[pcsx_rearmed.git] / deps / libchdr / deps / lzma-22.01 / src / CpuArch.c

/* CpuArch.c -- CPU specific code\r
2021-07-13 : Igor Pavlov : Public domain */\r
\r
#include "Precomp.h"\r
\r
#include "CpuArch.h"\r
\r
#ifdef MY_CPU_X86_OR_AMD64\r
\r
#if (defined(_MSC_VER) && !defined(MY_CPU_AMD64)) || defined(__GNUC__)\r
#define USE_ASM\r
#endif\r
\r
#if !defined(USE_ASM) && _MSC_VER >= 1500\r
#include <intrin.h>\r
#endif\r
\r
#if defined(USE_ASM) && !defined(MY_CPU_AMD64)\r
static UInt32 CheckFlag(UInt32 flag)\r
{\r
  #ifdef _MSC_VER\r
  __asm pushfd;\r
  __asm pop EAX;\r
  __asm mov EDX, EAX;\r
  __asm xor EAX, flag;\r
  __asm push EAX;\r
  __asm popfd;\r
  __asm pushfd;\r
  __asm pop EAX;\r
  __asm xor EAX, EDX;\r
  __asm push EDX;\r
  __asm popfd;\r
  __asm and flag, EAX;\r
  #else\r
  __asm__ __volatile__ (\r
    "pushf\n\t"\r
    "pop  %%EAX\n\t"\r
    "movl %%EAX,%%EDX\n\t"\r
    "xorl %0,%%EAX\n\t"\r
    "push %%EAX\n\t"\r
    "popf\n\t"\r
    "pushf\n\t"\r
    "pop  %%EAX\n\t"\r
    "xorl %%EDX,%%EAX\n\t"\r
    "push %%EDX\n\t"\r
    "popf\n\t"\r
    "andl %%EAX, %0\n\t":\r
    "=c" (flag) : "c" (flag) :\r
    "%eax", "%edx");\r
  #endif\r
  return flag;\r
}\r
#define CHECK_CPUID_IS_SUPPORTED if (CheckFlag(1 << 18) == 0 || CheckFlag(1 << 21) == 0) return False;\r
#else\r
#define CHECK_CPUID_IS_SUPPORTED\r
#endif\r
\r
#ifndef USE_ASM\r
  #ifdef _MSC_VER\r
    #if _MSC_VER >= 1600\r
      #define MY__cpuidex  __cpuidex\r
    #else\r
\r
/*\r
 __cpuid (function == 4) requires subfunction number in ECX.\r
  MSDN: The __cpuid intrinsic clears the ECX register before calling the cpuid instruction.\r
   __cpuid() in new MSVC clears ECX.\r
   __cpuid() in old MSVC (14.00) doesn't clear ECX\r
 We still can use __cpuid for low (function) values that don't require ECX,\r
 but __cpuid() in old MSVC will be incorrect for some function values: (function == 4).\r
 So here we use the hack for old MSVC to send (subFunction) in ECX register to cpuid instruction,\r
 where ECX value is first parameter for FAST_CALL / NO_INLINE function,\r
 So the caller of MY__cpuidex_HACK() sets ECX as subFunction, and\r
 old MSVC for __cpuid() doesn't change ECX and cpuid instruction gets (subFunction) value.\r
 \r
 DON'T remove MY_NO_INLINE and MY_FAST_CALL for MY__cpuidex_HACK() !!!\r
*/\r
\r
static\r
MY_NO_INLINE\r
void MY_FAST_CALL MY__cpuidex_HACK(UInt32 subFunction, int *CPUInfo, UInt32 function)\r
{\r
  UNUSED_VAR(subFunction);\r
  __cpuid(CPUInfo, function);\r
}\r
\r
      #define MY__cpuidex(info, func, func2)  MY__cpuidex_HACK(func2, info, func)\r
      #pragma message("======== MY__cpuidex_HACK WAS USED ========")\r
    #endif\r
  #else\r
     #define MY__cpuidex(info, func, func2)  __cpuid(info, func)\r
     #pragma message("======== (INCORRECT ?) cpuid WAS USED ========")\r
  #endif\r
#endif\r
\r
\r
\r
\r
void MyCPUID(UInt32 function, UInt32 *a, UInt32 *b, UInt32 *c, UInt32 *d)\r
{\r
  #ifdef USE_ASM\r
\r
  #ifdef _MSC_VER\r
\r
  UInt32 a2, b2, c2, d2;\r
  __asm xor EBX, EBX;\r
  __asm xor ECX, ECX;\r
  __asm xor EDX, EDX;\r
  __asm mov EAX, function;\r
  __asm cpuid;\r
  __asm mov a2, EAX;\r
  __asm mov b2, EBX;\r
  __asm mov c2, ECX;\r
  __asm mov d2, EDX;\r
\r
