//===-- cpu_model.c - Support for __cpu_model builtin  ------------*- C -*-===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
//  This file is based on LLVM's lib/Support/Host.cpp.
//  It implements the operating system Host concept and builtin
//  __cpu_model for the compiler_rt library, for x86 only.
//
//===----------------------------------------------------------------------===//

#if (defined(__i386__) || defined(_M_IX86) || \
     defined(__x86_64__) || defined(_M_X64)) && \
    (defined(__GNUC__) || defined(__clang__) || defined(_MSC_VER))

#include <assert.h>

#define bool int
#define true 1
#define false 0

#ifdef _MSC_VER
#include <intrin.h>
#endif

#ifndef __has_attribute
#define __has_attribute(attr) 0
#endif

enum VendorSignatures {
  SIG_INTEL = 0x756e6547 /* Genu */,
  SIG_AMD = 0x68747541 /* Auth */
};

enum ProcessorVendors {
  VENDOR_INTEL = 1,
  VENDOR_AMD,
  VENDOR_OTHER,
  VENDOR_MAX
};

enum ProcessorTypes {
  INTEL_BONNELL = 1,
  INTEL_CORE2,
  INTEL_COREI7,
  AMDFAM10H,
  AMDFAM15H,
  INTEL_SILVERMONT,
  INTEL_KNL,
  AMD_BTVER1,
  AMD_BTVER2,
  AMDFAM17H,
  CPU_TYPE_MAX
};

enum ProcessorSubtypes {
  INTEL_COREI7_NEHALEM = 1,
  INTEL_COREI7_WESTMERE,
  INTEL_COREI7_SANDYBRIDGE,
  AMDFAM10H_BARCELONA,
  AMDFAM10H_SHANGHAI,
  AMDFAM10H_ISTANBUL,
  AMDFAM15H_BDVER1,
  AMDFAM15H_BDVER2,
  AMDFAM15H_BDVER3,
  AMDFAM15H_BDVER4,
  AMDFAM17H_ZNVER1,
  INTEL_COREI7_IVYBRIDGE,
  INTEL_COREI7_HASWELL,
  INTEL_COREI7_BROADWELL,
  INTEL_COREI7_SKYLAKE,
  INTEL_COREI7_SKYLAKE_AVX512,
  CPU_SUBTYPE_MAX
};

enum ProcessorFeatures {
  FEATURE_CMOV = 0,
  FEATURE_MMX,
  FEATURE_POPCNT,
  FEATURE_SSE,
  FEATURE_SSE2,
  FEATURE_SSE3,
  FEATURE_SSSE3,
  FEATURE_SSE4_1,
  FEATURE_SSE4_2,
  FEATURE_AVX,
  FEATURE_AVX2,
  FEATURE_SSE4_A,
  FEATURE_FMA4,
  FEATURE_XOP,
  FEATURE_FMA,
  FEATURE_AVX512F,
  FEATURE_BMI,
  FEATURE_BMI2,
  FEATURE_AES,
  FEATURE_PCLMUL,
  FEATURE_AVX512VL,
  FEATURE_AVX512BW,
  FEATURE_AVX512DQ,
  FEATURE_AVX512CD,
  FEATURE_AVX512ER,
  FEATURE_AVX512PF,
  FEATURE_AVX512VBMI,
  FEATURE_AVX512IFMA,
  FEATURE_AVX5124VNNIW,
  FEATURE_AVX5124FMAPS,
  FEATURE_AVX512VPOPCNTDQ
};

// The check below for i386 was copied from clang's cpuid.h (__get_cpuid_max).
// Check motivated by bug reports for OpenSSL crashing on CPUs without CPUID
// support. Consequently, for i386, the presence of CPUID is checked first
// via the corresponding eflags bit.
static bool isCpuIdSupported() {
#if defined(__GNUC__) || defined(__clang__)
#if defined(__i386__)
  int __cpuid_supported;
  __asm__("  pushfl\n"
          "  popl   %%eax\n"
          "  movl   %%eax,%%ecx\n"
          "  xorl   $0x00200000,%%eax\n"
          "  pushl  %%eax\n"
          "  popfl\n"
          "  pushfl\n"
          "  popl   %%eax\n"
          "  movl   $0,%0\n"
          "  cmpl   %%eax,%%ecx\n"
          "  je     1f\n"
          "  movl   $1,%0\n"
          "1:"
          : "=r"(__cpuid_supported)
          :
          : "eax", "ecx");
  if (!__cpuid_supported)
    return false;
#endif
  return true;
#endif
  return true;
}

// This code is copied from lib/Support/Host.cpp.
// Changes to either file should be mirrored in the other.

