1 //===-- Host.cpp - Implement OS Host Concept --------------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the operating system Host concept.
12 //===----------------------------------------------------------------------===//
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"
25 // Include the platform-specific parts of this class.
27 #include "Unix/Host.inc"
30 #include "Windows/Host.inc"
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>
42 #define DEBUG_TYPE "host-detection"
44 //===----------------------------------------------------------------------===//
46 // Implementations of the CPU detection routines
48 //===----------------------------------------------------------------------===//
52 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
59 std::error_code EC = sys::fs::openFileForRead("/proc/cpuinfo", FD);
61 DEBUG(dbgs() << "Unable to open /proc/cpuinfo: " << EC.message() << "\n");
64 int Ret = read(FD, Buf, Size);
65 int CloseStatus = close(FD);
71 #if defined(i386) || defined(__i386__) || defined(__x86__) || defined(_M_IX86)\
72 || defined(__x86_64__) || defined(_M_AMD64) || defined (_M_X64)
74 /// GetX86CpuIDAndInfo - Execute the specified cpuid and return the 4 values in the
75 /// specified arguments. If we can't run cpuid on the host, return true.
76 static bool GetX86CpuIDAndInfo(unsigned value, unsigned *rEAX, unsigned *rEBX,
77 unsigned *rECX, unsigned *rEDX) {
78 #if defined(__GNUC__) || defined(__clang__)
79 #if defined(__x86_64__) || defined(_M_AMD64) || defined (_M_X64)
80 // gcc doesn't know cpuid would clobber ebx/rbx. Preseve it manually.
81 asm ("movq\t%%rbx, %%rsi\n\t"
83 "xchgq\t%%rbx, %%rsi\n\t"
90 #elif defined(i386) || defined(__i386__) || defined(__x86__) || defined(_M_IX86)
91 asm ("movl\t%%ebx, %%esi\n\t"
93 "xchgl\t%%ebx, %%esi\n\t"
100 // pedantic #else returns to appease -Wunreachable-code (so we don't generate
101 // postprocessed code that looks like "return true; return false;")
105 #elif defined(_MSC_VER)
106 // The MSVC intrinsic is portable across x86 and x64.
108 __cpuid(registers, value);
109 *rEAX = registers[0];
110 *rEBX = registers[1];
111 *rECX = registers[2];
112 *rEDX = registers[3];
119 /// GetX86CpuIDAndInfoEx - Execute the specified cpuid with subleaf and return the
120 /// 4 values in the specified arguments. If we can't run cpuid on the host,
122 static bool GetX86CpuIDAndInfoEx(unsigned value, unsigned subleaf,
123 unsigned *rEAX, unsigned *rEBX, unsigned *rECX,
125 #if defined(__x86_64__) || defined(_M_AMD64) || defined (_M_X64)
126 #if defined(__GNUC__)
127 // gcc doesn't know cpuid would clobber ebx/rbx. Preseve it manually.
128 asm ("movq\t%%rbx, %%rsi\n\t"
130 "xchgq\t%%rbx, %%rsi\n\t"
138 #elif defined(_MSC_VER)
139 // __cpuidex was added in MSVC++ 9.0 SP1
140 #if (_MSC_VER > 1500) || (_MSC_VER == 1500 && _MSC_FULL_VER >= 150030729)
142 __cpuidex(registers, value, subleaf);
143 *rEAX = registers[0];
144 *rEBX = registers[1];
145 *rECX = registers[2];
146 *rEDX = registers[3];
154 #elif defined(i386) || defined(__i386__) || defined(__x86__) || defined(_M_IX86)
155 #if defined(__GNUC__)
156 asm ("movl\t%%ebx, %%esi\n\t"
158 "xchgl\t%%ebx, %%esi\n\t"
166 #elif defined(_MSC_VER)
172 mov dword ptr [esi],eax
174 mov dword ptr [esi],ebx
176 mov dword ptr [esi],ecx
178 mov dword ptr [esi],edx
189 static bool OSHasAVXSupport() {
190 #if defined(__GNUC__)
191 // Check xgetbv; this uses a .byte sequence instead of the instruction
192 // directly because older assemblers do not include support for xgetbv and
193 // there is no easy way to conditionally compile based on the assembler used.
