCLMUL;
949   if ((ECX >> 9) & 1)
950     Features |= 1 << FEATURE_SSSE3;
951   if ((ECX >> 12) & 1)
952     Features |= 1 << FEATURE_FMA;
953   if ((ECX >> 19) & 1)
954     Features |= 1 << FEATURE_SSE4_1;
955   if ((ECX >> 20) & 1)
956     Features |= 1 << FEATURE_SSE4_2;
957   if ((ECX >> 23) & 1)
958     Features |= 1 << FEATURE_POPCNT;
959   if ((ECX >> 25) & 1)
960     Features |= 1 << FEATURE_AES;
961 
962   if ((ECX >> 22) & 1)
963     Features2 |= 1 << (FEATURE_MOVBE - 32);
964 
965   // If CPUID indicates support for XSAVE, XRESTORE and AVX, and XGETBV
966   // indicates that the AVX registers will be saved and restored on context
967   // switch, then we have full AVX support.
968   const unsigned AVXBits = (1 << 27) | (1 << 28);
969   bool HasAVX = ((ECX & AVXBits) == AVXBits) && !getX86XCR0(&EAX, &EDX) &&
970                 ((EAX & 0x6) == 0x6);
971   bool HasAVX512Save = HasAVX && ((EAX & 0xe0) == 0xe0);
972 
973   if (HasAVX)
974     Features |= 1 << FEATURE_AVX;
975 
976   bool HasLeaf7 =
977       MaxLeaf >= 0x7 && !getX86CpuIDAndInfoEx(0x7, 0x0, &EAX, &EBX, &ECX, &EDX);
978 
979   if (HasLeaf7 && ((EBX >> 3) & 1))
980     Features |= 1 << FEATURE_BMI;
981   if (HasLeaf7 && ((EBX >> 5) & 1) && HasAVX)
982     Features |= 1 << FEATURE_AVX2;
983   if (HasLeaf7 && ((EBX >> 9) & 1))
984     Features |= 1 << FEATURE_BMI2;
985   if (HasLeaf7 && ((EBX >> 16) & 1) && HasAVX512Save)
986     Features |= 1 << FEATURE_AVX512F;
987   if (HasLeaf7 && ((EBX >> 17) & 1) && HasAVX512Save)
988     Features |= 1 << FEATURE_AVX512DQ;
989   if (HasLeaf7 && ((EBX >> 19) & 1))
990     Features2 |= 1 << (FEATURE_ADX - 32);
991   if (HasLeaf7 && ((EBX >> 21) & 1) && HasAVX512Save)
992     Features |= 1 << FEATURE_AVX512IFMA;
993   if (HasLeaf7 && ((EBX >> 23) & 1))
994     Features2 |= 1 << (FEATURE_CLFLUSHOPT - 32);
995   if (HasLeaf7 && ((EBX >> 26) & 1) && HasAVX512Save)
996     Features |= 1 << FEATURE_AVX512PF;
997   if (HasLeaf7 && ((EBX >> 27) & 1) && HasAVX512Save)
998     Features |= 1 << FEATURE_AVX512ER;
999   if (HasLeaf7 && ((EBX >> 28) & 1) && HasAVX512Save)
1000     Features |= 1 << FEATURE_AVX512CD;
1001   if (HasLeaf7 && ((EBX >> 29) & 1))
1002     Features2 |= 1 << (FEATURE_SHA - 32);
1003   if (HasLeaf7 && ((EBX >> 30) & 1) && HasAVX512Save)
1004     Features |= 1 << FEATURE_AVX512BW;
1005   if (HasLeaf7 && ((EBX >> 31) & 1) && HasAVX512Save)
1006     Features |= 1 << FEATURE_AVX512VL;
1007 
1008   if (HasLeaf7 && ((ECX >> 1) & 1) && HasAVX512Save)
1009     Features |= 1 << FEATURE_AVX512VBMI;
1010   if (HasLeaf7 && ((ECX >> 14) & 1) && HasAVX512Save)
1011     Features |= 1 << FEATURE_AVX512VPOPCNTDQ;
1012 
1013   if (HasLeaf7 && ((EDX >> 2) & 1) && HasAVX512Save)
1014     Features |= 1 << FEATURE_AVX5124VNNIW;
1015   if (HasLeaf7 && ((EDX >> 3) & 1) && HasAVX512Save)
1016     Features |= 1 << FEATURE_AVX5124FMAPS;
1017 
1018   unsigned MaxExtLevel;
1019   getX86CpuIDAndInfo(0x80000000, &MaxExtLevel, &EBX, &ECX, &EDX);
1020 
1021   bool HasExtLeaf1 = MaxExtLevel >= 0x80000001 &&
1022                      !getX86CpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
1023   if (HasExtLeaf1 && ((ECX >> 6) & 1))
1024     Features |= 1 << FEATURE_SSE4_A;