/// getX86CpuIDAndInfo - Execute the specified cpuid and return the 4 values in
/// the specified arguments.  If we can't run cpuid on the host, return true.
static bool getX86CpuIDAndInfo(unsigned value, unsigned *rEAX, unsigned *rEBX,
                               unsigned *rECX, unsigned *rEDX) {
#if defined(__GNUC__) || defined(__clang__)
#if defined(__x86_64__)
  // gcc doesn't know cpuid would clobber ebx/rbx. Preserve it manually.
  // FIXME: should we save this for Clang?
  __asm__("movq\t%%rbx, %%rsi\n\t"
          "cpuid\n\t"
          "xchgq\t%%rbx, %%rsi\n\t"
          : "=a"(*rEAX), "=S"(*rEBX), "=c"(*rECX), "=d"(*rEDX)
          : "a"(value));
  return false;
#elif defined(__i386__)
  __asm__("movl\t%%ebx, %%esi\n\t"
          "cpuid\n\t"
          "xchgl\t%%ebx, %%esi\n\t"
          : "=a"(*rEAX), "=S"(*rEBX), "=c"(*rECX), "=d"(*rEDX)
          : "a"(value));
  return false;
#else
  return true;
#endif
#elif defined(_MSC_VER)
  // The MSVC intrinsic is portable across x86 and x64.
  int registers[4];
  __cpuid(registers, value);
  *rEAX = registers[0];
  *rEBX = registers[1];
  *rECX = registers[2];
  *rEDX = registers[3];
  return false;
#else
  return true;
#endif
}

/// getX86CpuIDAndInfoEx - Execute the specified cpuid with subleaf and return
/// the 4 values in the specified arguments.  If we can't run cpuid on the host,
/// return true.
static bool getX86CpuIDAndInfoEx(unsigned value, unsigned subleaf,
                                 unsigned *rEAX, unsigned *rEBX, unsigned *rECX,
                                 unsigned *rEDX) {
#if defined(__GNUC__) || defined(__clang__)
#if defined(__x86_64__)
  // gcc doesn't know cpuid would clobber ebx/rbx. Preserve it manually.
  // FIXME: should we save this for Clang?
  __asm__("movq\t%%rbx, %%rsi\n\t"
          "cpuid\n\t"
          "xchgq\t%%rbx, %%rsi\n\t"
          : "=a"(*rEAX), "=S"(*rEBX), "=c"(*rECX), "=d"(*rEDX)
          : "a"(value), "c"(subleaf));
  return false;
#elif defined(__i386__)
  __asm__("movl\t%%ebx, %%esi\n\t"
          "cpuid\n\t"
          "xchgl\t%%ebx, %%esi\n\t"
          : "=a"(*rEAX), "=S"(*rEBX), "=c"(*rECX), "=d"(*rEDX)
          : "a"(value), "c"(subleaf));
  return false;
#else
  return true;
#endif
#elif defined(_MSC_VER)
  int registers[4];
  __cpuidex(registers, value, subleaf);
  *rEAX = registers[0];
  *rEBX = registers[1];
  *rECX = registers[2];
  *rEDX = registers[3];
  return false;
#else
  return true;
#endif
}

// Read control register 0 (XCR0). Used to detect features such as AVX.
static bool getX86XCR0(unsigned *rEAX, unsigned *rEDX) {
#if defined(__GNUC__) || defined(__clang__)
  // Check xgetbv; this uses a .byte sequence instead of the instruction
  // directly because older assemblers do not include support for xgetbv and
  // there is no easy way to conditionally compile based on the assembler used.
  __asm__(".byte 0x0f, 0x01, 0xd0" : "=a"(*rEAX), "=d"(*rEDX) : "c"(0));
  return false;
#elif defined(_MSC_FULL_VER) && defined(_XCR_XFEATURE_ENABLED_MASK)
  unsigned long long Result = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
  *rEAX = Result;
  *rEDX = Result >> 32;
  return false;
#else
  return true;
#endif
}

static void detectX86FamilyModel(unsigned EAX, unsigned *Family,
                                 unsigned *Model) {
  *Family = (EAX >> 8) & 0xf; // Bits 8 - 11
  *Model = (EAX >> 4) & 0xf;  // Bits 4 - 7
  if (*Family == 6 || *Family == 0xf) {
    if (*Family == 0xf)
      // Examine extended family ID if family ID is F.
      *Family += (EAX >> 20) & 0xff; // Bits 20 - 27
    // Examine extended model ID if family ID is 6 or F.
    *Model += ((EAX >> 16) & 0xf) << 4; // Bits 16 - 19
  }
}