195 __asm__ (".byte 0x0f, 0x01, 0xd0" : "=a" (rEAX), "=d" (rEDX) : "c" (0));
196 #elif defined(_MSC_FULL_VER) && defined(_XCR_XFEATURE_ENABLED_MASK)
197 unsigned long long rEAX = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
199 int rEAX = 0; // Ensures we return false
201 return (rEAX & 6) == 6;
204 static void DetectX86FamilyModel(unsigned EAX, unsigned &Family,
206 Family = (EAX >> 8) & 0xf; // Bits 8 - 11
207 Model = (EAX >> 4) & 0xf; // Bits 4 - 7
208 if (Family == 6 || Family == 0xf) {
210 // Examine extended family ID if family ID is F.
211 Family += (EAX >> 20) & 0xff; // Bits 20 - 27
212 // Examine extended model ID if family ID is 6 or F.
213 Model += ((EAX >> 16) & 0xf) << 4; // Bits 16 - 19
217 StringRef sys::getHostCPUName() {
218 unsigned EAX = 0, EBX = 0, ECX = 0, EDX = 0;
219 if (GetX86CpuIDAndInfo(0x1, &EAX, &EBX, &ECX, &EDX))
223 DetectX86FamilyModel(EAX, Family, Model);
230 GetX86CpuIDAndInfo(0, &EAX, text.u+0, text.u+2, text.u+1);
232 unsigned MaxLeaf = EAX;
233 bool HasSSE3 = (ECX & 0x1);
234 bool HasSSE41 = (ECX & 0x80000);
235 // If CPUID indicates support for XSAVE, XRESTORE and AVX, and XGETBV
236 // indicates that the AVX registers will be saved and restored on context
237 // switch, then we have full AVX support.
238 const unsigned AVXBits = (1 << 27) | (1 << 28);
239 bool HasAVX = ((ECX & AVXBits) == AVXBits) && OSHasAVXSupport();
240 bool HasAVX2 = HasAVX && MaxLeaf >= 0x7 &&
241 !GetX86CpuIDAndInfoEx(0x7, 0x0, &EAX, &EBX, &ECX, &EDX) &&
243 GetX86CpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
244 bool Em64T = (EDX >> 29) & 0x1;
245 bool HasTBM = (ECX >> 21) & 0x1;
247 if (memcmp(text.c, "GenuineIntel", 12) == 0) {
253 case 0: // Intel486 DX processors
254 case 1: // Intel486 DX processors
255 case 2: // Intel486 SX processors
256 case 3: // Intel487 processors, IntelDX2 OverDrive processors,
257 // IntelDX2 processors
258 case 4: // Intel486 SL processor
259 case 5: // IntelSX2 processors
260 case 7: // Write-Back Enhanced IntelDX2 processors
261 case 8: // IntelDX4 OverDrive processors, IntelDX4 processors
262 default: return "i486";
266 case 1: // Pentium OverDrive processor for Pentium processor (60, 66),
267 // Pentium processors (60, 66)
268 case 2: // Pentium OverDrive processor for Pentium processor (75, 90,
269 // 100, 120, 133), Pentium processors (75, 90, 100, 120, 133,
271 case 3: // Pentium OverDrive processors for Intel486 processor-based
275 case 4: // Pentium OverDrive processor with MMX technology for Pentium
276 // processor (75, 90, 100, 120, 133), Pentium processor with
277 // MMX technology (166, 200)
278 return "pentium-mmx";
280 default: return "pentium";
284 case 1: // Pentium Pro processor
287 case 3: // Intel Pentium II OverDrive processor, Pentium II processor,
289 case 5: // Pentium II processor, model 05, Pentium II Xeon processor,
290 // model 05, and Intel Celeron processor, model 05
291 case 6: // Celeron processor, model 06
294 case 7: // Pentium III processor, model 07, and Pentium III Xeon
295 // processor, model 07
296 case 8: // Pentium III processor, model 08, Pentium III Xeon processor,
297 // model 08, and Celeron processor, model 08
298 case 10: // Pentium III Xeon processor, model 0Ah
299 case 11: // Pentium III processor, model 0Bh
302 case 9: // Intel Pentium M processor, Intel Celeron M processor model 09.