1025   if (HasExtLeaf1 && ((ECX >> 11) & 1))
1026     Features |= 1 << FEATURE_XOP;
1027   if (HasExtLeaf1 && ((ECX >> 16) & 1))
1028     Features |= 1 << FEATURE_FMA4;
1029 
1030   if (HasExtLeaf1 && ((EDX >> 29) & 1))
1031     Features2 |= 1 << (FEATURE_EM64T - 32);
1032 
1033   *FeaturesOut  = Features;
1034   *Features2Out = Features2;
1035 }
1036 
1037 StringRef sys::getHostCPUName() {
1038   unsigned EAX = 0, EBX = 0, ECX = 0, EDX = 0;
1039   unsigned MaxLeaf, Vendor;
1040 
1041 #if defined(__GNUC__) || defined(__clang__)
1042   //FIXME: include cpuid.h from clang or copy __get_cpuid_max here
1043   // and simplify it to not invoke __cpuid (like cpu_model.c in
1044   // compiler-rt/lib/builtins/cpu_model.c?
1045   // Opting for the second option.
1046   if(!isCpuIdSupported())
1047     return "generic";
1048 #endif
1049   if (getX86CpuIDAndInfo(0, &MaxLeaf, &Vendor, &ECX, &EDX) || MaxLeaf < 1)
1050     return "generic";
1051   getX86CpuIDAndInfo(0x1, &EAX, &EBX, &ECX, &EDX);
1052 
1053   unsigned Brand_id = EBX & 0xff;
1054   unsigned Family = 0, Model = 0;
1055   unsigned Features = 0, Features2 = 0;
1056   detectX86FamilyModel(EAX, &Family, &Model);
1057   getAvailableFeatures(ECX, EDX, MaxLeaf, &Features, &Features2);
1058 
1059   unsigned Type;
1060   unsigned Subtype;
1061 
1062   if (Vendor == SIG_INTEL) {
1063     getIntelProcessorTypeAndSubtype(Family, Model, Brand_id, Features,
1064                                     Features2, &Type, &Subtype);
1065     switch (Type) {
1066     case INTEL_i386:
1067       return "i386";
1068     case INTEL_i486:
1069       return "i486";
1070     case INTEL_PENTIUM:
1071       if (Subtype == INTEL_PENTIUM_MMX)
1072         return "pentium-mmx";
1073       return "pentium";
1074     case INTEL_PENTIUM_PRO:
1075       return "pentiumpro";
1076     case INTEL_PENTIUM_II:
1077       return "pentium2";
1078     case INTEL_PENTIUM_III:
1079       return "pentium3";
1080     case INTEL_PENTIUM_IV:
1081       return "pentium4";
1082     case INTEL_PENTIUM_M:
1083       return "pentium-m";
1084     case INTEL_CORE_DUO:
1085       return "yonah";
1086     case INTEL_CORE2:
1087       switch (Subtype) {
1088       case INTEL_CORE2_65:
1089         return "core2";
1090       case INTEL_CORE2_45:
1091         return "penryn";
1092       default:
1093         llvm_unreachable("Unexpected subtype!");
1094       }
1095     case INTEL_COREI7:
1096       switch (Subtype) {
1097       case INTEL_COREI7_NEHALEM:
1098         return "nehalem";
1099       case INTEL_COREI7_WESTMERE:
1100         return "westmere";
1101       case INTEL_COREI7_SANDYBRIDGE:
1102         return "sandybridge";
1103       case INTEL_COREI7_IVYBRIDGE:
1104         return "ivybridge";
1105       case INTEL_COREI7_HASWELL:
1106         return "haswell";
1107       case INTEL_COREI7_BROADWELL:
1108         return "broadwell";
1109       case INTEL_COREI7_SKYLAKE:
1110         return "skylake";
1111       case INTEL_COREI7_SKYLAKE_AVX512:
1112         return "skylake-avx512";
1113       default:
1114         llvm_unreachable("Unexpected subtype!");
1115       }
1116     case INTEL_BONNELL:
1117       return "bonnell";
1118     case INTEL_SILVERMONT:
1119       return "silvermont";
1120     case INTEL_GOLDMONT:
1121       return "goldmont";
1122     case INTEL_KNL:
1123       return "knl";
1124     case INTEL_X86_64:
1125       return "x86-64";
1126     case INTEL_NOCONA:
1127       return "nocona";
1128     case INTEL_PRESCOTT:
1129       return "prescott";
1130     default:
1131       break;
1132     }
1133   } else if (Vendor == SIG_AMD) {
1134     getAMDProcessorTypeAndSubtype(Family, Model, Features, &Type, &Subtype);
1135     switch (Type) {
1136     case AMD_i486:
1137       return "i486";
1138     case AMDPENTIUM:
1139       switch (Subtype) {
1140       case AMDPENTIUM_K6:
1141         return "k6";
1142       case AMDPENTIUM_K62:
1143         return "k6-2";
1144       case AMDPENTIUM_K63:
1145         return "k6-3";
1146       case AMDPENTIUM_GEODE:
1147         return "geode";
1148       default:
1149         return "pentium";
1150       }
1151     case AMDATHLON:
1152       switch (Subtype) {
1153       case AMDATHLON_CLASSIC:
1154         return "athlon";
1155       case AMDATHLON_XP:
1156         return "athlon-xp";
1157       case AMDATHLON_K8:
1158         return "k8";
1159       case AMDATHLON_K8SSE3:
1160         return "k8-sse3";
1161       default:
1162         llvm_unreachable("Unexpected subtype!");
1163       }
1164     case AMDFAM10H:
1165       return "amdfam10";
1166     case AMD_BTVER1:
1167       return "btver1";
1168     case AMDFAM15H:
1169       switch (Subtype) {
1170       default: // There are gaps in the subtype detection.
1171       case AMDFAM15H_BDVER1:
1172         return "bdver1";
1173       case AMDFAM15H_BDVER2:
1174         return "bdver2";
1175       case AMDFAM15H_BDVER3:
1176         return "bdver3";
1177       case AMDFAM15H_BDVER4:
1178         return "bdver4";
1179       }
1180     case AMD_BTVER2:
1181       return "btver2";
1182     case AMDFAM17H:
1183       return "znver1";
1184     default:
1185       break;
1186     }
1187   }
1188   return "generic";
1189 }
1190 
1191 #elif defined(__APPLE__) && (defined(__ppc__) || defined(__powerpc__))
1192 StringRef sys::getHostCPUName() {
1193   host_basic_info_data_t hostInfo;
1194   mach_msg_type_number_t infoCount;
1195 
1196   infoCount = HOST_BASIC_INFO_COUNT;
1197   host_info(mach_host_self(), HOST_BASIC_INFO, (host_info_t)&hostInfo,
1198             &infoCount);
1199 
1200   if (hostInfo.cpu_type != CPU_TYPE_POWERPC)
1201     return "generic";
1202 
1203   switch (hostInfo.cpu_subtype) {
1204   case CPU_SUBTYPE_POWERPC_601:
1205     return "601";
1206   case CPU_SUBTYPE_POWERPC_602:
1207     return "602";
1208   case CPU_SUBTYPE_POWERPC_603:
1209     return "603";
1210   case CPU_SUBTYPE_POWERPC_603e:
1211     return "603e";
1212   case CPU_SUBTYPE_POWERPC_603ev:
1213     return "603ev";
1214   case CPU_SUBTYPE_POWERPC_604:
1215     return "604";
1216   case CPU_SUBTYPE_POWERPC_604e:
1217     return "604e";
1218   case CPU_SUBTYPE_POWERPC_620:
1219     return "620";
1220   case CPU_SUBTYPE_POWERPC_750:
1221     return "750";
1222   case CPU_SUBTYPE_POWERPC_7400:
1223     return "7400";
1224   case CPU_SUBTYPE_POWERPC_7450:
1225     return "7450";
1226   case CPU_SUBTYPE_POWERPC_970:
1227     return "970";
1228   default:;
1229   }