static void
getIntelProcessorTypeAndSubtype(unsigned Family, unsigned Model,
                                unsigned Brand_id, unsigned Features,
                                unsigned *Type, unsigned *Subtype) {
  if (Brand_id != 0)
    return;
  switch (Family) {
  case 6:
    switch (Model) {
    case 0x0f: // Intel Core 2 Duo processor, Intel Core 2 Duo mobile
               // processor, Intel Core 2 Quad processor, Intel Core 2 Quad
               // mobile processor, Intel Core 2 Extreme processor, Intel
               // Pentium Dual-Core processor, Intel Xeon processor, model
               // 0Fh. All processors are manufactured using the 65 nm process.
    case 0x16: // Intel Celeron processor model 16h. All processors are
               // manufactured using the 65 nm process
    case 0x17: // Intel Core 2 Extreme processor, Intel Xeon processor, model
               // 17h. All processors are manufactured using the 45 nm process.
               //
               // 45nm: Penryn , Wolfdale, Yorkfield (XE)
    case 0x1d: // Intel Xeon processor MP. All processors are manufactured using
               // the 45 nm process.
      *Type = INTEL_CORE2; // "penryn"
      break;
    case 0x1a: // Intel Core i7 processor and Intel Xeon processor. All
               // processors are manufactured using the 45 nm process.
    case 0x1e: // Intel(R) Core(TM) i7 CPU         870  @ 2.93GHz.
               // As found in a Summer 2010 model iMac.
    case 0x1f:
    case 0x2e:             // Nehalem EX
      *Type = INTEL_COREI7; // "nehalem"
      *Subtype = INTEL_COREI7_NEHALEM;
      break;
    case 0x25: // Intel Core i7, laptop version.
    case 0x2c: // Intel Core i7 processor and Intel Xeon processor. All
               // processors are manufactured using the 32 nm process.
    case 0x2f: // Westmere EX
      *Type = INTEL_COREI7; // "westmere"
      *Subtype = INTEL_COREI7_WESTMERE;
      break;
    case 0x2a: // Intel Core i7 processor. All processors are manufactured
               // using the 32 nm process.
    case 0x2d:
      *Type = INTEL_COREI7; //"sandybridge"
      *Subtype = INTEL_COREI7_SANDYBRIDGE;
      break;
    case 0x3a:
    case 0x3e:             // Ivy Bridge EP
      *Type = INTEL_COREI7; // "ivybridge"
      *Subtype = INTEL_COREI7_IVYBRIDGE;
      break;

    // Haswell:
    case 0x3c:
    case 0x3f:
    case 0x45:
    case 0x46:
      *Type = INTEL_COREI7; // "haswell"
      *Subtype = INTEL_COREI7_HASWELL;
      break;

    // Broadwell:
    case 0x3d:
    case 0x47:
    case 0x4f:
    case 0x56:
      *Type = INTEL_COREI7; // "broadwell"
      *Subtype = INTEL_COREI7_BROADWELL;
      break;

    // Skylake:
    case 0x4e: // Skylake mobile
    case 0x5e: // Skylake desktop
    case 0x8e: // Kaby Lake mobile
    case 0x9e: // Kaby Lake desktop
      *Type = INTEL_COREI7; // "skylake"
      *Subtype = INTEL_COREI7_SKYLAKE;
      break;

    // Skylake Xeon:
    case 0x55:
      *Type = INTEL_COREI7;
      *Subtype = INTEL_COREI7_SKYLAKE_AVX512; // "skylake-avx512"
      break;

    case 0x1c: // Most 45 nm Intel Atom processors
    case 0x26: // 45 nm Atom Lincroft
    case 0x27: // 32 nm Atom Medfield
    case 0x35: // 32 nm Atom Midview
    case 0x36: // 32 nm Atom Midview
      *Type = INTEL_BONNELL;
      break; // "bonnell"