303 case 13: // Intel Pentium M processor, Intel Celeron M processor, model
304 // 0Dh. All processors are manufactured using the 90 nm process.
307 case 14: // Intel Core Duo processor, Intel Core Solo processor, model
308 // 0Eh. All processors are manufactured using the 65 nm process.
311 case 15: // Intel Core 2 Duo processor, Intel Core 2 Duo mobile
312 // processor, Intel Core 2 Quad processor, Intel Core 2 Quad
313 // mobile processor, Intel Core 2 Extreme processor, Intel
314 // Pentium Dual-Core processor, Intel Xeon processor, model
315 // 0Fh. All processors are manufactured using the 65 nm process.
316 case 22: // Intel Celeron processor model 16h. All processors are
317 // manufactured using the 65 nm process
320 case 21: // Intel EP80579 Integrated Processor and Intel EP80579
321 // Integrated Processor with Intel QuickAssist Technology
322 return "i686"; // FIXME: ???
324 case 23: // Intel Core 2 Extreme processor, Intel Xeon processor, model
325 // 17h. All processors are manufactured using the 45 nm process.
327 // 45nm: Penryn , Wolfdale, Yorkfield (XE)
328 // Not all Penryn processors support SSE 4.1 (such as the Pentium brand)
329 return HasSSE41 ? "penryn" : "core2";
331 case 26: // Intel Core i7 processor and Intel Xeon processor. All
332 // processors are manufactured using the 45 nm process.
333 case 29: // Intel Xeon processor MP. All processors are manufactured using
334 // the 45 nm process.
335 case 30: // Intel(R) Core(TM) i7 CPU 870 @ 2.93GHz.
336 // As found in a Summer 2010 model iMac.
337 case 37: // Intel Core i7, laptop version.
338 case 44: // Intel Core i7 processor and Intel Xeon processor. All
339 // processors are manufactured using the 32 nm process.
340 case 46: // Nehalem EX
341 case 47: // Westmere EX
345 case 42: // Intel Core i7 processor. All processors are manufactured
346 // using the 32 nm process.
348 // Not all Sandy Bridge processors support AVX (such as the Pentium
349 // versions instead of the i7 versions).
350 return HasAVX ? "corei7-avx" : "corei7";
354 case 62: // Ivy Bridge EP
355 // Not all Ivy Bridge processors support AVX (such as the Pentium
356 // versions instead of the i7 versions).
357 return HasAVX ? "core-avx-i" : "corei7";
364 // Not all Haswell processors support AVX too (such as the Pentium
365 // versions instead of the i7 versions).
366 return HasAVX2 ? "core-avx2" : "corei7";
368 case 28: // Most 45 nm Intel Atom processors
369 case 38: // 45 nm Atom Lincroft
370 case 39: // 32 nm Atom Medfield
371 case 53: // 32 nm Atom Midview
372 case 54: // 32 nm Atom Midview
375 // Atom Silvermont codes from the Intel software optimization guide.
381 default: return (Em64T) ? "x86-64" : "i686";
385 case 0: // Pentium 4 processor, Intel Xeon processor. All processors are
386 // model 00h and manufactured using the 0.18 micron process.
387 case 1: // Pentium 4 processor, Intel Xeon processor, Intel Xeon
388 // processor MP, and Intel Celeron processor. All processors are
389 // model 01h and manufactured using the 0.18 micron process.
390 case 2: // Pentium 4 processor, Mobile Intel Pentium 4 processor - M,
391 // Intel Xeon processor, Intel Xeon processor MP, Intel Celeron
392 // processor, and Mobile Intel Celeron processor. All processors
393 // are model 02h and manufactured using the 0.13 micron process.