  *a = a2;\r
  *b = b2;\r
  *c = c2;\r
  *d = d2;\r
\r
  #else\r
\r
  __asm__ __volatile__ (\r
  #if defined(MY_CPU_AMD64) && defined(__PIC__)\r
    "mov %%rbx, %%rdi;"\r
    "cpuid;"\r
    "xchg %%rbx, %%rdi;"\r
    : "=a" (*a) ,\r
      "=D" (*b) ,\r
  #elif defined(MY_CPU_X86) && defined(__PIC__)\r
    "mov %%ebx, %%edi;"\r
    "cpuid;"\r
    "xchgl %%ebx, %%edi;"\r
    : "=a" (*a) ,\r
      "=D" (*b) ,\r
  #else\r
    "cpuid"\r
    : "=a" (*a) ,\r
      "=b" (*b) ,\r
  #endif\r
      "=c" (*c) ,\r
      "=d" (*d)\r
    : "0" (function), "c"(0) ) ;\r
\r
  #endif\r
  \r
  #else\r
\r
  int CPUInfo[4];\r
\r
  MY__cpuidex(CPUInfo, (int)function, 0);\r
\r
  *a = (UInt32)CPUInfo[0];\r
  *b = (UInt32)CPUInfo[1];\r
  *c = (UInt32)CPUInfo[2];\r
  *d = (UInt32)CPUInfo[3];\r
\r
  #endif\r
}\r
\r
BoolInt x86cpuid_CheckAndRead(Cx86cpuid *p)\r
{\r
  CHECK_CPUID_IS_SUPPORTED\r
  MyCPUID(0, &p->maxFunc, &p->vendor[0], &p->vendor[2], &p->vendor[1]);\r
  MyCPUID(1, &p->ver, &p->b, &p->c, &p->d);\r
  return True;\r
}\r
\r
static const UInt32 kVendors[][3] =\r
{\r
  { 0x756E6547, 0x49656E69, 0x6C65746E},\r
  { 0x68747541, 0x69746E65, 0x444D4163},\r
  { 0x746E6543, 0x48727561, 0x736C7561}\r
};\r
\r
int x86cpuid_GetFirm(const Cx86cpuid *p)\r
{\r
  unsigned i;\r
  for (i = 0; i < sizeof(kVendors) / sizeof(kVendors[i]); i++)\r
  {\r
    const UInt32 *v = kVendors[i];\r
    if (v[0] == p->vendor[0] &&\r
        v[1] == p->vendor[1] &&\r
        v[2] == p->vendor[2])\r
      return (int)i;\r
  }\r
  return -1;\r
}\r
\r
BoolInt CPU_Is_InOrder()\r
{\r
  Cx86cpuid p;\r
  int firm;\r
  UInt32 family, model;\r
  if (!x86cpuid_CheckAndRead(&p))\r
    return True;\r
\r
  family = x86cpuid_GetFamily(p.ver);\r
  model = x86cpuid_GetModel(p.ver);\r
  \r
  firm = x86cpuid_GetFirm(&p);\r
\r
  switch (firm)\r
  {\r
    case CPU_FIRM_INTEL: return (family < 6 || (family == 6 && (\r
        /* In-Order Atom CPU */\r
           model == 0x1C  /* 45 nm, N4xx, D4xx, N5xx, D5xx, 230, 330 */\r
        || model == 0x26  /* 45 nm, Z6xx */\r
        || model == 0x27  /* 32 nm, Z2460 */\r
        || model == 0x35  /* 32 nm, Z2760 */\r
        || model == 0x36  /* 32 nm, N2xxx, D2xxx */\r
        )));\r
    case CPU_FIRM_AMD: return (family < 5 || (family == 5 && (model < 6 || model == 0xA)));\r
    case CPU_FIRM_VIA: return (family < 6 || (family == 6 && model < 0xF));\r
  }\r
  return True;\r
}\r
\r
#if !defined(MY_CPU_AMD64) && defined(_WIN32)\r
#include <windows.h>\r
static BoolInt CPU_Sys_Is_SSE_Supported()\r
{\r
  OSVERSIONINFO vi;\r
  vi.dwOSVersionInfoSize = sizeof(vi);\r
  if (!GetVersionEx(&vi))\r
    return False;\r
  return (vi.dwMajorVersion >= 5);\r
}\r
#define CHECK_SYS_SSE_SUPPORT if (!CPU_Sys_Is_SSE_Supported()) return False;\r
#else\r
#define CHECK_SYS_SSE_SUPPORT\r
#endif\r
\r
\r
static UInt32 X86_CPUID_ECX_Get_Flags()\r
{\r
  Cx86cpuid p;\r
  CHECK_SYS_SSE_SUPPORT\r
  if (!x86cpuid_CheckAndRead(&p))\r
    return 0;\r
  return p.c;\r
}\r
\r
BoolInt CPU_IsSupported_AES()\r
{\r
  return (X86_CPUID_ECX_Get_Flags() >> 25) & 1;\r
}\r
\r
BoolInt CPU_IsSupported_SSSE3()\r
{\r