    // Atom Silvermont codes from the Intel software optimization guide.
    case 0x37:
    case 0x4a:
    case 0x4d:
    case 0x5a:
    case 0x5d:
    case 0x4c: // really airmont
      *Type = INTEL_SILVERMONT;
      break; // "silvermont"

    case 0x57:
      *Type = INTEL_KNL; // knl
      break;

    default: // Unknown family 6 CPU.
      break;
    break;
    }
  default:
    break; // Unknown.
  }
}

static void getAMDProcessorTypeAndSubtype(unsigned Family, unsigned Model,
                                          unsigned Features, unsigned *Type,
                                          unsigned *Subtype) {
  // FIXME: this poorly matches the generated SubtargetFeatureKV table.  There
  // appears to be no way to generate the wide variety of AMD-specific targets
  // from the information returned from CPUID.
  switch (Family) {
  case 16:
    *Type = AMDFAM10H; // "amdfam10"
    switch (Model) {
    case 2:
      *Subtype = AMDFAM10H_BARCELONA;
      break;
    case 4:
      *Subtype = AMDFAM10H_SHANGHAI;
      break;
    case 8:
      *Subtype = AMDFAM10H_ISTANBUL;
      break;
    }
    break;
  case 20:
    *Type = AMD_BTVER1;
    break; // "btver1";
  case 21:
    *Type = AMDFAM15H;
    if (Model >= 0x60 && Model <= 0x7f) {
      *Subtype = AMDFAM15H_BDVER4;
      break; // "bdver4"; 60h-7Fh: Excavator
    }
    if (Model >= 0x30 && Model <= 0x3f) {
      *Subtype = AMDFAM15H_BDVER3;
      break; // "bdver3"; 30h-3Fh: Steamroller
    }
    if (Model >= 0x10 && Model <= 0x1f) {
      *Subtype = AMDFAM15H_BDVER2;
      break; // "bdver2"; 10h-1Fh: Piledriver
    }
    if (Model <= 0x0f) {
      *Subtype = AMDFAM15H_BDVER1;
      break; // "bdver1"; 00h-0Fh: Bulldozer
    }
    break;
  case 22:
    *Type = AMD_BTVER2;
    break; // "btver2"
  case 23:
    *Type = AMDFAM17H;
    *Subtype = AMDFAM17H_ZNVER1;
    break;
  default:
    break; // "generic"
  }
}

static void getAvailableFeatures(unsigned ECX, unsigned EDX, unsigned MaxLeaf,
                                 unsigned *FeaturesOut) {
  unsigned Features = 0;
  unsigned EAX, EBX;

  if ((EDX >> 15) & 1)
    Features |= 1 << FEATURE_CMOV;
  if ((EDX >> 23) & 1)
    Features |= 1 << FEATURE_MMX;
  if ((EDX >> 25) & 1)
    Features |= 1 << FEATURE_SSE;
  if ((EDX >> 26) & 1)
    Features |= 1 << FEATURE_SSE2;

  if ((ECX >> 0) & 1)
    Features |= 1 << FEATURE_SSE3;
  if ((ECX >> 1) & 1)
    Features |= 1 << FEATURE_PCLMUL;
  if ((ECX >> 9) & 1)
    Features |= 1 << FEATURE_SSSE3;
  if ((ECX >> 12) & 1)
    Features |= 1 << FEATURE_FMA;
  if ((ECX >> 19) & 1)
    Features |= 1 << FEATURE_SSE4_1;
  if ((ECX >> 20) & 1)
    Features |= 1 << FEATURE_SSE4_2;
  if ((ECX >> 23) & 1)
    Features |= 1 << FEATURE_POPCNT;
  if ((ECX >> 25) & 1)
    Features |= 1 << FEATURE_AES;

  // If CPUID indicates support for XSAVE, XRESTORE and AVX, and XGETBV
  // indicates that the AVX registers will be saved and restored on context
  // switch, then we have full AVX support.
  const unsigned AVXBits = (1 << 27) | (1 << 28);
  bool HasAVX = ((ECX & AVXBits) == AVXBits) && !getX86XCR0(&EAX, &EDX) &&
                ((EAX & 0x6) == 0x6);
  bool HasAVX512Save = HasAVX && ((EAX & 0xe0) == 0xe0);

  if (HasAVX)
    Features |= 1 << FEATURE_AVX;

  bool HasLeaf7 =
      MaxLeaf >= 0x7 && !getX86CpuIDAndInfoEx(0x7, 0x0, &EAX, &EBX, &ECX, &EDX);

  if (HasLeaf7 && ((EBX >> 3) & 1))
    Features |= 1 << FEATURE_BMI;
  if (HasLeaf7 && ((EBX >> 5) & 1) && HasAVX)
    Features |= 1 << FEATURE_AVX2;
  if (HasLeaf7 && ((EBX >> 9) & 1))
    Features |= 1 << FEATURE_BMI2;
  if (HasLeaf7 && ((EBX >> 16) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX512F;
  if (HasLeaf7 && ((EBX >> 17) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX512DQ;
  if (HasLeaf7 && ((EBX >> 21) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX512IFMA;
  if (HasLeaf7 && ((EBX >> 26) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX512PF;
  if (HasLeaf7 && ((EBX >> 27) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX512ER;
  if (HasLeaf7 && ((EBX >> 28) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX512CD;
  if (HasLeaf7 && ((EBX >> 30) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX512BW;
  if (HasLeaf7 && ((EBX >> 31) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX512VL;