1230 
1231   return "generic";
1232 }
1233 #elif defined(__linux__) && (defined(__ppc__) || defined(__powerpc__))
1234 StringRef sys::getHostCPUName() {
1235   std::unique_ptr<llvm::MemoryBuffer> P = getProcCpuinfoContent();
1236   const StringRef& Content = P ? P->getBuffer() : "";
1237   return detail::getHostCPUNameForPowerPC(Content);
1238 }
1239 #elif defined(__linux__) && (defined(__arm__) || defined(__aarch64__))
1240 StringRef sys::getHostCPUName() {
1241   std::unique_ptr<llvm::MemoryBuffer> P = getProcCpuinfoContent();
1242   const StringRef& Content = P ? P->getBuffer() : "";
1243   return detail::getHostCPUNameForARM(Content);
1244 }
1245 #elif defined(__linux__) && defined(__s390x__)
1246 StringRef sys::getHostCPUName() {
1247   std::unique_ptr<llvm::MemoryBuffer> P = getProcCpuinfoContent();
1248   const StringRef& Content = P ? P->getBuffer() : "";
1249   return detail::getHostCPUNameForS390x(Content);
1250 }
1251 #else
1252 StringRef sys::getHostCPUName() { return "generic"; }
1253 #endif
1254 
1255 #if defined(__linux__) && defined(__x86_64__)
1256 // On Linux, the number of physical cores can be computed from /proc/cpuinfo,
1257 // using the number of unique physical/core id pairs. The following
1258 // implementation reads the /proc/cpuinfo format on an x86_64 system.
1259 static int computeHostNumPhysicalCores() {
1260   // Read /proc/cpuinfo as a stream (until EOF reached). It cannot be
1261   // mmapped because it appears to have 0 size.
1262   llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> Text =
1263       llvm::MemoryBuffer::getFileAsStream("/proc/cpuinfo");
1264   if (std::error_code EC = Text.getError()) {
1265     llvm::errs() << "Can't read "
1266                  << "/proc/cpuinfo: " << EC.message() << "\n";
1267     return -1;
1268   }
1269   SmallVector<StringRef, 8> strs;
1270   (*Text)->getBuffer().split(strs, "\n", /*MaxSplit=*/-1,
1271                              /*KeepEmpty=*/false);
1272   int CurPhysicalId = -1;
1273   int CurCoreId = -1;
1274   SmallSet<std::pair<int, int>, 32> UniqueItems;
1275   for (auto &Line : strs) {
1276     Line = Line.trim();
1277     if (!Line.startswith("physical id") && !Line.startswith("core id"))
1278       continue;
1279     std::pair<StringRef, StringRef> Data = Line.split(':');
1280     auto Name = Data.first.trim();
1281     auto Val = Data.second.trim();
1282     if (Name == "physical id") {
1283       assert(CurPhysicalId == -1 &&
1284              "Expected a core id before seeing another physical id");
1285       Val.getAsInteger(10, CurPhysicalId);
1286     }
1287     if (Name == "core id") {
1288       assert(CurCoreId == -1 &&
1289              "Expected a physical id before seeing another core id");
1290       Val.getAsInteger(10, CurCoreId);
1291     }
1292     if (CurPhysicalId != -1 && CurCoreId != -1) {
1293       UniqueItems.insert(std::make_pair(CurPhysicalId, CurCoreId));
1294       CurPhysicalId = -1;
1295       CurCoreId = -1;
1296     }
1297   }
1298   return UniqueItems.size();
1299 }
1300 #elif defined(__APPLE__) && defined(__x86_64__)
1301 #include <sys/param.h>
1302 #include <sys/sysctl.h>
1303 
1304 // Gets the number of *physical cores* on the machine.