  return (X86_CPUID_ECX_Get_Flags() >> 9) & 1;\r
}\r
\r
BoolInt CPU_IsSupported_SSE41()\r
{\r
  return (X86_CPUID_ECX_Get_Flags() >> 19) & 1;\r
}\r
\r
BoolInt CPU_IsSupported_SHA()\r
{\r
  Cx86cpuid p;\r
  CHECK_SYS_SSE_SUPPORT\r
  if (!x86cpuid_CheckAndRead(&p))\r
    return False;\r
\r
  if (p.maxFunc < 7)\r
    return False;\r
  {\r
    UInt32 d[4] = { 0 };\r
    MyCPUID(7, &d[0], &d[1], &d[2], &d[3]);\r
    return (d[1] >> 29) & 1;\r
  }\r
}\r
\r
// #include <stdio.h>\r
\r
#ifdef _WIN32\r
#include <windows.h>\r
#endif\r
\r
BoolInt CPU_IsSupported_AVX2()\r
{\r
  Cx86cpuid p;\r
  CHECK_SYS_SSE_SUPPORT\r
\r
  #ifdef _WIN32\r
  #define MY__PF_XSAVE_ENABLED  17\r
  if (!IsProcessorFeaturePresent(MY__PF_XSAVE_ENABLED))\r
    return False;\r
  #endif\r
\r
  if (!x86cpuid_CheckAndRead(&p))\r
    return False;\r
  if (p.maxFunc < 7)\r
    return False;\r
  {\r
    UInt32 d[4] = { 0 };\r
    MyCPUID(7, &d[0], &d[1], &d[2], &d[3]);\r
    // printf("\ncpuid(7): ebx=%8x ecx=%8x\n", d[1], d[2]);\r
    return 1\r
      & (d[1] >> 5); // avx2\r
  }\r
}\r
\r
BoolInt CPU_IsSupported_VAES_AVX2()\r
{\r
  Cx86cpuid p;\r
  CHECK_SYS_SSE_SUPPORT\r
\r
  #ifdef _WIN32\r
  #define MY__PF_XSAVE_ENABLED  17\r
  if (!IsProcessorFeaturePresent(MY__PF_XSAVE_ENABLED))\r
    return False;\r
  #endif\r
\r
  if (!x86cpuid_CheckAndRead(&p))\r
    return False;\r
  if (p.maxFunc < 7)\r
    return False;\r
  {\r
    UInt32 d[4] = { 0 };\r
    MyCPUID(7, &d[0], &d[1], &d[2], &d[3]);\r
    // printf("\ncpuid(7): ebx=%8x ecx=%8x\n", d[1], d[2]);\r
    return 1\r
      & (d[1] >> 5) // avx2\r
      // & (d[1] >> 31) // avx512vl\r
      & (d[2] >> 9); // vaes // VEX-256/EVEX\r
  }\r
}\r
\r
BoolInt CPU_IsSupported_PageGB()\r
{\r
  Cx86cpuid cpuid;\r
  if (!x86cpuid_CheckAndRead(&cpuid))\r
    return False;\r
  {\r
    UInt32 d[4] = { 0 };\r
    MyCPUID(0x80000000, &d[0], &d[1], &d[2], &d[3]);\r
    if (d[0] < 0x80000001)\r
      return False;\r
  }\r
  {\r
    UInt32 d[4] = { 0 };\r
    MyCPUID(0x80000001, &d[0], &d[1], &d[2], &d[3]);\r
    return (d[3] >> 26) & 1;\r
  }\r
}\r
\r
\r
#elif defined(MY_CPU_ARM_OR_ARM64)\r
\r
#ifdef _WIN32\r
\r
#include <windows.h>\r
\r
BoolInt CPU_IsSupported_CRC32()  { return IsProcessorFeaturePresent(PF_ARM_V8_CRC32_INSTRUCTIONS_AVAILABLE) ? 1 : 0; }\r
BoolInt CPU_IsSupported_CRYPTO() { return IsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE) ? 1 : 0; }\r
BoolInt CPU_IsSupported_NEON()   { return IsProcessorFeaturePresent(PF_ARM_NEON_INSTRUCTIONS_AVAILABLE) ? 1 : 0; }\r
\r
#else\r
\r
#if defined(__APPLE__)\r
\r
/*\r
#include <stdio.h>\r
#include <string.h>\r
static void Print_sysctlbyname(const char *name)\r
{\r
  size_t bufSize = 256;\r
  char buf[256];\r
  int res = sysctlbyname(name, &buf, &bufSize, NULL, 0);\r
  {\r
    int i;\r
    printf("\nres = %d : %s : '%s' : bufSize = %d, numeric", res, name, buf, (unsigned)bufSize);\r
    for (i = 0; i < 20; i++)\r
      printf(" %2x", (unsigned)(Byte)buf[i]);\r
\r
  }\r
}\r
*/\r
\r
static BoolInt My_sysctlbyname_Get_BoolInt(const char *name)\r
{\r
  UInt32 val = 0;\r
  if (My_sysctlbyname_Get_UInt32(name, &val) == 0 && val == 1)\r
    return 1;\r
  return 0;\r
}\r
\r
  /*\r