  if (HasLeaf7 && ((ECX >> 1) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX512VBMI;
  if (HasLeaf7 && ((ECX >> 14) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX512VPOPCNTDQ;

  if (HasLeaf7 && ((EDX >> 2) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX5124VNNIW;
  if (HasLeaf7 && ((EDX >> 3) & 1) && HasAVX512Save)
    Features |= 1 << FEATURE_AVX5124FMAPS;

  unsigned MaxExtLevel;
  getX86CpuIDAndInfo(0x80000000, &MaxExtLevel, &EBX, &ECX, &EDX);

  bool HasExtLeaf1 = MaxExtLevel >= 0x80000001 &&
                     !getX86CpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
  if (HasExtLeaf1 && ((ECX >> 6) & 1))
    Features |= 1 << FEATURE_SSE4_A;
  if (HasExtLeaf1 && ((ECX >> 11) & 1))
    Features |= 1 << FEATURE_XOP;
  if (HasExtLeaf1 && ((ECX >> 16) & 1))
    Features |= 1 << FEATURE_FMA4;

  *FeaturesOut = Features;
}

#if defined(HAVE_INIT_PRIORITY)
#define CONSTRUCTOR_ATTRIBUTE __attribute__((__constructor__ 101))
#elif __has_attribute(__constructor__)
#define CONSTRUCTOR_ATTRIBUTE __attribute__((__constructor__))
#else
// FIXME: For MSVC, we should make a function pointer global in .CRT$X?? so that
// this runs during initialization.
#define CONSTRUCTOR_ATTRIBUTE
#endif

int __cpu_indicator_init(void) CONSTRUCTOR_ATTRIBUTE;

struct __processor_model {
  unsigned int __cpu_vendor;
  unsigned int __cpu_type;
  unsigned int __cpu_subtype;
  unsigned int __cpu_features[1];
} __cpu_model = {0, 0, 0, {0}};

/* A constructor function that is sets __cpu_model and __cpu_features with
   the right values.  This needs to run only once.  This constructor is
   given the highest priority and it should run before constructors without
   the priority set.  However, it still runs after ifunc initializers and
   needs to be called explicitly there.  */

int CONSTRUCTOR_ATTRIBUTE
__cpu_indicator_init(void) {
  unsigned EAX, EBX, ECX, EDX;
  unsigned MaxLeaf = 5;
  unsigned Vendor;
  unsigned Model, Family, Brand_id;
  unsigned Features = 0;

  /* This function needs to run just once.  */
  if (__cpu_model.__cpu_vendor)
    return 0;

  if (!isCpuIdSupported())
    return -1;

  /* Assume cpuid insn present. Run in level 0 to get vendor id. */
  if (getX86CpuIDAndInfo(0, &MaxLeaf, &Vendor, &ECX, &EDX) || MaxLeaf < 1) {
    __cpu_model.__cpu_vendor = VENDOR_OTHER;
    return -1;
  }
  getX86CpuIDAndInfo(1, &EAX, &EBX, &ECX, &EDX);
  detectX86FamilyModel(EAX, &Family, &Model);
  Brand_id = EBX & 0xff;

  /* Find available features. */
  getAvailableFeatures(ECX, EDX, MaxLeaf, &Features);
  __cpu_model.__cpu_features[0] = Features;

  if (Vendor == SIG_INTEL) {
    /* Get CPU type.  */
    getIntelProcessorTypeAndSubtype(Family, Model, Brand_id, Features,
                                    &(__cpu_model.__cpu_type),
                                    &(__cpu_model.__cpu_subtype));
    __cpu_model.__cpu_vendor = VENDOR_INTEL;
  } else if (Vendor == SIG_AMD) {
    /* Get CPU type.  */
    getAMDProcessorTypeAndSubtype(Family, Model, Features,
                                  &(__cpu_model.__cpu_type),
                                  &(__cpu_model.__cpu_subtype));
    __cpu_model.__cpu_vendor = VENDOR_AMD;
  } else
    __cpu_model.__cpu_vendor = VENDOR_OTHER;

  assert(__cpu_model.__cpu_vendor < VENDOR_MAX);
  assert(__cpu_model.__cpu_type < CPU_TYPE_MAX);
  assert(__cpu_model.__cpu_subtype < CPU_SUBTYPE_MAX);

  return 0;
}

#endif