1305 static int computeHostNumPhysicalCores() {
1306   uint32_t count;
1307   size_t len = sizeof(count);
1308   sysctlbyname("hw.physicalcpu", &count, &len, NULL, 0);
1309   if (count < 1) {
1310     int nm[2];
1311     nm[0] = CTL_HW;
1312     nm[1] = HW_AVAILCPU;
1313     sysctl(nm, 2, &count, &len, NULL, 0);
1314     if (count < 1)
1315       return -1;
1316   }
1317   return count;
1318 }
1319 #else
1320 // On other systems, return -1 to indicate unknown.
1321 static int computeHostNumPhysicalCores() { return -1; }
1322 #endif
1323 
1324 int sys::getHostNumPhysicalCores() {
1325   static int NumCores = computeHostNumPhysicalCores();
1326   return NumCores;
1327 }
1328 
1329 #if defined(__i386__) || defined(_M_IX86) || \
1330     defined(__x86_64__) || defined(_M_X64)
1331 bool sys::getHostCPUFeatures(StringMap<bool> &Features) {
1332   unsigned EAX = 0, EBX = 0, ECX = 0, EDX = 0;
1333   unsigned MaxLevel;
1334   union {
1335     unsigned u[3];
1336     char c[12];
1337   } text;
1338 
1339   if (getX86CpuIDAndInfo(0, &MaxLevel, text.u + 0, text.u + 2, text.u + 1) ||
1340       MaxLevel < 1)
1341     return false;
1342 
1343   getX86CpuIDAndInfo(1, &EAX, &EBX, &ECX, &EDX);
1344 
1345   Features["cmov"] = (EDX >> 15) & 1;
1346   Features["mmx"] = (EDX >> 23) & 1;
1347   Features["sse"] = (EDX >> 25) & 1;
1348   Features["sse2"] = (EDX >> 26) & 1;
1349   Features["sse3"] = (ECX >> 0) & 1;
1350   Features["ssse3"] = (ECX >> 9) & 1;
1351   Features["sse4.1"] = (ECX >> 19) & 1;
1352   Features["sse4.2"] = (ECX >> 20) & 1;
1353 
1354   Features["pclmul"] = (ECX >> 1) & 1;
1355   Features["cx16"] = (ECX >> 13) & 1;
1356   Features["movbe"] = (ECX >> 22) & 1;
1357   Features["popcnt"] = (ECX >> 23) & 1;
1358   Features["aes"] = (ECX >> 25) & 1;
1359   Features["rdrnd"] = (ECX >> 30) & 1;
1360 
1361   // If CPUID indicates support for XSAVE, XRESTORE and AVX, and XGETBV
1362   // indicates that the AVX registers will be saved and restored on context
1363   // switch, then we have full AVX support.
1364   bool HasAVXSave = ((ECX >> 27) & 1) && ((ECX >> 28) & 1) &&
1365                     !getX86XCR0(&EAX, &EDX) && ((EAX & 0x6) == 0x6);
1366   Features["avx"] = HasAVXSave;
1367   Features["fma"] = HasAVXSave && (ECX >> 12) & 1;
1368   Features["f16c"] = HasAVXSave && (ECX >> 29) & 1;
1369 
1370   // Only enable XSAVE if OS has enabled support for saving YMM state.
1371   Features["xsave"] = HasAVXSave && (ECX >> 26) & 1;
1372 
1373   // AVX512 requires additional context to be saved by the OS.