  Print_sysctlbyname("hw.pagesize");\r
  Print_sysctlbyname("machdep.cpu.brand_string");\r
  */\r
\r
BoolInt CPU_IsSupported_CRC32(void)\r
{\r
  return My_sysctlbyname_Get_BoolInt("hw.optional.armv8_crc32");\r
}\r
\r
BoolInt CPU_IsSupported_NEON(void)\r
{\r
  return My_sysctlbyname_Get_BoolInt("hw.optional.neon");\r
}\r
\r
#ifdef MY_CPU_ARM64\r
#define APPLE_CRYPTO_SUPPORT_VAL 1\r
#else\r
#define APPLE_CRYPTO_SUPPORT_VAL 0\r
#endif\r
\r
BoolInt CPU_IsSupported_SHA1(void) { return APPLE_CRYPTO_SUPPORT_VAL; }\r
BoolInt CPU_IsSupported_SHA2(void) { return APPLE_CRYPTO_SUPPORT_VAL; }\r
BoolInt CPU_IsSupported_AES (void) { return APPLE_CRYPTO_SUPPORT_VAL; }\r
\r
\r
#else // __APPLE__\r
\r
#if defined(__SWITCH__) || defined(__vita__)\r
\r
BoolInt CPU_IsSupported_CRC32(void) { return 0; }\r
BoolInt CPU_IsSupported_NEON(void) { return 1; }\r
BoolInt CPU_IsSupported_SHA1(void) { return 0; }\r
BoolInt CPU_IsSupported_SHA2(void) { return 0; }\r
BoolInt CPU_IsSupported_AES (void) { return 0; }\r
\r
#else\r
\r
#include <sys/auxv.h>\r
\r
#define USE_HWCAP\r
\r
#ifdef USE_HWCAP\r
\r
#include <asm/hwcap.h>\r
\r
  #define MY_HWCAP_CHECK_FUNC_2(name1, name2) \\r
  BoolInt CPU_IsSupported_ ## name1() { return (getauxval(AT_HWCAP)  & (HWCAP_  ## name2)) ? 1 : 0; }\r
\r
#ifdef MY_CPU_ARM64\r
  #define MY_HWCAP_CHECK_FUNC(name) \\r
  MY_HWCAP_CHECK_FUNC_2(name, name)\r
  MY_HWCAP_CHECK_FUNC_2(NEON, ASIMD)\r
// MY_HWCAP_CHECK_FUNC (ASIMD)\r
#elif defined(MY_CPU_ARM)\r
  #define MY_HWCAP_CHECK_FUNC(name) \\r
  BoolInt CPU_IsSupported_ ## name() { return (getauxval(AT_HWCAP2) & (HWCAP2_ ## name)) ? 1 : 0; }\r
  MY_HWCAP_CHECK_FUNC_2(NEON, NEON)\r
#endif\r
\r
#else // USE_HWCAP\r
\r
  #define MY_HWCAP_CHECK_FUNC(name) \\r
  BoolInt CPU_IsSupported_ ## name() { return 0; }\r
  MY_HWCAP_CHECK_FUNC(NEON)\r
\r
#endif // USE_HWCAP\r
\r
MY_HWCAP_CHECK_FUNC (CRC32)\r
MY_HWCAP_CHECK_FUNC (SHA1)\r
MY_HWCAP_CHECK_FUNC (SHA2)\r
MY_HWCAP_CHECK_FUNC (AES)\r
\r
#endif\r
#endif // __APPLE__\r
#endif // _WIN32\r
\r
#endif // MY_CPU_ARM_OR_ARM64\r
\r
\r
\r
#ifdef __APPLE__\r
\r
#include <sys/sysctl.h>\r
\r
int My_sysctlbyname_Get(const char *name, void *buf, size_t *bufSize)\r
{\r
  return sysctlbyname(name, buf, bufSize, NULL, 0);\r
}\r
\r
int My_sysctlbyname_Get_UInt32(const char *name, UInt32 *val)\r
{\r
  size_t bufSize = sizeof(*val);\r
  int res = My_sysctlbyname_Get(name, val, &bufSize);\r
  if (res == 0 && bufSize != sizeof(*val))\r
    return EFAULT;\r
  return res;\r
}\r
\r
#endif\r
Commit	Line	Data
9e052883	1	/* CpuArch.c -- CPU specific code\r
	2	2021-07-13 : Igor Pavlov : Public domain */\r
	3	\r
	4	#include "Precomp.h"\r
	5	\r
	6	#include "CpuArch.h"\r
	7	\r
	8	#ifdef MY_CPU_X86_OR_AMD64\r
	9	\r
	10	#if (defined(_MSC_VER) && !defined(MY_CPU_AMD64)) \|\| defined(__GNUC__)\r
	11	#define USE_ASM\r
	12	#endif\r
	13	\r
	14	#if !defined(USE_ASM) && _MSC_VER >= 1500\r
	15	#include <intrin.h>\r
	16	#endif\r
	17	\r
	18	#if defined(USE_ASM) && !defined(MY_CPU_AMD64)\r
	19	static UInt32 CheckFlag(UInt32 flag)\r
	20	{\r
	21	#ifdef _MSC_VER\r
	22	__asm pushfd;\r
	23	__asm pop EAX;\r
	24	__asm mov EDX, EAX;\r
	25	__asm xor EAX, flag;\r