394 return (Em64T) ? "x86-64" : "pentium4";
396 case 3: // Pentium 4 processor, Intel Xeon processor, Intel Celeron D
397 // processor. All processors are model 03h and manufactured using
398 // the 90 nm process.
399 case 4: // Pentium 4 processor, Pentium 4 processor Extreme Edition,
400 // Pentium D processor, Intel Xeon processor, Intel Xeon
401 // processor MP, Intel Celeron D processor. All processors are
402 // model 04h and manufactured using the 90 nm process.
403 case 6: // Pentium 4 processor, Pentium D processor, Pentium processor
404 // Extreme Edition, Intel Xeon processor, Intel Xeon processor
405 // MP, Intel Celeron D processor. All processors are model 06h
406 // and manufactured using the 65 nm process.
407 return (Em64T) ? "nocona" : "prescott";
410 return (Em64T) ? "x86-64" : "pentium4";
417 } else if (memcmp(text.c, "AuthenticAMD", 12) == 0) {
418 // FIXME: this poorly matches the generated SubtargetFeatureKV table. There
419 // appears to be no way to generate the wide variety of AMD-specific targets
420 // from the information returned from CPUID.
428 case 8: return "k6-2";
430 case 13: return "k6-3";
431 case 10: return "geode";
432 default: return "pentium";
436 case 4: return "athlon-tbird";
439 case 8: return "athlon-mp";
440 case 10: return "athlon-xp";
441 default: return "athlon";
447 case 1: return "opteron";
448 case 5: return "athlon-fx"; // also opteron
449 default: return "athlon64";
456 if (!HasAVX) // If the OS doesn't support AVX provide a sane fallback.
459 return "bdver4"; // 50h-6Fh: Excavator
461 return "bdver3"; // 30h-3Fh: Steamroller
462 if (Model >= 0x10 || HasTBM)
463 return "bdver2"; // 10h-1Fh: Piledriver
464 return "bdver1"; // 00h-0Fh: Bulldozer
466 if (!HasAVX) // If the OS doesn't support AVX provide a sane fallback.
475 #elif defined(__APPLE__) && (defined(__ppc__) || defined(__powerpc__))
476 StringRef sys::getHostCPUName() {
477 host_basic_info_data_t hostInfo;
478 mach_msg_type_number_t infoCount;
480 infoCount = HOST_BASIC_INFO_COUNT;
481 host_info(mach_host_self(), HOST_BASIC_INFO, (host_info_t)&hostInfo,
484 if (hostInfo.cpu_type != CPU_TYPE_POWERPC) return "generic";
486 switch(hostInfo.cpu_subtype) {
487 case CPU_SUBTYPE_POWERPC_601: return "601";
488 case CPU_SUBTYPE_POWERPC_602: return "602";
489 case CPU_SUBTYPE_POWERPC_603: return "603";
490 case CPU_SUBTYPE_POWERPC_603e: return "603e";
491 case CPU_SUBTYPE_POWERPC_603ev: return "603ev";
492 case CPU_SUBTYPE_POWERPC_604: return "604";
493 case CPU_SUBTYPE_POWERPC_604e: return "604e";
494 case CPU_SUBTYPE_POWERPC_620: return "620";
495 case CPU_SUBTYPE_POWERPC_750: return "750";
496 case CPU_SUBTYPE_POWERPC_7400: return "7400";
497 case CPU_SUBTYPE_POWERPC_7450: return "7450";
498 case CPU_SUBTYPE_POWERPC_970: return "970";
504 #elif defined(__linux__) && (defined(__ppc__) || defined(__powerpc__))
505 StringRef sys::getHostCPUName() {
506 // Access to the Processor Version Register (PVR) on PowerPC is privileged,
507 // and so we must use an operating-system interface to determine the current
508 // processor type. On Linux, this is exposed through the /proc/cpuinfo file.
509 const char *generic = "generic";
511 // The cpu line is second (after the 'processor: 0' line), so if this
512 // buffer is too small then something has changed (or is wrong).