1374   bool HasAVX512Save = HasAVXSave && ((EAX & 0xe0) == 0xe0);
1375 
1376   unsigned MaxExtLevel;
1377   getX86CpuIDAndInfo(0x80000000, &MaxExtLevel, &EBX, &ECX, &EDX);
1378 
1379   bool HasExtLeaf1 = MaxExtLevel >= 0x80000001 &&
1380                      !getX86CpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
1381   Features["lzcnt"] = HasExtLeaf1 && ((ECX >> 5) & 1);
1382   Features["sse4a"] = HasExtLeaf1 && ((ECX >> 6) & 1);
1383   Features["prfchw"] = HasExtLeaf1 && ((ECX >> 8) & 1);
1384   Features["xop"] = HasExtLeaf1 && ((ECX >> 11) & 1) && HasAVXSave;
1385   Features["lwp"] = HasExtLeaf1 && ((ECX >> 15) & 1);
1386   Features["fma4"] = HasExtLeaf1 && ((ECX >> 16) & 1) && HasAVXSave;
1387   Features["tbm"] = HasExtLeaf1 && ((ECX >> 21) & 1);
1388   Features["mwaitx"] = HasExtLeaf1 && ((ECX >> 29) & 1);
1389 
1390   bool HasExtLeaf8 = MaxExtLevel >= 0x80000008 &&
1391                      !getX86CpuIDAndInfoEx(0x80000008,0x0, &EAX, &EBX, &ECX, &EDX);
1392   Features["clzero"] = HasExtLeaf8 && ((EBX >> 0) & 1);
1393 
1394   bool HasLeaf7 =
1395       MaxLevel >= 7 && !getX86CpuIDAndInfoEx(0x7, 0x0, &EAX, &EBX, &ECX, &EDX);
1396 
1397   // AVX2 is only supported if we have the OS save support from AVX.
1398   Features["avx2"] = HasAVXSave && HasLeaf7 && ((EBX >> 5) & 1);
1399 
1400   Features["fsgsbase"] = HasLeaf7 && ((EBX >> 0) & 1);
1401   Features["sgx"] = HasLeaf7 && ((EBX >> 2) & 1);
1402   Features["bmi"] = HasLeaf7 && ((EBX >> 3) & 1);
1403   Features["bmi2"] = HasLeaf7 && ((EBX >> 8) & 1);
1404   Features["rtm"] = HasLeaf7 && ((EBX >> 11) & 1);
1405   Features["rdseed"] = HasLeaf7 && ((EBX >> 18) & 1);
1406   Features["adx"] = HasLeaf7 && ((EBX >> 19) & 1);
1407   Features["clflushopt"] = HasLeaf7 && ((EBX >> 23) & 1);
1408   Features["clwb"] = HasLeaf7 && ((EBX >> 24) & 1);
1409   Features["sha"] = HasLeaf7 && ((EBX >> 29) & 1);
1410 
1411   // AVX512 is only supported if the OS supports the context save for it.
1412   Features["avx512f"] = HasLeaf7 && ((EBX >> 16) & 1) && HasAVX512Save;
1413   Features["avx512dq"] = HasLeaf7 && ((EBX >> 17) & 1) && HasAVX512Save;
1414   Features["avx512ifma"] = HasLeaf7 && ((EBX >> 21) & 1) && HasAVX512Save;
1415   Features["avx512pf"] = HasLeaf7 && ((EBX >> 26) & 1) && HasAVX512Save;
1416   Features["avx512er"] = HasLeaf7 && ((EBX >> 27) & 1) && HasAVX512Save;
1417   Features["avx512cd"] = HasLeaf7 && ((EBX >> 28) & 1) && HasAVX512Save;
1418   Features["avx512bw"] = HasLeaf7 && ((EBX >> 30) & 1) && HasAVX512Save;
1419   Features["avx512vl"] = HasLeaf7 && ((EBX >> 31) & 1) && HasAVX512Save;
1420 
1421   Features["prefetchwt1"] = HasLeaf7 && (ECX & 1);
1422   Features["avx512vbmi"] = HasLeaf7 && ((ECX >> 1) & 1) && HasAVX512Save;
1423   Features["avx512vpopcntdq"] = HasLeaf7 && ((ECX >> 14) & 1) && HasAVX512Save;
1424   // Enable protection keys
1425   Features["pku"] = HasLeaf7 && ((ECX >> 4) & 1);
1426 
1427   bool HasLeafD = MaxLevel >= 0xd &&
1428                   !getX86CpuIDAndInfoEx(0xd, 0x1, &EAX, &EBX, &ECX, &EDX);
1429 
1430   // Only enable XSAVE if OS has enabled support for saving YMM state.