	26	__asm push EAX;\r
	27	__asm popfd;\r
	28	__asm pushfd;\r
	29	__asm pop EAX;\r
	30	__asm xor EAX, EDX;\r
	31	__asm push EDX;\r
	32	__asm popfd;\r
	33	__asm and flag, EAX;\r
	34	#else\r
	35	__asm__ __volatile__ (\r
	36	"pushf\n\t"\r
	37	"pop %%EAX\n\t"\r
	38	"movl %%EAX,%%EDX\n\t"\r
	39	"xorl %0,%%EAX\n\t"\r
	40	"push %%EAX\n\t"\r
	41	"popf\n\t"\r
	42	"pushf\n\t"\r
	43	"pop %%EAX\n\t"\r
	44	"xorl %%EDX,%%EAX\n\t"\r
	45	"push %%EDX\n\t"\r
	46	"popf\n\t"\r
	47	"andl %%EAX, %0\n\t":\r
	48	"=c" (flag) : "c" (flag) :\r
	49	"%eax", "%edx");\r
	50	#endif\r
	51	return flag;\r
	52	}\r
	53	#define CHECK_CPUID_IS_SUPPORTED if (CheckFlag(1 << 18) == 0 \|\| CheckFlag(1 << 21) == 0) return False;\r
	54	#else\r
	55	#define CHECK_CPUID_IS_SUPPORTED\r
	56	#endif\r
	57	\r
	58	#ifndef USE_ASM\r
	59	#ifdef _MSC_VER\r
	60	#if _MSC_VER >= 1600\r
	61	#define MY__cpuidex __cpuidex\r
	62	#else\r
	63	\r
	64	/*\r
65	__cpuid (function == 4) requires subfunction number in ECX.\r
66	MSDN: The __cpuid intrinsic clears the ECX register before calling the cpuid instruction.\r
67	__cpuid() in new MSVC clears ECX.\r
68	__cpuid() in old MSVC (14.00) doesn't clear ECX\r
69	We still can use __cpuid for low (function) values that don't require ECX,\r
70	but __cpuid() in old MSVC will be incorrect for some function values: (function == 4).\r
71	So here we use the hack for old MSVC to send (subFunction) in ECX register to cpuid instruction,\r
72	where ECX value is first parameter for FAST_CALL / NO_INLINE function,\r
73	So the caller of MY__cpuidex_HACK() sets ECX as subFunction, and\r
74	old MSVC for __cpuid() doesn't change ECX and cpuid instruction gets (subFunction) value.\r
75	\r
76	DON'T remove MY_NO_INLINE and MY_FAST_CALL for MY__cpuidex_HACK() !!!\r
77	*/\r
78	\r
79	static\r
80	MY_NO_INLINE\r
81	void MY_FAST_CALL MY__cpuidex_HACK(UInt32 subFunction, int *CPUInfo, UInt32 function)\r
82	{\r
83	UNUSED_VAR(subFunction);\r
84	__cpuid(CPUInfo, function);\r
85	}\r
86	\r
87	#define MY__cpuidex(info, func, func2) MY__cpuidex_HACK(func2, info, func)\r
88	#pragma message("======== MY__cpuidex_HACK WAS USED ========")\r
89	#endif\r
90	#else\r
91	#define MY__cpuidex(info, func, func2) __cpuid(info, func)\r
92	#pragma message("======== (INCORRECT ?) cpuid WAS USED ========")\r
93	#endif\r
94	#endif\r
95	\r
96	\r
97	\r
98	\r
99	void MyCPUID(UInt32 function, UInt32 a, UInt32 b, UInt32 c, UInt32 d)\r
100	{\r
101	#ifdef USE_ASM\r
102	\r
103	#ifdef _MSC_VER\r
104	\r
105	UInt32 a2, b2, c2, d2;\r
106	__asm xor EBX, EBX;\r
107	__asm xor ECX, ECX;\r
108	__asm xor EDX, EDX;\r
109	__asm mov EAX, function;\r
110	__asm cpuid;\r
111	__asm mov a2, EAX;\r
112	__asm mov b2, EBX;\r
113	__asm mov c2, ECX;\r
114	__asm mov d2, EDX;\r
115	\r
116	*a = a2;\r
117	*b = b2;\r
118	*c = c2;\r
119	*d = d2;\r
120	\r
121	#else\r
122	\r
123	__asm__ __volatile__ (\r
124	#if defined(MY_CPU_AMD64) && defined(__PIC__)\r
125	"mov %%rbx, %%rdi;"\r
126	"cpuid;"\r
127	"xchg %%rbx, %%rdi;"\r
128	: "=a" (*a) ,\r
129	"=D" (*b) ,\r
130	#elif defined(MY_CPU_X86) && defined(__PIC__)\r
131	"mov %%ebx, %%edi;"\r
132	"cpuid;"\r
133	"xchgl %%ebx, %%edi;"\r
134	: "=a" (*a) ,\r
135	"=D" (*b) ,\r
136	#else\r