514 ssize_t CPUInfoSize = readCpuInfo(buffer, sizeof(buffer));
515 if (CPUInfoSize == -1)
518 const char *CPUInfoStart = buffer;
519 const char *CPUInfoEnd = buffer + CPUInfoSize;
521 const char *CIP = CPUInfoStart;
523 const char *CPUStart = 0;
526 // We need to find the first line which starts with cpu, spaces, and a colon.
527 // After the colon, there may be some additional spaces and then the cpu type.
528 while (CIP < CPUInfoEnd && CPUStart == 0) {
529 if (CIP < CPUInfoEnd && *CIP == '\n')
532 if (CIP < CPUInfoEnd && *CIP == 'c') {
534 if (CIP < CPUInfoEnd && *CIP == 'p') {
536 if (CIP < CPUInfoEnd && *CIP == 'u') {
538 while (CIP < CPUInfoEnd && (*CIP == ' ' || *CIP == '\t'))
541 if (CIP < CPUInfoEnd && *CIP == ':') {
543 while (CIP < CPUInfoEnd && (*CIP == ' ' || *CIP == '\t'))
546 if (CIP < CPUInfoEnd) {
548 while (CIP < CPUInfoEnd && (*CIP != ' ' && *CIP != '\t' &&
549 *CIP != ',' && *CIP != '\n'))
551 CPULen = CIP - CPUStart;
559 while (CIP < CPUInfoEnd && *CIP != '\n')
566 return StringSwitch<const char *>(StringRef(CPUStart, CPULen))
567 .Case("604e", "604e")
569 .Case("7400", "7400")
570 .Case("7410", "7400")
571 .Case("7447", "7400")
572 .Case("7455", "7450")
574 .Case("POWER4", "970")
575 .Case("PPC970FX", "970")
576 .Case("PPC970MP", "970")
578 .Case("POWER5", "g5")
580 .Case("POWER6", "pwr6")
581 .Case("POWER7", "pwr7")
582 .Case("POWER8", "pwr8")
583 .Case("POWER8E", "pwr8")
586 #elif defined(__linux__) && defined(__arm__)
587 StringRef sys::getHostCPUName() {
588 // The cpuid register on arm is not accessible from user space. On Linux,
589 // it is exposed through the /proc/cpuinfo file.
591 // Read 1024 bytes from /proc/cpuinfo, which should contain the CPU part line
594 ssize_t CPUInfoSize = readCpuInfo(buffer, sizeof(buffer));
595 if (CPUInfoSize == -1)
598 StringRef Str(buffer, CPUInfoSize);
600 SmallVector<StringRef, 32> Lines;
601 Str.split(Lines, "\n");
603 // Look for the CPU implementer line.
604 StringRef Implementer;
605 for (unsigned I = 0, E = Lines.size(); I != E; ++I)
606 if (Lines[I].startswith("CPU implementer"))
607 Implementer = Lines[I].substr(15).ltrim("\t :");
609 if (Implementer == "0x41") // ARM Ltd.
610 // Look for the CPU part line.
611 for (unsigned I = 0, E = Lines.size(); I != E; ++I)
612 if (Lines[I].startswith("CPU part"))
613 // The CPU part is a 3 digit hexadecimal number with a 0x prefix. The
614 // values correspond to the "Part number" in the CP15/c0 register. The
615 // contents are specified in the various processor manuals.
616 return StringSwitch<const char *>(Lines[I].substr(8).ltrim("\t :"))
617 .Case("0x926", "arm926ej-s")
618 .Case("0xb02", "mpcore")
619 .Case("0xb36", "arm1136j-s")
620 .Case("0xb56", "arm1156t2-s")
621 .Case("0xb76", "arm1176jz-s")
622 .Case("0xc08", "cortex-a8")
623 .Case("0xc09", "cortex-a9")
624 .Case("0xc0f", "cortex-a15")
625 .Case("0xc20", "cortex-m0")
626 .Case("0xc23", "cortex-m3")
627 .Case("0xc24", "cortex-m4")
630 if (Implementer == "0x51") // Qualcomm Technologies, Inc.
631 // Look for the CPU part line.