1431   Features["xsaveopt"] = HasAVXSave && HasLeafD && ((EAX >> 0) & 1);
1432   Features["xsavec"] = HasAVXSave && HasLeafD && ((EAX >> 1) & 1);
1433   Features["xsaves"] = HasAVXSave && HasLeafD && ((EAX >> 3) & 1);
1434 
1435   return true;
1436 }
1437 #elif defined(__linux__) && (defined(__arm__) || defined(__aarch64__))
1438 bool sys::getHostCPUFeatures(StringMap<bool> &Features) {
1439   std::unique_ptr<llvm::MemoryBuffer> P = getProcCpuinfoContent();
1440   if (!P)
1441     return false;
1442 
1443   SmallVector<StringRef, 32> Lines;
1444   P->getBuffer().split(Lines, "\n");
1445 
1446   SmallVector<StringRef, 32> CPUFeatures;
1447 
1448   // Look for the CPU features.
1449   for (unsigned I = 0, E = Lines.size(); I != E; ++I)
1450     if (Lines[I].startswith("Features")) {
1451       Lines[I].split(CPUFeatures, ' ');
1452       break;
1453     }
1454 
1455 #if defined(__aarch64__)
1456   // Keep track of which crypto features we have seen
1457   enum { CAP_AES = 0x1, CAP_PMULL = 0x2, CAP_SHA1 = 0x4, CAP_SHA2 = 0x8 };
1458   uint32_t crypto = 0;
1459 #endif
1460 
1461   for (unsigned I = 0, E = CPUFeatures.size(); I != E; ++I) {
1462     StringRef LLVMFeatureStr = StringSwitch<StringRef>(CPUFeatures[I])
1463 #if defined(__aarch64__)
1464                                    .Case("asimd", "neon")
1465                                    .Case("fp", "fp-armv8")
1466                                    .Case("crc32", "crc")
1467 #else
1468                                    .Case("half", "fp16")
1469                                    .Case("neon", "neon")
1470                                    .Case("vfpv3", "vfp3")
1471                                    .Case("vfpv3d16", "d16")
1472                                    .Case("vfpv4", "vfp4")
1473                                    .Case("idiva", "hwdiv-arm")
1474                                    .Case("idivt", "hwdiv")
1475 #endif
1476                                    .Default("");
1477 
1478 #if defined(__aarch64__)
1479     // We need to check crypto separately since we need all of the crypto
1480     // extensions to enable the subtarget feature
1481     if (CPUFeatures[I] == "aes")
1482       crypto |= CAP_AES;
1483     else if (CPUFeatures[I] == "pmull")
1484       crypto |= CAP_PMULL;
1485     else if (CPUFeatures[I] == "sha1")
1486       crypto |= CAP_SHA1;
1487     else if (CPUFeatures[I] == "sha2")
1488       crypto |= CAP_SHA2;
1489 #endif
1490 
1491     if (LLVMFeatureStr != "")
1492       Features[LLVMFeatureStr] = true;
1493   }
1494 
1495 #if defined(__aarch64__)
1496   // If we have all crypto bits we can add the feature
1497   if (crypto == (CAP_AES | CAP_PMULL | CAP_SHA1 | CAP_SHA2))
1498     Features["crypto"] = true;
1499 #endif
1500 
1501   return true;
1502 }
1503 #else
1504 bool sys::getHostCPUFeatures(StringMap<bool> &Features) { return false; }
1505 #endif
1506 
1507 std::string sys::getProcessTriple() {
1508   std::string TargetTripleString = updateTripleOSVersion(LLVM_HOST_TRIPLE);
1509   Triple PT(Triple::normalize(TargetTripleString));
1510 
1511   if (sizeof(void *) == 8 && PT.isArch32Bit())
1512     PT = PT.get64BitArchVariant();
1513   if (sizeof(void *) == 4 && PT.isArch64Bit())
1514     PT = PT.get32BitArchVariant();
1515 
1516   return PT.str();
1517 }
1518