137	"cpuid"\r
138	: "=a" (*a) ,\r
139	"=b" (*b) ,\r
140	#endif\r
141	"=c" (*c) ,\r
142	"=d" (*d)\r
143	: "0" (function), "c"(0) ) ;\r
144	\r
145	#endif\r
146	\r
147	#else\r
148	\r
149	int CPUInfo[4];\r
150	\r
151	MY__cpuidex(CPUInfo, (int)function, 0);\r
152	\r
153	*a = (UInt32)CPUInfo[0];\r
154	*b = (UInt32)CPUInfo[1];\r
155	*c = (UInt32)CPUInfo[2];\r
156	*d = (UInt32)CPUInfo[3];\r
157	\r
158	#endif\r
159	}\r
160	\r
161	BoolInt x86cpuid_CheckAndRead(Cx86cpuid *p)\r
162	{\r
163	CHECK_CPUID_IS_SUPPORTED\r
164	MyCPUID(0, &p->maxFunc, &p->vendor[0], &p->vendor[2], &p->vendor[1]);\r
165	MyCPUID(1, &p->ver, &p->b, &p->c, &p->d);\r
166	return True;\r
167	}\r
168	\r
169	static const UInt32 kVendors[][3] =\r
170	{\r
171	{ 0x756E6547, 0x49656E69, 0x6C65746E},\r
172	{ 0x68747541, 0x69746E65, 0x444D4163},\r
173	{ 0x746E6543, 0x48727561, 0x736C7561}\r
174	};\r
175	\r
176	int x86cpuid_GetFirm(const Cx86cpuid *p)\r
177	{\r
178	unsigned i;\r
179	for (i = 0; i < sizeof(kVendors) / sizeof(kVendors[i]); i++)\r
180	{\r
181	const UInt32 *v = kVendors[i];\r
182	if (v[0] == p->vendor[0] &&\r
183	v[1] == p->vendor[1] &&\r
184	v[2] == p->vendor[2])\r
185	return (int)i;\r
186	}\r
187	return -1;\r
188	}\r
189	\r
190	BoolInt CPU_Is_InOrder()\r
191	{\r
192	Cx86cpuid p;\r
193	int firm;\r
194	UInt32 family, model;\r
195	if (!x86cpuid_CheckAndRead(&p))\r
196	return True;\r
197	\r
198	family = x86cpuid_GetFamily(p.ver);\r
199	model = x86cpuid_GetModel(p.ver);\r
200	\r
201	firm = x86cpuid_GetFirm(&p);\r
202	\r
203	switch (firm)\r
204	{\r
205	case CPU_FIRM_INTEL: return (family < 6 \|\| (family == 6 && (\r
206	/* In-Order Atom CPU */\r
207	model == 0x1C /* 45 nm, N4xx, D4xx, N5xx, D5xx, 230, 330 */\r
208	\|\| model == 0x26 /* 45 nm, Z6xx */\r
209	\|\| model == 0x27 /* 32 nm, Z2460 */\r
210	\|\| model == 0x35 /* 32 nm, Z2760 */\r
211	\|\| model == 0x36 /* 32 nm, N2xxx, D2xxx */\r
212	)));\r
213	case CPU_FIRM_AMD: return (family < 5 \|\| (family == 5 && (model < 6 \|\| model == 0xA)));\r
214	case CPU_FIRM_VIA: return (family < 6 \|\| (family == 6 && model < 0xF));\r
215	}\r
216	return True;\r
217	}\r
218	\r
219	#if !defined(MY_CPU_AMD64) && defined(_WIN32)\r
648db22b	220	#include <windows.h>\r
9e052883	221	static BoolInt CPU_Sys_Is_SSE_Supported()\r
	222	{\r
	223	OSVERSIONINFO vi;\r
	224	vi.dwOSVersionInfoSize = sizeof(vi);\r
	225	if (!GetVersionEx(&vi))\r
	226	return False;\r
	227	return (vi.dwMajorVersion >= 5);\r
	228	}\r
	229	#define CHECK_SYS_SSE_SUPPORT if (!CPU_Sys_Is_SSE_Supported()) return False;\r
	230	#else\r
	231	#define CHECK_SYS_SSE_SUPPORT\r
	232	#endif\r
	233	\r
	234	\r
	235	static UInt32 X86_CPUID_ECX_Get_Flags()\r
	236	{\r
	237	Cx86cpuid p;\r
	238	CHECK_SYS_SSE_SUPPORT\r
	239	if (!x86cpuid_CheckAndRead(&p))\r
	240	return 0;\r
	241	return p.c;\r
	242	}\r
	243	\r
	244	BoolInt CPU_IsSupported_AES()\r
	245	{\r
	246	return (X86_CPUID_ECX_Get_Flags() >> 25) & 1;\r
	247	}\r
	248	\r
	249	BoolInt CPU_IsSupported_SSSE3()\r
	250	{\r
	251	return (X86_CPUID_ECX_Get_Flags() >> 9) & 1;\r
	252	}\r
	253	\r
	254	BoolInt CPU_IsSupported_SSE41()\r
	255	{\r
	256	return (X86_CPUID_ECX_Get_Flags() >> 19) & 1;\r
	257	}\r
	258	\r