632 for (unsigned I = 0, E = Lines.size(); I != E; ++I)
633 if (Lines[I].startswith("CPU part"))
634 // The CPU part is a 3 digit hexadecimal number with a 0x prefix. The
635 // values correspond to the "Part number" in the CP15/c0 register. The
636 // contents are specified in the various processor manuals.
637 return StringSwitch<const char *>(Lines[I].substr(8).ltrim("\t :"))
638 .Case("0x06f", "krait") // APQ8064
643 #elif defined(__linux__) && defined(__s390x__)
644 StringRef sys::getHostCPUName() {
645 // STIDP is a privileged operation, so use /proc/cpuinfo instead.
647 // The "processor 0:" line comes after a fair amount of other information,
648 // including a cache breakdown, but this should be plenty.
650 ssize_t CPUInfoSize = readCpuInfo(buffer, sizeof(buffer));
651 if (CPUInfoSize == -1)
654 StringRef Str(buffer, CPUInfoSize);
655 SmallVector<StringRef, 32> Lines;
656 Str.split(Lines, "\n");
657 for (unsigned I = 0, E = Lines.size(); I != E; ++I) {
658 if (Lines[I].startswith("processor ")) {
659 size_t Pos = Lines[I].find("machine = ");
660 if (Pos != StringRef::npos) {
661 Pos += sizeof("machine = ") - 1;
663 if (!Lines[I].drop_front(Pos).getAsInteger(10, Id)) {
677 StringRef sys::getHostCPUName() {
682 #if defined(__linux__) && (defined(__arm__) || defined(__aarch64__))
683 bool sys::getHostCPUFeatures(StringMap<bool> &Features) {
684 // Read 1024 bytes from /proc/cpuinfo, which should contain the Features line
687 ssize_t CPUInfoSize = readCpuInfo(buffer, sizeof(buffer));
688 if (CPUInfoSize == -1)
691 StringRef Str(buffer, CPUInfoSize);
693 SmallVector<StringRef, 32> Lines;
694 Str.split(Lines, "\n");
696 SmallVector<StringRef, 32> CPUFeatures;
698 // Look for the CPU features.
699 for (unsigned I = 0, E = Lines.size(); I != E; ++I)
700 if (Lines[I].startswith("Features")) {
701 Lines[I].split(CPUFeatures, " ");
705 #if defined(__aarch64__)
706 // Keep track of which crypto features we have seen
716 for (unsigned I = 0, E = CPUFeatures.size(); I != E; ++I) {
717 StringRef LLVMFeatureStr = StringSwitch<StringRef>(CPUFeatures[I])
718 #if defined(__aarch64__)
719 .Case("asimd", "neon")
720 .Case("fp", "fp-armv8")
721 .Case("crc32", "crc")
723 .Case("half", "fp16")
724 .Case("neon", "neon")
725 .Case("vfpv3", "vfp3")
726 .Case("vfpv3d16", "d16")
727 .Case("vfpv4", "vfp4")
728 .Case("idiva", "hwdiv-arm")
729 .Case("idivt", "hwdiv")
733 #if defined(__aarch64__)
734 // We need to check crypto separately since we need all of the crypto
735 // extensions to enable the subtarget feature
736 if (CPUFeatures[I] == "aes")
738 else if (CPUFeatures[I] == "pmull")
740 else if (CPUFeatures[I] == "sha1")
742 else if (CPUFeatures[I] == "sha2")
746 if (LLVMFeatureStr != "")
747 Features[LLVMFeatureStr] = true;
750 #if defined(__aarch64__)
751 // If we have all crypto bits we can add the feature
752 if (crypto == (CAP_AES | CAP_PMULL | CAP_SHA1 | CAP_SHA2))
753 Features["crypto"] = true;
759 bool sys::getHostCPUFeatures(StringMap<bool> &Features){
764 std::string sys::getProcessTriple() {
765 Triple PT(Triple::normalize(LLVM_HOST_TRIPLE));
767 if (sizeof(void *) == 8 && PT.isArch32Bit())
768 PT = PT.get64BitArchVariant();
769 if (sizeof(void *) == 4 && PT.isArch64Bit())
770 PT = PT.get32BitArchVariant();