	259	BoolInt CPU_IsSupported_SHA()\r
	260	{\r
	261	Cx86cpuid p;\r
	262	CHECK_SYS_SSE_SUPPORT\r
	263	if (!x86cpuid_CheckAndRead(&p))\r
	264	return False;\r
	265	\r
	266	if (p.maxFunc < 7)\r
	267	return False;\r
	268	{\r
	269	UInt32 d[4] = { 0 };\r
	270	MyCPUID(7, &d[0], &d[1], &d[2], &d[3]);\r
	271	return (d[1] >> 29) & 1;\r
	272	}\r
	273	}\r
	274	\r
	275	// #include <stdio.h>\r
	276	\r
	277	#ifdef _WIN32\r
648db22b	278	#include <windows.h>\r
9e052883	279	#endif\r
	280	\r
	281	BoolInt CPU_IsSupported_AVX2()\r
	282	{\r
	283	Cx86cpuid p;\r
	284	CHECK_SYS_SSE_SUPPORT\r
	285	\r
	286	#ifdef _WIN32\r
	287	#define MY__PF_XSAVE_ENABLED 17\r
	288	if (!IsProcessorFeaturePresent(MY__PF_XSAVE_ENABLED))\r
	289	return False;\r
	290	#endif\r
	291	\r
	292	if (!x86cpuid_CheckAndRead(&p))\r
	293	return False;\r
	294	if (p.maxFunc < 7)\r
	295	return False;\r
	296	{\r
	297	UInt32 d[4] = { 0 };\r
	298	MyCPUID(7, &d[0], &d[1], &d[2], &d[3]);\r
	299	// printf("\ncpuid(7): ebx=%8x ecx=%8x\n", d[1], d[2]);\r
	300	return 1\r
	301	& (d[1] >> 5); // avx2\r
	302	}\r
	303	}\r
	304	\r
	305	BoolInt CPU_IsSupported_VAES_AVX2()\r
	306	{\r
	307	Cx86cpuid p;\r
	308	CHECK_SYS_SSE_SUPPORT\r
	309	\r
	310	#ifdef _WIN32\r
	311	#define MY__PF_XSAVE_ENABLED 17\r
	312	if (!IsProcessorFeaturePresent(MY__PF_XSAVE_ENABLED))\r
	313	return False;\r
	314	#endif\r
	315	\r
	316	if (!x86cpuid_CheckAndRead(&p))\r
	317	return False;\r
	318	if (p.maxFunc < 7)\r
	319	return False;\r
	320	{\r
	321	UInt32 d[4] = { 0 };\r
	322	MyCPUID(7, &d[0], &d[1], &d[2], &d[3]);\r
	323	// printf("\ncpuid(7): ebx=%8x ecx=%8x\n", d[1], d[2]);\r
	324	return 1\r
	325	& (d[1] >> 5) // avx2\r
	326	// & (d[1] >> 31) // avx512vl\r
	327	& (d[2] >> 9); // vaes // VEX-256/EVEX\r
	328	}\r
	329	}\r
	330	\r
	331	BoolInt CPU_IsSupported_PageGB()\r
	332	{\r
	333	Cx86cpuid cpuid;\r
	334	if (!x86cpuid_CheckAndRead(&cpuid))\r
	335	return False;\r
	336	{\r
	337	UInt32 d[4] = { 0 };\r
	338	MyCPUID(0x80000000, &d[0], &d[1], &d[2], &d[3]);\r
	339	if (d[0] < 0x80000001)\r
	340	return False;\r
	341	}\r
	342	{\r
343	UInt32 d[4] = { 0 };\r
344	MyCPUID(0x80000001, &d[0], &d[1], &d[2], &d[3]);\r
345	return (d[3] >> 26) & 1;\r
346	}\r
347	}\r
348	\r
349	\r
350	#elif defined(MY_CPU_ARM_OR_ARM64)\r
351	\r
352	#ifdef _WIN32\r
353	\r
648db22b	354	#include <windows.h>\r
9e052883	355	\r
	356	BoolInt CPU_IsSupported_CRC32() { return IsProcessorFeaturePresent(PF_ARM_V8_CRC32_INSTRUCTIONS_AVAILABLE) ? 1 : 0; }\r
	357	BoolInt CPU_IsSupported_CRYPTO() { return IsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE) ? 1 : 0; }\r
	358	BoolInt CPU_IsSupported_NEON() { return IsProcessorFeaturePresent(PF_ARM_NEON_INSTRUCTIONS_AVAILABLE) ? 1 : 0; }\r
	359	\r
	360	#else\r
	361	\r
	362	#if defined(__APPLE__)\r
	363	\r
	364	/*\r
	365	#include <stdio.h>\r
	366	#include <string.h>\r
	367	static void Print_sysctlbyname(const char *name)\r
	368	{\r
	369	size_t bufSize = 256;\r
	370	char buf[256];\r
	371	int res = sysctlbyname(name, &buf, &bufSize, NULL, 0);\r
	372	{\r
	373	int i;\r
	374	printf("\nres = %d : %s : '%s' : bufSize = %d, numeric", res, name, buf, (unsigned)bufSize);\r
	375	for (i = 0; i < 20; i++)\r
	376	printf(" %2x", (unsigned)(Byte)buf[i]);\r
	377	\r
	378	}\r
	379	}\r
	380	*/\r
	381	\r
	382	static BoolInt My_sysctlbyname_Get_BoolInt(const char *name)\r
	383	{\r
	384	UInt32 val = 0;\r
	385	if (My_sysctlbyname_Get_UInt32(name, &val) == 0 && val == 1)\r
	386	return 1;\r
	387	return 0;\r
	388	}\r
	389	\r
	390	/*\r
	391	Print_sysctlbyname("hw.pagesize");\r
	392	Print_sysctlbyname("machdep.cpu.brand_string");\r
	393	*/\r
	394	\r
	395	BoolInt CPU_IsSupported_CRC32(void)\r
	396	{\r
	397	return My_sysctlbyname_Get_BoolInt("hw.optional.armv8_crc32");\r
	398	}\r
	399	\r
	400	BoolInt CPU_IsSupported_NEON(void)\r
	401	{\r
	402	return My_sysctlbyname_Get_BoolInt("hw.optional.neon");\r
	403	}\r
	404	\r
	405	#ifdef MY_CPU_ARM64\r
	406	#define APPLE_CRYPTO_SUPPORT_VAL 1\r
	407	#else\r
	408	#define APPLE_CRYPTO_SUPPORT_VAL 0\r
	409	#endif\r
	410	\r
	411	BoolInt CPU_IsSupported_SHA1(void) { return APPLE_CRYPTO_SUPPORT_VAL; }\r
	412	BoolInt CPU_IsSupported_SHA2(void) { return APPLE_CRYPTO_SUPPORT_VAL; }\r
	413	BoolInt CPU_IsSupported_AES (void) { return APPLE_CRYPTO_SUPPORT_VAL; }\r
	414	\r
	415	\r
	416	#else // __APPLE__\r
	417	\r
	418	#if defined(__SWITCH__) \|\| defined(__vita__)\r
419	\r
420	BoolInt CPU_IsSupported_CRC32(void) { return 0; }\r
421	BoolInt CPU_IsSupported_NEON(void) { return 1; }\r
422	BoolInt CPU_IsSupported_SHA1(void) { return 0; }\r
423	BoolInt CPU_IsSupported_SHA2(void) { return 0; }\r
424	BoolInt CPU_IsSupported_AES (void) { return 0; }\r
425	\r
426	#else\r
427	\r
428	#include <sys/auxv.h>\r
429	\r
430	#define USE_HWCAP\r
431	\r
432	#ifdef USE_HWCAP\r
433	\r
434	#include <asm/hwcap.h>\r
435	\r
436	#define MY_HWCAP_CHECK_FUNC_2(name1, name2) \\r
437	BoolInt CPU_IsSupported_ ## name1() { return (getauxval(AT_HWCAP) & (HWCAP_ ## name2)) ? 1 : 0; }\r
438	\r
439	#ifdef MY_CPU_ARM64\r
440	#define MY_HWCAP_CHECK_FUNC(name) \\r
441	MY_HWCAP_CHECK_FUNC_2(name, name)\r
442	MY_HWCAP_CHECK_FUNC_2(NEON, ASIMD)\r
443	// MY_HWCAP_CHECK_FUNC (ASIMD)\r
444	#elif defined(MY_CPU_ARM)\r
445	#define MY_HWCAP_CHECK_FUNC(name) \\r
446	BoolInt CPU_IsSupported_ ## name() { return (getauxval(AT_HWCAP2) & (HWCAP2_ ## name)) ? 1 : 0; }\r
447	MY_HWCAP_CHECK_FUNC_2(NEON, NEON)\r
448	#endif\r
449	\r
450	#else // USE_HWCAP\r
451	\r
452	#define MY_HWCAP_CHECK_FUNC(name) \\r
453	BoolInt CPU_IsSupported_ ## name() { return 0; }\r
454	MY_HWCAP_CHECK_FUNC(NEON)\r
455	\r
456	#endif // USE_HWCAP\r
457	\r
458	MY_HWCAP_CHECK_FUNC (CRC32)\r
459	MY_HWCAP_CHECK_FUNC (SHA1)\r
460	MY_HWCAP_CHECK_FUNC (SHA2)\r
461	MY_HWCAP_CHECK_FUNC (AES)\r
462	\r
463	#endif\r
464	#endif // __APPLE__\r
465	#endif // _WIN32\r
466	\r
467	#endif // MY_CPU_ARM_OR_ARM64\r
468	\r
469	\r
470	\r
471	#ifdef __APPLE__\r
472	\r
473	#include <sys/sysctl.h>\r
474	\r
475	int My_sysctlbyname_Get(const char name, void buf, size_t *bufSize)\r
476	{\r
477	return sysctlbyname(name, buf, bufSize, NULL, 0);\r
478	}\r
479	\r
480	int My_sysctlbyname_Get_UInt32(const char name, UInt32 val)\r
481	{\r
482	size_t bufSize = sizeof(*val);\r
483	int res = My_sysctlbyname_Get(name, val, &bufSize);\r
484	if (res == 0 && bufSize != sizeof(*val))\r
485	return EFAULT;\r
486	return res;\r
487	}\r
488	\r
489	#endif\r